renderer: simplify dynamic params

Co-authored-by: Senaxx <Senaxx@users.noreply.github.com>

3rdparty/renderdoc/renderdoc_app.h

@@ -0,0 +1,875 @@
+/******************************************************************************
+ * The MIT License (MIT)
+ *
+ * Copyright (c) 2015-2026 Baldur Karlsson
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a copy
+ * of this software and associated documentation files (the "Software"), to deal
+ * in the Software without restriction, including without limitation the rights
+ * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+ * copies of the Software, and to permit persons to whom the Software is
+ * furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+ * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+ * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+ * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
+ * THE SOFTWARE.
+ ******************************************************************************/
+
+#pragma once
+
+//////////////////////////////////////////////////////////////////////////////////////////////////
+//
+// Documentation for the API is available at https://renderdoc.org/docs/in_application_api.html
+//
+
+#if !defined(RENDERDOC_NO_STDINT)
+#include <stdint.h>
+#endif
+
+#if defined(WIN32) || defined(__WIN32__) || defined(_WIN32) || defined(_MSC_VER)
+#define RENDERDOC_CC __cdecl
+#elif defined(__linux__) || defined(__FreeBSD__) || defined(__sun__) || defined(__OpenBSD__)
+#define RENDERDOC_CC
+#elif defined(__APPLE__)
+#define RENDERDOC_CC
+#else
+#error "Unknown platform"
+#endif
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+//////////////////////////////////////////////////////////////////////////////////////////////////
+// Constants not used directly in below API
+
+// This is a GUID/magic value used for when applications pass a path where shader debug
+// information can be found to match up with a stripped shader.
+// the define can be used like so: const GUID RENDERDOC_ShaderDebugMagicValue =
+// RENDERDOC_ShaderDebugMagicValue_value
+#define RENDERDOC_ShaderDebugMagicValue_struct                                \
+  {                                                                           \
+    0xeab25520, 0x6670, 0x4865, 0x84, 0x29, 0x6c, 0x8, 0x51, 0x54, 0x00, 0xff \
+  }
+
+// as an alternative when you want a byte array (assuming x86 endianness):
+#define RENDERDOC_ShaderDebugMagicValue_bytearray                                                 \
+  {                                                                                               \
+    0x20, 0x55, 0xb2, 0xea, 0x70, 0x66, 0x65, 0x48, 0x84, 0x29, 0x6c, 0x8, 0x51, 0x54, 0x00, 0xff \
+  }
+
+// truncated version when only a uint64_t is available (e.g. Vulkan tags):
+#define RENDERDOC_ShaderDebugMagicValue_truncated 0x48656670eab25520ULL
+
+// this is a magic value for vulkan user tags to indicate which dispatchable API objects are which
+// for object annotations
+#define RENDERDOC_APIObjectAnnotationHelper 0xfbb3b337b664d0adULL
+
+//////////////////////////////////////////////////////////////////////////////////////////////////
+// RenderDoc capture options
+//
+
+typedef enum RENDERDOC_CaptureOption
+{
+  // Allow the application to enable vsync
+  //
+  // Default - enabled
+  //
+  // 1 - The application can enable or disable vsync at will
+  // 0 - vsync is force disabled
+  eRENDERDOC_Option_AllowVSync = 0,
+
+  // Allow the application to enable fullscreen
+  //
+  // Default - enabled
+  //
+  // 1 - The application can enable or disable fullscreen at will
+  // 0 - fullscreen is force disabled
+  eRENDERDOC_Option_AllowFullscreen = 1,
+
+  // Record API debugging events and messages
+  //
+  // Default - disabled
+  //
+  // 1 - Enable built-in API debugging features and records the results into
+  //     the capture, which is matched up with events on replay
+  // 0 - no API debugging is forcibly enabled
+  eRENDERDOC_Option_APIValidation = 2,
+  eRENDERDOC_Option_DebugDeviceMode = 2,    // deprecated name of this enum
+
+  // Capture CPU callstacks for API events
+  //
+  // Default - disabled
+  //
+  // 1 - Enables capturing of callstacks
+  // 0 - no callstacks are captured
+  eRENDERDOC_Option_CaptureCallstacks = 3,
+
+  // When capturing CPU callstacks, only capture them from actions.
+  // This option does nothing without the above option being enabled
+  //
+  // Default - disabled
+  //
+  // 1 - Only captures callstacks for actions.
+  //     Ignored if CaptureCallstacks is disabled
+  // 0 - Callstacks, if enabled, are captured for every event.
+  eRENDERDOC_Option_CaptureCallstacksOnlyDraws = 4,
+  eRENDERDOC_Option_CaptureCallstacksOnlyActions = 4,
+
+  // Specify a delay in seconds to wait for a debugger to attach, after
+  // creating or injecting into a process, before continuing to allow it to run.
+  //
+  // 0 indicates no delay, and the process will run immediately after injection
+  //
+  // Default - 0 seconds
+  //
+  eRENDERDOC_Option_DelayForDebugger = 5,
+
+  // Verify buffer access. This includes checking the memory returned by a Map() call to
+  // detect any out-of-bounds modification, as well as initialising buffers with undefined contents
+  // to a marker value to catch use of uninitialised memory.
+  //
+  // NOTE: This option is only valid for OpenGL and D3D11. Explicit APIs such as D3D12 and Vulkan do
+  // not do the same kind of interception & checking and undefined contents are really undefined.
+  //
+  // Default - disabled
+  //
+  // 1 - Verify buffer access
+  // 0 - No verification is performed, and overwriting bounds may cause crashes or corruption in
+  //     RenderDoc.
+  eRENDERDOC_Option_VerifyBufferAccess = 6,
+
+  // The old name for eRENDERDOC_Option_VerifyBufferAccess was eRENDERDOC_Option_VerifyMapWrites.
+  // This option now controls the filling of uninitialised buffers with 0xdddddddd which was
+  // previously always enabled
+  eRENDERDOC_Option_VerifyMapWrites = eRENDERDOC_Option_VerifyBufferAccess,
+
+  // Hooks any system API calls that create child processes, and injects
+  // RenderDoc into them recursively with the same options.
+  //
+  // Default - disabled
+  //
+  // 1 - Hooks into spawned child processes
+  // 0 - Child processes are not hooked by RenderDoc
+  eRENDERDOC_Option_HookIntoChildren = 7,
+
+  // By default RenderDoc only includes resources in the final capture necessary
+  // for that frame, this allows you to override that behaviour.
+  //
+  // Default - disabled
+  //
+  // 1 - all live resources at the time of capture are included in the capture
+  //     and available for inspection
+  // 0 - only the resources referenced by the captured frame are included
+  eRENDERDOC_Option_RefAllResources = 8,
+
+  // **NOTE**: As of RenderDoc v1.1 this option has been deprecated. Setting or
+  // getting it will be ignored, to allow compatibility with older versions.
+  // In v1.1 the option acts as if it's always enabled.
+  //
+  // By default RenderDoc skips saving initial states for resources where the
+  // previous contents don't appear to be used, assuming that writes before
+  // reads indicate previous contents aren't used.
+  //
+  // Default - disabled
+  //
+  // 1 - initial contents at the start of each captured frame are saved, even if
+  //     they are later overwritten or cleared before being used.
+  // 0 - unless a read is detected, initial contents will not be saved and will
+  //     appear as black or empty data.
+  eRENDERDOC_Option_SaveAllInitials = 9,
+
+  // In APIs that allow for the recording of command lists to be replayed later,
+  // RenderDoc may choose to not capture command lists before a frame capture is
+  // triggered, to reduce overheads. This means any command lists recorded once
+  // and replayed many times will not be available and may cause a failure to
+  // capture.
+  //
+  // NOTE: This is only true for APIs where multithreading is difficult or
+  // discouraged. Newer APIs like Vulkan and D3D12 will ignore this option
+  // and always capture all command lists since the API is heavily oriented
+  // around it and the overheads have been reduced by API design.
+  //
+  // 1 - All command lists are captured from the start of the application
+  // 0 - Command lists are only captured if their recording begins during
+  //     the period when a frame capture is in progress.
+  eRENDERDOC_Option_CaptureAllCmdLists = 10,
+
+  // Mute API debugging output when the API validation mode option is enabled
+  //
+  // Default - enabled
+  //
+  // 1 - Mute any API debug messages from being displayed or passed through
+  // 0 - API debugging is displayed as normal
+  eRENDERDOC_Option_DebugOutputMute = 11,
+
+  // Option to allow vendor extensions to be used even when they may be
+  // incompatible with RenderDoc and cause corrupted replays or crashes.
+  //
+  // Default - inactive
+  //
+  // No values are documented, this option should only be used when absolutely
+  // necessary as directed by a RenderDoc developer.
+  eRENDERDOC_Option_AllowUnsupportedVendorExtensions = 12,
+
+  // Define a soft memory limit which some APIs may aim to keep overhead under where
+  // possible. Anything above this limit will where possible be saved directly to disk during
+  // capture.
+  // This will cause increased disk space use (which may cause a capture to fail if disk space is
+  // exhausted) as well as slower capture times.
+  //
+  // Not all memory allocations may be deferred like this so it is not a guarantee of a memory
+  // limit.
+  //
+  // Units are in MBs, suggested values would range from 200MB to 1000MB.
+  //
+  // Default - 0 Megabytes
+  eRENDERDOC_Option_SoftMemoryLimit = 13,
+} RENDERDOC_CaptureOption;
+
+// Sets an option that controls how RenderDoc behaves on capture.
+//
+// Returns 1 if the option and value are valid
+// Returns 0 if either is invalid and the option is unchanged
+typedef int(RENDERDOC_CC *pRENDERDOC_SetCaptureOptionU32)(RENDERDOC_CaptureOption opt, uint32_t val);
+typedef int(RENDERDOC_CC *pRENDERDOC_SetCaptureOptionF32)(RENDERDOC_CaptureOption opt, float val);
+
+// Gets the current value of an option as a uint32_t
+//
+// If the option is invalid, 0xffffffff is returned
+typedef uint32_t(RENDERDOC_CC *pRENDERDOC_GetCaptureOptionU32)(RENDERDOC_CaptureOption opt);
+
+// Gets the current value of an option as a float
+//
+// If the option is invalid, -FLT_MAX is returned
+typedef float(RENDERDOC_CC *pRENDERDOC_GetCaptureOptionF32)(RENDERDOC_CaptureOption opt);
+
+typedef enum RENDERDOC_InputButton
+{
+  // '0' - '9' matches ASCII values
+  eRENDERDOC_Key_0 = 0x30,
+  eRENDERDOC_Key_1 = 0x31,
+  eRENDERDOC_Key_2 = 0x32,
+  eRENDERDOC_Key_3 = 0x33,
+  eRENDERDOC_Key_4 = 0x34,
+  eRENDERDOC_Key_5 = 0x35,
+  eRENDERDOC_Key_6 = 0x36,
+  eRENDERDOC_Key_7 = 0x37,
+  eRENDERDOC_Key_8 = 0x38,
+  eRENDERDOC_Key_9 = 0x39,
+
+  // 'A' - 'Z' matches ASCII values
+  eRENDERDOC_Key_A = 0x41,
+  eRENDERDOC_Key_B = 0x42,
+  eRENDERDOC_Key_C = 0x43,
+  eRENDERDOC_Key_D = 0x44,
+  eRENDERDOC_Key_E = 0x45,
+  eRENDERDOC_Key_F = 0x46,
+  eRENDERDOC_Key_G = 0x47,
+  eRENDERDOC_Key_H = 0x48,
+  eRENDERDOC_Key_I = 0x49,
+  eRENDERDOC_Key_J = 0x4A,
+  eRENDERDOC_Key_K = 0x4B,
+  eRENDERDOC_Key_L = 0x4C,
+  eRENDERDOC_Key_M = 0x4D,
+  eRENDERDOC_Key_N = 0x4E,
+  eRENDERDOC_Key_O = 0x4F,
+  eRENDERDOC_Key_P = 0x50,
+  eRENDERDOC_Key_Q = 0x51,
+  eRENDERDOC_Key_R = 0x52,
+  eRENDERDOC_Key_S = 0x53,
+  eRENDERDOC_Key_T = 0x54,
+  eRENDERDOC_Key_U = 0x55,
+  eRENDERDOC_Key_V = 0x56,
+  eRENDERDOC_Key_W = 0x57,
+  eRENDERDOC_Key_X = 0x58,
+  eRENDERDOC_Key_Y = 0x59,
+  eRENDERDOC_Key_Z = 0x5A,
+
+  // leave the rest of the ASCII range free
+  // in case we want to use it later
+  eRENDERDOC_Key_NonPrintable = 0x100,
+
+  eRENDERDOC_Key_Divide,
+  eRENDERDOC_Key_Multiply,
+  eRENDERDOC_Key_Subtract,
+  eRENDERDOC_Key_Plus,
+
+  eRENDERDOC_Key_F1,
+  eRENDERDOC_Key_F2,
+  eRENDERDOC_Key_F3,
+  eRENDERDOC_Key_F4,
+  eRENDERDOC_Key_F5,
+  eRENDERDOC_Key_F6,
+  eRENDERDOC_Key_F7,
+  eRENDERDOC_Key_F8,
+  eRENDERDOC_Key_F9,
+  eRENDERDOC_Key_F10,
+  eRENDERDOC_Key_F11,
+  eRENDERDOC_Key_F12,
+
+  eRENDERDOC_Key_Home,
+  eRENDERDOC_Key_End,
+  eRENDERDOC_Key_Insert,
+  eRENDERDOC_Key_Delete,
+  eRENDERDOC_Key_PageUp,
+  eRENDERDOC_Key_PageDn,
+
+  eRENDERDOC_Key_Backspace,
+  eRENDERDOC_Key_Tab,
+  eRENDERDOC_Key_PrtScrn,
+  eRENDERDOC_Key_Pause,
+
+  eRENDERDOC_Key_Max,
+} RENDERDOC_InputButton;
+
+// Sets which key or keys can be used to toggle focus between multiple windows
+//
+// If keys is NULL or num is 0, toggle keys will be disabled
+typedef void(RENDERDOC_CC *pRENDERDOC_SetFocusToggleKeys)(RENDERDOC_InputButton *keys, int num);
+
+// Sets which key or keys can be used to capture the next frame
+//
+// If keys is NULL or num is 0, captures keys will be disabled
+typedef void(RENDERDOC_CC *pRENDERDOC_SetCaptureKeys)(RENDERDOC_InputButton *keys, int num);
+
+typedef enum RENDERDOC_OverlayBits
+{
+  // This single bit controls whether the overlay is enabled or disabled globally
+  eRENDERDOC_Overlay_Enabled = 0x1,
+
+  // Show the average framerate over several seconds as well as min/max
+  eRENDERDOC_Overlay_FrameRate = 0x2,
+
+  // Show the current frame number
+  eRENDERDOC_Overlay_FrameNumber = 0x4,
+
+  // Show a list of recent captures, and how many captures have been made
+  eRENDERDOC_Overlay_CaptureList = 0x8,
+
+  // Default values for the overlay mask
+  eRENDERDOC_Overlay_Default = (eRENDERDOC_Overlay_Enabled | eRENDERDOC_Overlay_FrameRate |
+                                eRENDERDOC_Overlay_FrameNumber | eRENDERDOC_Overlay_CaptureList),
+
+  // Enable all bits
+  eRENDERDOC_Overlay_All = 0x7ffffff,
+
+  // Disable all bits
+  eRENDERDOC_Overlay_None = 0,
+} RENDERDOC_OverlayBits;
+
+// returns the overlay bits that have been set
+typedef uint32_t(RENDERDOC_CC *pRENDERDOC_GetOverlayBits)(void);
+// sets the overlay bits with an and & or mask
+typedef void(RENDERDOC_CC *pRENDERDOC_MaskOverlayBits)(uint32_t And, uint32_t Or);
+
+// this function will attempt to remove RenderDoc's hooks in the application.
+//
+// Note: that this can only work correctly if done immediately after
+// the module is loaded, before any API work happens. RenderDoc will remove its
+// injected hooks and shut down. Behaviour is undefined if this is called
+// after any API functions have been called, and there is still no guarantee of
+// success.
+typedef void(RENDERDOC_CC *pRENDERDOC_RemoveHooks)(void);
+
+// DEPRECATED: compatibility for code compiled against pre-1.4.1 headers.
+typedef pRENDERDOC_RemoveHooks pRENDERDOC_Shutdown;
+
+// This function will unload RenderDoc's crash handler.
+//
+// If you use your own crash handler and don't want RenderDoc's handler to
+// intercede, you can call this function to unload it and any unhandled
+// exceptions will pass to the next handler.
+typedef void(RENDERDOC_CC *pRENDERDOC_UnloadCrashHandler)(void);
+
+// Sets the capture file path template
+//
+// pathtemplate is a UTF-8 string that gives a template for how captures will be named
+// and where they will be saved.
+//
+// Any extension is stripped off the path, and captures are saved in the directory
+// specified, and named with the filename and the frame number appended. If the
+// directory does not exist it will be created, including any parent directories.
+//
+// If pathtemplate is NULL, the template will remain unchanged
+//
+// Example:
+//
+// SetCaptureFilePathTemplate("my_captures/example");
+//
+// Capture #1 -> my_captures/example_frame123.rdc
+// Capture #2 -> my_captures/example_frame456.rdc
+typedef void(RENDERDOC_CC *pRENDERDOC_SetCaptureFilePathTemplate)(const char *pathtemplate);
+
+// returns the current capture path template, see SetCaptureFileTemplate above, as a UTF-8 string
+typedef const char *(RENDERDOC_CC *pRENDERDOC_GetCaptureFilePathTemplate)(void);
+
+// DEPRECATED: compatibility for code compiled against pre-1.1.2 headers.
+typedef pRENDERDOC_SetCaptureFilePathTemplate pRENDERDOC_SetLogFilePathTemplate;
+typedef pRENDERDOC_GetCaptureFilePathTemplate pRENDERDOC_GetLogFilePathTemplate;
+
+// returns the number of captures that have been made
+typedef uint32_t(RENDERDOC_CC *pRENDERDOC_GetNumCaptures)(void);
+
+// This function returns the details of a capture, by index. New captures are added
+// to the end of the list.
+//
+// filename will be filled with the absolute path to the capture file, as a UTF-8 string
+// pathlength will be written with the length in bytes of the filename string
+// timestamp will be written with the time of the capture, in seconds since the Unix epoch
+//
+// Any of the parameters can be NULL and they'll be skipped.
+//
+// The function will return 1 if the capture index is valid, or 0 if the index is invalid
+// If the index is invalid, the values will be unchanged
+//
+// Note: when captures are deleted in the UI they will remain in this list, so the
+// capture path may not exist anymore.
+typedef uint32_t(RENDERDOC_CC *pRENDERDOC_GetCapture)(uint32_t idx, char *filename,
+                                                      uint32_t *pathlength, uint64_t *timestamp);
+
+// Sets the comments associated with a capture file. These comments are displayed in the
+// UI program when opening.
+//
+// filePath should be a path to the capture file to add comments to. If set to NULL or ""
+// the most recent capture file created made will be used instead.
+// comments should be a NULL-terminated UTF-8 string to add as comments.
+//
+// Any existing comments will be overwritten.
+typedef void(RENDERDOC_CC *pRENDERDOC_SetCaptureFileComments)(const char *filePath,
+                                                              const char *comments);
+
+// returns 1 if the RenderDoc UI is connected to this application, 0 otherwise
+typedef uint32_t(RENDERDOC_CC *pRENDERDOC_IsTargetControlConnected)(void);
+
+// DEPRECATED: compatibility for code compiled against pre-1.1.1 headers.
+// This was renamed to IsTargetControlConnected in API 1.1.1, the old typedef is kept here for
+// backwards compatibility with old code, it is castable either way since it's ABI compatible
+// as the same function pointer type.
+typedef pRENDERDOC_IsTargetControlConnected pRENDERDOC_IsRemoteAccessConnected;
+
+// This function will launch the Replay UI associated with the RenderDoc library injected
+// into the running application.
+//
+// if connectTargetControl is 1, the Replay UI will be launched with a command line parameter
+// to connect to this application
+// cmdline is the rest of the command line, as a UTF-8 string. E.g. a captures to open
+// if cmdline is NULL, the command line will be empty.
+//
+// returns the PID of the replay UI if successful, 0 if not successful.
+typedef uint32_t(RENDERDOC_CC *pRENDERDOC_LaunchReplayUI)(uint32_t connectTargetControl,
+                                                          const char *cmdline);
+
+// RenderDoc can return a higher version than requested if it's backwards compatible,
+// this function returns the actual version returned. If a parameter is NULL, it will be
+// ignored and the others will be filled out.
+typedef void(RENDERDOC_CC *pRENDERDOC_GetAPIVersion)(int *major, int *minor, int *patch);
+
+// Requests that the replay UI show itself (if hidden or not the current top window). This can be
+// used in conjunction with IsTargetControlConnected and LaunchReplayUI to intelligently handle
+// showing the UI after making a capture.
+//
+// This will return 1 if the request was successfully passed on, though it's not guaranteed that
+// the UI will be on top in all cases depending on OS rules. It will return 0 if there is no current
+// target control connection to make such a request, or if there was another error
+typedef uint32_t(RENDERDOC_CC *pRENDERDOC_ShowReplayUI)(void);
+
+//////////////////////////////////////////////////////////////////////////
+// Capturing functions
+//
+
+// A device pointer is a pointer to the API's root handle.
+//
+// This would be an ID3D11Device, HGLRC/GLXContext, ID3D12Device, etc
+typedef void *RENDERDOC_DevicePointer;
+
+// A window handle is the OS's native window handle
+//
+// This would be an HWND, GLXDrawable, etc
+typedef void *RENDERDOC_WindowHandle;
+
+// A helper macro for Vulkan, where the device handle cannot be used directly.
+//
+// Passing the VkInstance to this macro will return the RENDERDOC_DevicePointer to use.
+//
+// Specifically, the value needed is the dispatch table pointer, which sits as the first
+// pointer-sized object in the memory pointed to by the VkInstance. Thus we cast to a void** and
+// indirect once.
+#define RENDERDOC_DEVICEPOINTER_FROM_VKINSTANCE(inst) (*((void **)(inst)))
+
+// This sets the RenderDoc in-app overlay in the API/window pair as 'active' and it will
+// respond to keypresses. Neither parameter can be NULL
+typedef void(RENDERDOC_CC *pRENDERDOC_SetActiveWindow)(RENDERDOC_DevicePointer device,
+                                                       RENDERDOC_WindowHandle wndHandle);
+
+// capture the next frame on whichever window and API is currently considered active
+typedef void(RENDERDOC_CC *pRENDERDOC_TriggerCapture)(void);
+
+// capture the next N frames on whichever window and API is currently considered active
+typedef void(RENDERDOC_CC *pRENDERDOC_TriggerMultiFrameCapture)(uint32_t numFrames);
+
+// When choosing either a device pointer or a window handle to capture, you can pass NULL.
+// Passing NULL specifies a 'wildcard' match against anything. This allows you to specify
+// any API rendering to a specific window, or a specific API instance rendering to any window,
+// or in the simplest case of one window and one API, you can just pass NULL for both.
+//
+// In either case, if there are two or more possible matching (device,window) pairs it
+// is undefined which one will be captured.
+//
+// Note: for headless rendering you can pass NULL for the window handle and either specify
+// a device pointer or leave it NULL as above.
+
+// Immediately starts capturing API calls on the specified device pointer and window handle.
+//
+// If there is no matching thing to capture (e.g. no supported API has been initialised),
+// this will do nothing.
+//
+// The results are undefined (including crashes) if two captures are started overlapping,
+// even on separate devices and/oror windows.
+typedef void(RENDERDOC_CC *pRENDERDOC_StartFrameCapture)(RENDERDOC_DevicePointer device,
+                                                         RENDERDOC_WindowHandle wndHandle);
+
+// Returns whether or not a frame capture is currently ongoing anywhere.
+//
+// This will return 1 if a capture is ongoing, and 0 if there is no capture running
+typedef uint32_t(RENDERDOC_CC *pRENDERDOC_IsFrameCapturing)(void);
+
+// Ends capturing immediately.
+//
+// This will return 1 if the capture succeeded, and 0 if there was an error capturing.
+typedef uint32_t(RENDERDOC_CC *pRENDERDOC_EndFrameCapture)(RENDERDOC_DevicePointer device,
+                                                           RENDERDOC_WindowHandle wndHandle);
+
+// Ends capturing immediately and discard any data stored without saving to disk.
+//
+// This will return 1 if the capture was discarded, and 0 if there was an error or no capture
+// was in progress
+typedef uint32_t(RENDERDOC_CC *pRENDERDOC_DiscardFrameCapture)(RENDERDOC_DevicePointer device,
+                                                               RENDERDOC_WindowHandle wndHandle);
+
+// Only valid to be called between a call to StartFrameCapture and EndFrameCapture. Gives a custom
+// title to the capture produced which will be displayed in the UI.
+//
+// If multiple captures are ongoing, this title will be applied to the first capture to end after
+// this call. The second capture to end will have no title, unless this function is called again.
+//
+// Calling this function has no effect if no capture is currently running, and if it is called
+// multiple times only the last title will be used.
+typedef void(RENDERDOC_CC *pRENDERDOC_SetCaptureTitle)(const char *title);
+
+// Annotations API:
+//
+// These functions allow you to specify annotations either on a per-command level, or a per-object
+// level.
+//
+// Basic types of annotations are supported, as well as vector versions and references to API objects.
+//
+// The annotations are stored as keys, with the key being a dot-separated path allowing arbitrary
+// nesting and user organisation. The keys are sorted in human order so `foo.2.bar` will be displayed
+// before `foo.10.bar` to allow creation of arrays if desired.
+//
+// Deleting an annotation can be done by assigning an empty value to it.
+
+// the type of an annotation value, or Empty to delete an annotation
+typedef enum RENDERDOC_AnnotationType
+{
+  eRENDERDOC_Empty,
+  eRENDERDOC_Bool,
+  eRENDERDOC_Int32,
+  eRENDERDOC_UInt32,
+  eRENDERDOC_Int64,
+  eRENDERDOC_UInt64,
+  eRENDERDOC_Float,
+  eRENDERDOC_Double,
+  eRENDERDOC_String,
+  eRENDERDOC_APIObject,
+  eRENDERDOC_AnnotationMax = 0x7FFFFFFF,
+} RENDERDOC_AnnotationType;
+
+// a union with vector annotation value data
+typedef union RENDERDOC_AnnotationVectorValue
+{
+  bool boolean[4];
+  int32_t int32[4];
+  int64_t int64[4];
+  uint32_t uint32[4];
+  uint64_t uint64[4];
+  float float32[4];
+  double float64[4];
+} RENDERDOC_AnnotationVectorValue;
+
+// a union with scalar annotation value data
+typedef union RENDERDOC_AnnotationValue
+{
+  bool boolean;
+  int32_t int32;
+  int64_t int64;
+  uint32_t uint32;
+  uint64_t uint64;
+  float float32;
+  double float64;
+
+  RENDERDOC_AnnotationVectorValue vector;
+
+  const char *string;
+  void *apiObject;
+} RENDERDOC_AnnotationValue;
+
+// a struct for specifying a GL object, as we don't have pointers we can use so instead we specify a
+// pointer to this struct giving both the type and the name
+typedef struct RENDERDOC_GLResourceReference
+{
+  // this is the same GLenum identifier as passed to glObjectLabel
+  uint32_t identifier;
+  uint32_t name;
+} GLResourceReference;
+
+// simple C++ helpers to avoid the need for a temporary objects for value passing and GL object specification
+#ifdef __cplusplus
+struct RDGLObjectHelper
+{
+  RENDERDOC_GLResourceReference gl;
+
+  RDGLObjectHelper(uint32_t identifier, uint32_t name)
+  {
+    gl.identifier = identifier;
+    gl.name = name;
+  }
+
+  operator RENDERDOC_GLResourceReference *() { return &gl; }
+};
+
+struct RDAnnotationHelper
+{
+  RENDERDOC_AnnotationValue val;
+
+  RDAnnotationHelper(bool b) { val.boolean = b; }
+  RDAnnotationHelper(int32_t i) { val.int32 = i; }
+  RDAnnotationHelper(int64_t i) { val.int64 = i; }
+  RDAnnotationHelper(uint32_t i) { val.uint32 = i; }
+  RDAnnotationHelper(uint64_t i) { val.uint64 = i; }
+  RDAnnotationHelper(float f) { val.float32 = f; }
+  RDAnnotationHelper(double d) { val.float64 = d; }
+  RDAnnotationHelper(const char *s) { val.string = s; }
+
+  operator RENDERDOC_AnnotationValue *() { return &val; }
+};
+#endif
+
+// The device is specified in the same way as other API calls that take a RENDERDOC_DevicePointer
+// to specify the device.
+//
+// The object or queue/commandbuffer will depend on the graphics API in question.
+//
+// Return value:
+// 0 - The annotation was applied successfully.
+// 1 - The device is unknown/invalid
+// 2 - The device is valid but the annotation is not supported for API-specific reasons, such as an
+//     unrecognised or invalid object or queue/commandbuffer
+// 3 - The call is ill-formed or invalid e.g. empty is specified with a value pointer, or non-empty
+//     is specified with a NULL value pointer
+typedef uint32_t(RENDERDOC_CC *pRENDERDOC_SetObjectAnnotation)(RENDERDOC_DevicePointer device,
+                                                               void *object, const char *key,
+                                                               RENDERDOC_AnnotationType valueType,
+                                                               uint32_t valueVectorWidth,
+                                                               const RENDERDOC_AnnotationValue *value);
+
+typedef uint32_t(RENDERDOC_CC *pRENDERDOC_SetCommandAnnotation)(
+    RENDERDOC_DevicePointer device, void *queueOrCommandBuffer, const char *key,
+    RENDERDOC_AnnotationType valueType, uint32_t valueVectorWidth,
+    const RENDERDOC_AnnotationValue *value);
+
+//////////////////////////////////////////////////////////////////////////////////////////////////
+// RenderDoc API versions
+//
+
+// RenderDoc uses semantic versioning (http://semver.org/).
+//
+// MAJOR version is incremented when incompatible API changes happen.
+// MINOR version is incremented when functionality is added in a backwards-compatible manner.
+// PATCH version is incremented when backwards-compatible bug fixes happen.
+//
+// Note that this means the API returned can be higher than the one you might have requested.
+// e.g. if you are running against a newer RenderDoc that supports 1.0.1, it will be returned
+// instead of 1.0.0. You can check this with the GetAPIVersion entry point
+typedef enum RENDERDOC_Version
+{
+  eRENDERDOC_API_Version_1_0_0 = 10000,    // RENDERDOC_API_1_0_0 = 1 00 00
+  eRENDERDOC_API_Version_1_0_1 = 10001,    // RENDERDOC_API_1_0_1 = 1 00 01
+  eRENDERDOC_API_Version_1_0_2 = 10002,    // RENDERDOC_API_1_0_2 = 1 00 02
+  eRENDERDOC_API_Version_1_1_0 = 10100,    // RENDERDOC_API_1_1_0 = 1 01 00
+  eRENDERDOC_API_Version_1_1_1 = 10101,    // RENDERDOC_API_1_1_1 = 1 01 01
+  eRENDERDOC_API_Version_1_1_2 = 10102,    // RENDERDOC_API_1_1_2 = 1 01 02
+  eRENDERDOC_API_Version_1_2_0 = 10200,    // RENDERDOC_API_1_2_0 = 1 02 00
+  eRENDERDOC_API_Version_1_3_0 = 10300,    // RENDERDOC_API_1_3_0 = 1 03 00
+  eRENDERDOC_API_Version_1_4_0 = 10400,    // RENDERDOC_API_1_4_0 = 1 04 00
+  eRENDERDOC_API_Version_1_4_1 = 10401,    // RENDERDOC_API_1_4_1 = 1 04 01
+  eRENDERDOC_API_Version_1_4_2 = 10402,    // RENDERDOC_API_1_4_2 = 1 04 02
+  eRENDERDOC_API_Version_1_5_0 = 10500,    // RENDERDOC_API_1_5_0 = 1 05 00
+  eRENDERDOC_API_Version_1_6_0 = 10600,    // RENDERDOC_API_1_6_0 = 1 06 00
+  eRENDERDOC_API_Version_1_7_0 = 10700,    // RENDERDOC_API_1_7_0 = 1 07 00
+} RENDERDOC_Version;
+
+// API version changelog:
+//
+// 1.0.0 - initial release
+// 1.0.1 - Bugfix: IsFrameCapturing() was returning false for captures that were triggered
+//         by keypress or TriggerCapture, instead of Start/EndFrameCapture.
+// 1.0.2 - Refactor: Renamed eRENDERDOC_Option_DebugDeviceMode to eRENDERDOC_Option_APIValidation
+// 1.1.0 - Add feature: TriggerMultiFrameCapture(). Backwards compatible with 1.0.x since the new
+//         function pointer is added to the end of the struct, the original layout is identical
+// 1.1.1 - Refactor: Renamed remote access to target control (to better disambiguate from remote
+//         replay/remote server concept in replay UI)
+// 1.1.2 - Refactor: Renamed "log file" in function names to just capture, to clarify that these
+//         are captures and not debug logging files. This is the first API version in the v1.0
+//         branch.
+// 1.2.0 - Added feature: SetCaptureFileComments() to add comments to a capture file that will be
+//         displayed in the UI program on load.
+// 1.3.0 - Added feature: New capture option eRENDERDOC_Option_AllowUnsupportedVendorExtensions
+//         which allows users to opt-in to allowing unsupported vendor extensions to function.
+//         Should be used at the user's own risk.
+//         Refactor: Renamed eRENDERDOC_Option_VerifyMapWrites to
+//         eRENDERDOC_Option_VerifyBufferAccess, which now also controls initialisation to
+//         0xdddddddd of uninitialised buffer contents.
+// 1.4.0 - Added feature: DiscardFrameCapture() to discard a frame capture in progress and stop
+//         capturing without saving anything to disk.
+// 1.4.1 - Refactor: Renamed Shutdown to RemoveHooks to better clarify what is happening
+// 1.4.2 - Refactor: Renamed 'draws' to 'actions' in callstack capture option.
+// 1.5.0 - Added feature: ShowReplayUI() to request that the replay UI show itself if connected
+// 1.6.0 - Added feature: SetCaptureTitle() which can be used to set a title for a
+//         capture made with StartFrameCapture() or EndFrameCapture()
+// 1.7.0 - Added feature: SetObjectAnnotation() / SetCommandAnnotation() for adding rich
+//         annotations to objects and command streams
+
+typedef struct RENDERDOC_API_1_7_0
+{
+  pRENDERDOC_GetAPIVersion GetAPIVersion;
+
+  pRENDERDOC_SetCaptureOptionU32 SetCaptureOptionU32;
+  pRENDERDOC_SetCaptureOptionF32 SetCaptureOptionF32;
+
+  pRENDERDOC_GetCaptureOptionU32 GetCaptureOptionU32;
+  pRENDERDOC_GetCaptureOptionF32 GetCaptureOptionF32;
+
+  pRENDERDOC_SetFocusToggleKeys SetFocusToggleKeys;
+  pRENDERDOC_SetCaptureKeys SetCaptureKeys;
+
+  pRENDERDOC_GetOverlayBits GetOverlayBits;
+  pRENDERDOC_MaskOverlayBits MaskOverlayBits;
+
+  // Shutdown was renamed to RemoveHooks in 1.4.1.
+  // These unions allow old code to continue compiling without changes
+  union
+  {
+    pRENDERDOC_Shutdown Shutdown;
+    pRENDERDOC_RemoveHooks RemoveHooks;
+  };
+  pRENDERDOC_UnloadCrashHandler UnloadCrashHandler;
+
+  // Get/SetLogFilePathTemplate was renamed to Get/SetCaptureFilePathTemplate in 1.1.2.
+  // These unions allow old code to continue compiling without changes
+  union
+  {
+    // deprecated name
+    pRENDERDOC_SetLogFilePathTemplate SetLogFilePathTemplate;
+    // current name
+    pRENDERDOC_SetCaptureFilePathTemplate SetCaptureFilePathTemplate;
+  };
+  union
+  {
+    // deprecated name
+    pRENDERDOC_GetLogFilePathTemplate GetLogFilePathTemplate;
+    // current name
+    pRENDERDOC_GetCaptureFilePathTemplate GetCaptureFilePathTemplate;
+  };
+
+  pRENDERDOC_GetNumCaptures GetNumCaptures;
+  pRENDERDOC_GetCapture GetCapture;
+
+  pRENDERDOC_TriggerCapture TriggerCapture;
+
+  // IsRemoteAccessConnected was renamed to IsTargetControlConnected in 1.1.1.
+  // This union allows old code to continue compiling without changes
+  union
+  {
+    // deprecated name
+    pRENDERDOC_IsRemoteAccessConnected IsRemoteAccessConnected;
+    // current name
+    pRENDERDOC_IsTargetControlConnected IsTargetControlConnected;
+  };
+  pRENDERDOC_LaunchReplayUI LaunchReplayUI;
+
+  pRENDERDOC_SetActiveWindow SetActiveWindow;
+
+  pRENDERDOC_StartFrameCapture StartFrameCapture;
+  pRENDERDOC_IsFrameCapturing IsFrameCapturing;
+  pRENDERDOC_EndFrameCapture EndFrameCapture;
+
+  // new function in 1.1.0
+  pRENDERDOC_TriggerMultiFrameCapture TriggerMultiFrameCapture;
+
+  // new function in 1.2.0
+  pRENDERDOC_SetCaptureFileComments SetCaptureFileComments;
+
+  // new function in 1.4.0
+  pRENDERDOC_DiscardFrameCapture DiscardFrameCapture;
+
+  // new function in 1.5.0
+  pRENDERDOC_ShowReplayUI ShowReplayUI;
+
+  // new function in 1.6.0
+  pRENDERDOC_SetCaptureTitle SetCaptureTitle;
+
+  // new functions in 1.7.0
+  pRENDERDOC_SetObjectAnnotation SetObjectAnnotation;
+  pRENDERDOC_SetCommandAnnotation SetCommandAnnotation;
+} RENDERDOC_API_1_7_0;
+
+typedef RENDERDOC_API_1_7_0 RENDERDOC_API_1_0_0;
+typedef RENDERDOC_API_1_7_0 RENDERDOC_API_1_0_1;
+typedef RENDERDOC_API_1_7_0 RENDERDOC_API_1_0_2;
+typedef RENDERDOC_API_1_7_0 RENDERDOC_API_1_1_0;
+typedef RENDERDOC_API_1_7_0 RENDERDOC_API_1_1_1;
+typedef RENDERDOC_API_1_7_0 RENDERDOC_API_1_1_2;
+typedef RENDERDOC_API_1_7_0 RENDERDOC_API_1_2_0;
+typedef RENDERDOC_API_1_7_0 RENDERDOC_API_1_3_0;
+typedef RENDERDOC_API_1_7_0 RENDERDOC_API_1_4_0;
+typedef RENDERDOC_API_1_7_0 RENDERDOC_API_1_4_1;
+typedef RENDERDOC_API_1_7_0 RENDERDOC_API_1_4_2;
+typedef RENDERDOC_API_1_7_0 RENDERDOC_API_1_5_0;
+typedef RENDERDOC_API_1_7_0 RENDERDOC_API_1_6_0;
+
+//////////////////////////////////////////////////////////////////////////////////////////////////
+// RenderDoc API entry point
+//
+// This entry point can be obtained via GetProcAddress/dlsym if RenderDoc is available.
+//
+// The name is the same as the typedef - "RENDERDOC_GetAPI"
+//
+// This function is not thread safe, and should not be called on multiple threads at once.
+// Ideally, call this once as early as possible in your application's startup, before doing
+// any API work, since some configuration functionality etc has to be done also before
+// initialising any APIs.
+//
+// Parameters:
+//   version is a single value from the RENDERDOC_Version above.
+//
+//   outAPIPointers will be filled out with a pointer to the corresponding struct of function
+//   pointers.
+//
+// Returns:
+//   1 - if the outAPIPointers has been filled with a pointer to the API struct requested
+//   0 - if the requested version is not supported or the arguments are invalid.
+//
+typedef int(RENDERDOC_CC *pRENDERDOC_GetAPI)(RENDERDOC_Version version, void **outAPIPointers);
+
+#ifdef __cplusplus
+}    // extern "C"
+#endif

CMakeLists.txt

@@ -203,14 +203,21 @@ file(GLOB kyty_emulator_src CONFIGURE_DEPENDS
 	"${KYTY_SOURCE_DIR}/graphics/shader/*.h"
 	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/*.cpp"
 	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/*.h"
-	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/cfg/*.cpp"
-	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/cfg/*.h"
-	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/decompiler/*.cpp"
-	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/decompiler/*.h"
-	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/emitter/*.cpp"
-	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/emitter/*.h"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/frontend/cfg/*.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/frontend/cfg/*.h"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/frontend/decode/*.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/frontend/decode/*.h"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/frontend/translate/*.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/frontend/translate/*.h"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/backend/spirv/*.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/backend/spirv/*.h"
 	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/*.cpp"
 	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/*.h"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/opcodes/*.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/opcodes/*.h"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/opcodes/*.inc"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/passes/*.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/passes/*.h"
 	"${KYTY_SOURCE_DIR}/graphics/presentation/*.cpp"
 	"${KYTY_SOURCE_DIR}/graphics/presentation/*.h"
 	"${KYTY_SOURCE_DIR}/graphics/presentation/window/*.cpp"
@@ -322,6 +329,7 @@ set(inc_headers
 	${KYTY_THIRD_PARTY_DIR}/fmt/include
 	${KYTY_THIRD_PARTY_DIR}/xxhash
 	${KYTY_THIRD_PARTY_DIR}/cpuinfo/include
+	${KYTY_THIRD_PARTY_DIR}/renderdoc
 	${KYTY_THIRD_PARTY_DIR}/stb
 )
 
@@ -367,9 +375,8 @@ add_kyty_full_emulator_test(shader_cfg_tests "${KYTY_TESTS_DIR}/shaderCfgTests.c
 add_executable(scalar_provenance_tests EXCLUDE_FROM_ALL
 	"${KYTY_TESTS_DIR}/ScalarProvenanceTests.cpp"
 	"${KYTY_SOURCE_DIR}/graphics/host_gpu/hostMemory.cpp"
-	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/ReadLaneElimination.cpp"
-	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/ScalarProvenance.cpp"
-	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/SrtWalker.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/passes/ReadLaneElimination.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/passes/SrtWalker.cpp"
 )
 target_link_libraries(scalar_provenance_tests fmt::fmt)
 target_include_directories(scalar_provenance_tests PRIVATE ${inc_headers})
@@ -378,6 +385,7 @@ add_executable(page_manager_tests EXCLUDE_FROM_ALL
 	"${KYTY_TESTS_DIR}/PageManagerTests.cpp"
 	"${KYTY_SOURCE_DIR}/graphics/host_gpu/pageManager.cpp"
 )
+target_link_libraries(page_manager_tests common)
 target_include_directories(page_manager_tests PRIVATE ${inc_headers})
 
 add_executable(bit_array_tests EXCLUDE_FROM_ALL
@@ -387,6 +395,7 @@ target_include_directories(bit_array_tests PRIVATE ${inc_headers})
 
 add_executable(ime_dialog_tests EXCLUDE_FROM_ALL
 	"${KYTY_TESTS_DIR}/ImeDialogTests.cpp"
+	"${KYTY_SOURCE_DIR}/libs/imeCommon.cpp"
 	"${KYTY_SOURCE_DIR}/libs/imeDialog.cpp"
 )
 target_include_directories(ime_dialog_tests PRIVATE ${inc_headers})
@@ -411,9 +420,8 @@ add_executable(shader_stage_runtime_tests EXCLUDE_FROM_ALL
 	"${KYTY_SOURCE_DIR}/graphics/guest_gpu/gpu_format.cpp"
 	"${KYTY_SOURCE_DIR}/graphics/host_gpu/hostMemory.cpp"
 	"${KYTY_SOURCE_DIR}/graphics/shader/shaderStageRuntime.cpp"
-	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/ResourceMaterialization.cpp"
-	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/ScalarProvenance.cpp"
-	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/SrtWalker.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/passes/ResourceMaterialization.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/passes/SrtWalker.cpp"
 )
 target_link_libraries(shader_stage_runtime_tests fmt::fmt)
 target_include_directories(shader_stage_runtime_tests PRIVATE ${inc_headers})
@@ -422,13 +430,11 @@ add_executable(resource_tracking_tests EXCLUDE_FROM_ALL
 	"${KYTY_TESTS_DIR}/ResourceTrackingTests.cpp"
 	"${KYTY_SOURCE_DIR}/graphics/guest_gpu/gpu_format.cpp"
 	"${KYTY_SOURCE_DIR}/graphics/host_gpu/hostMemory.cpp"
-	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/ScalarProvenance.cpp"
-	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/SrtWalker.cpp"
-	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/SrtPatcher.cpp"
-	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/ResourceTracking.cpp"
-	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/ResourceMaterialization.cpp"
-	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/ShaderInfoCollection.cpp"
-	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/BindingLayout.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/passes/SrtWalker.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/passes/ResourceTracking.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/passes/ResourceMaterialization.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/passes/ShaderInfoCollection.cpp"
+	"${KYTY_SOURCE_DIR}/graphics/shader/recompiler/ir/passes/BindingLayout.cpp"
 )
 target_link_libraries(resource_tracking_tests fmt::fmt)
 target_include_directories(resource_tracking_tests PRIVATE ${inc_headers})
@@ -449,6 +455,9 @@ add_executable(event_queue_lifetime_tests EXCLUDE_FROM_ALL
 target_link_libraries(event_queue_lifetime_tests common fmt::fmt)
 target_include_directories(event_queue_lifetime_tests PRIVATE ${inc_headers})
 
+add_kyty_full_emulator_test(kernel_file_system_tests
+	"${KYTY_TESTS_DIR}/KernelFileSystemTests.cpp")
+
 add_executable(sync_on_address_tests EXCLUDE_FROM_ALL
 	"${KYTY_TESTS_DIR}/SyncOnAddressTests.cpp"
 	"${KYTY_SOURCE_DIR}/kernel/syncOnAddress.cpp"
@@ -509,9 +518,12 @@ if(BUILD_TESTING)
 	add_test(NAME shader_stage_runtime COMMAND $<TARGET_FILE:shader_stage_runtime_tests>)
 	add_test(NAME resource_tracking COMMAND $<TARGET_FILE:resource_tracking_tests>)
 	add_test(NAME event_queue_lifetime COMMAND $<TARGET_FILE:event_queue_lifetime_tests>)
+	add_test(NAME kernel_file_system COMMAND $<TARGET_FILE:kernel_file_system_tests>)
 	add_test(NAME sync_on_address COMMAND $<TARGET_FILE:sync_on_address_tests>)
 	add_test(NAME audio_out2_port COMMAND $<TARGET_FILE:audio_out2_port_tests>)
 	add_test(NAME shader_recompiler_compute COMMAND $<TARGET_FILE:shader_recompiler_compute_tests>)
+	add_test(NAME shader_recompiler_alignbyte
+		COMMAND $<TARGET_FILE:shader_recompiler_compute_tests> --alignbyte-only)
 	add_test(NAME virtual_memory_allocation
 		COMMAND $<TARGET_FILE:virtual_memory_allocation_tests>)
 	add_test(NAME guest_red_zone_patcher
@@ -560,9 +572,11 @@ if(BUILD_TESTING)
 		shader_stage_runtime_tests
 		resource_tracking_tests
 		event_queue_lifetime_tests
+		kernel_file_system_tests
 		sync_on_address_tests
 		shader_recompiler_compute_tests
 		virtual_memory_allocation_tests
+		audio_out2_port_tests
 	)
 endif()
 

README.md

@@ -55,7 +55,7 @@ graphical glitches, low compatibility, and poor performance.
   <tr>
     <td align="center">
       <strong>Neptunia ReVerse</strong><br>
-      <img src="docs/screenshots/ps5-04.png" width="300" alt="Minecraft Legends running in KytyPS5">
+      <img src="docs/screenshots/ps5-04.png" width="300" alt="Neptunia ReVerse running in KytyPS5">
     </td>
     <td align="center">
       <strong>SILENT HILL: The Short Message</strong><br>
@@ -65,11 +65,11 @@ graphical glitches, low compatibility, and poor performance.
   <tr>
     <td align="center">
       <strong>Hellboy</strong><br>
-      <img src="docs/screenshots/ps5-02.png" width="300" alt="Disgaea 6 running in KytyPS5">
+      <img src="docs/screenshots/ps5-02.png" width="300" alt="Hellboy running in KytyPS5">
     </td>
     <td align="center">
       <strong>Paleo Pines</strong><br>
-      <img src="docs/screenshots/ps5-06.png" width="300" alt="Dreaming Sarah running in KytyPS5">
+      <img src="docs/screenshots/ps5-06.png" width="300" alt="Paleo Pines running in KytyPS5">
     </td>
   </tr>
 </table>

src/common/assert.cpp

@@ -35,7 +35,7 @@ static std::string BuildFatalReport(const char* title, std::string_view text, co
 
 static int DbgReport(const char* title, std::string_view text, const char* file, int line) {
 	Log::WriteFatal(BuildFatalReport(title, text, file, line));
-	SubsystemsListSingleton::Instance()->ShutdownAll();
+	Subsystems::EmergencyShutdownActive();
 	return 1;
 }
 

src/common/commonSubsystem.cpp

@@ -1,21 +0,0 @@
-#include "common/commonSubsystem.h"
-
-#include "common/byteBuffer.h" // IWYU pragma: associated
-#include "common/common.h"     // IWYU pragma: associated
-#include "common/debug.h"      // IWYU pragma: associated
-#include "common/hash.h"       // IWYU pragma: associated
-#include "common/magicEnum.h"  // IWYU pragma: associated
-#include "common/singleton.h"  // IWYU pragma: associated
-#include "common/virtualMemory.h"
-
-namespace Common {
-
-KYTY_SUBSYSTEM_INIT(Common) {
-	VirtualMemory::Init();
-}
-
-KYTY_SUBSYSTEM_UNEXPECTED_SHUTDOWN(Common) {}
-
-KYTY_SUBSYSTEM_DESTROY(Common) {}
-
-} // namespace Common

src/common/commonSubsystem.h

@@ -1,12 +0,0 @@
-#ifndef KYTY_COMMON_COMMON_SUBSYSTEM_H_
-#define KYTY_COMMON_COMMON_SUBSYSTEM_H_
-
-#include "common/subsystems.h"
-
-namespace Common {
-
-KYTY_SUBSYSTEM_DEFINE(Common);
-
-} // namespace Common
-
-#endif /* KYTY_COMMON_COMMON_SUBSYSTEM_H_ */

src/common/emulatorConfig.cpp

@@ -9,15 +9,15 @@ namespace Config {
 
 static std::unique_ptr<ConfigOptions> g_config;
 
-KYTY_SUBSYSTEM_INIT(Config) {
+void Initialize() {
 	EXIT_IF(g_config != nullptr);
 
 	g_config = std::make_unique<ConfigOptions>();
 }
 
-KYTY_SUBSYSTEM_UNEXPECTED_SHUTDOWN(Config) {}
-
-KYTY_SUBSYSTEM_DESTROY(Config) {}
+void Shutdown() {
+	g_config.reset();
+}
 
 void Load(const ConfigOptions& cfg) {
 	EXIT_IF(g_config == nullptr);
@@ -93,6 +93,10 @@ bool SpirvDebugPrintfEnabled() {
 	return g_config->spirv_debug_printf_enabled;
 }
 
+bool GpuAssistedValidationEnabled() {
+	return g_config->gpu_assisted_validation_enabled && g_config->vulkan_validation_enabled;
+}
+
 bool RenderDocEnabled() {
 	return g_config->renderdoc_enabled;
 }
@@ -101,6 +105,10 @@ bool ReadbackLinearImagesEnabled() {
 	return g_config->readback_linear_images;
 }
 
+bool PlayGoHackEnabled() {
+	return g_config->playgo_hack_enabled;
+}
+
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
 bool RedZoneProtectionEnabled() {
 	return g_config->red_zone_protection_enabled;

src/common/emulatorConfig.h

@@ -2,7 +2,6 @@
 #define KYTY_COMMON_EMULATOR_CONFIG_H_
 
 #include "common/common.h"
-#include "common/subsystems.h"
 
 #include <filesystem>
 #include <string>
@@ -10,7 +9,14 @@
 
 namespace Config {
 
-KYTY_SUBSYSTEM_DEFINE(Config);
+void Initialize();
+void Shutdown();
+
+struct Lifecycle {
+	static constexpr const char* name       = "Config";
+	static constexpr auto        initialize = Config::Initialize;
+	static constexpr auto        shutdown   = Config::Shutdown;
+};
 
 enum class ShaderOptimizationType { None, Size, Performance };
 
@@ -43,8 +49,10 @@ struct ConfigOptions {
 	std::filesystem::path  printf_output_file          = "_kyty.txt";
 	ProfilerDirection      profiler_direction          = ProfilerDirection::None;
 	bool                   spirv_debug_printf_enabled  = false;
+	bool                   gpu_assisted_validation_enabled = false;
 	bool                   renderdoc_enabled           = false;
 	bool                   readback_linear_images      = false;
+	bool                   playgo_hack_enabled         = false;
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
 	bool red_zone_protection_enabled = false;
 #endif
@@ -77,8 +85,11 @@ ProfilerDirection GetProfilerDirection();
 
 bool SpirvDebugPrintfEnabled();
 
+bool GpuAssistedValidationEnabled();
+
 bool RenderDocEnabled();
 bool ReadbackLinearImagesEnabled();
+bool PlayGoHackEnabled();
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
 bool RedZoneProtectionEnabled();
 #endif

src/common/hostException.cpp

@@ -2,7 +2,6 @@
 
 #include <atomic>
 #include <cstdio>
-#include <cstdlib>
 
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
 #include <windows.h> // IWYU pragma: keep
@@ -28,54 +27,14 @@ namespace Common::HostException {
 
 static std::atomic<Handler> g_handler {nullptr};
 static std::atomic_uint32_t g_install_state {0};
-static thread_local bool    g_in_exception_filter = false;
 
 static_assert(decltype(g_handler)::is_always_lock_free);
 static_assert(decltype(g_install_state)::is_always_lock_free);
-
-[[noreturn]] static void FailFast(const char* reason) noexcept {
-	std::fputs("HostException fail-fast: ", stderr);
-	std::fputs(reason != nullptr ? reason : "unspecified", stderr);
-	std::fputc('\n', stderr);
-	std::fflush(stderr);
-#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
-	TerminateProcess(GetCurrentProcess(), static_cast<UINT>(EXCEPTION_NONCONTINUABLE_EXCEPTION));
-#endif
-	std::_Exit(321);
-}
-
-class FilterScope final {
-public:
-	FilterScope() noexcept {
-		if (g_in_exception_filter) {
-			FailFast("nested exception while resolving a host fault");
-		}
-		g_in_exception_filter = true;
-	}
-
-	~FilterScope() { g_in_exception_filter = false; }
-
-	KYTY_CLASS_NO_COPY(FilterScope);
-};
-
-static Handler LoadInstalledHandler() noexcept {
-	if (g_install_state.load(std::memory_order_acquire) == 0) {
-		FailFast("host exception handler is not installed");
-	}
-
-	const auto handler = g_handler.load(std::memory_order_acquire);
-	if (handler == nullptr) {
-		FailFast("host exception callback is null");
-	}
-	return handler;
-}
 #endif
 
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
 
-static LONG WINAPI ExceptionFilter(PEXCEPTION_POINTERS exception) {
-	FilterScope filter_scope;
-
+static LONG WINAPI ExceptionFilter(PEXCEPTION_POINTERS exception) noexcept {
 	auto* exception_record = exception->ExceptionRecord;
 
 	if (exception_record->ExceptionCode == DBG_PRINTEXCEPTION_C ||
@@ -105,12 +64,6 @@ static LONG WINAPI ExceptionFilter(PEXCEPTION_POINTERS exception) {
 	} else if (exception_record->ExceptionCode == EXCEPTION_ILLEGAL_INSTRUCTION) {
 		info.type = ExceptionType::IllegalInstruction;
 	} else {
-		printf("Unhandled win exception: code=0x%08" PRIx32 ", addr=0x%016" PRIx64
-		       ", rip=0x%016" PRIx64 ", rsp=0x%016" PRIx64 ", rbp=0x%016" PRIx64 "\n",
-		       static_cast<uint32_t>(exception_record->ExceptionCode),
-		       reinterpret_cast<uint64_t>(exception_record->ExceptionAddress),
-		       exception->ContextRecord->Rip, exception->ContextRecord->Rsp,
-		       exception->ContextRecord->Rbp);
 		return EXCEPTION_CONTINUE_SEARCH;
 	}
 
@@ -131,28 +84,21 @@ static LONG WINAPI ExceptionFilter(PEXCEPTION_POINTERS exception) {
 	info.r14 = exception->ContextRecord->R14;
 	info.r15 = exception->ContextRecord->R15;
 
-	const auto handler = LoadInstalledHandler();
-
-	return handler(info) ? EXCEPTION_CONTINUE_EXECUTION : EXCEPTION_CONTINUE_SEARCH;
+	const auto handler = g_handler.load(std::memory_order_acquire);
+	if (handler != nullptr && handler(info)) {
+		return EXCEPTION_CONTINUE_EXECUTION;
+	}
+	return EXCEPTION_CONTINUE_SEARCH;
 }
 
 #elif defined(__APPLE__)
 
 static std::atomic<Handler> g_handler {nullptr};
 static std::atomic_uint32_t g_install_state {0};
-static thread_local bool    g_in_exception_filter = false;
 
 static_assert(decltype(g_handler)::is_always_lock_free);
 static_assert(decltype(g_install_state)::is_always_lock_free);
 
-[[noreturn]] static void FailFast(const char* reason) noexcept {
-	std::fputs("HostException fail-fast: ", stderr);
-	std::fputs(reason != nullptr ? reason : "unspecified", stderr);
-	std::fputc('\n', stderr);
-	std::fflush(stderr);
-	std::_Exit(321);
-}
-
 // Translate the x86-64 page-fault error code (mcontext __es.__err) into an access type.
 // bit 1 (0x2) = write, bit 4 (0x10) = instruction fetch, otherwise a read.
 static AccessViolationType DecodeAccess(uint64_t err) {
@@ -170,11 +116,6 @@ static AccessViolationType DecodeAccess(uint64_t err) {
 // re-executing the faulting instruction against the now-fixed protection. An unresolved
 // fault restores the default disposition so the retry terminates the process.
 static void SignalHandler(int sig, siginfo_t* si, void* uctx) {
-	if (g_in_exception_filter) {
-		FailFast("nested exception while resolving a host fault");
-	}
-	g_in_exception_filter = true;
-
 	auto*       uc = static_cast<ucontext_t*>(uctx);
 	const auto* mc = uc->uc_mcontext;
 	const auto& ss = mc->__ss;
@@ -210,14 +151,7 @@ static void SignalHandler(int sig, siginfo_t* si, void* uctx) {
 	info.r15 = ss.__r15;
 
 	const auto handler = g_handler.load(std::memory_order_acquire);
-	if (handler == nullptr) {
-		FailFast("host exception callback is null");
-	}
-
-	const bool resolved   = handler(info);
-	g_in_exception_filter = false;
-
-	if (resolved) {
+	if (handler != nullptr && handler(info)) {
 		return; // retry the faulting instruction against the fixed mapping
 	}
 
@@ -244,8 +178,6 @@ static void ChainToDefault(int signal_number) noexcept {
 }
 
 static void SignalHandler(int signal_number, siginfo_t* signal_info, void* native_context) {
-	FilterScope filter_scope;
-
 	auto* context = static_cast<ucontext_t*>(native_context);
 	auto* gregs   = context->uc_mcontext.gregs;
 
@@ -289,9 +221,8 @@ static void SignalHandler(int signal_number, siginfo_t* signal_info, void* nativ
 	info.r14 = static_cast<uint64_t>(gregs[REG_R14]);
 	info.r15 = static_cast<uint64_t>(gregs[REG_R15]);
 
-	const auto handler = LoadInstalledHandler();
-
-	if (handler(info)) {
+	const auto handler = g_handler.load(std::memory_order_acquire);
+	if (handler != nullptr && handler(info)) {
 		return;
 	}
 
@@ -313,7 +244,7 @@ bool InstallHandler(Handler handler) {
 	g_handler.store(handler, std::memory_order_release);
 
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
-	if (AddVectoredExceptionHandler(1, ExceptionFilter) == nullptr) {
+	if (AddVectoredExceptionHandler(0, ExceptionFilter) == nullptr) {
 		g_handler.store(nullptr, std::memory_order_release);
 		g_install_state.store(0, std::memory_order_release);
 		printf("AddVectoredExceptionHandler() failed\n");

src/common/logging/log.cpp

@@ -149,7 +149,7 @@ void WriteFatal(fmt::text_style style, std::string_view text) {
 	Flush();
 }
 
-KYTY_SUBSYSTEM_INIT(Log) {
+void Initialize() {
 	g_initialized = true;
 	switch (Config::GetPrintfDirection()) {
 		case Config::OutputDirection::Silent: g_direction = Direction::Silent; break;
@@ -161,13 +161,7 @@ KYTY_SUBSYSTEM_INIT(Log) {
 	SetupLogger();
 }
 
-KYTY_SUBSYSTEM_UNEXPECTED_SHUTDOWN(Log) {
-	Flush();
-	std::lock_guard lock(g_logger_mutex);
-	g_logger.reset();
-}
-
-KYTY_SUBSYSTEM_DESTROY(Log) {
+void Shutdown() {
 	Flush();
 	std::lock_guard lock(g_logger_mutex);
 	g_logger.reset();

src/common/logging/log.h

@@ -2,7 +2,6 @@
 #define KYTY_COMMON_LOGGING_LOG_H_
 
 #include "common/common.h"
-#include "common/subsystems.h"
 
 #include <fmt/color.h>
 #include <fmt/printf.h>
@@ -10,7 +9,15 @@
 
 namespace Log {
 
-KYTY_SUBSYSTEM_DEFINE(Log);
+void Initialize();
+void Shutdown();
+
+struct Lifecycle {
+	static constexpr const char* name               = "Log";
+	static constexpr auto        initialize         = Log::Initialize;
+	static constexpr auto        shutdown           = Log::Shutdown;
+	static constexpr auto        emergency_shutdown = Log::Shutdown;
+};
 
 enum class Direction { Silent, Console, File };
 

src/common/platform/sysWindowsDbg.cpp

@@ -15,56 +15,8 @@
 #include "common/assert.h"
 #include "common/platform/sysDbg.h"
 
-struct FrameS {
-	struct FrameS* next;
-	void*          ret_addr;
-};
-
-int WalkStack(int z_stack_depth, void** z_stack_trace) {
-	CONTEXT context;
-	//	KNONVOLATILE_CONTEXT_POINTERS NvContext;
-	PRUNTIME_FUNCTION runtime_function  = nullptr;
-	PVOID             handler_data      = nullptr;
-	ULONG64           establisher_frame = 0;
-	ULONG64           image_base        = 0;
-
-	RtlCaptureContext(&context);
-
-	int frame = 0;
-
-	while (true) {
-		if (frame >= z_stack_depth) {
-			break;
-		}
-
-		z_stack_trace[frame] = reinterpret_cast<void*>(context.Rip);
-
-		frame++;
-
-		runtime_function = RtlLookupFunctionEntry(context.Rip, &image_base, nullptr);
-
-		if (runtime_function == nullptr) {
-			break;
-		}
-
-		// RtlZeroMemory(&NvContext, sizeof(KNONVOLATILE_CONTEXT_POINTERS));
-		RtlVirtualUnwind(0, image_base, context.Rip, runtime_function, &context, &handler_data,
-		                 &establisher_frame, nullptr /*&NvContext*/);
-
-		if (context.Rip == 0u) {
-			break;
-		}
-	}
-
-	return frame;
-}
-
 void SysStackWalk(void** stack, int* depth) {
-	int n  = WalkStack(*depth, stack);
-	*depth = n;
-	//	unwind_info_t info = {stack, 0, *depth};
-	//	 _Unwind_Backtrace(&trace_fcn, &info);
-	//	 *depth = info.depth;
+	*depth = static_cast<int>(CaptureStackBackTrace(0, static_cast<DWORD>(*depth), stack, nullptr));
 }
 
 void SysStackUsagePrint(sys_dbg_stack_info_t& stack) {

src/common/profiler.cpp

@@ -1,7 +1,6 @@
 #include "common/profiler.h"
 
 #include "common/emulatorConfig.h"
-#include "common/subsystems.h"
 
 #include <algorithm>
 #include <common/TracyProtocol.hpp>
@@ -55,13 +54,7 @@ void SetThreadName(const char* name) {
 	}
 }
 
-void Close() {
-	if (tracy::ProfilerAvailable()) {
-		tracy::ShutdownProfiler();
-	}
-}
-
-KYTY_SUBSYSTEM_INIT(Profiler) {
+void Initialize() {
 	switch (Config::GetProfilerDirection()) {
 		case Config::ProfilerDirection::Network:
 			if (!tracy::ProfilerAvailable()) {
@@ -78,12 +71,10 @@ KYTY_SUBSYSTEM_INIT(Profiler) {
 	}
 }
 
-KYTY_SUBSYSTEM_UNEXPECTED_SHUTDOWN(Profiler) {
-	Close();
-}
-
-KYTY_SUBSYSTEM_DESTROY(Profiler) {
-	Close();
+void Shutdown() {
+	if (tracy::ProfilerAvailable()) {
+		tracy::ShutdownProfiler();
+	}
 }
 
 } // namespace Profiler

src/common/profiler.h

@@ -2,7 +2,6 @@
 #define KYTY_COMMON_PROFILER_H_
 
 #include "common/common.h"
-#include "common/subsystems.h"
 
 #include <cstdint>
 #include <optional>
@@ -42,7 +41,15 @@ private:
 void EndBlock();
 void SetThreadName(const char* name);
 
-KYTY_SUBSYSTEM_DEFINE(Profiler);
+void Initialize();
+void Shutdown();
+
+struct Lifecycle {
+	static constexpr const char* name               = "Profiler";
+	static constexpr auto        initialize         = Profiler::Initialize;
+	static constexpr auto        shutdown           = Profiler::Shutdown;
+	static constexpr auto        emergency_shutdown = Profiler::Shutdown;
+};
 
 } // namespace Profiler
 

src/common/slotVector.h

@@ -0,0 +1,132 @@
+#ifndef EMULATOR_SRC_COMMON_SLOTVECTOR_H_
+#define EMULATOR_SRC_COMMON_SLOTVECTOR_H_
+
+#include "common/assert.h"
+#include "common/abi.h"
+
+#include <compare>
+#include <cstdint>
+#include <deque>
+#include <functional>
+#include <limits>
+#include <optional>
+#include <utility>
+#include <vector>
+
+namespace Common {
+
+struct SlotId {
+	static constexpr uint32_t INVALID_INDEX = std::numeric_limits<uint32_t>::max();
+
+	constexpr SlotId() noexcept = default;
+	constexpr SlotId(uint32_t value) noexcept: index(value), generation(1) {}
+	constexpr SlotId(uint32_t value, uint32_t slot_generation) noexcept
+	    : index(value), generation(slot_generation) {}
+
+	[[nodiscard]] constexpr explicit operator bool() const noexcept { return index != INVALID_INDEX; }
+	constexpr auto operator<=>(const SlotId&) const noexcept = default;
+
+	uint32_t index = INVALID_INDEX;
+	uint32_t generation = 0;
+};
+
+// Stable-address slot storage for cache resources that are intentionally non-movable.
+template <typename T>
+class SlotVector {
+public:
+	SlotVector() = default;
+	KYTY_CLASS_NO_COPY(SlotVector);
+
+	[[nodiscard]] T& operator[](SlotId id) noexcept {
+		EXIT_IF(!is_allocated(id));
+		return *m_values[id.index].value;
+	}
+
+	[[nodiscard]] const T& operator[](SlotId id) const noexcept {
+		EXIT_IF(!is_allocated(id));
+		return *m_values[id.index].value;
+	}
+
+	[[nodiscard]] T* try_get(SlotId id) noexcept {
+		return is_allocated(id) ? &*m_values[id.index].value : nullptr;
+	}
+
+	[[nodiscard]] const T* try_get(SlotId id) const noexcept {
+		return is_allocated(id) ? &*m_values[id.index].value : nullptr;
+	}
+
+	[[nodiscard]] bool is_allocated(SlotId id) const noexcept {
+		return id && id.index < m_values.size() &&
+		       m_values[id.index].generation == id.generation && m_values[id.index].value.has_value();
+	}
+
+	template <typename... Args>
+	[[nodiscard]] SlotId insert(Args&&... args) {
+		uint32_t index = 0;
+		if (m_free_list.empty()) {
+			index = static_cast<uint32_t>(m_values.size());
+			m_values.emplace_back();
+			m_values.back().value.emplace(std::forward<Args>(args)...);
+		} else {
+			index = m_free_list.back();
+			m_free_list.pop_back();
+			EXIT_IF(m_values[index].value.has_value());
+			m_values[index].value.emplace(std::forward<Args>(args)...);
+		}
+		++m_size;
+		return SlotId {index, m_values[index].generation};
+	}
+
+	void erase(SlotId id) noexcept {
+		EXIT_IF(!is_allocated(id));
+		auto& slot = m_values[id.index];
+		slot.value.reset();
+		if (++slot.generation == 0) {
+			slot.generation = 1;
+		}
+		m_free_list.push_back(id.index);
+		--m_size;
+	}
+
+	[[nodiscard]] size_t size() const noexcept { return m_size; }
+	[[nodiscard]] size_t capacity() const noexcept { return m_values.size(); }
+
+	template <typename F>
+	void ForEach(F&& fn) {
+		for (uint32_t index = 0; index < m_values.size(); ++index) {
+			if (m_values[index].value) {
+				fn(SlotId {index, m_values[index].generation}, *m_values[index].value);
+			}
+		}
+	}
+
+	template <typename F>
+	void ForEach(F&& fn) const {
+		for (uint32_t index = 0; index < m_values.size(); ++index) {
+			if (m_values[index].value) {
+				fn(SlotId {index, m_values[index].generation}, *m_values[index].value);
+			}
+		}
+	}
+
+private:
+	struct Slot {
+		std::optional<T> value;
+		uint32_t         generation = 1;
+	};
+
+	std::deque<Slot>             m_values;
+	std::vector<uint32_t>        m_free_list;
+	size_t                       m_size = 0;
+};
+
+} // namespace Common
+
+template <>
+struct std::hash<Common::SlotId> {
+	[[nodiscard]] size_t operator()(Common::SlotId id) const noexcept {
+		return std::hash<uint64_t> {}((static_cast<uint64_t>(id.generation) << 32u) | id.index);
+	}
+};
+
+#endif // EMULATOR_SRC_COMMON_SLOTVECTOR_H_

src/common/subsystems.cpp

@@ -1,232 +1,43 @@
 #include "common/subsystems.h"
 
-#include "common/assert.h"
-
-#include <cstdarg>
-#include <cstdio>
-#include <cstring>
-#include <memory>
-#include <string>
-
 namespace Common {
 
-class SubsystemPrivate {
-public:
-	SubsystemPrivate()          = default;
-	virtual ~SubsystemPrivate() = default;
+static Subsystems* g_active_subsystems = nullptr;
 
-	KYTY_CLASS_NO_COPY(SubsystemPrivate);
+Subsystems::Subsystems(bool print): m_print(print) {
+	m_active.reserve(16);
+	g_active_subsystems = this;
+}
 
-	bool        failed {false};
-	std::string fail_msg;
-};
-
-class SubsystemsListPrivate {
-public:
-	explicit SubsystemsListPrivate(SubsystemsList& p): parent(p) {}
-
-	virtual ~SubsystemsListPrivate() = default;
-
-	void SetArgs(int argc, char** argv) {
-		this->m_argc = argc;
-		this->m_argv = argv;
+Subsystems::~Subsystems() {
+	Destroy();
+	if (g_active_subsystems == this) {
+		g_active_subsystems = nullptr;
 	}
+}
 
-	void Add(Subsystem* s, std::initializer_list<Subsystem*> deps) {
-		EXIT_IF(!s);
-
-		const char* name = s->Id();
-
-		EXIT_IF(FindByName(name) != nullptr);
-
-		auto nl = std::make_unique<SubsListStruct>();
-
-		nl->s         = s;
-		nl->name      = name;
-		nl->deps      = nullptr;
-		nl->next      = std::move(list);
-		nl->prev_init = nullptr;
-
-		auto* node = nl.get();
-		list       = std::move(nl);
-
-		for (auto* dep: deps) {
-			const char* str = dep->Id();
-
-			auto l      = std::make_unique<DepsListStruct>();
-			l->dep_name = str;
-			l->next     = std::move(node->deps);
-
-			node->deps = std::move(l);
+void Subsystems::Destroy() {
+	for (auto it = m_active.rbegin(); it != m_active.rend(); ++it) {
+		if (it->shutdown != nullptr) {
+			it->shutdown();
 		}
-
-		node->initialized = false;
 	}
+	m_active.clear();
+}
 
-	bool InitAll(bool print_msg) {
-		while (SubsListStruct* n = FindNextToInitialize()) {
-			n->s->Init(&parent);
-
-			if (n->s->m_p->failed) {
-				fail_msg  = n->s->m_p->fail_msg.c_str();
-				fail_name = n->name;
-				return false;
-			}
-
-			if (print_msg) {
-				printf("Initialized: %s\n", n->name);
-			}
-
-			n->initialized = true;
-
-			SubsListStruct* last = last_init;
-			last_init            = n;
-			n->prev_init         = last;
+void Subsystems::EmergencyShutdown() {
+	for (auto it = m_active.rbegin(); it != m_active.rend(); ++it) {
+		if (it->emergency_shutdown != nullptr) {
+			it->emergency_shutdown();
 		}
-
-		return true;
 	}
+	m_active.clear();
+}
 
-	void DestroyAll(bool print_msg) {
-		for (SubsListStruct* n = last_init; n != nullptr; n = n->prev_init) {
-			n->s->Destroy(&parent);
-			n->initialized = false;
-
-			if (print_msg) {
-				printf("Destroyed: %s\n", n->name);
-			}
-		}
-
-		last_init = nullptr;
+void Subsystems::EmergencyShutdownActive() {
+	if (g_active_subsystems != nullptr) {
+		g_active_subsystems->EmergencyShutdown();
 	}
-
-	void ShutdownAll() {
-		for (SubsListStruct* n = last_init; n != nullptr; n = n->prev_init) {
-			n->s->UnexpectedShutdown(&parent);
-			n->initialized = false;
-		}
-
-		last_init = nullptr;
-	}
-
-	struct DepsListStruct {
-		const char*                     dep_name;
-		std::unique_ptr<DepsListStruct> next;
-	};
-
-	struct SubsListStruct {
-		Subsystem*                      s;
-		const char*                     name;
-		std::unique_ptr<DepsListStruct> deps;
-		std::unique_ptr<SubsListStruct> next;
-		SubsListStruct*                 prev_init;
-		bool                            initialized;
-	};
-
-	[[nodiscard]] SubsListStruct* FindByName(const char* name) const {
-		for (SubsListStruct* n = list.get(); n != nullptr; n = n->next.get()) {
-			if (std::strcmp(n->name, name) == 0) {
-				return n;
-			}
-		}
-
-		return nullptr;
-	}
-
-	[[nodiscard]] SubsListStruct* FindNextToInitialize() const {
-		for (SubsListStruct* n = list.get(); n != nullptr; n = n->next.get()) {
-			if (n->initialized) {
-				continue;
-			}
-
-			DepsListStruct* d = n->deps.get();
-			for (; d != nullptr; d = d->next.get()) {
-				SubsListStruct* s = FindByName(d->dep_name);
-				if ((s == nullptr) || !s->initialized) {
-					break;
-				}
-			}
-
-			if (d == nullptr) {
-				return n;
-			}
-		}
-
-		return nullptr;
-	}
-
-	KYTY_CLASS_NO_COPY(SubsystemsListPrivate);
-
-	std::unique_ptr<SubsListStruct> list;
-	SubsListStruct*                 last_init = nullptr;
-	int                             m_argc    = 0;
-	char**                          m_argv    = nullptr;
-	const char*                     fail_msg  = nullptr;
-	const char*                     fail_name = nullptr;
-	SubsystemsList&                 parent;
-};
-
-SubsystemsList::SubsystemsList(): m_p(std::make_unique<SubsystemsListPrivate>(*this)) {}
-
-SubsystemsList::~SubsystemsList() = default;
-
-void SubsystemsList::Add(Subsystem* s, std::initializer_list<Subsystem*> deps) {
-	m_p->Add(s, deps);
-}
-
-bool SubsystemsList::InitAll(bool print_msg) {
-	return m_p->InitAll(print_msg);
-}
-
-void SubsystemsList::DestroyAll(bool print_msg) {
-	m_p->DestroyAll(print_msg);
-}
-
-int* SubsystemsList::GetArgc() {
-	return &m_p->m_argc;
-}
-
-char** SubsystemsList::GetArgv() {
-	return m_p->m_argv;
-}
-
-Subsystem::Subsystem(): m_p(std::make_unique<SubsystemPrivate>()) {}
-
-Subsystem::~Subsystem() = default;
-
-void Subsystem::Fail(const char* format, ...) {
-	va_list args {};
-	va_start(args, format);
-
-	va_list args_copy {};
-	va_copy(args_copy, args);
-	int len = std::vsnprintf(nullptr, 0, format, args_copy);
-	va_end(args_copy);
-
-	if (len > 0) {
-		std::string msg(static_cast<size_t>(len) + 1, '\0');
-		std::vsnprintf(msg.data(), msg.size(), format, args);
-		m_p->fail_msg = msg.c_str();
-		m_p->failed   = true;
-	}
-
-	va_end(args);
-}
-
-const char* SubsystemsList::GetFailName() const {
-	return m_p->fail_name;
-}
-
-const char* SubsystemsList::GetFailMsg() const {
-	return m_p->fail_msg;
-}
-
-void SubsystemsList::SetArgs(int argc, char* argv[]) {
-	m_p->SetArgs(argc, argv);
-}
-
-void SubsystemsList::ShutdownAll() {
-	m_p->ShutdownAll();
 }
 
 } // namespace Common

src/common/subsystems.h

@@ -2,90 +2,61 @@
 #define KYTY_COMMON_SUBSYSTEMS_H_
 
 #include "common/common.h"
-#include "common/singleton.h"
 
-#include <initializer_list>
-#include <memory>
+#include <cstdio>
+#include <vector>
 
 namespace Common {
 
-class Subsystem;
-class SubsystemsListPrivate;
-class SubsystemPrivate;
-
-class SubsystemsList {
+class Subsystems {
 public:
-	SubsystemsList();
-	virtual ~SubsystemsList();
-	void SetArgs(int argc, char* argv[]);
+	using Callback = void (*)();
 
-	// void Add(Subsystem* s, const char* name, ...);
-	void Add(Subsystem* s, std::initializer_list<Subsystem*> deps);
+	explicit Subsystems(bool print = false);
+	~Subsystems();
 
-	bool InitAll(bool print_msg = false);
-	void DestroyAll(bool print_msg = false);
+	template <typename Lifecycle>
+	void Initialize() {
+		Lifecycle::initialize();
+		m_active.push_back({ShutdownCallback<Lifecycle>(), EmergencyCallback<Lifecycle>()});
+		if (m_print) {
+			std::printf("Initialized: %s\n", Lifecycle::name);
+		}
+	}
 
-	int*   GetArgc();
-	char** GetArgv();
+	void Destroy();
+	void EmergencyShutdown();
 
-	[[nodiscard]] const char* GetFailName() const;
-	[[nodiscard]] const char* GetFailMsg() const;
+	static void EmergencyShutdownActive();
 
-	void ShutdownAll();
-
-	static SubsystemsList* Instance() { return Common::Singleton<SubsystemsList>::Instance(); }
-
-	KYTY_CLASS_NO_COPY(SubsystemsList);
+	KYTY_CLASS_NO_COPY(Subsystems);
 
 private:
-	std::unique_ptr<SubsystemsListPrivate> m_p;
+	template <typename Lifecycle>
+	static consteval Callback ShutdownCallback() {
+		if constexpr (requires { Lifecycle::shutdown; }) {
+			return Lifecycle::shutdown;
+		}
+		return nullptr;
+	}
+
+	template <typename Lifecycle>
+	static consteval Callback EmergencyCallback() {
+		if constexpr (requires { Lifecycle::emergency_shutdown; }) {
+			return Lifecycle::emergency_shutdown;
+		}
+		return nullptr;
+	}
+
+	struct Entry {
+		Callback shutdown;
+		Callback emergency_shutdown;
+	};
+
+	std::vector<Entry> m_active;
+	bool               m_print;
 };
 
-using SubsystemsListSingleton = Common::Singleton<SubsystemsList>;
-
-class Subsystem {
-public:
-	Subsystem();
-	virtual ~Subsystem();
-
-	virtual const char* Id()                                       = 0;
-	virtual void        Init(SubsystemsList* parent)               = 0;
-	virtual void        Destroy(SubsystemsList* parent)            = 0;
-	virtual void        UnexpectedShutdown(SubsystemsList* parent) = 0;
-
-	friend class SubsystemsListPrivate;
-
-	KYTY_CLASS_NO_COPY(Subsystem);
-
-protected:
-	void Fail(const char* format, ...) KYTY_FORMAT_PRINTF(2, 3);
-
-private:
-	std::unique_ptr<SubsystemPrivate> m_p;
-};
-
-#define KYTY_SUBSYSTEM_DEFINE(s)                                                                   \
-	class s##Subsystem: public Common::Subsystem {                                                 \
-	public:                                                                                        \
-		static Subsystem* Instance() { return Common::Singleton<s##Subsystem>::Instance(); }       \
-		const char*       Id() { return #s; }                                                      \
-		void              Init(Common::SubsystemsList* parent);                                    \
-		void              Destroy(Common::SubsystemsList* parent);                                 \
-		void              UnexpectedShutdown(Common::SubsystemsList* parent);                      \
-	};                                                                                             \
-	typedef Common::Singleton<s##Subsystem> s##SubsystemSingleton;
-
-#define KYTY_SUBSYSTEM_INIT(s)                                                                     \
-	void s##Subsystem::Init([[maybe_unused]] Common::SubsystemsList* parent)
-#define KYTY_SUBSYSTEM_DESTROY(s)                                                                  \
-	void s##Subsystem::Destroy([[maybe_unused]] Common::SubsystemsList* parent)
-#define KYTY_SUBSYSTEM_UNEXPECTED_SHUTDOWN(s)                                                      \
-	void s##Subsystem::UnexpectedShutdown([[maybe_unused]] Common::SubsystemsList* parent)
-
-// #define KYTY_SUBSYSTEM_ADD(list, s, ...) list.push_back(s##SubsystemSingleton::Instance(), #s,
-// __VA_ARGS__); #define KYTY_SUBSYSTEM_ADD2(list, ...) list.Add2(s##SubsystemSingleton::Instance(),
-// __VA_ARGS__);
-
 } // namespace Common
 
 #endif /* KYTY_COMMON_SUBSYSTEMS_H_ */

src/common/threads.cpp

@@ -252,15 +252,11 @@ static thread_id_t      g_main_thread;
 static int              g_main_thread_int;
 static std::atomic<int> g_thread_counter = 0;
 
-KYTY_SUBSYSTEM_INIT(Threads) {
+void InitializeThreads() {
 	g_main_thread     = std::this_thread::get_id();
 	g_main_thread_int = Thread::GetThreadIdUnique();
 }
 
-KYTY_SUBSYSTEM_UNEXPECTED_SHUTDOWN(Threads) {}
-
-KYTY_SUBSYSTEM_DESTROY(Threads) {}
-
 Thread::Thread(thread_func_t func, void* arg)
     : m_thread(std::make_unique<ThreadPrivate>(func, arg)) {
 	while (!m_thread->started) {

src/common/threads.h

@@ -2,14 +2,13 @@
 #define KYTY_COMMON_THREADS_H_
 
 #include "common/common.h"
-#include "common/subsystems.h"
 
 #include <memory>
 #include <string>
 
 namespace Common {
 
-KYTY_SUBSYSTEM_DEFINE(Threads);
+void InitializeThreads();
 
 using thread_func_t    = void (*)(void*);
 using wait_poll_func_t = void (*)();

src/emulator.cpp

@@ -2,7 +2,6 @@
 
 #include "common/abi.h"
 #include "common/assert.h"
-#include "common/commonSubsystem.h"
 #include "common/emulatorConfig.h"
 #include "common/file.h"
 #include "common/logging/log.h"
@@ -43,7 +42,7 @@ static void KytyClose() {
 
 	LOGF("done!\n");
 
-	Common::SubsystemsListSingleton::Instance()->ShutdownAll();
+	Common::Subsystems::EmergencyShutdownActive();
 }
 
 static void MountOrCreateDir(const std::filesystem::path& dir, const std::string& point) {
@@ -105,30 +104,13 @@ static void ClearDebugTextureFolder() {
 	}
 }
 
-static void Init(const Config::ConfigOptions& cfg, const std::filesystem::path& param_json) {
+static void Init(const Config::ConfigOptions& cfg, const std::filesystem::path& param_json,
+                 Common::Subsystems& subsystems) {
 	EXIT_IF(!Common::Thread::IsMainThread());
 
-	auto* slist = Common::SubsystemsList::Instance();
-
-	auto* audio       = Libs::Audio::AudioSubsystem::Instance();
-	auto* config      = Config::ConfigSubsystem::Instance();
-	auto* controller  = Libs::Controller::ControllerSubsystem::Instance();
-	auto* core        = Common::CommonSubsystem::Instance();
-	auto* file_system = Libs::LibKernel::FileSystem::FileSystemSubsystem::Instance();
-	auto* graphics    = Libs::Graphics::GraphicsSubsystem::Instance();
-	auto* log         = Log::LogSubsystem::Instance();
-	auto* memory      = Libs::LibKernel::Memory::MemorySubsystem::Instance();
-	auto* network     = Libs::Network::NetworkSubsystem::Instance();
-	auto* profiler    = Profiler::ProfilerSubsystem::Instance();
-	auto* pthread     = Libs::LibKernel::PthreadSubsystem::Instance();
-	auto* timer       = Loader::Timer::TimerSubsystem::Instance();
-
-	slist->Add(config, {core});
-	slist->InitAll(true);
-
+	subsystems.Initialize<Config::Lifecycle>();
 	Config::Load(cfg);
-	slist->Add(log, {core, config});
-	slist->InitAll(true);
+	subsystems.Initialize<Log::Lifecycle>();
 
 	if (Common::File::IsFileExisting(param_json)) {
 		Loader::SystemContentLoadParamSfo(param_json);
@@ -138,17 +120,16 @@ static void Init(const Config::ConfigOptions& cfg, const std::filesystem::path&
 		}
 	}
 
-	slist->Add(audio, {core, log, pthread, memory});
-	slist->Add(controller, {core, log, config});
-	slist->Add(file_system, {core, log, pthread});
-	slist->Add(graphics, {core, log, pthread, memory, config, profiler, controller});
-	slist->Add(memory, {core, log});
-	slist->Add(network, {core, log, pthread});
-	slist->Add(profiler, {core, config});
-	slist->Add(pthread, {core, log, timer});
-	slist->Add(timer, {core, log});
-
-	slist->InitAll(true);
+	// Initialization order is explicit; destruction is automatic and reversed.
+	subsystems.Initialize<Loader::Timer::Lifecycle>();
+	subsystems.Initialize<Libs::LibKernel::PthreadLifecycle>();
+	subsystems.Initialize<Profiler::Lifecycle>();
+	subsystems.Initialize<Libs::Network::Lifecycle>();
+	subsystems.Initialize<Libs::LibKernel::Memory::Lifecycle>();
+	subsystems.Initialize<Libs::LibKernel::FileSystem::Lifecycle>();
+	subsystems.Initialize<Libs::Controller::Lifecycle>();
+	subsystems.Initialize<Libs::Audio::Lifecycle>();
+	subsystems.Initialize<Libs::Graphics::Lifecycle>();
 }
 
 static void LoadElf(const std::filesystem::path& elf, bool dbg_print_reloc = false,
@@ -189,8 +170,9 @@ void Run(const RunOptions& options) {
 		EXIT("ELF is required\n");
 	}
 
-	const auto param_json = options.app0_dir / "sce_sys" / "param.json";
-	Init(options.config, param_json);
+	const auto         param_json = options.app0_dir / "sce_sys" / "param.json";
+	Common::Subsystems subsystems(true);
+	Init(options.config, param_json, subsystems);
 
 	ClearDebugTextureFolder();
 

src/graphics/guest_gpu/command_processor/commandProcessor.h

@@ -103,7 +103,7 @@ public:
 	                    uint32_t mode);
 	void DispatchIndirect(uint32_t data_offset, uint32_t mode);
 	void WaitFlipDone(uint32_t video_out_handle, uint32_t display_buffer_index);
-	void TriggerEvent(uint32_t event_type, uint32_t event_index);
+	void TriggerEvent(uint32_t event_type, uint32_t event_index, uint64_t event_address = 0);
 
 	void SetUserDataMarker(HW::UserSgprType type) { m_user_data_marker = type; }
 	[[nodiscard]] HW::UserSgprType GetUserDataMarker() const { return m_user_data_marker; }
@@ -112,6 +112,7 @@ public:
 	void SetCeComplete(bool complete) { m_ce_complete = complete; }
 	void WaitCe();
 	void WaitDeDiff(uint32_t diff);
+	void WaitForRewind(bool valid);
 	void IncrementDe();
 	void IncrementCe();
 
@@ -177,8 +178,9 @@ private:
 	uint32_t m_const_ram[0x3000] = {0};
 
 	FlipInfo m_flip;
-	uint64_t m_submit_id      = 0;
-	bool     m_predicate_skip = false;
+	uint64_t m_submit_id                   = 0;
+	uint64_t m_synthetic_occlusion_counter = 0;
+	bool     m_predicate_skip              = false;
 };
 
 } // namespace Libs::Graphics

src/graphics/guest_gpu/command_processor/pm4Dispatch.cpp

@@ -35,7 +35,6 @@ constexpr auto MakeContextDispatchTable() {
 	}
 	g_hw_ctx_func[Pm4::CB_TARGET_MASK]                 = HwCtxSetRenderTargetMask;
 	g_hw_ctx_func[Pm4::PA_SC_GENERIC_SCISSOR_TL]       = HwCtxSetGenericScissor;
-	g_hw_ctx_func[Pm4::CB_BLEND_RED]                   = HwCtxSetBlendColor;
 	g_hw_ctx_func[Pm4::CB_DCC_CONTROL]                 = HwCtxSetCbDccControl;
 	g_hw_ctx_func[Pm4::DB_STENCIL_CONTROL]             = HwCtxSetStencilControl;
 	g_hw_ctx_func[Pm4::DB_STENCILREFMASK]              = HwCtxSetStencilMask;
@@ -82,7 +81,6 @@ constexpr auto MakeContextDispatchTable() {
 	g_hw_ctx_func[Pm4::VGT_PRIMITIVEID_RESET]                 = HwCtxSetPrimitiveIdReset;
 	g_hw_ctx_func[Pm4::PA_CL_OBJPRIM_ID_CNTL]                 = HwCtxSetObjprimIdControl;
 	g_hw_ctx_func[Pm4::VGT_SHADER_STAGES_EN]                  = HwCtxSetShaderStages;
-	g_hw_ctx_func[Pm4::PA_CL_GB_VERT_CLIP_ADJ]                = HwCtxSetGuardBands;
 
 	for (uint32_t slot = 0; slot < 8; slot++) {
 		g_hw_ctx_func[Pm4::CB_COLOR0_BASE + slot * 15] = HwCtxSetRenderTarget;
@@ -217,6 +215,7 @@ constexpr auto MakeOpcodeDispatchTable() {
 	g_cp_op_func[Pm4::IT_COND_EXEC]                 = CpOpCondExec;
 	g_cp_op_func[Pm4::IT_SET_PREDICATION]           = CpOpSetPredication;
 	g_cp_op_func[Pm4::IT_WRITE_DATA]                = CpOpWriteData;
+	g_cp_op_func[Pm4::IT_WAIT_REG_MEM]              = CpOpWaitRegMem32;
 	g_cp_op_func[Pm4::IT_INDIRECT_BUFFER]           = CpOpIndirectBuffer;
 	g_cp_op_func[Pm4::IT_COPY_DATA]                 = CpOpCopyData;
 	g_cp_op_func[Pm4::IT_EVENT_WRITE]               = CpOpEventWrite;
@@ -224,6 +223,7 @@ constexpr auto MakeOpcodeDispatchTable() {
 	g_cp_op_func[Pm4::IT_EVENT_WRITE_EOS]           = CpOpEventWriteEos;
 	g_cp_op_func[Pm4::IT_DMA_DATA]                  = CpOpDmaData;
 	g_cp_op_func[Pm4::IT_ACQUIRE_MEM]               = CpOpAcquireMem;
+	g_cp_op_func[Pm4::IT_REWIND]                    = CpOpRewind;
 	g_cp_op_func[Pm4::IT_SET_CONTEXT_REG]           = CpOpSetContextReg;
 	g_cp_op_func[Pm4::IT_SET_SH_REG]                = CpOpSetShaderReg;
 	g_cp_op_func[Pm4::IT_SET_UCONFIG_REG]           = CpOpSetUconfigReg;
@@ -238,6 +238,7 @@ constexpr auto MakeOpcodeDispatchTable() {
 	g_cp_op_func[Pm4::IT_WAIT_ON_CE_COUNTER]        = CpOpWaitOnCeCounter;
 	g_cp_op_func[Pm4::IT_WAIT_ON_DE_COUNTER_DIFF]   = CpOpWaitOnDeCounterDiff;
 	g_cp_op_func[Pm4::IT_GET_LOD_STATS]             = CpOpGetLodStats;
+	g_cp_op_func[Pm4::IT_WAIT_REG_MEM_64]           = CpOpWaitRegMem64;
 
 	return g_cp_op_func;
 }
@@ -246,12 +247,10 @@ constexpr auto MakeCustomOpcodeDispatchTable() {
 	std::array<cp_op_parser_func_t, Pm4::R_NUM> g_cp_op_custom_func {};
 
 	g_cp_op_custom_func[Pm4::R_DISPATCH_RESET] = CpOpDispatchReset;
-	g_cp_op_custom_func[Pm4::R_WAIT_MEM_32]    = CpOpWaitRegMem32;
 	g_cp_op_custom_func[Pm4::R_WAIT_FLIP_DONE] = CpOpWaitFlipDone;
 	g_cp_op_custom_func[Pm4::R_PUSH_MARKER]    = CpOpPushMarker;
 	g_cp_op_custom_func[Pm4::R_POP_MARKER]     = CpOpPopMarker;
 	g_cp_op_custom_func[Pm4::R_ACQUIRE_MEM]    = CpOpAcquireMem;
-	g_cp_op_custom_func[Pm4::R_WAIT_MEM_64]    = CpOpWaitRegMem64;
 	g_cp_op_custom_func[Pm4::R_FLIP]           = CpOpFlip;
 	g_cp_op_custom_func[Pm4::R_RELEASE_MEM]    = CpOpReleaseMem;
 	g_cp_op_custom_func[Pm4::R_CONTEXT_STATE]  = CpOpContextState;

src/graphics/guest_gpu/command_processor/pm4Dispatch.h

@@ -56,7 +56,6 @@ uint32_t HwCtxSetWindowScissor(CommandProcessor&, uint32_t, uint32_t, const uint
 uint32_t HwCtxSetClipRect(CommandProcessor&, uint32_t, uint32_t, const uint32_t*, uint32_t);
 uint32_t HwCtxSetRenderTargetMask(CommandProcessor&, uint32_t, uint32_t, const uint32_t*, uint32_t);
 uint32_t HwCtxSetGenericScissor(CommandProcessor&, uint32_t, uint32_t, const uint32_t*, uint32_t);
-uint32_t HwCtxSetBlendColor(CommandProcessor&, uint32_t, uint32_t, const uint32_t*, uint32_t);
 uint32_t HwCtxSetCbDccControl(CommandProcessor&, uint32_t, uint32_t, const uint32_t*, uint32_t);
 uint32_t HwCtxSetStencilControl(CommandProcessor&, uint32_t, uint32_t, const uint32_t*, uint32_t);
 uint32_t HwCtxSetStencilMask(CommandProcessor&, uint32_t, uint32_t, const uint32_t*, uint32_t);
@@ -90,7 +89,6 @@ uint32_t HwCtxSetDrawPayloadControl(CommandProcessor&, uint32_t, uint32_t, const
 uint32_t HwCtxSetPrimitiveIdReset(CommandProcessor&, uint32_t, uint32_t, const uint32_t*, uint32_t);
 uint32_t HwCtxSetObjprimIdControl(CommandProcessor&, uint32_t, uint32_t, const uint32_t*, uint32_t);
 uint32_t HwCtxSetShaderStages(CommandProcessor&, uint32_t, uint32_t, const uint32_t*, uint32_t);
-uint32_t HwCtxSetGuardBands(CommandProcessor&, uint32_t, uint32_t, const uint32_t*, uint32_t);
 uint32_t HwCtxSetRenderTarget(CommandProcessor&, uint32_t, uint32_t, const uint32_t*, uint32_t);
 uint32_t HwCtxSetColorInfo(CommandProcessor&, uint32_t, uint32_t, const uint32_t*, uint32_t);
 uint32_t HwCtxSetBlendControl(CommandProcessor&, uint32_t, uint32_t, const uint32_t*, uint32_t);
@@ -133,6 +131,7 @@ uint32_t CpOpDrawIndirectMulti(CommandProcessor&, uint32_t, const uint32_t*, uin
 uint32_t CpOpDrawIndex(CommandProcessor&, uint32_t, const uint32_t*, uint32_t, uint32_t);
 uint32_t CpOpDrawIndexOffset(CommandProcessor&, uint32_t, const uint32_t*, uint32_t, uint32_t);
 uint32_t CpOpPfpSyncMe(CommandProcessor&, uint32_t, const uint32_t*, uint32_t, uint32_t);
+uint32_t CpOpRewind(CommandProcessor&, uint32_t, const uint32_t*, uint32_t, uint32_t);
 uint32_t CpOpIndexType(CommandProcessor&, uint32_t, const uint32_t*, uint32_t, uint32_t);
 uint32_t CpOpNumInstances(CommandProcessor&, uint32_t, const uint32_t*, uint32_t, uint32_t);
 uint32_t CpOpDrawIndexAuto(CommandProcessor&, uint32_t, const uint32_t*, uint32_t, uint32_t);

src/graphics/guest_gpu/command_processor/pm4Handlers.cpp

@@ -71,19 +71,6 @@ constexpr uint32_t NormalizeRegisterOffset(uint32_t raw_offset) {
 	return raw_offset & ~RegisterSelectorMask;
 }
 
-// Indirect Cx descriptors retain their selector. Selector 1 offsets 0..31 address the
-// SPI_PS_INPUT_CNTL register bank; ordinary context-register offsets remain unchanged.
-constexpr uint32_t DecodeIndirectCxRegisterOffset(uint32_t raw_offset) {
-	const auto offset = NormalizeRegisterOffset(raw_offset);
-	return (raw_offset & RegisterSelectorMask) == Pm4::CX_PS_SHADER_USAGE_BASE && offset < 32u
-	           ? Pm4::SPI_PS_INPUT_CNTL_0 + offset
-	           : offset;
-}
-
-static_assert(DecodeIndirectCxRegisterOffset(Pm4::CX_PS_SHADER_USAGE_BASE + 2u) ==
-              Pm4::SPI_PS_INPUT_CNTL_0 + 2u);
-static_assert(DecodeIndirectCxRegisterOffset(Pm4::DB_Z_INFO) == Pm4::DB_Z_INFO);
-
 bool ReleaseMemGcrNeedsBarrier(uint32_t eop_event_type, uint32_t gcr_cntl) {
 	return eop_event_type != 0x28u ||
 	       (gcr_cntl & (GcrGl2MetadataInvalidate | GcrGl0VectorInvalidate | GcrGl1Invalidate |
@@ -103,6 +90,128 @@ void LogUnknownReleaseMemGcr(uint32_t gcr_cntl) {
 
 } // namespace
 
+static HW::BlendControl DecodeBlendControl(uint32_t value) {
+	HW::BlendControl r {};
+
+	r.color_srcblend       = KYTY_PM4_GET(value, CB_BLEND0_CONTROL, COLOR_SRCBLEND);
+	r.color_comb_fcn       = KYTY_PM4_GET(value, CB_BLEND0_CONTROL, COLOR_COMB_FCN);
+	r.color_destblend      = KYTY_PM4_GET(value, CB_BLEND0_CONTROL, COLOR_DESTBLEND);
+	r.alpha_srcblend       = KYTY_PM4_GET(value, CB_BLEND0_CONTROL, ALPHA_SRCBLEND);
+	r.alpha_comb_fcn       = KYTY_PM4_GET(value, CB_BLEND0_CONTROL, ALPHA_COMB_FCN);
+	r.alpha_destblend      = KYTY_PM4_GET(value, CB_BLEND0_CONTROL, ALPHA_DESTBLEND);
+	r.separate_alpha_blend = KYTY_PM4_GET(value, CB_BLEND0_CONTROL, SEPARATE_ALPHA_BLEND) != 0;
+	r.enable               = KYTY_PM4_GET(value, CB_BLEND0_CONTROL, ENABLE) != 0;
+
+	return r;
+}
+
+static HW::ClipControl DecodeClipControl(uint32_t value) {
+	HW::ClipControl r {};
+
+	r.user_clip_planes          = KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, UCP_ENA);
+	r.user_clip_plane_mode      = KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, PS_UCP_MODE);
+	r.dx_clip_space             = KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, DX_CLIP_SPACE_DEF) != 0;
+	r.vertex_kill_any           = KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, VTX_KILL_OR) != 0;
+	r.min_z_clip_disable        = KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, ZCLIP_NEAR_DISABLE) != 0;
+	r.max_z_clip_disable        = KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, ZCLIP_FAR_DISABLE) != 0;
+	r.user_clip_plane_negate_y  = KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, PS_UCP_Y_SCALE_NEG) != 0;
+	r.clip_disable              = KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, CLIP_DISABLE) != 0;
+	r.user_clip_plane_cull_only = KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, UCP_CULL_ONLY_ENA) != 0;
+	r.cull_on_clipping_error_disable =
+	    KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, DIS_CLIP_ERR_DETECT) != 0;
+	r.linear_attribute_clip_enable =
+	    KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, DX_LINEAR_ATTR_CLIP_ENA) != 0;
+	r.force_viewport_index_from_vs_enable =
+	    KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, VTE_VPORT_PROVOKE_DISABLE) != 0;
+
+	return r;
+}
+
+static HW::ColorControl DecodeColorControl(uint32_t value) {
+	HW::ColorControl r {};
+	r.mode = KYTY_PM4_GET(value, CB_COLOR_CONTROL, MODE);
+	r.op   = KYTY_PM4_GET(value, CB_COLOR_CONTROL, ROP3);
+	return r;
+}
+
+static HW::DepthControl DecodeDepthControl(uint32_t value) {
+	HW::DepthControl r {};
+
+	r.stencil_enable      = KYTY_PM4_GET(value, DB_DEPTH_CONTROL, STENCIL_ENABLE) != 0;
+	r.z_enable            = KYTY_PM4_GET(value, DB_DEPTH_CONTROL, Z_ENABLE) != 0;
+	r.z_write_enable      = KYTY_PM4_GET(value, DB_DEPTH_CONTROL, Z_WRITE_ENABLE) != 0;
+	r.depth_bounds_enable = KYTY_PM4_GET(value, DB_DEPTH_CONTROL, DEPTH_BOUNDS_ENABLE) != 0;
+	r.zfunc               = KYTY_PM4_GET(value, DB_DEPTH_CONTROL, ZFUNC);
+	r.backface_enable     = KYTY_PM4_GET(value, DB_DEPTH_CONTROL, BACKFACE_ENABLE) != 0;
+	r.stencilfunc         = KYTY_PM4_GET(value, DB_DEPTH_CONTROL, STENCILFUNC);
+	r.stencilfunc_bf      = KYTY_PM4_GET(value, DB_DEPTH_CONTROL, STENCILFUNC_BF);
+	r.color_writes_on_depth_fail_enable =
+	    KYTY_PM4_GET(value, DB_DEPTH_CONTROL, ENABLE_COLOR_WRITES_ON_DEPTH_FAIL) != 0;
+	r.color_writes_on_depth_pass_disable =
+	    KYTY_PM4_GET(value, DB_DEPTH_CONTROL, DISABLE_COLOR_WRITES_ON_DEPTH_PASS) != 0;
+
+	return r;
+}
+
+static HW::ModeControl DecodeModeControl(uint32_t value) {
+	HW::ModeControl r {};
+
+	r.cull_front           = KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, CULL_FRONT) != 0;
+	r.cull_back            = KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, CULL_BACK) != 0;
+	r.face                 = KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, FACE) != 0;
+	r.poly_mode            = KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, POLY_MODE);
+	r.polymode_front_ptype = KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, POLYMODE_FRONT_PTYPE);
+	r.polymode_back_ptype  = KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, POLYMODE_BACK_PTYPE);
+	r.poly_offset_front_enable =
+	    KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, POLY_OFFSET_FRONT_ENABLE) != 0;
+	r.poly_offset_back_enable =
+	    KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, POLY_OFFSET_BACK_ENABLE) != 0;
+	r.vtx_window_offset_enable =
+	    KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, VTX_WINDOW_OFFSET_ENABLE) != 0;
+	r.provoking_vtx_last = KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, PROVOKING_VTX_LAST) != 0;
+	r.persp_corr_dis     = KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, PERSP_CORR_DIS) != 0;
+
+	return r;
+}
+
+static HW::RenderControl DecodeRenderControl(uint32_t value) {
+	HW::RenderControl r {};
+
+	r.depth_clear_enable   = KYTY_PM4_GET(value, DB_RENDER_CONTROL, DEPTH_CLEAR_ENABLE) != 0;
+	r.stencil_clear_enable = KYTY_PM4_GET(value, DB_RENDER_CONTROL, STENCIL_CLEAR_ENABLE) != 0;
+	r.resummarize_enable   = KYTY_PM4_GET(value, DB_RENDER_CONTROL, RESUMMARIZE_ENABLE) != 0;
+	r.stencil_compress_disable =
+	    KYTY_PM4_GET(value, DB_RENDER_CONTROL, STENCIL_COMPRESS_DISABLE) != 0;
+	r.depth_compress_disable = KYTY_PM4_GET(value, DB_RENDER_CONTROL, DEPTH_COMPRESS_DISABLE) != 0;
+	r.copy_centroid          = KYTY_PM4_GET(value, DB_RENDER_CONTROL, COPY_CENTROID) != 0;
+	r.copy_sample            = KYTY_PM4_GET(value, DB_RENDER_CONTROL, COPY_SAMPLE);
+
+	return r;
+}
+
+static HW::ScanModeControl DecodeScanModeControl(uint32_t value) {
+	HW::ScanModeControl r {};
+
+	r.msaa_enable          = KYTY_PM4_GET(value, PA_SC_MODE_CNTL_0, MSAA_ENABLE) != 0;
+	r.vport_scissor_enable = KYTY_PM4_GET(value, PA_SC_MODE_CNTL_0, VPORT_SCISSOR_ENABLE) != 0;
+	r.line_stipple_enable  = KYTY_PM4_GET(value, PA_SC_MODE_CNTL_0, LINE_STIPPLE_ENABLE) != 0;
+
+	return r;
+}
+
+static HW::StencilControl DecodeStencilControl(uint32_t value) {
+	HW::StencilControl r {};
+
+	r.stencil_fail     = KYTY_PM4_GET(value, DB_STENCIL_CONTROL, STENCILFAIL);
+	r.stencil_zpass    = KYTY_PM4_GET(value, DB_STENCIL_CONTROL, STENCILZPASS);
+	r.stencil_zfail    = KYTY_PM4_GET(value, DB_STENCIL_CONTROL, STENCILZFAIL);
+	r.stencil_fail_bf  = KYTY_PM4_GET(value, DB_STENCIL_CONTROL, STENCILFAIL_BF);
+	r.stencil_zpass_bf = KYTY_PM4_GET(value, DB_STENCIL_CONTROL, STENCILZPASS_BF);
+	r.stencil_zfail_bf = KYTY_PM4_GET(value, DB_STENCIL_CONTROL, STENCILZFAIL_BF);
+
+	return r;
+}
+
 static HW::AaConfig ParseAaConfig(uint32_t value) {
 	HW::AaConfig r;
 
@@ -408,47 +517,12 @@ KYTY_HW_CTX_PARSER(HwCtxSetAaMask) {
 	return num_values;
 }
 
-KYTY_HW_CTX_PARSER(HwCtxSetBlendColor) {
-	EXIT_NOT_IMPLEMENTED(cmd_id != 0xc0046900);
-	EXIT_NOT_IMPLEMENTED(cmd_offset != Pm4::CB_BLEND_RED);
-
-	HW::BlendColor r;
-
-	r.red   = *reinterpret_cast<const float*>(&buffer[0]);
-	r.green = *reinterpret_cast<const float*>(&buffer[1]);
-	r.blue  = *reinterpret_cast<const float*>(&buffer[2]);
-	r.alpha = *reinterpret_cast<const float*>(&buffer[3]);
-
-	cp.GetCtx().SetBlendColor(r);
-
-	return 4;
-}
-
 KYTY_HW_CTX_PARSER(HwCtxSetBlendControl) {
 	EXIT_NOT_IMPLEMENTED(cmd_id != 0xC0016900);
 
 	uint32_t param = (cmd_offset - Pm4::CB_BLEND0_CONTROL) / 1;
 
-	HW::BlendControl r;
-
-	r.color_srcblend       = (buffer[0] >> Pm4::CB_BLEND0_CONTROL_COLOR_SRCBLEND_SHIFT) &
-	                         Pm4::CB_BLEND0_CONTROL_COLOR_SRCBLEND_MASK;
-	r.color_comb_fcn       = (buffer[0] >> Pm4::CB_BLEND0_CONTROL_COLOR_COMB_FCN_SHIFT) &
-	                         Pm4::CB_BLEND0_CONTROL_COLOR_COMB_FCN_MASK;
-	r.color_destblend      = (buffer[0] >> Pm4::CB_BLEND0_CONTROL_COLOR_DESTBLEND_SHIFT) &
-	                         Pm4::CB_BLEND0_CONTROL_COLOR_DESTBLEND_MASK;
-	r.alpha_srcblend       = (buffer[0] >> Pm4::CB_BLEND0_CONTROL_ALPHA_SRCBLEND_SHIFT) &
-	                         Pm4::CB_BLEND0_CONTROL_ALPHA_SRCBLEND_MASK;
-	r.alpha_comb_fcn       = (buffer[0] >> Pm4::CB_BLEND0_CONTROL_ALPHA_COMB_FCN_SHIFT) &
-	                         Pm4::CB_BLEND0_CONTROL_ALPHA_COMB_FCN_MASK;
-	r.alpha_destblend      = (buffer[0] >> Pm4::CB_BLEND0_CONTROL_ALPHA_DESTBLEND_SHIFT) &
-	                         Pm4::CB_BLEND0_CONTROL_ALPHA_DESTBLEND_MASK;
-	r.separate_alpha_blend = ((buffer[0] >> Pm4::CB_BLEND0_CONTROL_SEPARATE_ALPHA_BLEND_SHIFT) &
-	                          Pm4::CB_BLEND0_CONTROL_SEPARATE_ALPHA_BLEND_MASK) != 0;
-	r.enable               = ((buffer[0] >> Pm4::CB_BLEND0_CONTROL_ENABLE_SHIFT) &
-	                          Pm4::CB_BLEND0_CONTROL_ENABLE_MASK) != 0;
-
-	cp.GetCtx().SetBlendControl(param, r);
+	cp.GetCtx().SetBlendControl(param, DecodeBlendControl(buffer[0]));
 
 	return 1;
 }
@@ -457,38 +531,7 @@ KYTY_HW_CTX_PARSER(HwCtxSetClipControl) {
 	EXIT_NOT_IMPLEMENTED(cmd_id != 0xC0016900);
 	EXIT_NOT_IMPLEMENTED(cmd_offset != Pm4::PA_CL_CLIP_CNTL);
 
-	HW::ClipControl r;
-
-	//	r.user_clip_planes                    = buffer[0] & 0x3fu;
-	//	r.user_clip_plane_mode                = (buffer[0] >> 14u) & 0x3u;
-	//	r.clip_space                          = (buffer[0] >> 19u) & 0x1u;
-	//	r.vertex_kill_mode                    = (buffer[0] >> 21u) & 0x1u;
-	//	r.min_z_clip_enable                   = (buffer[0] >> 26u) & 0x1u;
-	//	r.max_z_clip_enable                   = (buffer[0] >> 27u) & 0x1u;
-	//	r.user_clip_plane_negate_y            = (buffer[0] & 0x00002000u) != 0;
-	//	r.clip_enable                         = (buffer[0] & 0x00010000u) != 0;
-	//	r.user_clip_plane_cull_only           = (buffer[0] & 0x00020000u) != 0;
-	//	r.cull_on_clipping_error_disable      = (buffer[0] & 0x00100000u) != 0;
-	//	r.linear_attribute_clip_enable        = (buffer[0] & 0x01000000u) != 0;
-	//	r.force_viewport_index_from_vs_enable = (buffer[0] & 0x02000000u) != 0;
-
-	r.user_clip_planes          = KYTY_PM4_GET(buffer[0], PA_CL_CLIP_CNTL, UCP_ENA);
-	r.user_clip_plane_mode      = KYTY_PM4_GET(buffer[0], PA_CL_CLIP_CNTL, PS_UCP_MODE);
-	r.dx_clip_space             = KYTY_PM4_GET(buffer[0], PA_CL_CLIP_CNTL, DX_CLIP_SPACE_DEF) != 0;
-	r.vertex_kill_any           = KYTY_PM4_GET(buffer[0], PA_CL_CLIP_CNTL, VTX_KILL_OR) != 0;
-	r.min_z_clip_disable        = KYTY_PM4_GET(buffer[0], PA_CL_CLIP_CNTL, ZCLIP_NEAR_DISABLE) != 0;
-	r.max_z_clip_disable        = KYTY_PM4_GET(buffer[0], PA_CL_CLIP_CNTL, ZCLIP_FAR_DISABLE) != 0;
-	r.user_clip_plane_negate_y  = KYTY_PM4_GET(buffer[0], PA_CL_CLIP_CNTL, PS_UCP_Y_SCALE_NEG) != 0;
-	r.clip_disable              = KYTY_PM4_GET(buffer[0], PA_CL_CLIP_CNTL, CLIP_DISABLE) != 0;
-	r.user_clip_plane_cull_only = KYTY_PM4_GET(buffer[0], PA_CL_CLIP_CNTL, UCP_CULL_ONLY_ENA) != 0;
-	r.cull_on_clipping_error_disable =
-	    KYTY_PM4_GET(buffer[0], PA_CL_CLIP_CNTL, DIS_CLIP_ERR_DETECT) != 0;
-	r.linear_attribute_clip_enable =
-	    KYTY_PM4_GET(buffer[0], PA_CL_CLIP_CNTL, DX_LINEAR_ATTR_CLIP_ENA) != 0;
-	r.force_viewport_index_from_vs_enable =
-	    KYTY_PM4_GET(buffer[0], PA_CL_CLIP_CNTL, VTE_VPORT_PROVOKE_DISABLE) != 0;
-
-	cp.GetCtx().SetClipControl(r);
+	cp.GetCtx().SetClipControl(DecodeClipControl(buffer[0]));
 
 	return 1;
 }
@@ -497,12 +540,7 @@ KYTY_HW_CTX_PARSER(HwCtxSetColorControl) {
 	EXIT_NOT_IMPLEMENTED(cmd_id != 0xc0016900);
 	EXIT_NOT_IMPLEMENTED(cmd_offset != Pm4::CB_COLOR_CONTROL);
 
-	HW::ColorControl r;
-
-	r.mode = KYTY_PM4_GET(buffer[0], CB_COLOR_CONTROL, MODE);
-	r.op   = KYTY_PM4_GET(buffer[0], CB_COLOR_CONTROL, ROP3);
-
-	cp.GetCtx().SetColorControl(r);
+	cp.GetCtx().SetColorControl(DecodeColorControl(buffer[0]));
 
 	return 1;
 }
@@ -526,9 +564,12 @@ KYTY_HW_CTX_PARSER(HwCtxSetColorInfo) {
 	//	r.format                         = (buffer[4] >> 2u) & 0x1fu;
 	//	r.channel_type                   = (buffer[4] >> 8u) & 0x7u;
 	//	r.channel_order                  = (buffer[4] >> 11u) & 0x3u;
-	r.format                   = KYTY_PM4_GET(buffer[0], CB_COLOR0_INFO, FORMAT);
-	r.channel_type             = KYTY_PM4_GET(buffer[0], CB_COLOR0_INFO, NUMBER_TYPE);
-	r.channel_order            = KYTY_PM4_GET(buffer[0], CB_COLOR0_INFO, COMP_SWAP);
+	r.format =
+	    static_cast<Prospero::ChannelLayout>(KYTY_PM4_GET(buffer[0], CB_COLOR0_INFO, FORMAT));
+	r.channel_type =
+	    static_cast<Prospero::ChannelType>(KYTY_PM4_GET(buffer[0], CB_COLOR0_INFO, NUMBER_TYPE));
+	r.channel_order =
+	    static_cast<Prospero::ChannelOrder>(KYTY_PM4_GET(buffer[0], CB_COLOR0_INFO, COMP_SWAP));
 	r.cmask_fast_clear_enable  = KYTY_PM4_GET(buffer[0], CB_COLOR0_INFO, FAST_CLEAR) != 0;
 	r.fmask_compression_enable = KYTY_PM4_GET(buffer[0], CB_COLOR0_INFO, COMPRESSION) != 0;
 	r.blend_clamp              = KYTY_PM4_GET(buffer[0], CB_COLOR0_INFO, BLEND_CLAMP) != 0;
@@ -560,22 +601,7 @@ KYTY_HW_CTX_PARSER(HwCtxSetDepthControl) {
 	EXIT_NOT_IMPLEMENTED(cmd_id != 0xC0016900);
 	EXIT_NOT_IMPLEMENTED(cmd_offset != Pm4::DB_DEPTH_CONTROL);
 
-	HW::DepthControl r;
-
-	r.stencil_enable      = KYTY_PM4_GET(buffer[0], DB_DEPTH_CONTROL, STENCIL_ENABLE) != 0;
-	r.z_enable            = KYTY_PM4_GET(buffer[0], DB_DEPTH_CONTROL, Z_ENABLE) != 0;
-	r.z_write_enable      = KYTY_PM4_GET(buffer[0], DB_DEPTH_CONTROL, Z_WRITE_ENABLE) != 0;
-	r.depth_bounds_enable = KYTY_PM4_GET(buffer[0], DB_DEPTH_CONTROL, DEPTH_BOUNDS_ENABLE) != 0;
-	r.zfunc               = KYTY_PM4_GET(buffer[0], DB_DEPTH_CONTROL, ZFUNC);
-	r.backface_enable     = KYTY_PM4_GET(buffer[0], DB_DEPTH_CONTROL, BACKFACE_ENABLE) != 0;
-	r.stencilfunc         = KYTY_PM4_GET(buffer[0], DB_DEPTH_CONTROL, STENCILFUNC);
-	r.stencilfunc_bf      = KYTY_PM4_GET(buffer[0], DB_DEPTH_CONTROL, STENCILFUNC_BF);
-	r.color_writes_on_depth_fail_enable =
-	    KYTY_PM4_GET(buffer[0], DB_DEPTH_CONTROL, ENABLE_COLOR_WRITES_ON_DEPTH_FAIL) != 0;
-	r.color_writes_on_depth_pass_disable =
-	    KYTY_PM4_GET(buffer[0], DB_DEPTH_CONTROL, DISABLE_COLOR_WRITES_ON_DEPTH_PASS) != 0;
-
-	cp.GetCtx().SetDepthControl(r);
+	cp.GetCtx().SetDepthControl(DecodeDepthControl(buffer[0]));
 
 	return 1;
 }
@@ -804,20 +830,6 @@ KYTY_HW_CTX_PARSER(HwCtxSetClipRect) {
 	return value_count;
 }
 
-KYTY_HW_CTX_PARSER(HwCtxSetGuardBands) {
-	EXIT_NOT_IMPLEMENTED(cmd_id != 0xC0046900);
-	EXIT_NOT_IMPLEMENTED(cmd_offset != Pm4::PA_CL_GB_VERT_CLIP_ADJ);
-
-	auto vert_clip    = *reinterpret_cast<const float*>(&buffer[0]); // PA_CL_GB_VERT_CLIP_ADJ
-	auto vert_discard = *reinterpret_cast<const float*>(&buffer[1]); // PA_CL_GB_VERT_DISC_ADJ
-	auto horz_clip    = *reinterpret_cast<const float*>(&buffer[2]); // PA_CL_GB_HORZ_CLIP_ADJ
-	auto horz_discard = *reinterpret_cast<const float*>(&buffer[3]); // PA_CL_GB_HORZ_DISC_ADJ
-
-	cp.GetCtx().SetGuardBands(horz_clip, vert_clip, horz_discard, vert_discard);
-
-	return 4;
-}
-
 KYTY_HW_CTX_PARSER(HwCtxSetHardwareScreenOffset) {
 	EXIT_NOT_IMPLEMENTED(cmd_id != 0xC0016900);
 	EXIT_NOT_IMPLEMENTED(cmd_offset != Pm4::PA_SU_HARDWARE_SCREEN_OFFSET);
@@ -879,24 +891,7 @@ KYTY_HW_CTX_PARSER(HwCtxSetModeControl) {
 	EXIT_NOT_IMPLEMENTED(cmd_id != 0xC0016900);
 	EXIT_NOT_IMPLEMENTED(cmd_offset != Pm4::PA_SU_SC_MODE_CNTL);
 
-	HW::ModeControl r;
-
-	r.cull_front           = KYTY_PM4_GET(buffer[0], PA_SU_SC_MODE_CNTL, CULL_FRONT) != 0;
-	r.cull_back            = KYTY_PM4_GET(buffer[0], PA_SU_SC_MODE_CNTL, CULL_BACK) != 0;
-	r.face                 = KYTY_PM4_GET(buffer[0], PA_SU_SC_MODE_CNTL, FACE) != 0;
-	r.poly_mode            = KYTY_PM4_GET(buffer[0], PA_SU_SC_MODE_CNTL, POLY_MODE);
-	r.polymode_front_ptype = KYTY_PM4_GET(buffer[0], PA_SU_SC_MODE_CNTL, POLYMODE_FRONT_PTYPE);
-	r.polymode_back_ptype  = KYTY_PM4_GET(buffer[0], PA_SU_SC_MODE_CNTL, POLYMODE_BACK_PTYPE);
-	r.poly_offset_front_enable =
-	    KYTY_PM4_GET(buffer[0], PA_SU_SC_MODE_CNTL, POLY_OFFSET_FRONT_ENABLE) != 0;
-	r.poly_offset_back_enable =
-	    KYTY_PM4_GET(buffer[0], PA_SU_SC_MODE_CNTL, POLY_OFFSET_BACK_ENABLE) != 0;
-	r.vtx_window_offset_enable =
-	    KYTY_PM4_GET(buffer[0], PA_SU_SC_MODE_CNTL, VTX_WINDOW_OFFSET_ENABLE) != 0;
-	r.provoking_vtx_last = KYTY_PM4_GET(buffer[0], PA_SU_SC_MODE_CNTL, PROVOKING_VTX_LAST) != 0;
-	r.persp_corr_dis     = KYTY_PM4_GET(buffer[0], PA_SU_SC_MODE_CNTL, PERSP_CORR_DIS) != 0;
-
-	cp.GetCtx().SetModeControl(r);
+	cp.GetCtx().SetModeControl(DecodeModeControl(buffer[0]));
 
 	return 1;
 }
@@ -962,19 +957,7 @@ KYTY_HW_CTX_PARSER(HwCtxSetRenderControl) {
 	EXIT_NOT_IMPLEMENTED(cmd_id != 0xC0016900);
 	EXIT_NOT_IMPLEMENTED(cmd_offset != Pm4::DB_RENDER_CONTROL);
 
-	HW::RenderControl r;
-
-	r.depth_clear_enable   = KYTY_PM4_GET(buffer[0], DB_RENDER_CONTROL, DEPTH_CLEAR_ENABLE) != 0;
-	r.stencil_clear_enable = KYTY_PM4_GET(buffer[0], DB_RENDER_CONTROL, STENCIL_CLEAR_ENABLE) != 0;
-	r.resummarize_enable   = KYTY_PM4_GET(buffer[0], DB_RENDER_CONTROL, RESUMMARIZE_ENABLE) != 0;
-	r.stencil_compress_disable =
-	    KYTY_PM4_GET(buffer[0], DB_RENDER_CONTROL, STENCIL_COMPRESS_DISABLE) != 0;
-	r.depth_compress_disable =
-	    KYTY_PM4_GET(buffer[0], DB_RENDER_CONTROL, DEPTH_COMPRESS_DISABLE) != 0;
-	r.copy_centroid = KYTY_PM4_GET(buffer[0], DB_RENDER_CONTROL, COPY_CENTROID) != 0;
-	r.copy_sample   = KYTY_PM4_GET(buffer[0], DB_RENDER_CONTROL, COPY_SAMPLE);
-
-	cp.GetCtx().SetRenderControl(r);
+	cp.GetCtx().SetRenderControl(DecodeRenderControl(buffer[0]));
 
 	return 1;
 }
@@ -1042,9 +1025,12 @@ KYTY_HW_CTX_PARSER(HwCtxSetRenderTarget) {
 	//	info.format                  = (buffer[4] >> 2u) & 0x1fu;
 	//	info.channel_type            = (buffer[4] >> 8u) & 0x7u;
 	//	info.channel_order           = (buffer[4] >> 11u) & 0x3u;
-	info.format                   = KYTY_PM4_GET(buffer[4], CB_COLOR0_INFO, FORMAT);
-	info.channel_type             = KYTY_PM4_GET(buffer[4], CB_COLOR0_INFO, NUMBER_TYPE);
-	info.channel_order            = KYTY_PM4_GET(buffer[4], CB_COLOR0_INFO, COMP_SWAP);
+	info.format =
+	    static_cast<Prospero::ChannelLayout>(KYTY_PM4_GET(buffer[4], CB_COLOR0_INFO, FORMAT));
+	info.channel_type =
+	    static_cast<Prospero::ChannelType>(KYTY_PM4_GET(buffer[4], CB_COLOR0_INFO, NUMBER_TYPE));
+	info.channel_order =
+	    static_cast<Prospero::ChannelOrder>(KYTY_PM4_GET(buffer[4], CB_COLOR0_INFO, COMP_SWAP));
 	info.cmask_fast_clear_enable  = KYTY_PM4_GET(buffer[4], CB_COLOR0_INFO, FAST_CLEAR) != 0;
 	info.fmask_compression_enable = KYTY_PM4_GET(buffer[4], CB_COLOR0_INFO, COMPRESSION) != 0;
 	info.blend_clamp              = KYTY_PM4_GET(buffer[4], CB_COLOR0_INFO, BLEND_CLAMP) != 0;
@@ -1066,10 +1052,12 @@ KYTY_HW_CTX_PARSER(HwCtxSetRenderTarget) {
 	//	attrib.num_fragments            = (buffer[5] >> 15u) & 0x3u;
 	attrib.force_dest_alpha_to_one =
 	    KYTY_PM4_GET(buffer[5], CB_COLOR0_ATTRIB, FORCE_DST_ALPHA_1) != 0;
-	attrib.tile_mode       = KYTY_PM4_GET(buffer[5], CB_COLOR0_ATTRIB, TILE_MODE_INDEX);
-	attrib.fmask_tile_mode = KYTY_PM4_GET(buffer[5], CB_COLOR0_ATTRIB, FMASK_TILE_MODE_INDEX);
-	attrib.num_samples     = KYTY_PM4_GET(buffer[5], CB_COLOR0_ATTRIB, NUM_SAMPLES);
-	attrib.num_fragments   = KYTY_PM4_GET(buffer[5], CB_COLOR0_ATTRIB, NUM_FRAGMENTS);
+	attrib.tile_mode =
+	    static_cast<Prospero::TileMode>(KYTY_PM4_GET(buffer[5], CB_COLOR0_ATTRIB, TILE_MODE_INDEX));
+	attrib.fmask_tile_mode = static_cast<Prospero::TileMode>(
+	    KYTY_PM4_GET(buffer[5], CB_COLOR0_ATTRIB, FMASK_TILE_MODE_INDEX));
+	attrib.num_samples   = KYTY_PM4_GET(buffer[5], CB_COLOR0_ATTRIB, NUM_SAMPLES);
+	attrib.num_fragments = KYTY_PM4_GET(buffer[5], CB_COLOR0_ATTRIB, NUM_FRAGMENTS);
 
 	//	dcc.max_uncompressed_block_size = (buffer[6] >> 2u) & 0x3u;
 	//	dcc.max_compressed_block_size   = (buffer[6] >> 5u) & 0x3u;
@@ -1162,13 +1150,7 @@ KYTY_HW_CTX_PARSER(HwCtxSetScanModeControl) {
 	EXIT_NOT_IMPLEMENTED(cmd_id != 0xc0016900);
 	EXIT_NOT_IMPLEMENTED(cmd_offset != Pm4::PA_SC_MODE_CNTL_0);
 
-	HW::ScanModeControl r;
-
-	r.msaa_enable          = KYTY_PM4_GET(buffer[0], PA_SC_MODE_CNTL_0, MSAA_ENABLE) != 0;
-	r.vport_scissor_enable = KYTY_PM4_GET(buffer[0], PA_SC_MODE_CNTL_0, VPORT_SCISSOR_ENABLE) != 0;
-	r.line_stipple_enable  = KYTY_PM4_GET(buffer[0], PA_SC_MODE_CNTL_0, LINE_STIPPLE_ENABLE) != 0;
-
-	cp.GetCtx().SetScanModeControl(r);
+	cp.GetCtx().SetScanModeControl(DecodeScanModeControl(buffer[0]));
 
 	return 1;
 }
@@ -1208,16 +1190,7 @@ KYTY_HW_CTX_PARSER(HwCtxSetStencilControl) {
 	EXIT_NOT_IMPLEMENTED(cmd_id != 0xC0016900);
 	EXIT_NOT_IMPLEMENTED(cmd_offset != Pm4::DB_STENCIL_CONTROL);
 
-	HW::StencilControl r;
-
-	r.stencil_fail     = KYTY_PM4_GET(buffer[0], DB_STENCIL_CONTROL, STENCILFAIL);
-	r.stencil_zpass    = KYTY_PM4_GET(buffer[0], DB_STENCIL_CONTROL, STENCILZPASS);
-	r.stencil_zfail    = KYTY_PM4_GET(buffer[0], DB_STENCIL_CONTROL, STENCILZFAIL);
-	r.stencil_fail_bf  = KYTY_PM4_GET(buffer[0], DB_STENCIL_CONTROL, STENCILFAIL_BF);
-	r.stencil_zpass_bf = KYTY_PM4_GET(buffer[0], DB_STENCIL_CONTROL, STENCILZPASS_BF);
-	r.stencil_zfail_bf = KYTY_PM4_GET(buffer[0], DB_STENCIL_CONTROL, STENCILZFAIL_BF);
-
-	cp.GetCtx().SetStencilControl(r);
+	cp.GetCtx().SetStencilControl(DecodeStencilControl(buffer[0]));
 
 	return 1;
 }
@@ -1232,23 +1205,16 @@ KYTY_HW_CTX_PARSER(HwCtxSetStencilInfo) {
 }
 
 KYTY_HW_CTX_PARSER(HwCtxSetStencilMask) {
-	EXIT_NOT_IMPLEMENTED(cmd_id != 0xc0026900);
-	EXIT_NOT_IMPLEMENTED(cmd_offset != Pm4::DB_STENCILREFMASK);
+	auto num_values = KYTY_PM4_LEN(cmd_id) - 2u;
+	EXIT_NOT_IMPLEMENTED(num_values == 0);
+	EXIT_NOT_IMPLEMENTED(cmd_offset < Pm4::DB_STENCILREFMASK);
+	EXIT_NOT_IMPLEMENTED(cmd_offset + num_values - 1u > Pm4::DB_STENCILREFMASK_BF);
 
-	HW::StencilMask r;
+	for (uint32_t i = 0; i < num_values; i++) {
+		g_hw_ctx_indirect_func[(cmd_offset + i) & (Pm4::CX_NUM - 1)](cp, cmd_offset + i, buffer[i]);
+	}
 
-	r.stencil_testval      = KYTY_PM4_GET(buffer[0], DB_STENCILREFMASK, STENCILTESTVAL);
-	r.stencil_mask         = KYTY_PM4_GET(buffer[0], DB_STENCILREFMASK, STENCILMASK);
-	r.stencil_writemask    = KYTY_PM4_GET(buffer[0], DB_STENCILREFMASK, STENCILWRITEMASK);
-	r.stencil_opval        = KYTY_PM4_GET(buffer[0], DB_STENCILREFMASK, STENCILOPVAL);
-	r.stencil_testval_bf   = KYTY_PM4_GET(buffer[1], DB_STENCILREFMASK_BF, STENCILTESTVAL_BF);
-	r.stencil_mask_bf      = KYTY_PM4_GET(buffer[1], DB_STENCILREFMASK_BF, STENCILMASK_BF);
-	r.stencil_writemask_bf = KYTY_PM4_GET(buffer[1], DB_STENCILREFMASK_BF, STENCILWRITEMASK_BF);
-	r.stencil_opval_bf     = KYTY_PM4_GET(buffer[1], DB_STENCILREFMASK_BF, STENCILOPVAL_BF);
-
-	cp.GetCtx().SetStencilMask(r);
-
-	return 2;
+	return num_values;
 }
 
 KYTY_HW_CTX_PARSER(HwCtxSetViewportScaleOffset) {
@@ -1351,10 +1317,6 @@ static void HwShSetCsRegister(CommandProcessor& cp, uint32_t cmd_offset, uint32_
 			                         Pm4::COMPUTE_PGM_RSRC1_BULKY_MASK) != 0u;
 			cs_regs.threadgroup_configuration = ((value >> Pm4::COMPUTE_PGM_RSRC1_WGP_MODE_SHIFT) &
 			                                     Pm4::COMPUTE_PGM_RSRC1_WGP_MODE_MASK) != 0u;
-			cs_regs.wave_size                 = (((value >> Pm4::COMPUTE_PGM_RSRC1_W32_EN_SHIFT) &
-			                                      Pm4::COMPUTE_PGM_RSRC1_W32_EN_MASK) != 0u)
-			                                        ? 32u
-			                                        : 64u;
 			break;
 		case Pm4::COMPUTE_PGM_RSRC2:
 			cs_regs.scratch_en     = ((value >> Pm4::COMPUTE_PGM_RSRC2_SCRATCH_EN_SHIFT) &
@@ -1568,7 +1530,7 @@ KYTY_HW_UC_PARSER(HwUcSetPrimitiveType) {
 
 	uint32_t prim_type = KYTY_PM4_GET(buffer[0], VGT_PRIMITIVE_TYPE, PRIM_TYPE);
 
-	cp.GetUcfg().SetPrimitiveType(prim_type);
+	cp.GetUcfg().SetPrimitiveType(static_cast<Prospero::PrimitiveType>(prim_type));
 
 	return 1;
 }
@@ -1611,16 +1573,10 @@ KYTY_HW_UC_PARSER(HwUcSetIaMultiVgtParam) {
 	return num_values;
 }
 
-static void HwIgnoreMultiPrimIbReset([[maybe_unused]] uint32_t value) {}
-
 KYTY_HW_UC_PARSER(HwUcSetMultiPrimIbReset) {
-	auto num_values = KYTY_PM4_LEN(cmd_id) - 2u;
-
-	for (uint32_t i = 0; i < num_values; i++) {
-		HwIgnoreMultiPrimIbReset(buffer[i]);
-	}
-
-	return num_values;
+	EXIT_NOT_IMPLEMENTED(cmd_offset != Pm4::GE_MULTI_PRIM_IB_RESET_EN);
+	cp.GetUcfg().SetPrimitiveResetControl(buffer[0]);
+	return 1;
 }
 
 static void HwUcIgnoreBorderColorTableAddr([[maybe_unused]] uint32_t cmd_offset,
@@ -1820,6 +1776,17 @@ KYTY_CP_OP_PARSER(CpOpPfpSyncMe) {
 	return 1;
 }
 
+KYTY_CP_OP_PARSER(CpOpRewind) {
+	KYTY_PROFILER_FUNCTION();
+
+	EXIT_NOT_IMPLEMENTED(cmd_id != 0xc0005900);
+	EXIT_NOT_IMPLEMENTED((buffer[0] & ~0x81000000u) != 0);
+
+	cp.WaitForRewind((buffer[0] & 0x80000000u) != 0);
+
+	return 1;
+}
+
 KYTY_CP_OP_PARSER(CpOpSetPredication) {
 	KYTY_PROFILER_FUNCTION();
 
@@ -2180,12 +2147,19 @@ KYTY_CP_OP_PARSER(CpOpEventWrite) {
 
 	EXIT_NOT_IMPLEMENTED(((cmd_id >> 8u) & 0xffu) != Pm4::IT_EVENT_WRITE);
 
-	uint32_t event_index = (buffer[0] >> 8u) & 0x7u;
-	uint32_t event_type  = (buffer[0]) & 0x3fu;
+	const auto packet_size_dw = KYTY_PM4_LEN(cmd_id);
+	uint32_t   event_index    = (buffer[0] >> 8u) & 0x7u;
+	uint32_t   event_type     = (buffer[0]) & 0x3fu;
+	uint64_t   event_address  = 0;
 
-	cp.TriggerEvent(event_type, event_index);
+	if (event_type == 0x39u) {
+		EXIT_NOT_IMPLEMENTED(packet_size_dw != 4u);
+		event_address = buffer[1] | (static_cast<uint64_t>(buffer[2]) << 32u);
+	}
 
-	return KYTY_PM4_LEN(cmd_id) - 1u;
+	cp.TriggerEvent(event_type, event_index, event_address);
+
+	return packet_size_dw - 1u;
 }
 
 KYTY_CP_OP_PARSER(CpOpEventWriteEop) {
@@ -2390,10 +2364,10 @@ KYTY_CP_OP_PARSER(CpOpIndirectCxRegs) {
 		EXIT("indirect CX registers have null address, num_regs = %" PRIu32 "\n", indirect_num_dw);
 	}
 	for (uint32_t i = 0; i < indirect_num_dw; i++, indirect_buffer += 2) {
-		// Keep the encoded offset for packet control values, and use the decoded offset only
-		// for register dispatch.
+		// Keep the encoded offset for packet control values, and use the normalized offset
+		// only for register dispatch.
 		auto raw_cmd_offset = indirect_buffer[0];
-		auto cmd_offset     = DecodeIndirectCxRegisterOffset(raw_cmd_offset);
+		auto cmd_offset     = NormalizeRegisterOffset(raw_cmd_offset);
 		auto value          = indirect_buffer[1];
 
 		if (HwCtxTrySetFakeRegister(cmd_offset, value)) {
@@ -2843,6 +2817,20 @@ KYTY_CP_OP_PARSER(CpOpSetShaderReg) {
 	auto pfunc = g_hw_sh_func[cmd_offset];
 
 	if (pfunc == nullptr) {
+		auto num_values = KYTY_PM4_LEN(cmd_id) - 2u;
+		bool handled    = num_values != 0;
+		for (uint32_t i = 0; i < num_values; i++) {
+			if (cmd_offset + i >= Pm4::SH_NUM || g_hw_sh_indirect_func[cmd_offset + i] == nullptr) {
+				handled = false;
+				break;
+			}
+		}
+		if (handled) {
+			for (uint32_t i = 0; i < num_values; i++) {
+				g_hw_sh_indirect_func[cmd_offset + i](cp, cmd_offset + i, buffer[1 + i]);
+			}
+			return num_values + 1u;
+		}
 		EXIT("unknown shader register\n\t%05" PRIx32 ":\n\tcmd_id = %08" PRIx32
 		     "\n\tcmd_offset = %08" PRIx32 "\n",
 		     num_dw - dw, cmd_id, cmd_offset);
@@ -2856,6 +2844,10 @@ KYTY_CP_OP_PARSER(CpOpSetShaderReg) {
 KYTY_CP_OP_PARSER(CpOpSetUconfigReg) {
 	KYTY_PROFILER_FUNCTION();
 
+	if (Gen5::AgcIsInternalDataPacket(cmd_id, buffer)) {
+		return KYTY_PM4_LEN(cmd_id) - 1u;
+	}
+
 	auto raw_cmd_offset = buffer[0];
 	auto cmd_offset     = NormalizeRegisterOffset(raw_cmd_offset);
 
@@ -2946,19 +2938,21 @@ template <typename T>
 static uint32_t CpOpWaitRegMemSized(CommandProcessor& cp, uint32_t cmd_id, const uint32_t* buffer) {
 	static_assert(sizeof(T) == sizeof(uint32_t) || sizeof(T) == sizeof(uint64_t));
 
-	constexpr auto value_dw   = static_cast<uint32_t>(sizeof(T) / sizeof(uint32_t));
+	constexpr auto value_dw = static_cast<uint32_t>(sizeof(T) / sizeof(uint32_t));
 	constexpr auto payload_dw = 4u + value_dw * 2u;
 	constexpr auto packet_dw  = payload_dw + 1u;
-	constexpr auto register_id =
-	    (sizeof(T) == sizeof(uint32_t) ? Pm4::R_WAIT_MEM_32 : Pm4::R_WAIT_MEM_64);
+	constexpr auto opcode =
+	    (sizeof(T) == sizeof(uint32_t) ? Pm4::IT_WAIT_REG_MEM : Pm4::IT_WAIT_REG_MEM_64);
 
-	EXIT_NOT_IMPLEMENTED(KYTY_PM4_R(cmd_id) != register_id);
+	EXIT_NOT_IMPLEMENTED(((cmd_id >> 8u) & 0xffu) != opcode || KYTY_PM4_R(cmd_id) != 0u);
 	EXIT_NOT_IMPLEMENTED(KYTY_PM4_LEN(cmd_id) != packet_dw);
 
-	auto* addr = reinterpret_cast<const T*>(buffer[0] | (static_cast<uint64_t>(buffer[1]) << 32u));
-	auto  mask = CpOpWaitRegMemReadValue<T>(buffer + 2u);
-	auto  ref  = CpOpWaitRegMemReadValue<T>(buffer + 2u + value_dw);
-	auto  ctrl = buffer[2u + value_dw * 2u];
+	constexpr auto address_align_mask = (sizeof(T) == sizeof(uint32_t) ? 0x3u : 0x7u);
+	auto  ctrl = buffer[0];
+	auto* addr = reinterpret_cast<const T*>((buffer[1] & ~address_align_mask) |
+	                                        (static_cast<uint64_t>(buffer[2] & 0x3ffffu) << 32u));
+	auto  ref  = CpOpWaitRegMemReadValue<T>(buffer + 3u);
+	auto  mask = CpOpWaitRegMemReadValue<T>(buffer + 3u + value_dw);
 	auto  poll = buffer[3u + value_dw * 2u];
 
 	EXIT_NOT_IMPLEMENTED((ctrl & 0x10u) == 0);
@@ -3106,9 +3100,12 @@ void GraphicsInitJmpTablesCxIndirect() {
 		g_hw_ctx_indirect_func[cmd_offset] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
 			uint32_t      slot = (cmd_offset - Pm4::CB_COLOR0_INFO) / 15;
 			HW::ColorInfo info;
-			info.format                   = KYTY_PM4_GET(value, CB_COLOR0_INFO, FORMAT);
-			info.channel_type             = KYTY_PM4_GET(value, CB_COLOR0_INFO, NUMBER_TYPE);
-			info.channel_order            = KYTY_PM4_GET(value, CB_COLOR0_INFO, COMP_SWAP);
+			info.format =
+			    static_cast<Prospero::ChannelLayout>(KYTY_PM4_GET(value, CB_COLOR0_INFO, FORMAT));
+			info.channel_type = static_cast<Prospero::ChannelType>(
+			    KYTY_PM4_GET(value, CB_COLOR0_INFO, NUMBER_TYPE));
+			info.channel_order =
+			    static_cast<Prospero::ChannelOrder>(KYTY_PM4_GET(value, CB_COLOR0_INFO, COMP_SWAP));
 			info.cmask_fast_clear_enable  = KYTY_PM4_GET(value, CB_COLOR0_INFO, FAST_CLEAR) != 0;
 			info.fmask_compression_enable = KYTY_PM4_GET(value, CB_COLOR0_INFO, COMPRESSION) != 0;
 			info.blend_clamp              = KYTY_PM4_GET(value, CB_COLOR0_INFO, BLEND_CLAMP) != 0;
@@ -3133,10 +3130,12 @@ void GraphicsInitJmpTablesCxIndirect() {
 			HW::ColorAttrib attrib;
 			attrib.force_dest_alpha_to_one =
 			    KYTY_PM4_GET(value, CB_COLOR0_ATTRIB, FORCE_DST_ALPHA_1) != 0;
-			attrib.tile_mode       = KYTY_PM4_GET(value, CB_COLOR0_ATTRIB, TILE_MODE_INDEX);
-			attrib.fmask_tile_mode = KYTY_PM4_GET(value, CB_COLOR0_ATTRIB, FMASK_TILE_MODE_INDEX);
-			attrib.num_samples     = KYTY_PM4_GET(value, CB_COLOR0_ATTRIB, NUM_SAMPLES);
-			attrib.num_fragments   = KYTY_PM4_GET(value, CB_COLOR0_ATTRIB, NUM_FRAGMENTS);
+			attrib.tile_mode = static_cast<Prospero::TileMode>(
+			    KYTY_PM4_GET(value, CB_COLOR0_ATTRIB, TILE_MODE_INDEX));
+			attrib.fmask_tile_mode = static_cast<Prospero::TileMode>(
+			    KYTY_PM4_GET(value, CB_COLOR0_ATTRIB, FMASK_TILE_MODE_INDEX));
+			attrib.num_samples   = KYTY_PM4_GET(value, CB_COLOR0_ATTRIB, NUM_SAMPLES);
+			attrib.num_fragments = KYTY_PM4_GET(value, CB_COLOR0_ATTRIB, NUM_FRAGMENTS);
 			cp.GetCtx().SetColorAttrib(slot, attrib);
 		};
 	}
@@ -3290,8 +3289,9 @@ void GraphicsInitJmpTablesCxIndirect() {
 		g_hw_ctx_indirect_func[cmd_offset] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
 			uint32_t         slot = (cmd_offset - Pm4::CB_COLOR0_ATTRIB3);
 			HW::ColorAttrib3 attrib3;
-			attrib3.depth              = KYTY_PM4_GET(value, CB_COLOR0_ATTRIB3, MIP0_DEPTH);
-			attrib3.tile_mode          = KYTY_PM4_GET(value, CB_COLOR0_ATTRIB3, COLOR_SW_MODE);
+			attrib3.depth     = KYTY_PM4_GET(value, CB_COLOR0_ATTRIB3, MIP0_DEPTH);
+			attrib3.tile_mode = static_cast<Prospero::TileMode>(
+			    KYTY_PM4_GET(value, CB_COLOR0_ATTRIB3, COLOR_SW_MODE));
 			attrib3.dimension          = KYTY_PM4_GET(value, CB_COLOR0_ATTRIB3, RESOURCE_TYPE);
 			attrib3.cmask_pipe_aligned = KYTY_PM4_GET(value, CB_COLOR0_ATTRIB3, CMASK_PIPE_ALIGNED);
 			attrib3.dcc_pipe_aligned   = KYTY_PM4_GET(value, CB_COLOR0_ATTRIB3, DCC_PIPE_ALIGNED);
@@ -3524,7 +3524,7 @@ void GraphicsInitJmpTablesCxIndirect() {
 		cp.GetCtx().SetReuseOff(value);
 	};
 	g_hw_ctx_indirect_func[Pm4::VGT_MULTI_PRIM_IB_RESET_INDX] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HwIgnoreMultiPrimIbReset(value);
+		cp.GetCtx().SetPrimitiveResetIndex(value);
 	};
 	g_hw_ctx_indirect_func[Pm4::VGT_TESS_DISTRIBUTION] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
 		cp.GetCtx().SetTessDistribution(value);
@@ -3691,39 +3691,12 @@ void GraphicsInitJmpTablesCxIndirect() {
 	for (uint32_t slot = 0; slot < 8; slot++) {
 		g_hw_ctx_indirect_func[Pm4::CB_BLEND0_CONTROL + slot] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
 			uint32_t param = cmd_offset - Pm4::CB_BLEND0_CONTROL;
-
-			HW::BlendControl r;
-			r.color_srcblend  = KYTY_PM4_GET(value, CB_BLEND0_CONTROL, COLOR_SRCBLEND);
-			r.color_comb_fcn  = KYTY_PM4_GET(value, CB_BLEND0_CONTROL, COLOR_COMB_FCN);
-			r.color_destblend = KYTY_PM4_GET(value, CB_BLEND0_CONTROL, COLOR_DESTBLEND);
-			r.alpha_srcblend  = KYTY_PM4_GET(value, CB_BLEND0_CONTROL, ALPHA_SRCBLEND);
-			r.alpha_comb_fcn  = KYTY_PM4_GET(value, CB_BLEND0_CONTROL, ALPHA_COMB_FCN);
-			r.alpha_destblend = KYTY_PM4_GET(value, CB_BLEND0_CONTROL, ALPHA_DESTBLEND);
-			r.separate_alpha_blend =
-			    KYTY_PM4_GET(value, CB_BLEND0_CONTROL, SEPARATE_ALPHA_BLEND) != 0;
-			r.enable = KYTY_PM4_GET(value, CB_BLEND0_CONTROL, ENABLE) != 0;
-
-			cp.GetCtx().SetBlendControl(param, r);
+			cp.GetCtx().SetBlendControl(param, DecodeBlendControl(value));
 		};
 	}
 
 	g_hw_ctx_indirect_func[Pm4::PA_SU_SC_MODE_CNTL] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HW::ModeControl r;
-		r.cull_front           = KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, CULL_FRONT) != 0;
-		r.cull_back            = KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, CULL_BACK) != 0;
-		r.face                 = KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, FACE) != 0;
-		r.poly_mode            = KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, POLY_MODE);
-		r.polymode_front_ptype = KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, POLYMODE_FRONT_PTYPE);
-		r.polymode_back_ptype  = KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, POLYMODE_BACK_PTYPE);
-		r.poly_offset_front_enable =
-		    KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, POLY_OFFSET_FRONT_ENABLE) != 0;
-		r.poly_offset_back_enable =
-		    KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, POLY_OFFSET_BACK_ENABLE) != 0;
-		r.vtx_window_offset_enable =
-		    KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, VTX_WINDOW_OFFSET_ENABLE) != 0;
-		r.provoking_vtx_last = KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, PROVOKING_VTX_LAST) != 0;
-		r.persp_corr_dis     = KYTY_PM4_GET(value, PA_SU_SC_MODE_CNTL, PERSP_CORR_DIS) != 0;
-		cp.GetCtx().SetModeControl(r);
+		cp.GetCtx().SetModeControl(DecodeModeControl(value));
 	};
 
 	g_hw_ctx_indirect_func[Pm4::PA_SU_POLY_OFFSET_DB_FMT_CNTL] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
@@ -3891,28 +3864,11 @@ void GraphicsInitJmpTablesCxIndirect() {
 	};
 
 	g_hw_ctx_indirect_func[Pm4::DB_STENCIL_CONTROL] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HW::StencilControl r;
-		r.stencil_fail     = KYTY_PM4_GET(value, DB_STENCIL_CONTROL, STENCILFAIL);
-		r.stencil_zpass    = KYTY_PM4_GET(value, DB_STENCIL_CONTROL, STENCILZPASS);
-		r.stencil_zfail    = KYTY_PM4_GET(value, DB_STENCIL_CONTROL, STENCILZFAIL);
-		r.stencil_fail_bf  = KYTY_PM4_GET(value, DB_STENCIL_CONTROL, STENCILFAIL_BF);
-		r.stencil_zpass_bf = KYTY_PM4_GET(value, DB_STENCIL_CONTROL, STENCILZPASS_BF);
-		r.stencil_zfail_bf = KYTY_PM4_GET(value, DB_STENCIL_CONTROL, STENCILZFAIL_BF);
-		cp.GetCtx().SetStencilControl(r);
+		cp.GetCtx().SetStencilControl(DecodeStencilControl(value));
 	};
 
 	g_hw_ctx_indirect_func[Pm4::DB_RENDER_CONTROL] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HW::RenderControl r;
-		r.depth_clear_enable   = KYTY_PM4_GET(value, DB_RENDER_CONTROL, DEPTH_CLEAR_ENABLE) != 0;
-		r.stencil_clear_enable = KYTY_PM4_GET(value, DB_RENDER_CONTROL, STENCIL_CLEAR_ENABLE) != 0;
-		r.resummarize_enable   = KYTY_PM4_GET(value, DB_RENDER_CONTROL, RESUMMARIZE_ENABLE) != 0;
-		r.stencil_compress_disable =
-		    KYTY_PM4_GET(value, DB_RENDER_CONTROL, STENCIL_COMPRESS_DISABLE) != 0;
-		r.depth_compress_disable =
-		    KYTY_PM4_GET(value, DB_RENDER_CONTROL, DEPTH_COMPRESS_DISABLE) != 0;
-		r.copy_centroid = KYTY_PM4_GET(value, DB_RENDER_CONTROL, COPY_CENTROID) != 0;
-		r.copy_sample   = KYTY_PM4_GET(value, DB_RENDER_CONTROL, COPY_SAMPLE);
-		cp.GetCtx().SetRenderControl(r);
+		cp.GetCtx().SetRenderControl(DecodeRenderControl(value));
 	};
 
 	g_hw_ctx_indirect_func[Pm4::DB_STENCILREFMASK] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
@@ -3934,23 +3890,7 @@ void GraphicsInitJmpTablesCxIndirect() {
 	};
 
 	g_hw_ctx_indirect_func[Pm4::PA_CL_CLIP_CNTL] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HW::ClipControl r;
-		r.user_clip_planes          = KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, UCP_ENA);
-		r.user_clip_plane_mode      = KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, PS_UCP_MODE);
-		r.dx_clip_space             = KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, DX_CLIP_SPACE_DEF) != 0;
-		r.vertex_kill_any           = KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, VTX_KILL_OR) != 0;
-		r.min_z_clip_disable        = KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, ZCLIP_NEAR_DISABLE) != 0;
-		r.max_z_clip_disable        = KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, ZCLIP_FAR_DISABLE) != 0;
-		r.user_clip_plane_negate_y  = KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, PS_UCP_Y_SCALE_NEG) != 0;
-		r.clip_disable              = KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, CLIP_DISABLE) != 0;
-		r.user_clip_plane_cull_only = KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, UCP_CULL_ONLY_ENA) != 0;
-		r.cull_on_clipping_error_disable =
-		    KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, DIS_CLIP_ERR_DETECT) != 0;
-		r.linear_attribute_clip_enable =
-		    KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, DX_LINEAR_ATTR_CLIP_ENA) != 0;
-		r.force_viewport_index_from_vs_enable =
-		    KYTY_PM4_GET(value, PA_CL_CLIP_CNTL, VTE_VPORT_PROVOKE_DISABLE) != 0;
-		cp.GetCtx().SetClipControl(r);
+		cp.GetCtx().SetClipControl(DecodeClipControl(value));
 	};
 
 	g_hw_ctx_indirect_func[Pm4::PA_SU_LINE_CNTL] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
@@ -3978,11 +3918,7 @@ void GraphicsInitJmpTablesCxIndirect() {
 	};
 
 	g_hw_ctx_indirect_func[Pm4::PA_SC_MODE_CNTL_0] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HW::ScanModeControl r;
-		r.msaa_enable          = KYTY_PM4_GET(value, PA_SC_MODE_CNTL_0, MSAA_ENABLE) != 0;
-		r.vport_scissor_enable = KYTY_PM4_GET(value, PA_SC_MODE_CNTL_0, VPORT_SCISSOR_ENABLE) != 0;
-		r.line_stipple_enable  = KYTY_PM4_GET(value, PA_SC_MODE_CNTL_0, LINE_STIPPLE_ENABLE) != 0;
-		cp.GetCtx().SetScanModeControl(r);
+		cp.GetCtx().SetScanModeControl(DecodeScanModeControl(value));
 	};
 	g_hw_ctx_indirect_func[Pm4::PA_SC_MODE_CNTL_1] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
 		HwCtxIgnoreScanModeControl1(value);
@@ -4041,27 +3977,11 @@ void GraphicsInitJmpTablesCxIndirect() {
 		HwCtxIgnoreAlphaToMaskRegister(value);
 	};
 	g_hw_ctx_indirect_func[Pm4::CB_COLOR_CONTROL] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HW::ColorControl r;
-		r.mode = KYTY_PM4_GET(value, CB_COLOR_CONTROL, MODE);
-		r.op   = KYTY_PM4_GET(value, CB_COLOR_CONTROL, ROP3);
-		cp.GetCtx().SetColorControl(r);
+		cp.GetCtx().SetColorControl(DecodeColorControl(value));
 	};
 
 	g_hw_ctx_indirect_func[Pm4::DB_DEPTH_CONTROL] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
-		HW::DepthControl r;
-		r.stencil_enable      = KYTY_PM4_GET(value, DB_DEPTH_CONTROL, STENCIL_ENABLE) != 0;
-		r.z_enable            = KYTY_PM4_GET(value, DB_DEPTH_CONTROL, Z_ENABLE) != 0;
-		r.z_write_enable      = KYTY_PM4_GET(value, DB_DEPTH_CONTROL, Z_WRITE_ENABLE) != 0;
-		r.depth_bounds_enable = KYTY_PM4_GET(value, DB_DEPTH_CONTROL, DEPTH_BOUNDS_ENABLE) != 0;
-		r.zfunc               = KYTY_PM4_GET(value, DB_DEPTH_CONTROL, ZFUNC);
-		r.backface_enable     = KYTY_PM4_GET(value, DB_DEPTH_CONTROL, BACKFACE_ENABLE) != 0;
-		r.stencilfunc         = KYTY_PM4_GET(value, DB_DEPTH_CONTROL, STENCILFUNC);
-		r.stencilfunc_bf      = KYTY_PM4_GET(value, DB_DEPTH_CONTROL, STENCILFUNC_BF);
-		r.color_writes_on_depth_fail_enable =
-		    KYTY_PM4_GET(value, DB_DEPTH_CONTROL, ENABLE_COLOR_WRITES_ON_DEPTH_FAIL) != 0;
-		r.color_writes_on_depth_pass_disable =
-		    KYTY_PM4_GET(value, DB_DEPTH_CONTROL, DISABLE_COLOR_WRITES_ON_DEPTH_PASS) != 0;
-		cp.GetCtx().SetDepthControl(r);
+		cp.GetCtx().SetDepthControl(DecodeDepthControl(value));
 	};
 
 	g_hw_ctx_indirect_func[Pm4::DB_EQAA] = [](KYTY_HW_CTX_INDIRECT_ARGS) {
@@ -4467,7 +4387,7 @@ void GraphicsInitJmpTablesUcIndirect() {
 
 	g_hw_uc_indirect_func[Pm4::VGT_PRIMITIVE_TYPE] = [](KYTY_HW_UC_INDIRECT_ARGS) {
 		uint32_t prim_type = KYTY_PM4_GET(value, VGT_PRIMITIVE_TYPE, PRIM_TYPE);
-		cp.GetUcfg().SetPrimitiveType(prim_type);
+		cp.GetUcfg().SetPrimitiveType(static_cast<Prospero::PrimitiveType>(prim_type));
 	};
 	g_hw_uc_indirect_func[Pm4::VGT_INDEX_TYPE] = [](KYTY_HW_UC_INDIRECT_ARGS) {
 		cp.SetIndexType(value & 0x3u);
@@ -4482,7 +4402,7 @@ void GraphicsInitJmpTablesUcIndirect() {
 		HwUcIgnoreIaMultiVgtParam(value);
 	};
 	g_hw_uc_indirect_func[Pm4::GE_MULTI_PRIM_IB_RESET_EN] = [](KYTY_HW_UC_INDIRECT_ARGS) {
-		HwIgnoreMultiPrimIbReset(value);
+		cp.GetUcfg().SetPrimitiveResetControl(value);
 	};
 	g_hw_uc_indirect_func[Pm4::TA_CS_BC_BASE_ADDR] = [](KYTY_HW_UC_INDIRECT_ARGS) {
 		HwUcIgnoreBorderColorTableAddr(cmd_offset, value);

src/graphics/guest_gpu/gpu_defs.h

@@ -3,15 +3,8 @@
 
 #include "common/common.h"
 
-#include <type_traits>
-
 namespace Libs::Graphics::Prospero {
 
-template <typename T>
-constexpr auto GpuEnumValue(T value) {
-	return static_cast<std::underlying_type_t<T>>(value);
-}
-
 enum class PrimitiveType : uint32_t {
 	kNone               = 0,
 	kPointList          = 1,
@@ -115,32 +108,52 @@ enum class ChannelOrder : uint32_t {
 	kAltReversed = 3,
 };
 
-// Maps a shader's logical RGBA export to the host color-attachment components. Two bits per
-// component keep this cheap to carry in shader and pipeline keys.
+// Selects the logical shader component written to each physical color-attachment component.
+// Two bits per component keep this cheap to carry in shader and pipeline keys.
 struct ColorComponentMapping {
-	uint8_t packed = 0xe4u; // RGBA: 0, 1, 2, 3
+	static constexpr uint8_t kIdentity = 0xe4u;
+
+	uint8_t packed = kIdentity;
 
 	[[nodiscard]] constexpr uint32_t Map(uint32_t component) const {
 		return component < 4u ? (packed >> (component * 2u)) & 0x3u : component;
 	}
 
-	[[nodiscard]] constexpr uint32_t ApplyMask(uint32_t mask) const {
+	[[nodiscard]] constexpr uint32_t ApplyMask(uint32_t logical_mask) const {
 		uint32_t mapped = 0;
-		for (uint32_t component = 0; component < 4u; component++) {
-			mapped |= ((mask >> component) & 1u) << Map(component);
+		for (uint32_t physical_component = 0; physical_component < 4u; physical_component++) {
+			mapped |= ((logical_mask >> Map(physical_component)) & 1u) << physical_component;
 		}
 		return mapped;
 	}
 
-	[[nodiscard]] constexpr bool IsIdentity() const { return packed == 0xe4u; }
+	// This mapping selects an intermediate component; next selects the final logical component.
+	[[nodiscard]] constexpr ColorComponentMapping Then(ColorComponentMapping next) const {
+		ColorComponentMapping result {0u};
+		for (uint32_t physical_component = 0; physical_component < 4u; physical_component++) {
+			result.packed |= static_cast<uint8_t>(next.Map(Map(physical_component))
+			                                      << (physical_component * 2u));
+		}
+		return result;
+	}
+
+	[[nodiscard]] constexpr bool IsIdentity() const { return packed == kIdentity; }
 
 	bool operator==(const ColorComponentMapping&) const = default;
 };
 
 static_assert(sizeof(ColorComponentMapping) == sizeof(uint8_t));
 
-inline constexpr ColorComponentMapping ColorMappingAbgr {0x1bu}; // 3, 2, 1, 0
+inline constexpr ColorComponentMapping ColorMappingRgba {ColorComponentMapping::kIdentity};
+inline constexpr ColorComponentMapping ColorMappingGr {0xe1u};   // 1, 0, 2, 3
+inline constexpr ColorComponentMapping ColorMappingRabg {0x6cu}; // 0, 3, 2, 1
+inline constexpr ColorComponentMapping ColorMappingRgab {0xb4u}; // 0, 1, 3, 2
 inline constexpr ColorComponentMapping ColorMappingBgra {0xc6u}; // 2, 1, 0, 3
+inline constexpr ColorComponentMapping ColorMappingAbgr {0x1bu}; // 3, 2, 1, 0
+inline constexpr ColorComponentMapping ColorMappingAgba {0x27u}; // 3, 1, 2, 0
+inline constexpr ColorComponentMapping ColorMappingArbg {0x63u}; // 3, 0, 2, 1
+inline constexpr ColorComponentMapping ColorMappingAgbr {0x87u}; // 3, 1, 0, 2
+inline constexpr ColorComponentMapping ColorMappingArgb {0x93u}; // 3, 0, 1, 2
 
 enum class DepthFormat : uint32_t {
 	kInvalid = 0,
@@ -388,9 +401,11 @@ enum class BufferFormat : uint32_t {
 	k8Srgb              = 128,
 	k8_8Srgb            = 129,
 	k8_8_8_8Srgb        = 130,
+	k10_10_10_2Float    = 131,
 	k9_9_9_5Float       = 132,
 	k5_6_5UNorm         = 133,
 	k5_5_5_1UNorm       = 134,
+	k1_5_5_5UNorm       = 135,
 	k4_4_4_4UNorm       = 136,
 	kFmask8_S4_F4       = 161,
 	kBc1UNorm           = 169,

src/graphics/guest_gpu/gpu_format.cpp

@@ -3,129 +3,221 @@
 #include "graphics/guest_gpu/gpu_defs.h"
 
 #include <array>
+#include <bit>
 
 namespace Libs::Graphics::Prospero {
 namespace {
 
 struct FormatInfo {
-	uint32_t format;
-	uint32_t bytes_per_element;
-	uint32_t block_compressed_bytes_per_block;
-	uint32_t render_target_bytes_per_element;
-	bool     supported_sampled_texture_format;
-	bool     unsigned_integer_texture_format;
+	BufferFormat       format;
+	uint32_t           bytes_per_element;
+	uint32_t           block_compressed_bytes_per_block;
+	uint32_t           render_target_bytes_per_element;
+	bool               sampled_texture;
+	bool               uint_texture;
 };
 
 constexpr FormatInfo kFormatInfo[] = {
-    {GpuEnumValue(BufferFormat::k8UNorm), 1, 0, 1, true, false},
-    {GpuEnumValue(BufferFormat::k8SNorm), 0, 0, 1, false, false},
-    {GpuEnumValue(BufferFormat::k8UInt), 1, 0, 1, true, true},
-    {GpuEnumValue(BufferFormat::k16UNorm), 2, 0, 2, true, false},
-    {GpuEnumValue(BufferFormat::k16SNorm), 2, 0, 2, true, false},
-    {GpuEnumValue(BufferFormat::k16UInt), 2, 0, 2, true, true},
-    {GpuEnumValue(BufferFormat::k16SInt), 2, 0, 2, false, false},
-    {GpuEnumValue(BufferFormat::k16Float), 2, 0, 2, true, false},
-    {GpuEnumValue(BufferFormat::k8_8UNorm), 2, 0, 2, true, false},
-    {GpuEnumValue(BufferFormat::k8_8SNorm), 2, 0, 2, true, false},
-    {GpuEnumValue(BufferFormat::k8_8UInt), 2, 0, 2, true, true},
-    {GpuEnumValue(BufferFormat::k8_8SInt), 2, 0, 2, false, false},
-    {GpuEnumValue(BufferFormat::k32UInt), 4, 0, 4, true, true},
-    {GpuEnumValue(BufferFormat::k32SInt), 4, 0, 4, false, false},
-    {GpuEnumValue(BufferFormat::k32Float), 4, 0, 4, true, false},
-    {GpuEnumValue(BufferFormat::k16_16UNorm), 4, 0, 4, true, false},
-    {GpuEnumValue(BufferFormat::k16_16SNorm), 4, 0, 4, true, false},
-    {GpuEnumValue(BufferFormat::k16_16UInt), 4, 0, 4, true, true},
-    {GpuEnumValue(BufferFormat::k16_16SInt), 4, 0, 4, false, false},
-    {GpuEnumValue(BufferFormat::k16_16Float), 4, 0, 4, true, false},
-    {GpuEnumValue(BufferFormat::k11_11_10Float), 4, 0, 4, true, false},
-    {GpuEnumValue(BufferFormat::k10_10_10_2UNorm), 4, 0, 4, true, false},
-    {GpuEnumValue(BufferFormat::k10_10_10_2UInt), 4, 0, 4, true, true},
-    {GpuEnumValue(BufferFormat::k8_8_8_8UNorm), 4, 0, 4, true, false},
-    {GpuEnumValue(BufferFormat::k8_8_8_8SNorm), 4, 0, 4, true, false},
-    {GpuEnumValue(BufferFormat::k8_8_8_8UInt), 4, 0, 4, true, true},
-    {GpuEnumValue(BufferFormat::k8_8_8_8SInt), 4, 0, 4, false, false},
-    {GpuEnumValue(BufferFormat::k32_32UInt), 8, 0, 8, true, true},
-    {GpuEnumValue(BufferFormat::k32_32SInt), 8, 0, 8, false, false},
-    {GpuEnumValue(BufferFormat::k32_32Float), 8, 0, 8, true, false},
-    {GpuEnumValue(BufferFormat::k16_16_16_16UNorm), 8, 0, 8, true, false},
-    {GpuEnumValue(BufferFormat::k16_16_16_16SNorm), 8, 0, 8, true, false},
-    {GpuEnumValue(BufferFormat::k16_16_16_16UInt), 8, 0, 8, true, true},
-    {GpuEnumValue(BufferFormat::k16_16_16_16SInt), 8, 0, 8, false, false},
-    {GpuEnumValue(BufferFormat::k16_16_16_16Float), 8, 0, 8, true, false},
-    {GpuEnumValue(BufferFormat::k32_32_32UInt), 12, 0, 12, true, true},
-    {GpuEnumValue(BufferFormat::k32_32_32SInt), 12, 0, 12, false, false},
-    {GpuEnumValue(BufferFormat::k32_32_32Float), 12, 0, 12, true, false},
-    {GpuEnumValue(BufferFormat::k32_32_32_32UInt), 16, 0, 16, true, true},
-    {GpuEnumValue(BufferFormat::k32_32_32_32SInt), 16, 0, 16, false, false},
-    {GpuEnumValue(BufferFormat::k32_32_32_32Float), 16, 0, 16, true, false},
-    {GpuEnumValue(BufferFormat::k8Srgb), 1, 0, 0, true, false},
-    {GpuEnumValue(BufferFormat::k8_8Srgb), 2, 0, 0, true, false},
-    {GpuEnumValue(BufferFormat::k8_8_8_8Srgb), 4, 0, 4, true, false},
-    {GpuEnumValue(BufferFormat::k9_9_9_5Float), 4, 0, 0, true, false},
-    {GpuEnumValue(BufferFormat::k5_6_5UNorm), 2, 0, 2, true, false},
-    {GpuEnumValue(BufferFormat::k5_5_5_1UNorm), 2, 0, 2, true, false},
-    {GpuEnumValue(BufferFormat::k4_4_4_4UNorm), 2, 0, 2, true, false},
-    {GpuEnumValue(BufferFormat::kFmask8_S4_F4), 1, 0, 1, true, false},
-    {GpuEnumValue(BufferFormat::kBc1UNorm), 0, 8, 0, true, false},
-    {GpuEnumValue(BufferFormat::kBc1Srgb), 0, 8, 0, true, false},
-    {GpuEnumValue(BufferFormat::kBc2UNorm), 0, 16, 0, true, false},
-    {GpuEnumValue(BufferFormat::kBc2Srgb), 0, 16, 0, true, false},
-    {GpuEnumValue(BufferFormat::kBc3UNorm), 0, 16, 0, true, false},
-    {GpuEnumValue(BufferFormat::kBc3Srgb), 0, 16, 0, true, false},
-    {GpuEnumValue(BufferFormat::kBc4UNorm), 0, 8, 0, true, false},
-    {GpuEnumValue(BufferFormat::kBc4SNorm), 0, 8, 0, true, false},
-    {GpuEnumValue(BufferFormat::kBc5UNorm), 0, 16, 0, true, false},
-    {GpuEnumValue(BufferFormat::kBc5SNorm), 0, 16, 0, true, false},
-    {GpuEnumValue(BufferFormat::kBc6UFloat), 0, 16, 0, true, false},
-    {GpuEnumValue(BufferFormat::kBc6SFloat), 0, 16, 0, true, false},
-    {GpuEnumValue(BufferFormat::kBc7UNorm), 0, 16, 0, true, false},
-    {GpuEnumValue(BufferFormat::kBc7Srgb), 0, 16, 0, true, false},
+	{BufferFormat::k8UNorm, 1, 0, 1, true, false},
+	{BufferFormat::k8SNorm, 0, 0, 1, false, false},
+	{BufferFormat::k8UInt, 1, 0, 1, true, true},
+	{BufferFormat::k16UNorm, 2, 0, 2, true, false},
+	{BufferFormat::k16SNorm, 2, 0, 2, true, false},
+	{BufferFormat::k16UInt, 2, 0, 2, true, true},
+	{BufferFormat::k16SInt, 2, 0, 2, false, false},
+	{BufferFormat::k16Float, 2, 0, 2, true, false},
+	{BufferFormat::k8_8UNorm, 2, 0, 2, true, false},
+	{BufferFormat::k8_8SNorm, 2, 0, 2, true, false},
+	{BufferFormat::k8_8UInt, 2, 0, 2, true, true},
+	{BufferFormat::k8_8SInt, 2, 0, 2, false, false},
+	{BufferFormat::k32UInt, 4, 0, 4, true, true},
+	{BufferFormat::k32SInt, 4, 0, 4, false, false},
+	{BufferFormat::k32Float, 4, 0, 4, true, false},
+	{BufferFormat::k16_16UNorm, 4, 0, 4, true, false},
+	{BufferFormat::k16_16SNorm, 4, 0, 4, true, false},
+	{BufferFormat::k16_16UInt, 4, 0, 4, true, true},
+	{BufferFormat::k16_16SInt, 4, 0, 4, false, false},
+	{BufferFormat::k16_16Float, 4, 0, 4, true, false},
+	{BufferFormat::k11_11_10UInt, 4, 0, 4, true, true},
+	{BufferFormat::k11_11_10Float, 4, 0, 4, true, false},
+	{BufferFormat::k10_10_10_2UNorm, 4, 0, 4, true, false},
+	{BufferFormat::k10_10_10_2UInt, 4, 0, 4, true, true},
+	{BufferFormat::k8_8_8_8UNorm, 4, 0, 4, true, false},
+	{BufferFormat::k8_8_8_8SNorm, 4, 0, 4, true, false},
+	{BufferFormat::k8_8_8_8UInt, 4, 0, 4, true, true},
+	{BufferFormat::k8_8_8_8SInt, 4, 0, 4, false, false},
+	{BufferFormat::k32_32UInt, 8, 0, 8, true, true},
+	{BufferFormat::k32_32SInt, 8, 0, 8, false, false},
+	{BufferFormat::k32_32Float, 8, 0, 8, true, false},
+	{BufferFormat::k16_16_16_16UNorm, 8, 0, 8, true, false},
+	{BufferFormat::k16_16_16_16SNorm, 8, 0, 8, true, false},
+	{BufferFormat::k16_16_16_16UInt, 8, 0, 8, true, true},
+	{BufferFormat::k16_16_16_16SInt, 8, 0, 8, false, false},
+	{BufferFormat::k16_16_16_16Float, 8, 0, 8, true, false},
+	{BufferFormat::k32_32_32UInt, 12, 0, 12, true, true},
+	{BufferFormat::k32_32_32SInt, 12, 0, 12, false, false},
+	{BufferFormat::k32_32_32Float, 12, 0, 12, true, false},
+	{BufferFormat::k32_32_32_32UInt, 16, 0, 16, true, true},
+	{BufferFormat::k32_32_32_32SInt, 16, 0, 16, false, false},
+	{BufferFormat::k32_32_32_32Float, 16, 0, 16, true, false},
+	{BufferFormat::k8Srgb, 1, 0, 0, true, false},
+	{BufferFormat::k8_8Srgb, 2, 0, 0, true, false},
+	{BufferFormat::k8_8_8_8Srgb, 4, 0, 4, true, false},
+	{BufferFormat::k9_9_9_5Float, 4, 0, 0, true, false},
+	{BufferFormat::k5_6_5UNorm, 2, 0, 2, true, false},
+	{BufferFormat::k5_5_5_1UNorm, 2, 0, 2, true, false},
+	{BufferFormat::k1_5_5_5UNorm, 0, 0, 2, false, false},
+	{BufferFormat::k4_4_4_4UNorm, 2, 0, 2, true, false},
+	{BufferFormat::kFmask8_S4_F4, 1, 0, 1, true, false},
+	{BufferFormat::kBc1UNorm, 0, 8, 0, true, false},
+	{BufferFormat::kBc1Srgb, 0, 8, 0, true, false},
+	{BufferFormat::kBc2UNorm, 0, 16, 0, true, false},
+	{BufferFormat::kBc2Srgb, 0, 16, 0, true, false},
+	{BufferFormat::kBc3UNorm, 0, 16, 0, true, false},
+	{BufferFormat::kBc3Srgb, 0, 16, 0, true, false},
+	{BufferFormat::kBc4UNorm, 0, 8, 0, true, false},
+	{BufferFormat::kBc4SNorm, 0, 8, 0, true, false},
+	{BufferFormat::kBc5UNorm, 0, 16, 0, true, false},
+	{BufferFormat::kBc5SNorm, 0, 16, 0, true, false},
+	{BufferFormat::kBc6UFloat, 0, 16, 0, true, false},
+	{BufferFormat::kBc6SFloat, 0, 16, 0, true, false},
+	{BufferFormat::kBc7UNorm, 0, 16, 0, true, false},
+	{BufferFormat::kBc7Srgb, 0, 16, 0, true, false},
 };
 
+constexpr uint8_t RenderTargetTypeBit(ChannelType type) {
+	return static_cast<uint8_t>(1u << static_cast<uint32_t>(type));
+}
+
+constexpr auto kUNormType = RenderTargetTypeBit(ChannelType::kUNorm);
+constexpr auto kIntegerTypes =
+    RenderTargetTypeBit(ChannelType::kUInt) | RenderTargetTypeBit(ChannelType::kSInt);
+constexpr auto kNormalizedIntegerTypes =
+    kUNormType | RenderTargetTypeBit(ChannelType::kSNorm) | kIntegerTypes;
+
+struct RenderTargetLayoutInfo {
+	ChannelLayout       layout;
+	BufferFormat        first_linear_format;
+	uint8_t             linear_type_mask;
+	uint8_t             components;
+	BufferFormat        extra_format  = BufferFormat::kInvalid;
+	ChannelType         extra_type    = ChannelType::kUNorm;
+	ChannelOrderSupport order_support = ChannelOrderSupport::kAll;
+};
+
+constexpr RenderTargetLayoutInfo kRenderTargetLayouts[] = {
+    {ChannelLayout::k8, BufferFormat::k8UNorm, kNormalizedIntegerTypes, 1, BufferFormat::k8Srgb,
+     ChannelType::kSrgb},
+    {ChannelLayout::k16, BufferFormat::k16UNorm, kNormalizedIntegerTypes, 1, BufferFormat::k16Float,
+     ChannelType::kFloat},
+    {ChannelLayout::k8_8, BufferFormat::k8_8UNorm, kNormalizedIntegerTypes, 2,
+     BufferFormat::k8_8Srgb, ChannelType::kSrgb},
+    {ChannelLayout::k32, BufferFormat::k32UInt, kIntegerTypes, 1, BufferFormat::k32Float,
+     ChannelType::kFloat},
+    {ChannelLayout::k16_16, BufferFormat::k16_16UNorm, kNormalizedIntegerTypes, 2,
+     BufferFormat::k16_16Float, ChannelType::kFloat},
+    {ChannelLayout::k11_11_10, BufferFormat::k11_11_10UNorm, kNormalizedIntegerTypes, 3,
+     BufferFormat::k11_11_10Float, ChannelType::kFloat},
+    {ChannelLayout::k10_11_11, BufferFormat::k10_11_11UNorm, kNormalizedIntegerTypes, 3,
+     BufferFormat::k10_11_11Float, ChannelType::kFloat},
+    {ChannelLayout::k2_10_10_10, BufferFormat::k2_10_10_10UNorm, kNormalizedIntegerTypes, 4},
+    {ChannelLayout::k10_10_10_2, BufferFormat::k10_10_10_2UNorm, kNormalizedIntegerTypes, 4},
+    {ChannelLayout::k8_8_8_8, BufferFormat::k8_8_8_8UNorm, kNormalizedIntegerTypes, 4,
+     BufferFormat::k8_8_8_8Srgb, ChannelType::kSrgb},
+    {ChannelLayout::k32_32, BufferFormat::k32_32UInt, kIntegerTypes, 2, BufferFormat::k32_32Float,
+     ChannelType::kFloat},
+    {ChannelLayout::k16_16_16_16, BufferFormat::k16_16_16_16UNorm, kNormalizedIntegerTypes, 4,
+     BufferFormat::k16_16_16_16Float, ChannelType::kFloat},
+    {ChannelLayout::k32_32_32_32, BufferFormat::k32_32_32_32UInt, kIntegerTypes, 4,
+     BufferFormat::k32_32_32_32Float, ChannelType::kFloat},
+    {ChannelLayout::k5_6_5, BufferFormat::k5_6_5UNorm, kUNormType, 3},
+    {ChannelLayout::k5_5_5_1, BufferFormat::k5_5_5_1UNorm, kUNormType, 4},
+    {ChannelLayout::k1_5_5_5, BufferFormat::k1_5_5_5UNorm, kUNormType, 4},
+    {ChannelLayout::k4_4_4_4, BufferFormat::k4_4_4_4UNorm, kUNormType, 4},
+    {ChannelLayout::k10_10_10_2Float, BufferFormat::kInvalid, 0, 4, BufferFormat::k10_10_10_2Float,
+     ChannelType::kFloat, ChannelOrderSupport::kStandardOnly},
+};
+
+const RenderTargetLayoutInfo* FindRenderTargetLayout(ChannelLayout layout) {
+	for (const auto& info: kRenderTargetLayouts) {
+		if (info.layout == layout) {
+			return &info;
+		}
+	}
+	return nullptr;
+}
+
 constexpr auto MakeFormatInfoLookup() {
-	constexpr uint32_t                            kMaxFormat = GpuEnumValue(BufferFormat::kBc7Srgb);
+	constexpr size_t kMaxFormat = static_cast<size_t>(BufferFormat::kBc7Srgb);
 	std::array<const FormatInfo*, kMaxFormat + 1> lookup {};
 	for (const auto& info: kFormatInfo) {
-		lookup[info.format] = &info;
+		lookup[static_cast<size_t>(info.format)] = &info;
 	}
 	return lookup;
 }
 
 constexpr auto kFormatInfoLookup = MakeFormatInfoLookup();
 
-const FormatInfo* FindFormatInfo(uint32_t format) {
-	return format < kFormatInfoLookup.size() ? kFormatInfoLookup[format] : nullptr;
+const FormatInfo* FindFormatInfo(BufferFormat format) {
+	const auto index = static_cast<size_t>(format);
+	return index < kFormatInfoLookup.size() ? kFormatInfoLookup[index] : nullptr;
 }
 
 } // namespace
 
-uint32_t NumBytesPerElement(uint32_t format) {
+RenderTargetFormatEncoding ResolveRenderTargetFormat(ChannelLayout layout, ChannelType type) {
+	const auto* layout_info = FindRenderTargetLayout(layout);
+	const auto  raw_type    = static_cast<uint32_t>(type);
+	if (layout_info == nullptr || raw_type >= 8u) {
+		return {};
+	}
+
+	auto format = BufferFormat::kInvalid;
+	if (layout_info->extra_format != BufferFormat::kInvalid && type == layout_info->extra_type) {
+		format = layout_info->extra_format;
+	} else if ((layout_info->linear_type_mask & (1u << raw_type)) != 0) {
+		const auto first_type =
+		    std::countr_zero(static_cast<uint32_t>(layout_info->linear_type_mask));
+		format = static_cast<BufferFormat>(static_cast<uint32_t>(layout_info->first_linear_format) +
+		                                   raw_type - first_type);
+	}
+	if (format == BufferFormat::kInvalid) {
+		return {};
+	}
+	return {format, layout_info->components, layout_info->order_support};
+}
+
+uint32_t NumBytesPerElement(BufferFormat format) {
 	const auto* info = FindFormatInfo(format);
 	return info != nullptr ? info->bytes_per_element : 0;
 }
 
-uint32_t BlockCompressedBytesPerBlock(uint32_t format) {
+uint32_t BlockCompressedBytesPerBlock(BufferFormat format) {
 	const auto* info = FindFormatInfo(format);
 	return info != nullptr ? info->block_compressed_bytes_per_block : 0;
 }
 
-uint32_t RenderTargetBytesPerElement(uint32_t format) {
+uint32_t RenderTargetBytesPerElement(BufferFormat format) {
 	const auto* info = FindFormatInfo(format);
 	return info != nullptr ? info->render_target_bytes_per_element : 0;
 }
 
-bool IsSupportedTextureFormat(uint32_t format) {
-	const auto* info = FindFormatInfo(format);
-	return info != nullptr && info->supported_sampled_texture_format;
+bool IsFmaskTextureFormat(BufferFormat format) {
+	return format == BufferFormat::kFmask8_S4_F4;
 }
 
-bool IsUintTextureFormat(uint32_t format) {
+bool IsSampledTextureFormat(BufferFormat format) {
 	const auto* info = FindFormatInfo(format);
-	return info != nullptr && info->unsigned_integer_texture_format;
+	return info != nullptr && info->sampled_texture;
 }
 
-bool IsFmaskTextureFormat(uint32_t format) {
-	return format == GpuEnumValue(BufferFormat::kFmask8_S4_F4);
+bool IsUintTextureFormat(BufferFormat format) {
+	const auto* info = FindFormatInfo(format);
+	return info != nullptr && info->uint_texture;
+}
+
+BufferFormat RemapTextureFormat(BufferFormat format) {
+	return format == BufferFormat::k11_11_10UInt ? BufferFormat::k32UInt : format;
 }
 
 } // namespace Libs::Graphics::Prospero

src/graphics/guest_gpu/gpu_format.h

@@ -1,16 +1,50 @@
 #ifndef EMULATOR_INCLUDE_EMULATOR_GRAPHICS_GUEST_GPU_GPU_FORMAT_H_
 #define EMULATOR_INCLUDE_EMULATOR_GRAPHICS_GUEST_GPU_GPU_FORMAT_H_
 
-#include "common/common.h"
+#include "graphics/guest_gpu/gpu_defs.h"
 
 namespace Libs::Graphics::Prospero {
 
-uint32_t NumBytesPerElement(uint32_t format);
-uint32_t BlockCompressedBytesPerBlock(uint32_t format);
-uint32_t RenderTargetBytesPerElement(uint32_t format);
-bool     IsSupportedTextureFormat(uint32_t format);
-bool     IsUintTextureFormat(uint32_t format);
-bool     IsFmaskTextureFormat(uint32_t format);
+enum class ChannelOrderSupport : uint8_t {
+	kNone,
+	kStandardOnly,
+	kAll,
+};
+
+struct RenderTargetFormatEncoding {
+	BufferFormat        buffer_format = BufferFormat::kInvalid;
+	uint8_t             components    = 0;
+	ChannelOrderSupport order_support = ChannelOrderSupport::kNone;
+
+	[[nodiscard]] constexpr bool IsValid() const {
+		return buffer_format != BufferFormat::kInvalid && components >= 1u && components <= 4u &&
+		       order_support != ChannelOrderSupport::kNone;
+	}
+
+	[[nodiscard]] constexpr bool SupportsOrder(ChannelOrder order) const {
+		switch (order_support) {
+			case ChannelOrderSupport::kStandardOnly: return order == ChannelOrder::kStandard;
+			case ChannelOrderSupport::kAll:
+				switch (order) {
+					case ChannelOrder::kStandard:
+					case ChannelOrder::kAlt:
+					case ChannelOrder::kReversed:
+					case ChannelOrder::kAltReversed: return true;
+				}
+				return false;
+			default: return false;
+		}
+	}
+};
+
+RenderTargetFormatEncoding ResolveRenderTargetFormat(ChannelLayout layout, ChannelType type);
+uint32_t                   NumBytesPerElement(BufferFormat format);
+uint32_t                   BlockCompressedBytesPerBlock(BufferFormat format);
+uint32_t                   RenderTargetBytesPerElement(BufferFormat format);
+bool                       IsFmaskTextureFormat(BufferFormat format);
+bool                       IsSampledTextureFormat(BufferFormat format);
+bool                       IsUintTextureFormat(BufferFormat format);
+BufferFormat               RemapTextureFormat(BufferFormat format);
 
 } // namespace Libs::Graphics::Prospero
 

src/graphics/guest_gpu/graphicsRun.cpp

@@ -13,6 +13,7 @@
 #include "graphics/host_gpu/renderer/render.h"
 #include "graphics/host_gpu/renderer/renderContext.h"
 #include "graphics/host_gpu/renderer/sync.h"
+#include "graphics/presentation/renderDoc.h"
 #include "graphics/presentation/videoOut.h"
 #include "graphics/presentation/window.h"
 #include "graphics/shader/shader.h"
@@ -26,6 +27,7 @@
 #include <cstdio>
 #include <deque>
 #include <memory>
+#include <mutex>
 #include <semaphore>
 #include <thread>
 #include <vector>
@@ -36,7 +38,7 @@ static thread_local CommandProcessor* g_current_processor = nullptr;
 static thread_local Pm4Execution*     g_current_execution = nullptr;
 static thread_local bool              g_gpu_mutex_owned   = false;
 static thread_local bool              g_gpu_thread        = false;
-static thread_local GpuState*         g_gpu_state         = nullptr;
+static thread_local GuestGpu*         g_gpu_state         = nullptr;
 
 class GpuMutexLock final {
 public:
@@ -59,115 +61,30 @@ private:
 	Common::Mutex& m_mutex;
 };
 
-struct OwnedCmdBuffer {
-	OwnedCmdBuffer() = default;
-	explicit OwnedCmdBuffer(const uint32_t* data, uint32_t count) {
-		EXIT_IF(data == nullptr && count != 0);
-		if (count != 0) {
-			m_words.assign(data, data + count);
-		}
-	}
-
-	[[nodiscard]] bool      Empty() const noexcept { return m_words.empty(); }
-	[[nodiscard]] uint32_t  Size() const noexcept { return static_cast<uint32_t>(m_words.size()); }
-	[[nodiscard]] uint32_t* Data() noexcept { return m_words.data(); }
-
-private:
-	std::vector<uint32_t> m_words;
-};
-
-class GpuState {
-public:
-	static constexpr uint32_t ComputePipeCount     = 7;
-	static constexpr uint32_t QueuesPerComputePipe = 8;
-	static constexpr uint32_t ComputeQueueCount    = ComputePipeCount * QueuesPerComputePipe;
-	static constexpr uint32_t QueueCount           = 1 + ComputeQueueCount;
-
-	explicit GpuState(RenderContext& renderer): m_renderer(renderer) {
-		EXIT_NOT_IMPLEMENTED(!Common::Thread::IsMainThread());
-		m_gfx_cp = std::make_unique<CommandProcessor>(renderer);
-		m_thread = std::jthread(ThreadRun, this);
-	}
-	~GpuState();
-
-	KYTY_CLASS_NO_COPY(GpuState);
-
-	void Submit(uint32_t* cmd_draw_buffer, uint32_t num_draw_dw, uint32_t* cmd_const_buffer,
-	            uint32_t num_const_dw, bool trigger_agc_interrupt_on_done);
-	void SubmitCompute(uint32_t queue, uint32_t* cmd_buffer, uint32_t num_dw,
-	                   bool trigger_agc_interrupt_on_done);
-	void SubmitFlipPreparation(uint64_t request_id);
-	void Done();
-	void Shutdown();
-	[[nodiscard]] bool        IsStopping();
-	void                      SendCommand(Common::UniqueFunction<void>&& command);
-	void                      SendCommandSync(Common::UniqueFunction<void>&& command);
-	int                       GetFrameNum();
-	[[nodiscard]] static bool IsGpuThread() noexcept { return g_gpu_thread; }
-
-private:
-	enum class SubmissionType { Graphics, Compute, FlipPreparation };
-
-	struct Submission {
-		SubmissionType type     = SubmissionType::Graphics;
-		uint32_t       queue_id = 0;
-		OwnedCmdBuffer commands;
-		OwnedCmdBuffer constant_commands;
-		Pm4Execution   command_execution;
-		Pm4Execution   constant_execution;
-		bool           trigger_agc_interrupt_on_done = false;
-		bool           reset_processor               = false;
-		bool           started                       = false;
-		bool           command_complete              = false;
-		bool           constant_complete             = false;
-		bool           blocked                       = false;
-		uint64_t       flip_request_id               = 0;
-	};
-
-	void              Enqueue(Submission submission);
-	void              WaitForIdle();
-	void              ProcessCommands();
-	bool              Process(Submission& submission);
-	static void       ThreadRun(void* data);
-	CommandProcessor& GetProcessor(uint32_t queue_id);
-
-	RenderContext&                                 m_renderer;
-	Common::Mutex                                  m_submission_mutex;
-	Common::Mutex                                  m_queue_mutex;
-	std::mutex                                     m_shutdown_mutex;
-	Common::CondVar                                m_work_available;
-	Common::CondVar                                m_idle;
-	std::array<std::deque<Submission>, QueueCount> m_queues;
-	std::deque<Common::UniqueFunction<void>>       m_commands;
-	std::atomic_uint32_t                           m_pending_commands {0};
-	uint32_t                                       m_next_queue        = 0;
-	uint32_t                                       m_submission_count  = 0;
-	bool                                           m_processing        = false;
-	bool                                           m_graphics_done     = true;
-	bool                                           m_accepting         = true;
-	bool                                           m_stopping          = false;
-	bool                                           m_shutdown_complete = false;
-
-	std::unique_ptr<CommandProcessor>                                m_gfx_cp;
-	std::array<std::unique_ptr<CommandProcessor>, ComputeQueueCount> m_compute_cp;
-
-	uint64_t        m_submit_id = 0;
-	std::atomic_int m_done_num  = 0;
-	std::jthread    m_thread;
-
-	friend class CommandProcessor;
-};
-
 static bool GraphicsRunDebugDumpEnabled() {
 	return Config::GraphicsDebugDumpEnabled() &&
 	       Config::GetPrintfDirection() != Config::OutputDirection::Silent;
 }
 
-GpuState::~GpuState() {
+GuestGpu::OwnedCmdBuffer::OwnedCmdBuffer(const uint32_t* data, uint32_t count) {
+	EXIT_IF(data == nullptr && count != 0);
+	if (count != 0) {
+		m_words.assign(data, data + count);
+	}
+}
+
+GuestGpu::GuestGpu(RenderContext& renderer): m_renderer(renderer) {
+	EXIT_NOT_IMPLEMENTED(!Common::Thread::IsMainThread());
+	GraphicsInitJmpTables();
+	m_gfx_cp = std::make_unique<CommandProcessor>(renderer);
+	m_thread = std::jthread(ThreadRun, this);
+}
+
+GuestGpu::~GuestGpu() {
 	Shutdown();
 }
 
-void GpuState::Shutdown() {
+void GuestGpu::Shutdown() {
 	std::lock_guard shutdown_lock(m_shutdown_mutex);
 	if (m_shutdown_complete) {
 		return;
@@ -184,12 +101,12 @@ void GpuState::Shutdown() {
 	m_shutdown_complete = true;
 }
 
-bool GpuState::IsStopping() {
+bool GuestGpu::IsStopping() {
 	Common::LockGuard lock(m_queue_mutex);
 	return m_stopping;
 }
 
-void GpuState::SendCommand(Common::UniqueFunction<void>&& command) {
+void GuestGpu::SendCommand(Common::UniqueFunction<void>&& command) {
 	EXIT_IF(!command);
 	if (IsGpuThread()) {
 		command();
@@ -202,7 +119,7 @@ void GpuState::SendCommand(Common::UniqueFunction<void>&& command) {
 	m_work_available.Signal();
 }
 
-void GpuState::ProcessCommands() {
+void GuestGpu::ProcessCommands() {
 	EXIT_IF(!IsGpuThread());
 	while (m_pending_commands.load(std::memory_order_acquire) != 0) {
 		Common::UniqueFunction<void> command;
@@ -217,7 +134,7 @@ void GpuState::ProcessCommands() {
 	}
 }
 
-void GpuState::SendCommandSync(Common::UniqueFunction<void>&& command) {
+void GuestGpu::SendCommandSync(Common::UniqueFunction<void>&& command) {
 	EXIT_IF(!command);
 	if (IsGpuThread()) {
 		command();
@@ -231,8 +148,12 @@ void GpuState::SendCommandSync(Common::UniqueFunction<void>&& command) {
 	done.acquire();
 }
 
-void GpuState::Submit(uint32_t* cmd_draw_buffer, uint32_t num_draw_dw, uint32_t* cmd_const_buffer,
+void GuestGpu::Submit(uint32_t* cmd_draw_buffer, uint32_t num_draw_dw, uint32_t* cmd_const_buffer,
                       uint32_t num_const_dw, bool trigger_agc_interrupt_on_done) {
+	if (num_draw_dw == 0) {
+		return;
+	}
+	EXIT_IF(cmd_draw_buffer == nullptr);
 	GpuMutexLock lock(m_submission_mutex);
 	Submission   submission;
 	submission.type                          = SubmissionType::Graphics;
@@ -245,8 +166,9 @@ void GpuState::Submit(uint32_t* cmd_draw_buffer, uint32_t num_draw_dw, uint32_t*
 	Enqueue(std::move(submission));
 }
 
-void GpuState::SubmitCompute(uint32_t queue, uint32_t* cmd_buffer, uint32_t num_dw,
+void GuestGpu::SubmitCompute(uint32_t queue, uint32_t* cmd_buffer, uint32_t num_dw,
                              bool trigger_agc_interrupt_on_done) {
+	EXIT_IF(cmd_buffer == nullptr || num_dw == 0);
 	GpuMutexLock lock(m_submission_mutex);
 
 	constexpr uint32_t compute_queue_base = 0x20u;
@@ -262,7 +184,7 @@ void GpuState::SubmitCompute(uint32_t queue, uint32_t* cmd_buffer, uint32_t num_
 	Enqueue(std::move(submission));
 }
 
-void GpuState::SubmitFlipPreparation(uint64_t request_id) {
+void GuestGpu::SubmitFlipPreparation(uint64_t request_id) {
 	GpuMutexLock lock(m_submission_mutex);
 	Submission   submission;
 	submission.type            = SubmissionType::FlipPreparation;
@@ -273,7 +195,7 @@ void GpuState::SubmitFlipPreparation(uint64_t request_id) {
 	Enqueue(std::move(submission));
 }
 
-void GpuState::Done() {
+void GuestGpu::Done() {
 	GpuMutexLock lock(m_submission_mutex);
 	if (!IsGpuThread()) {
 		WaitForIdle();
@@ -282,11 +204,11 @@ void GpuState::Done() {
 	m_done_num++;
 }
 
-int GpuState::GetFrameNum() {
+int GuestGpu::GetFrameNum() const {
 	return m_done_num;
 }
 
-CommandProcessor& GpuState::GetProcessor(uint32_t queue_id) {
+CommandProcessor& GuestGpu::GetProcessor(uint32_t queue_id) {
 	EXIT_IF(queue_id >= QueueCount);
 	if (queue_id == 0) {
 		return *m_gfx_cp;
@@ -374,6 +296,12 @@ void CommandProcessor::WaitDeDiff(uint32_t diff) {
 	}
 }
 
+void CommandProcessor::WaitForRewind(bool valid) {
+	if (!valid) {
+		SuspendPm4();
+	}
+}
+
 void CommandProcessor::IncrementDe() {
 	m_de_count++;
 }
@@ -426,10 +354,8 @@ template void CommandProcessor::WaitRegMem<uint64_t>(uint32_t, const uint64_t*,
 
 void CommandProcessor::WriteData(uint32_t* dst, const uint32_t* src, uint32_t dw_num,
                                  uint32_t write_control) {
-	const uint32_t dst_sel      = ((write_control >> 30u) & 0x1u) | ((write_control >> 7u) & 0x1eu);
-	const uint32_t cache_policy = (write_control >> 25u) & 0x3u;
-	const uint32_t increment    = (write_control >> 16u) & 0x1u;
-	const uint32_t write_confirm = (write_control >> 20u) & 0x1u;
+	const uint32_t dst_sel = ((write_control >> 30u) & 0x1u) | ((write_control >> 7u) & 0x1eu);
+	const bool     write_one_address = ((write_control >> 16u) & 0x1u) != 0;
 
 	switch (dst_sel) {
 		case 0:
@@ -439,16 +365,17 @@ void CommandProcessor::WriteData(uint32_t* dst, const uint32_t* src, uint32_t dw
 		case 6: break;
 		default: EXIT("unsupported writeData destination selector 0x%02" PRIx32 "\n", dst_sel);
 	}
-	EXIT_NOT_IMPLEMENTED(increment != 0);
-
-	if (cache_policy > 3 || write_confirm > 1) {
-		LOGF("\t warning: unexpected write_data control 0x%08" PRIx32 "\n", write_control);
-	}
 	if (dw_num == 0) {
 		return;
 	}
 
-	memcpy(dst, src, static_cast<size_t>(dw_num) * sizeof(uint32_t));
+	if (write_one_address) {
+		for (uint32_t i = 0; i < dw_num; i++) {
+			dst[0] = src[i];
+		}
+	} else {
+		memcpy(dst, src, static_cast<size_t>(dw_num) * sizeof(uint32_t));
+	}
 }
 
 void CommandProcessor::WriteReferenceClock(uint64_t dst_address, uint32_t num_bytes) {
@@ -489,6 +416,13 @@ void CommandProcessor::DmaData(uint8_t engine, uint8_t dst_sel, uint8_t dst_cach
 			default: return false;
 		}
 	};
+	if (dst_sel == 2) {
+		// kNowhere discards the GL2 prefetch destination without a guest-visible write.
+		if (src_sel != 3) {
+			EXIT("unsupported dmaData nowhere source selector 0x%02" PRIx8 "\n", src_sel);
+		}
+		return;
+	}
 	bool dst_gds = false;
 	if (!decode_gds(dst_sel, dst_gds)) {
 		EXIT("unsupported dmaData destination selector 0x%02" PRIx8 "\n", dst_sel);
@@ -511,7 +445,7 @@ void CommandProcessor::DmaData(uint8_t engine, uint8_t dst_sel, uint8_t dst_cach
 	                        dst_gds, src_gds);
 }
 
-void GpuState::Enqueue(Submission submission) {
+void GuestGpu::Enqueue(Submission submission) {
 	EXIT_IF(submission.queue_id >= QueueCount);
 	Common::LockGuard lock(m_queue_mutex);
 	EXIT_IF(!m_accepting);
@@ -520,15 +454,15 @@ void GpuState::Enqueue(Submission submission) {
 	m_work_available.Signal();
 }
 
-void GpuState::WaitForIdle() {
+void GuestGpu::WaitForIdle() {
 	Common::LockGuard lock(m_queue_mutex);
 	while (m_processing || !m_commands.empty() || m_submission_count != 0) {
 		m_idle.Wait(&m_queue_mutex);
 	}
 }
 
-void GpuState::ThreadRun(void* data) {
-	auto* gpu = static_cast<GpuState*>(data);
+void GuestGpu::ThreadRun(void* data) {
+	auto* gpu = static_cast<GuestGpu*>(data);
 	EXIT_IF(gpu == nullptr);
 	KYTY_PROFILER_THREAD("Thread_Gpu");
 	g_gpu_thread = true;
@@ -624,9 +558,13 @@ void GpuState::ThreadRun(void* data) {
 	}
 }
 
-bool GpuState::Process(Submission& submission) {
-	auto&      cp          = GetProcessor(submission.queue_id);
+bool GuestGpu::Process(Submission& submission) {
 	const bool first_slice = !submission.started;
+	if (first_slice && RenderDocCaptureRequested()) {
+		Common::LockGuard render_lock(m_renderer.GetMutex());
+		RenderDocStartCapture();
+	}
+	auto& cp = GetProcessor(submission.queue_id);
 
 	if (first_slice && submission.reset_processor) {
 		cp.Reset();
@@ -1163,6 +1101,8 @@ void CommandProcessor::DrawIndirectMulti(uint32_t data_offset, uint32_t max_coun
 
 void CommandProcessor::DispatchDirect(uint32_t thread_group_x, uint32_t thread_group_y,
                                       uint32_t thread_group_z, uint32_t mode) {
+	m_sh_ctx.SetCsWaveSize(Pm4::ComputeWaveSize(mode));
+
 	uint32_t frame_num = 0;
 	// uint32_t local_x   = 1;
 	// uint32_t local_y   = 1;
@@ -1336,7 +1276,7 @@ void CommandProcessor::WriteAtEndOfPipe(uint32_t cache_policy, uint32_t event_wr
 				if (eop_event_type == 0x2f && cache_action == 0x00 && event_index == 0x06) {
 					auto* dst = static_cast<uint32_t*>(dst_gpu_addr);
 					SynchronizeGpu();
-					Sync::ReadGds(m_renderer.GetBufferCache().GetGdsBuffer(), dst, value & 0xffffu,
+					Sync::ReadGds(*m_renderer.GetBufferCache().GetGdsBuffer(), dst, value & 0xffffu,
 					              value >> 16u);
 					Sync::WriteAtEndOfPipeGds32(m_submit_id, CurrentBuffer(), dst, value & 0xffffu,
 					                            value >> 16u);
@@ -1384,9 +1324,14 @@ void CommandProcessor::WriteAtEndOfPipe(uint32_t cache_policy, uint32_t event_wr
 					case 0x00:
 						switch (eop_event_type) {
 							case 0x04:
+								if (event_index == 0x05) {
+									write64(false);
+									return;
+								}
+								break;
+							case 0x14:
 							case 0x28:
-								if ((eop_event_type == 0x04 && event_index == 0x05) ||
-								    (eop_event_type == 0x28 && event_index == 0x00)) {
+								if (event_index == 0x00) {
 									write64(false);
 									return;
 								}
@@ -1523,12 +1468,14 @@ void CommandProcessor::TriggerEopEventAtEndOfPipe(uint32_t interrupt_context_id)
 	Sync::TriggerEopEventAtEndOfPipe(CurrentBuffer(), interrupt_context_id);
 }
 
-void CommandProcessor::TriggerEvent(uint32_t event_type, uint32_t event_index) {
+void CommandProcessor::TriggerEvent(uint32_t event_type, uint32_t event_index,
+                                    uint64_t event_address) {
 	if (GraphicsRunDebugDumpEnabled()) {
 		LOGF("CommandProcessor::TriggerEvent()\n"
 		     "\t event_type  = 0x%08" PRIx32 "\n"
-		     "\t event_index = 0x%08" PRIx32 "\n",
-		     event_type, event_index);
+		     "\t event_index = 0x%08" PRIx32 "\n"
+		     "\t address     = 0x%016" PRIx64 "\n",
+		     event_type, event_index, event_address);
 	}
 
 	const auto valid_cache_event_index = event_index == 0x00000000 || event_index == 0x00000007;
@@ -1565,12 +1512,36 @@ void CommandProcessor::TriggerEvent(uint32_t event_type, uint32_t event_index) {
 		case 0x0000001a:
 		case 0x0000001b:
 		case 0x00000038:
-		case 0x00000039:
 		case 0x0000003a:
 			LOGF("\t temporary: ignoring unsupported event_write type 0x%08" PRIx32
 			     ", index 0x%08" PRIx32 "\n",
 			     event_type, event_index);
 			break;
+		case 0x00000039: {
+			if (event_index != 0x00000001 || event_address == 0 || (event_address & 0x7u) != 0) {
+				EXIT("invalid occlusion-counter dump: index=0x%08" PRIx32 ", address=0x%016" PRIx64
+				     "\n",
+				     event_index, event_address);
+			}
+			static std::once_flag warning_once;
+			std::call_once(warning_once, [] {
+				std::printf("Warning: game uses occlusion queries, which are currently treated as "
+				            "always visible; GPU usage may be higher and FPS may be lower.\n");
+			});
+
+			// Until host occlusion queries are implemented, publish an always-visible result. The
+			// PS5 layout contains one interleaved begin/end pair per DB, and bit 63 marks a result
+			// ready.
+			constexpr uint64_t ready_bit    = 1ull << 63u;
+			constexpr uint64_t counter_mask = ready_bit - 1u;
+			auto*              results      = reinterpret_cast<volatile uint64_t*>(event_address);
+			const auto         value        = ready_bit | m_synthetic_occlusion_counter;
+			for (uint32_t db = 0; db < 16u; db++) {
+				results[db * 2u] = value;
+			}
+			m_synthetic_occlusion_counter = (m_synthetic_occlusion_counter + 1u) & counter_mask;
+			break;
+		}
 		default:
 			EXIT("unknown event type: 0x%08" PRIx32 ", 0x%08" PRIx32 "\n", event_type, event_index);
 	}
@@ -1657,57 +1628,8 @@ void CommandProcessor::SynchronizeGpu() {
 	GetScheduler().FinishCurrent();
 }
 
-Gpu::Gpu(RenderContext& renderer) {
-	EXIT_NOT_IMPLEMENTED(!Common::Thread::IsMainThread());
-	GraphicsInitJmpTables();
-	m_state = std::make_unique<GpuState>(renderer);
-}
-
-Gpu::~Gpu() = default;
-
-void Gpu::Shutdown() {
-	m_state->Shutdown();
-}
-
-bool Gpu::IsStopping() {
-	return m_state->IsStopping();
-}
-
-void Gpu::SendCommand(Common::UniqueFunction<void>&& command) {
-	m_state->SendCommand(std::move(command));
-}
-
-void Gpu::SendCommandSync(Common::UniqueFunction<void>&& command) {
-	m_state->SendCommandSync(std::move(command));
-}
-
-void Gpu::Submit(uint32_t* draw_commands, uint32_t draw_size_dw, uint32_t* constant_commands,
-                 uint32_t constant_size_dw, bool trigger_agc_interrupt_on_done) {
-	EXIT_IF(draw_commands == nullptr || draw_size_dw == 0);
-	m_state->Submit(draw_commands, draw_size_dw, constant_commands, constant_size_dw,
-	                trigger_agc_interrupt_on_done);
-}
-
-void Gpu::SubmitCompute(uint32_t queue, uint32_t* commands, uint32_t size_dw,
-                        bool trigger_agc_interrupt_on_done) {
-	EXIT_IF(commands == nullptr || size_dw == 0);
-	m_state->SubmitCompute(queue, commands, size_dw, trigger_agc_interrupt_on_done);
-}
-
-void Gpu::SubmitFlipPreparation(uint64_t request_id) {
-	m_state->SubmitFlipPreparation(request_id);
-}
-
-void Gpu::Done() {
-	m_state->Done();
-}
-
-int Gpu::GetFrameNum() const {
-	return m_state->GetFrameNum();
-}
-
-bool Gpu::IsGpuThread() noexcept {
-	return GpuState::IsGpuThread();
+bool GuestGpu::IsGpuThread() noexcept {
+	return g_gpu_thread;
 }
 
 } // namespace Libs::Graphics

src/graphics/guest_gpu/graphicsRun.h

@@ -3,20 +3,28 @@
 
 #include "common/abi.h"
 #include "common/common.h"
+#include "common/threads.h"
 #include "common/uniqueFunction.h"
+#include "graphics/guest_gpu/command_processor/commandProcessor.h"
 
+#include <array>
+#include <atomic>
+#include <cstdint>
+#include <deque>
 #include <memory>
+#include <mutex>
+#include <thread>
+#include <vector>
 
 namespace Libs::Graphics {
 
-class GpuState;
 class RenderContext;
 
-class Gpu final {
+class GuestGpu final {
 public:
-	explicit Gpu(RenderContext& renderer);
-	~Gpu();
-	KYTY_CLASS_NO_COPY(Gpu);
+	explicit GuestGpu(RenderContext& renderer);
+	~GuestGpu();
+	KYTY_CLASS_NO_COPY(GuestGpu);
 
 	void               Shutdown();
 	[[nodiscard]] bool IsStopping();
@@ -34,7 +42,76 @@ public:
 	[[nodiscard]] static bool IsGpuThread() noexcept;
 
 private:
-	std::unique_ptr<GpuState> m_state;
+	static constexpr uint32_t ComputePipeCount     = 7;
+	static constexpr uint32_t QueuesPerComputePipe = 8;
+	static constexpr uint32_t ComputeQueueCount    = ComputePipeCount * QueuesPerComputePipe;
+	static constexpr uint32_t QueueCount           = 1 + ComputeQueueCount;
+
+	class OwnedCmdBuffer {
+	public:
+		OwnedCmdBuffer() = default;
+		explicit OwnedCmdBuffer(const uint32_t* data, uint32_t count);
+
+		[[nodiscard]] bool     Empty() const noexcept { return m_words.empty(); }
+		[[nodiscard]] uint32_t Size() const noexcept {
+			return static_cast<uint32_t>(m_words.size());
+		}
+		[[nodiscard]] uint32_t* Data() noexcept { return m_words.data(); }
+
+	private:
+		std::vector<uint32_t> m_words;
+	};
+
+	enum class SubmissionType { Graphics, Compute, FlipPreparation };
+
+	struct Submission {
+		SubmissionType type     = SubmissionType::Graphics;
+		uint32_t       queue_id = 0;
+		OwnedCmdBuffer commands;
+		OwnedCmdBuffer constant_commands;
+		Pm4Execution   command_execution;
+		Pm4Execution   constant_execution;
+		bool           trigger_agc_interrupt_on_done = false;
+		bool           reset_processor               = false;
+		bool           started                       = false;
+		bool           command_complete              = false;
+		bool           constant_complete             = false;
+		bool           blocked                       = false;
+		uint64_t       flip_request_id               = 0;
+	};
+
+	void              Enqueue(Submission submission);
+	void              WaitForIdle();
+	void              ProcessCommands();
+	bool              Process(Submission& submission);
+	static void       ThreadRun(void* data);
+	CommandProcessor& GetProcessor(uint32_t queue_id);
+
+	RenderContext&                                 m_renderer;
+	Common::Mutex                                  m_submission_mutex;
+	Common::Mutex                                  m_queue_mutex;
+	std::mutex                                     m_shutdown_mutex;
+	Common::CondVar                                m_work_available;
+	Common::CondVar                                m_idle;
+	std::array<std::deque<Submission>, QueueCount> m_queues;
+	std::deque<Common::UniqueFunction<void>>       m_commands;
+	std::atomic_uint32_t                           m_pending_commands {0};
+	uint32_t                                       m_next_queue        = 0;
+	uint32_t                                       m_submission_count  = 0;
+	bool                                           m_processing        = false;
+	bool                                           m_graphics_done     = true;
+	bool                                           m_accepting         = true;
+	bool                                           m_stopping          = false;
+	bool                                           m_shutdown_complete = false;
+
+	std::unique_ptr<CommandProcessor>                                m_gfx_cp;
+	std::array<std::unique_ptr<CommandProcessor>, ComputeQueueCount> m_compute_cp;
+
+	uint64_t        m_submit_id = 0;
+	std::atomic_int m_done_num  = 0;
+	std::jthread    m_thread;
+
+	friend class CommandProcessor;
 };
 } // namespace Libs::Graphics
 

src/graphics/guest_gpu/hardwareContext.h

@@ -29,27 +29,27 @@ struct ColorView {
 struct ColorInfo {
 	bool fmask_compression_enable = false;
 	// uint32_t fmask_compression_mode   = 0;
-	bool     fmask_data_compression_disable = false;
-	bool     fmask_one_frag_mode            = false;
-	bool     cmask_fast_clear_enable        = false;
-	bool     dcc_compression_enable         = false;
-	bool     blend_clamp                    = false;
-	bool     blend_bypass                   = false;
-	bool     round_mode                     = false;
-	uint32_t cmask_is_linear                = 0;
-	uint32_t cmask_addr_type                = 0;
-	bool     alt_tile_mode                  = false;
-	uint32_t format                         = 0;
-	uint32_t channel_type                   = 0;
-	uint32_t channel_order                  = 0;
+	bool                    fmask_data_compression_disable = false;
+	bool                    fmask_one_frag_mode            = false;
+	bool                    cmask_fast_clear_enable        = false;
+	bool                    dcc_compression_enable         = false;
+	bool                    blend_clamp                    = false;
+	bool                    blend_bypass                   = false;
+	bool                    round_mode                     = false;
+	uint32_t                cmask_is_linear                = 0;
+	uint32_t                cmask_addr_type                = 0;
+	bool                    alt_tile_mode                  = false;
+	Prospero::ChannelLayout format                         = Prospero::ChannelLayout::kInvalid;
+	Prospero::ChannelType   channel_type                   = Prospero::ChannelType::kUNorm;
+	Prospero::ChannelOrder  channel_order                  = Prospero::ChannelOrder::kStandard;
 };
 
 struct ColorAttrib {
-	bool     force_dest_alpha_to_one = false;
-	uint32_t tile_mode               = 0;
-	uint32_t fmask_tile_mode         = 0;
-	uint32_t num_samples             = 0;
-	uint32_t num_fragments           = 0;
+	bool               force_dest_alpha_to_one = false;
+	Prospero::TileMode tile_mode               = Prospero::TileMode::kLinear;
+	Prospero::TileMode fmask_tile_mode         = Prospero::TileMode::kLinear;
+	uint32_t           num_samples             = 0;
+	uint32_t           num_fragments           = 0;
 };
 
 struct ColorAttrib2 {
@@ -59,11 +59,11 @@ struct ColorAttrib2 {
 };
 
 struct ColorAttrib3 {
-	uint32_t depth              = 0;
-	uint32_t tile_mode          = 0;
-	uint32_t dimension          = 0;
-	bool     cmask_pipe_aligned = false;
-	bool     dcc_pipe_aligned   = false;
+	uint32_t           depth              = 0;
+	Prospero::TileMode tile_mode          = Prospero::TileMode::kLinear;
+	uint32_t           dimension          = 0;
+	bool               cmask_pipe_aligned = false;
+	bool               dcc_pipe_aligned   = false;
 };
 
 struct ColorDccControl {
@@ -132,7 +132,7 @@ struct RenderTarget {
 };
 
 struct DepthZInfo {
-	uint32_t                                    format      = 0;
+	Prospero::DepthFormat                       format      = Prospero::DepthFormat::kInvalid;
 	uint32_t                                    num_samples = 0;
 	Prospero::TextureCompatiblePlaneCompression texture_compatibility =
 	    Prospero::TextureCompatiblePlaneCompression::kDisable;
@@ -144,16 +144,16 @@ struct DepthZInfo {
 
 	[[nodiscard]] static DepthZInfo Decode(uint32_t value) {
 		DepthZInfo info;
-		info.format                = value & 0x3u;
+		info.format                = static_cast<Prospero::DepthFormat>(value & 0x3u);
 		info.num_samples           = (value >> 2u) & 0x3u;
 		info.texture_compatibility = static_cast<Prospero::TextureCompatiblePlaneCompression>(
-		    value & Prospero::GpuEnumValue(Prospero::TextureCompatiblePlaneCompression::kBitMask));
+		    value & static_cast<uint32_t>(Prospero::TextureCompatiblePlaneCompression::kBitMask));
 		info.partially_resident = (value & 0x00001000u) != 0;
 		info.max_mip_level      = static_cast<uint8_t>((value >> 16u) & 0x0fu);
 		info.expclear_enabled   = (value & 0x08000000u) != 0;
 		info.htile_acceleration = (value & 0x20000000u) != 0;
 		info.z_compare_base     = static_cast<Prospero::ZCompareBase>(
-		    value & Prospero::GpuEnumValue(Prospero::ZCompareBase::kBitMask));
+		    value & static_cast<uint32_t>(Prospero::ZCompareBase::kBitMask));
 		return info;
 	}
 
@@ -167,7 +167,7 @@ struct DepthZInfo {
 };
 
 struct DepthStencilInfo {
-	uint32_t                           format = 0;
+	Prospero::StencilFormat            format = Prospero::StencilFormat::kInvalid;
 	Prospero::TextureCompatibleStencil texture_compatibility =
 	    Prospero::TextureCompatibleStencil::kDisable;
 	bool expclear_enabled       = false;
@@ -176,9 +176,9 @@ struct DepthStencilInfo {
 
 	[[nodiscard]] static DepthStencilInfo Decode(uint32_t value) {
 		DepthStencilInfo info;
-		info.format                = value & 0x1u;
+		info.format                = static_cast<Prospero::StencilFormat>(value & 0x1u);
 		info.texture_compatibility = static_cast<Prospero::TextureCompatibleStencil>(
-		    value & Prospero::GpuEnumValue(Prospero::TextureCompatibleStencil::kBitMask));
+		    value & static_cast<uint32_t>(Prospero::TextureCompatibleStencil::kBitMask));
 		info.partially_resident     = (value & 0x00001000u) != 0;
 		info.expclear_enabled       = (value & 0x08000000u) != 0;
 		info.htile_stencil_disabled = (value & 0x20000000u) != 0;
@@ -295,10 +295,7 @@ struct ClipControl {
 	bool    linear_attribute_clip_enable        = false;
 	bool    force_viewport_index_from_vs_enable = false;
 
-	[[nodiscard]] bool IsZClipModeRepresentable() const {
-		return min_z_clip_disable == max_z_clip_disable;
-	}
-	[[nodiscard]] bool IsZClipEnabled() const { return !min_z_clip_disable; }
+	[[nodiscard]] bool IsZClipEnabled() const { return !min_z_clip_disable && !max_z_clip_disable; }
 };
 
 struct DepthControl {
@@ -941,8 +938,10 @@ public:
 	[[nodiscard]] uint8_t GetStencilClearValue() const { return m_stencil_clear_value; }
 	void SetStencilClearValue(uint8_t clear_value) { m_stencil_clear_value = clear_value; }
 
-	[[nodiscard]] float GetLineWidth() const { return m_line_width; }
-	void                SetLineWidth(float width) { m_line_width = width; }
+	[[nodiscard]] float    GetLineWidth() const { return m_line_width; }
+	void                   SetLineWidth(float width) { m_line_width = width; }
+	[[nodiscard]] uint32_t GetPrimitiveResetIndex() const { return m_primitive_reset_index; }
+	void SetPrimitiveResetIndex(uint32_t index) { m_primitive_reset_index = index; }
 
 	[[nodiscard]] const ShaderRegisters& GetShaderRegisters() const { return m_sh_regs; }
 
@@ -987,7 +986,8 @@ public:
 	void SetScShaderControl(uint32_t value) { m_sh_regs.m_paScShaderControl = value; }
 
 private:
-	float m_line_width = 1.0f;
+	float    m_line_width            = 1.0f;
+	uint32_t m_primitive_reset_index = 0xffffffffu;
 
 	BlendControl    m_blend_control[8];
 	BlendColor      m_blend_color;
@@ -1029,12 +1029,14 @@ public:
 
 	void Reset() { *this = UserConfig(); }
 
-	void                   SetPrimitiveType(uint32_t prim_type) { m_prim_type = prim_type; }
-	[[nodiscard]] uint32_t GetPrimType() const { return m_prim_type; }
+	void SetPrimitiveType(Prospero::PrimitiveType prim_type) { m_prim_type = prim_type; }
+	[[nodiscard]] Prospero::PrimitiveType GetPrimType() const { return m_prim_type; }
 	void                   SetIndexOffset(uint32_t index_offset) { m_index_offset = index_offset; }
 	[[nodiscard]] uint32_t GetIndexOffset() const { return m_index_offset; }
 	void                   SetObjectId(uint32_t object_id) { m_object_id = object_id; }
 	[[nodiscard]] uint32_t GetObjectId() const { return m_object_id; }
+	void SetPrimitiveResetControl(uint32_t control) { m_primitive_reset_control = control; }
+	[[nodiscard]] uint32_t GetPrimitiveResetControl() const { return m_primitive_reset_control; }
 
 	[[nodiscard]] const GeControl& GetGeControl() const { return m_ge_cntl; }
 	void                           SetGeControl(const GeControl& control) { m_ge_cntl = control; }
@@ -1049,9 +1051,10 @@ public:
 	void SetGdsOaAddress(uint32_t value) { m_gds_oa.counters[m_gds_oa.GetIndex()].address = value; }
 
 private:
-	uint32_t m_prim_type    = 0;
-	uint32_t m_index_offset = 0;
-	uint32_t m_object_id    = 0;
+	Prospero::PrimitiveType m_prim_type               = Prospero::PrimitiveType::kNone;
+	uint32_t                m_index_offset            = 0;
+	uint32_t                m_object_id               = 0;
+	uint32_t                m_primitive_reset_control = 0;
 
 	GeControl    m_ge_cntl;
 	GeUserVgprEn m_ge_user_vgpr_en;
@@ -1086,6 +1089,7 @@ public:
 	void SetPsShaderChksum(uint32_t value) { m_ps.ps_regs.chksum = value; }
 
 	void SetCsShader(const CsStageRegisters& cs_regs) { m_cs.cs_regs = cs_regs; }
+	void SetCsWaveSize(uint8_t wave_size) { m_cs.cs_regs.wave_size = wave_size; }
 
 	void SetPsUserSgpr(uint32_t id, uint32_t value, UserSgprType type) {
 		m_ps.ps_user_sgpr.value[id] = value;

src/graphics/guest_gpu/pm4.cpp

@@ -21,12 +21,10 @@ constexpr auto MakeRegisterNames() {
 	names[R_DRAW_RESET]     = "R_DRAW_RESET";
 	names[R_WAIT_FLIP_DONE] = "R_WAIT_FLIP_DONE";
 	names[R_DISPATCH_RESET] = "R_DISPATCH_RESET";
-	names[R_WAIT_MEM_32]    = "R_WAIT_MEM_32";
 	names[R_PUSH_MARKER]    = "R_PUSH_MARKER";
 	names[R_POP_MARKER]     = "R_POP_MARKER";
 	names[R_ACQUIRE_MEM]    = "R_ACQUIRE_MEM";
 	names[R_WRITE_DATA]     = "R_WRITE_DATA";
-	names[R_WAIT_MEM_64]    = "R_WAIT_MEM_64";
 	names[R_FLIP]           = "R_FLIP";
 	names[R_RELEASE_MEM]    = "R_RELEASE_MEM";
 	names[R_DMA_DATA]       = "R_DMA_DATA";
@@ -62,6 +60,7 @@ constexpr auto MakeOpcodeNames() {
 	names[IT_WRITE_DATA]                = "IT_WRITE_DATA";
 	names[IT_MEM_SEMAPHORE]             = "IT_MEM_SEMAPHORE";
 	names[IT_DRAW_INDEX_INDIRECT_MULTI] = "IT_DRAW_INDEX_INDIRECT_MULTI";
+	names[IT_WAIT_REG_MEM]               = "IT_WAIT_REG_MEM";
 	names[IT_INDIRECT_BUFFER]           = "IT_INDIRECT_BUFFER";
 	names[IT_COPY_DATA]                 = "IT_COPY_DATA";
 	names[IT_CP_DMA]                    = "IT_CP_DMA";
@@ -91,6 +90,7 @@ constexpr auto MakeOpcodeNames() {
 	names[IT_DISPATCH_DRAW_PREAMBLE]    = "IT_DISPATCH_DRAW_PREAMBLE";
 	names[IT_DISPATCH_DRAW]             = "IT_DISPATCH_DRAW";
 	names[IT_GET_LOD_STATS]             = "IT_GET_LOD_STATS";
+	names[IT_WAIT_REG_MEM_64]           = "IT_WAIT_REG_MEM_64";
 	names[IT_SET_CONTEXT_REG_INDIRECT]  = "IT_SET_CONTEXT_REG_INDIRECT";
 	return names;
 }

src/graphics/guest_gpu/pm4.h

@@ -42,6 +42,7 @@ constexpr uint32_t IT_WRITE_DATA                = 0x37;
 constexpr uint32_t IT_MEM_SEMAPHORE             = 0x39;
 constexpr uint32_t IT_DRAW_INDEX_INDIRECT_MULTI = 0x38;
 constexpr uint32_t IT_DISPATCH_DRAW_PREAMBLE    = 0x3A;
+constexpr uint32_t IT_WAIT_REG_MEM              = 0x3C;
 constexpr uint32_t IT_INDIRECT_BUFFER           = 0x3F;
 constexpr uint32_t IT_COPY_DATA                 = 0x40;
 constexpr uint32_t IT_CP_DMA                    = 0x41;
@@ -70,6 +71,7 @@ constexpr uint32_t IT_WAIT_ON_CE_COUNTER        = 0x86;
 constexpr uint32_t IT_WAIT_ON_DE_COUNTER_DIFF   = 0x88;
 constexpr uint32_t IT_DISPATCH_DRAW             = 0x8D;
 constexpr uint32_t IT_GET_LOD_STATS             = 0x8E;
+constexpr uint32_t IT_WAIT_REG_MEM_64           = 0x93;
 constexpr uint32_t IT_SET_CONTEXT_REG_INDIRECT  = 0x9F;
 
 /* Custom codes. Implemented via IT_NOP */
@@ -78,12 +80,10 @@ constexpr uint32_t R_ZERO           = 0x00;
 constexpr uint32_t R_DRAW_RESET     = 0x05;
 constexpr uint32_t R_WAIT_FLIP_DONE = 0x06;
 constexpr uint32_t R_DISPATCH_RESET = 0x09;
-constexpr uint32_t R_WAIT_MEM_32    = 0x0A;
 constexpr uint32_t R_PUSH_MARKER    = 0x0B;
 constexpr uint32_t R_POP_MARKER     = 0x0C;
 constexpr uint32_t R_ACQUIRE_MEM    = 0x14;
 constexpr uint32_t R_WRITE_DATA     = 0x15;
-constexpr uint32_t R_WAIT_MEM_64    = 0x16;
 constexpr uint32_t R_FLIP           = 0x17;
 constexpr uint32_t R_RELEASE_MEM    = 0x18;
 constexpr uint32_t R_DMA_DATA       = 0x19;
@@ -317,8 +317,8 @@ constexpr uint32_t FSR_RECURSIONS1              = 0xF2;
 constexpr uint32_t VGT_MULTI_PRIM_IB_RESET_INDX = 0x103;
 constexpr uint32_t CB_RMI_GL2_CACHE_CONTROL     = 0x104;
 constexpr uint32_t CB_BLEND_RED                 = 0x105;
-constexpr uint32_t CB_BLEND_GREEN               = 0x106;
-constexpr uint32_t CB_BLEND_BLUE                = 0x107;
+constexpr uint32_t CB_BLEND_BLUE                = 0x106;
+constexpr uint32_t CB_BLEND_GREEN               = 0x107;
 constexpr uint32_t CB_BLEND_ALPHA               = 0x108;
 constexpr uint32_t CB_DCC_CONTROL               = 0x109;
 
@@ -387,9 +387,6 @@ constexpr uint32_t SPI_SHADER_POS_FORMAT  = 0x1C3;
 constexpr uint32_t SPI_SHADER_Z_FORMAT    = 0x1C4;
 constexpr uint32_t SPI_SHADER_COL_FORMAT  = 0x1C5;
 
-// Indirect Cx descriptor selector for the 32-entry PS input-control register bank.
-constexpr uint32_t CX_PS_SHADER_USAGE_BASE = 0x10000000u;
-
 constexpr uint32_t CB_BLEND0_CONTROL                            = 0x1E0;
 constexpr uint32_t CB_BLEND0_CONTROL_COLOR_SRCBLEND_SHIFT       = 0;
 constexpr uint32_t CB_BLEND0_CONTROL_COLOR_SRCBLEND_MASK        = 0x1F;
@@ -984,8 +981,16 @@ constexpr uint32_t COMPUTE_PGM_RSRC1_FP16_OVFL_SHIFT  = 26;
 constexpr uint32_t COMPUTE_PGM_RSRC1_FP16_OVFL_MASK   = 0x1;
 constexpr uint32_t COMPUTE_PGM_RSRC1_WGP_MODE_SHIFT   = 29;
 constexpr uint32_t COMPUTE_PGM_RSRC1_WGP_MODE_MASK    = 0x1;
-constexpr uint32_t COMPUTE_PGM_RSRC1_W32_EN_SHIFT     = 30;
-constexpr uint32_t COMPUTE_PGM_RSRC1_W32_EN_MASK      = 0x1;
+
+constexpr uint32_t COMPUTE_DISPATCH_INITIATOR_CS_W32_EN_SHIFT = 15;
+constexpr uint32_t COMPUTE_DISPATCH_INITIATOR_CS_W32_EN_MASK  = 0x1;
+
+[[nodiscard]] constexpr uint8_t ComputeWaveSize(uint32_t dispatch_initiator) {
+	return ((dispatch_initiator >> COMPUTE_DISPATCH_INITIATOR_CS_W32_EN_SHIFT) &
+	        COMPUTE_DISPATCH_INITIATOR_CS_W32_EN_MASK) != 0u
+	           ? 32u
+	           : 64u;
+}
 
 constexpr uint32_t COMPUTE_PGM_RSRC2                      = 0x213;
 constexpr uint32_t COMPUTE_PGM_RSRC2_SCRATCH_EN_SHIFT     = 0;

src/graphics/guest_gpu/tile.cpp

@@ -265,7 +265,7 @@ static bool GetMipTailLayout(const TileBlockLayout& block, MipTailLayout& out) {
 	return false;
 }
 
-bool TileGetTextureElementLayout(uint32_t format, TileTextureElementLayout& out) {
+bool TileGetTextureElementLayout(Prospero::BufferFormat format, TileTextureElementLayout& out) {
 	if (const auto bytes = Prospero::NumBytesPerElement(format); std::has_single_bit(bytes)) {
 		out = {bytes, 1, 1};
 		return true;
@@ -277,10 +277,10 @@ bool TileGetTextureElementLayout(uint32_t format, TileTextureElementLayout& out)
 	return false;
 }
 
-bool TileGetTextureBlockLayout(uint32_t format, uint32_t tile, bool volume,
+bool TileGetTextureBlockLayout(Prospero::BufferFormat format, Prospero::TileMode tile, bool volume,
                                TileTextureBlockLayout& out) {
 	TileBlockFamily family = TileBlockFamily::Count;
-	switch (static_cast<Prospero::TileMode>(tile)) {
+	switch (tile) {
 		case Prospero::TileMode::kStandard256B:
 			if (!volume) family = TileBlockFamily::Standard256B;
 			break;
@@ -1130,15 +1130,18 @@ static bool TileGetHtileSize(uint32_t width, uint32_t height, TileSizeAlign& hti
 	return true;
 }
 
-bool TileGetDepthSize(uint32_t width, uint32_t height, uint32_t pitch, uint32_t z_format,
-                      uint32_t stencil_format, bool htile, TileSizeAlign& stencil_size,
-                      TileSizeAlign& htile_size, TileSizeAlign& depth_size,
-                      uint32_t num_fragments_log2) {
+bool TileGetDepthSize(uint32_t width, uint32_t height, uint32_t pitch,
+                      Prospero::DepthFormat z_format, Prospero::StencilFormat stencil_format,
+                      bool htile, TileSizeAlign& stencil_size, TileSizeAlign& htile_size,
+                      TileSizeAlign& depth_size, uint32_t num_fragments_log2) {
 	EXIT_IF(pitch != 0);
 	// Prospero derives uncompressed depth/stencil as independent 64 KiB block surfaces.
 	if (width > 0 && width <= 16384 && height > 0 && height <= 16384 &&
-	    (z_format == 1 || z_format == 3) && stencil_format <= 1 && num_fragments_log2 <= 3) {
-		const uint32_t depth_bytes          = z_format == 1 ? 2u : 4u;
+	    (z_format == Prospero::DepthFormat::kZ16 || z_format == Prospero::DepthFormat::kZ32F) &&
+	    (stencil_format == Prospero::StencilFormat::kInvalid ||
+	     stencil_format == Prospero::StencilFormat::k8UInt) &&
+	    num_fragments_log2 <= 3) {
+		const uint32_t depth_bytes          = z_format == Prospero::DepthFormat::kZ16 ? 2u : 4u;
 		uint32_t       depth_block_width    = 0;
 		uint32_t       depth_block_height   = 0;
 		uint32_t       stencil_block_width  = 0;
@@ -1146,7 +1149,7 @@ bool TileGetDepthSize(uint32_t width, uint32_t height, uint32_t pitch, uint32_t
 		const bool     valid_blocks =
 		    Gen5Msaa64KBBlockSizeFromElementBytes(depth_bytes, num_fragments_log2,
 		                                          &depth_block_width, &depth_block_height) &&
-		    (stencil_format == 0 ||
+		    (stencil_format == Prospero::StencilFormat::kInvalid ||
 		     Gen5Msaa64KBBlockSizeFromElementBytes(1, num_fragments_log2, &stencil_block_width,
 		                                           &stencil_block_height));
 		const uint32_t fragments = 1u << num_fragments_log2;
@@ -1155,7 +1158,7 @@ bool TileGetDepthSize(uint32_t width, uint32_t height, uint32_t pitch, uint32_t
 		                       AlignUp(height, depth_block_height) * depth_bytes * fragments
 		                 : 0;
 		const uint64_t stencil_bytes_total =
-		    stencil_format == 1 && valid_blocks
+		    stencil_format == Prospero::StencilFormat::k8UInt && valid_blocks
 		        ? static_cast<uint64_t>(AlignUp(width, stencil_block_width)) *
 		              AlignUp(height, stencil_block_height) * fragments
 		        : 0;
@@ -1163,7 +1166,7 @@ bool TileGetDepthSize(uint32_t width, uint32_t height, uint32_t pitch, uint32_t
 		const bool    htile_valid = !htile || TileGetHtileSize(width, height, calculated_htile);
 		if (depth_bytes_total <= UINT32_MAX && stencil_bytes_total <= UINT32_MAX && htile_valid) {
 			depth_size   = {static_cast<uint32_t>(depth_bytes_total), 65536};
-			stencil_size = stencil_format == 1
+			stencil_size = stencil_format == Prospero::StencilFormat::k8UInt
 			                   ? TileSizeAlign {static_cast<uint32_t>(stencil_bytes_total), 65536}
 			                   : TileSizeAlign {};
 			htile_size   = calculated_htile;
@@ -1236,31 +1239,29 @@ bool TileGetRenderTargetMipLayout(uint32_t width, uint32_t height, uint32_t pitc
 	if (levels > max_levels) {
 		return false;
 	}
-	uint32_t format = 0;
+	auto format = Prospero::BufferFormat::kInvalid;
 	switch (bytes_per_element) {
-		case 1: format = Prospero::GpuEnumValue(Prospero::BufferFormat::k8UNorm); break;
-		case 2: format = Prospero::GpuEnumValue(Prospero::BufferFormat::k16UNorm); break;
-		case 4: format = Prospero::GpuEnumValue(Prospero::BufferFormat::k32Float); break;
-		case 8: format = Prospero::GpuEnumValue(Prospero::BufferFormat::k16_16_16_16Float); break;
-		case 16: format = Prospero::GpuEnumValue(Prospero::BufferFormat::k32_32_32_32Float); break;
+		case 1: format = Prospero::BufferFormat::k8UNorm; break;
+		case 2: format = Prospero::BufferFormat::k16UNorm; break;
+		case 4: format = Prospero::BufferFormat::k32Float; break;
+		case 8: format = Prospero::BufferFormat::k16_16_16_16Float; break;
+		case 16: format = Prospero::BufferFormat::k32_32_32_32Float; break;
 		default: return false;
 	}
-	TileGetTextureSize(format, width, height, levels,
-	                   Prospero::GpuEnumValue(Prospero::TileMode::kRenderTarget), &total_size,
-	                   level_sizes, padded_size);
+	TileGetTextureSize(format, width, height, levels, Prospero::TileMode::kRenderTarget,
+	                   &total_size, level_sizes, padded_size);
 	return total_size.size != 0 && total_size.align == 65536;
 }
 
-void TileGetTextureSize(uint32_t format, uint32_t width, uint32_t height, uint32_t levels,
-                        uint32_t tile, TileSizeAlign* total_size, TileSizeOffset* level_sizes,
-                        TilePaddedSize* padded_size) {
+void TileGetTextureSize(Prospero::BufferFormat format, uint32_t width, uint32_t height,
+                        uint32_t levels, Prospero::TileMode tile, TileSizeAlign* total_size,
+                        TileSizeOffset* level_sizes, TilePaddedSize* padded_size) {
 	KYTY_PROFILER_FUNCTION();
 
 	EXIT_IF(levels == 0 || levels > 16);
 
 	TileTextureElementLayout element {};
-	if (static_cast<Prospero::TileMode>(tile) == Prospero::TileMode::kLinear &&
-	    TileGetTextureElementLayout(format, element)) {
+	if (tile == Prospero::TileMode::kLinear && TileGetTextureElementLayout(format, element)) {
 		uint32_t mip_pitch[16] {};
 		uint32_t mip_height[16] {};
 		uint32_t mip_size[16] {};
@@ -1302,26 +1303,27 @@ void TileGetTextureSize(uint32_t format, uint32_t width, uint32_t height, uint32
 	}
 	if (total_size != nullptr && total_size->size == 0) {
 		std::vector<std::string> list;
-		list.push_back(fmt::format("format = {}", format));
+		list.push_back(fmt::format("format = {}", static_cast<uint32_t>(format)));
 		list.push_back(fmt::format("width  = {}", width));
 		list.push_back(fmt::format("height = {}", height));
 		list.push_back(fmt::format("levels = {}", levels));
-		list.push_back(fmt::format("tile   = {}", tile));
+		list.push_back(fmt::format("tile   = {}", static_cast<uint32_t>(tile)));
 		EXIT("unknown format:\n%s\n", Common::Concat(list, '\n').c_str());
 	}
 }
 
-void TileGetTextureTotalSize(uint32_t format, uint32_t width, uint32_t height, uint32_t depth,
-                             uint32_t levels, uint32_t tile, bool volume_texture,
-                             TileSizeAlign& total_size) {
+void TileGetTextureTotalSize(Prospero::BufferFormat format, uint32_t width, uint32_t height,
+                             uint32_t depth, uint32_t levels, Prospero::TileMode tile,
+                             bool volume_texture, TileSizeAlign& total_size) {
 	EXIT_NOT_IMPLEMENTED(depth == 0);
-	if (volume_texture && static_cast<Prospero::TileMode>(tile) != Prospero::TileMode::kLinear) {
+	if (volume_texture && tile != Prospero::TileMode::kLinear) {
 		TileSurfaceLayout            layout {};
 		const TileSurfaceDescription description {
 		    format, tile, TileSurfaceDimension::Dim3D, width, height, depth, levels, 1};
 		if (!TileGetTiledTextureLayout(description, layout)) {
 			EXIT("unsupported 3D texture layout: format=%u tile=%u extent=%ux%ux%u levels=%u\n",
-			     format, tile, width, height, depth, levels);
+			     static_cast<uint32_t>(format), static_cast<uint32_t>(tile), width, height, depth,
+			     levels);
 		}
 		EXIT_NOT_IMPLEMENTED(layout.total_size > UINT32_MAX);
 		total_size.size  = static_cast<uint32_t>(layout.total_size);
@@ -1337,9 +1339,10 @@ void TileGetTextureTotalSize(uint32_t format, uint32_t width, uint32_t height, u
 	total_size.size = static_cast<uint32_t>(total);
 }
 
-uint32_t TileGetTexturePitch(uint32_t format, uint32_t width, uint32_t tile) {
+uint32_t TileGetTexturePitch(Prospero::BufferFormat format, uint32_t width,
+                             Prospero::TileMode tile) {
 	uint32_t pitch = width;
-	switch (static_cast<Prospero::TileMode>(tile)) {
+	switch (tile) {
 		case Prospero::TileMode::kLinear: {
 			TileTextureElementLayout element {};
 			if (TileGetTextureElementLayout(format, element) && element.texel_width == 1 &&
@@ -1348,6 +1351,7 @@ uint32_t TileGetTexturePitch(uint32_t format, uint32_t width, uint32_t tile) {
 			}
 			break;
 		}
+		case Prospero::TileMode::kStandard4KB:
 		case Prospero::TileMode::kStandard64KB:
 		case Prospero::TileMode::kPrt:
 		case Prospero::TileMode::kDepth:

src/graphics/guest_gpu/tile.h

@@ -3,6 +3,7 @@
 
 #include "common/abi.h"
 #include "common/common.h"
+#include "graphics/guest_gpu/gpu_defs.h"
 
 namespace Libs::Graphics {
 
@@ -65,14 +66,14 @@ enum class TileSurfaceDimension : uint32_t {
 };
 
 struct TileSurfaceDescription {
-	uint32_t             format    = 0;
-	uint32_t             tile_mode = 0;
-	TileSurfaceDimension dimension = TileSurfaceDimension::Dim2D;
-	uint32_t             width     = 0;
-	uint32_t             height    = 0;
-	uint32_t             depth     = 1;
-	uint32_t             levels    = 1;
-	uint32_t             layers    = 1;
+	Prospero::BufferFormat format    = Prospero::BufferFormat::kInvalid;
+	Prospero::TileMode     tile_mode = Prospero::TileMode::kLinear;
+	TileSurfaceDimension   dimension = TileSurfaceDimension::Dim2D;
+	uint32_t               width     = 0;
+	uint32_t               height    = 0;
+	uint32_t               depth     = 1;
+	uint32_t               levels    = 1;
+	uint32_t               layers    = 1;
 };
 
 // Width/height fields are measured in addressable elements, not texture texels.
@@ -98,8 +99,8 @@ struct TileSurfaceLayout {
 
 bool TileGetBlockLayout(TileBlockFamily family, uint32_t bytes_per_element,
                         TileBlockLayout& layout);
-bool TileGetTextureElementLayout(uint32_t format, TileTextureElementLayout& layout);
-bool TileGetTextureBlockLayout(uint32_t format, uint32_t tile, bool volume,
+bool TileGetTextureElementLayout(Prospero::BufferFormat format, TileTextureElementLayout& layout);
+bool TileGetTextureBlockLayout(Prospero::BufferFormat format, Prospero::TileMode tile, bool volume,
                                TileTextureBlockLayout& layout);
 bool TileGetTiledTextureLayout(const TileSurfaceDescription& description,
                                TileSurfaceLayout&            layout);
@@ -110,10 +111,10 @@ bool TileGetBlockXor(const TileBlockLayout& layout, uint32_t block_x, uint32_t b
 bool TileGetBlockXor(const TileBlockLayout& layout, uint32_t block_x, uint32_t block_y,
                      uint32_t block_z, uint32_t& byte_offset);
 
-bool     TileGetDepthSize(uint32_t width, uint32_t height, uint32_t pitch, uint32_t z_format,
-                          uint32_t stencil_format, bool htile, TileSizeAlign& stencil_size,
-                          TileSizeAlign& htile_size, TileSizeAlign& depth_size,
-                          uint32_t num_fragments_log2 = 0);
+bool     TileGetDepthSize(uint32_t width, uint32_t height, uint32_t pitch,
+                          Prospero::DepthFormat z_format, Prospero::StencilFormat stencil_format,
+                          bool htile, TileSizeAlign& stencil_size, TileSizeAlign& htile_size,
+                          TileSizeAlign& depth_size, uint32_t num_fragments_log2 = 0);
 uint32_t TileGetRenderTargetPitch(uint32_t width, uint32_t bytes_per_element,
                                   uint32_t num_fragments_log2 = 0);
 uint32_t TileGetDepthPitch(uint32_t width, uint32_t bytes_per_element,
@@ -125,13 +126,14 @@ bool     TileGetRenderTargetMipLayout(uint32_t width, uint32_t height, uint32_t
                                       uint32_t bytes_per_element, uint32_t levels,
                                       TileSizeAlign& total_size, TileSizeOffset* level_sizes,
                                       TilePaddedSize* padded_size);
-void     TileGetTextureSize(uint32_t format, uint32_t width, uint32_t height, uint32_t levels,
-                            uint32_t tile, TileSizeAlign* total_size, TileSizeOffset* level_sizes,
-                            TilePaddedSize* padded_size);
-void     TileGetTextureTotalSize(uint32_t format, uint32_t width, uint32_t height, uint32_t depth,
-                                 uint32_t levels, uint32_t tile, bool volume_texture,
-                                 TileSizeAlign& total_size);
-uint32_t TileGetTexturePitch(uint32_t format, uint32_t width, uint32_t tile);
+void     TileGetTextureSize(Prospero::BufferFormat format, uint32_t width, uint32_t height,
+                            uint32_t levels, Prospero::TileMode tile, TileSizeAlign* total_size,
+                            TileSizeOffset* level_sizes, TilePaddedSize* padded_size);
+void     TileGetTextureTotalSize(Prospero::BufferFormat format, uint32_t width, uint32_t height,
+                                 uint32_t depth, uint32_t levels, Prospero::TileMode tile,
+                                 bool volume_texture, TileSizeAlign& total_size);
+uint32_t TileGetTexturePitch(Prospero::BufferFormat format, uint32_t width,
+                             Prospero::TileMode tile);
 
 } // namespace Libs::Graphics
 

src/graphics/host_gpu/graphicContext.h

@@ -24,7 +24,6 @@ struct GraphicContext: public VulkanInstance {
 	[[nodiscard]] uint64_t GetDeviceMemoryUsage() const;
 	[[nodiscard]] uint64_t GetTotalMemoryBudget() const;
 	void                   CreateBuffer(uint64_t size, VulkanBuffer& buffer);
-	void                   DeleteBuffer(VulkanBuffer& buffer);
 	[[nodiscard]] bool     CreateImage(const vk::ImageCreateInfo& info, VulkanImage& image);
 	void                   DeleteImage(VulkanImage& image);
 	void                   MapMemory(VulkanMemory& memory, void*& data);

src/graphics/host_gpu/pageManager.cpp

@@ -1,5 +1,6 @@
 #include "graphics/host_gpu/pageManager.h"
 
+#include "common/platform/sysDbg.h"
 #include "graphics/host_gpu/regionDefinitions.h"
 #include "kernel/memory.h"
 
@@ -52,20 +53,20 @@ constexpr uint32_t READ_WRITE_PROTECTION = PAGE_READWRITE;
 	std::fputs("PageManager fail-fast: ", stderr);
 	std::fputs(reason != nullptr ? reason : "invalid page state", stderr);
 	std::fputc('\n', stderr);
+#if !defined(__APPLE__)
+	void* frames[16] {};
+	int   frame_count = static_cast<int>(std::size(frames));
+	SysStackWalk(frames, &frame_count);
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
-	void*      frames[16] {};
-	const auto frame_count =
-	    CaptureStackBackTrace(0, static_cast<DWORD>(std::size(frames)), frames, nullptr);
 	const auto image_base = reinterpret_cast<uintptr_t>(GetModuleHandleW(nullptr));
-	for (uint16_t i = 0; i < frame_count; i++) {
+	for (int i = 0; i < frame_count; i++) {
 		const auto address = reinterpret_cast<uintptr_t>(frames[i]);
-		std::fprintf(stderr, "  frame[%u]=0x%016" PRIxPTR " image_rva=0x%016" PRIxPTR "\n", i,
+		std::fprintf(stderr, "  frame[%d]=0x%016" PRIxPTR " image_rva=0x%016" PRIxPTR "\n", i,
 		             address, address >= image_base ? address - image_base : 0);
 	}
-#elif !defined(__APPLE__)
-	void*     frames[16] {};
-	const int frame_count = ::backtrace(frames, static_cast<int>(std::size(frames)));
+#else
 	::backtrace_symbols_fd(frames, frame_count, STDERR_FILENO);
+#endif
 #endif
 	std::fflush(stderr);
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS

src/graphics/host_gpu/renderer/cache/bufferCache.cpp

@@ -9,6 +9,7 @@
 #include "graphics/host_gpu/renderer/commandScheduler.h"
 #include "graphics/host_gpu/renderer/render.h"
 #include "graphics/host_gpu/renderer/renderContext.h"
+#include "graphics/host_gpu/vulkanCommon.h"
 #include "kernel/memory.h"
 
 #include <algorithm>
@@ -27,60 +28,25 @@ constexpr uint64_t GdsBufferSize = 64 * 1024;
 
 } // namespace
 
-uint64_t BufferCache::AlignDown(uint64_t value) noexcept {
-	return value & ~(CACHING_PAGE_SIZE - 1);
-}
-
-uint64_t BufferCache::AlignUp(uint64_t value) {
-	if (value > UINT64_MAX - (CACHING_PAGE_SIZE - 1)) {
-		EXIT("BufferCache: address alignment overflow\n");
-	}
-	return (value + CACHING_PAGE_SIZE - 1) & ~(CACHING_PAGE_SIZE - 1);
-}
-
-void BufferCache::Upload(CommandBuffer& command, Buffer& destination, uint64_t destination_offset,
-                         const void* source, uint64_t size) {
+void BufferCache::WriteDataBuffer(Buffer& buffer, uint64_t address, const void* source,
+                                  uint64_t size) {
 	auto* bytes = static_cast<const uint8_t*>(source);
 	while (size != 0) {
-		const auto chunk        = std::min(size, m_staging_buffer.Size());
-		const auto stage_offset = m_staging_buffer.Copy(bytes, chunk, 4);
-		destination.CopyFrom(command, m_staging_buffer, stage_offset, destination_offset, chunk,
-		                     vk::AccessFlagBits::eHostWrite);
+		const auto chunk  = std::min(size, m_staging_buffer.Size());
+		const auto offset = m_staging_buffer.Copy(bytes, chunk, 4);
+		buffer.CopyFrom(m_scheduler.Current(), m_staging_buffer, offset, buffer.Offset(address),
+		                chunk, vk::AccessFlagBits::eHostWrite);
 		bytes += chunk;
-		destination_offset += chunk;
+		address += chunk;
 		size -= chunk;
 	}
 }
 
-bool BufferCache::ResolveOverlap(CacheRange& merged, CacheRange candidate) noexcept {
-	if (merged.address == 0 || merged.size == 0 || candidate.address == 0 || candidate.size == 0 ||
-	    merged.size > UINT64_MAX - merged.address ||
-	    candidate.size > UINT64_MAX - candidate.address) {
-		EXIT("BufferCache: invalid overlap-merge range\n");
-	}
-	const auto merged_end    = merged.address + merged.size;
-	const auto candidate_end = candidate.address + candidate.size;
-	if (merged.address >= candidate_end || candidate.address >= merged_end) {
-		return false;
-	}
-	const auto address = std::min(merged.address, candidate.address);
-	const auto end     = std::max(merged_end, candidate_end);
-	merged             = {.address = address, .size = end - address};
-	return true;
-}
-
-struct BufferCache::CachedBuffer {
-	uint64_t                vaddr = 0;
-	uint64_t                size  = 0;
-	std::shared_ptr<Buffer> buffer;
-	uint64_t                tick_accessed_last = 0;
-};
-
 struct BufferCache::DownloadCopy {
-	std::shared_ptr<Buffer> owner;
-	uint64_t                source_offset = 0;
-	uint64_t                address       = 0;
-	uint64_t                size          = 0;
+	Buffer*  buffer        = nullptr;
+	uint64_t source_offset = 0;
+	uint64_t address       = 0;
+	uint64_t size          = 0;
 };
 
 struct BufferCache::DownloadRange {
@@ -89,15 +55,37 @@ struct BufferCache::DownloadRange {
 	uint64_t offset  = 0;
 };
 
-struct BufferCache::RetiredBuffer {
-	uint64_t                address = 0;
-	uint64_t                size    = 0;
-	std::shared_ptr<Buffer> owner;
-};
+void BufferCache::Unregister(BufferId id) {
+	auto& buffer = m_slot_buffers[id];
+	if (buffer.is_deleted) {
+		return;
+	}
+	const auto found = m_buffers.find(buffer.CpuAddress());
+	EXIT_IF(found == m_buffers.end() || found->second != id);
+	m_buffers.erase(found);
+	if (buffer.Size() > m_total_used_memory) {
+		EXIT("BufferCache: allocation accounting underflow\n");
+	}
+	m_total_used_memory -= buffer.Size();
+	buffer.is_deleted = true;
+}
+
+void BufferCache::DeleteBuffer(BufferId id) {
+	auto* buffer = m_slot_buffers.try_get(id);
+	if (buffer == nullptr || buffer->is_deleted) {
+		return;
+	}
+	Unregister(id);
+	if (m_scheduler.Active()) {
+		m_scheduler.DeferOperation([this, id] { m_slot_buffers.erase(id); });
+	} else {
+		m_slot_buffers.erase(id);
+	}
+}
 
 std::pair<uint64_t, uint64_t> BufferCache::DownloadEnvelope(const DownloadCopy& copy) {
-	if (copy.owner == nullptr || copy.size == 0 || copy.source_offset > copy.owner->Size() ||
-	    copy.size > copy.owner->Size() - copy.source_offset) {
+	if (copy.buffer == nullptr || copy.size == 0 || copy.source_offset > copy.buffer->Size() ||
+	    copy.size > copy.buffer->Size() - copy.source_offset) {
 		EXIT("BufferCache: invalid download copy\n");
 	}
 	const auto begin = copy.source_offset & ~uint64_t {3};
@@ -106,7 +94,7 @@ std::pair<uint64_t, uint64_t> BufferCache::DownloadEnvelope(const DownloadCopy&
 		EXIT("BufferCache: download copy alignment overflow\n");
 	}
 	const auto end = (copy.source_offset + copy.size + 3) & ~uint64_t {3};
-	if (end > copy.owner->Size()) {
+	if (end > copy.buffer->Size()) {
 		EXIT("BufferCache: aligned download copy exceeds its owner\n");
 	}
 	return {begin, end - begin};
@@ -143,7 +131,7 @@ BufferCache::RecordDownloads(std::span<const DownloadCopy> copies) {
 	for (const auto& copy: copies) {
 		const auto [source_begin, envelope_size] = DownloadEnvelope(copy);
 		const auto prefix                        = copy.source_offset - source_begin;
-		download.CopyFrom(m_scheduler.Current(), *copy.owner, source_begin, base_offset + cursor,
+		download.CopyFrom(m_scheduler.Current(), *copy.buffer, source_begin, base_offset + cursor,
 		                  envelope_size, vk::AccessFlagBits::eMemoryWrite, vk::AccessFlags {},
 		                  vk::AccessFlagBits::eMemoryRead | vk::AccessFlagBits::eMemoryWrite,
 		                  vk::AccessFlagBits::eHostRead);
@@ -162,27 +150,29 @@ void BufferCache::PublishDownloads(std::span<const DownloadRange> downloads) {
 	}
 }
 
-void BufferCache::QueueGarbageDownload(std::span<const DownloadCopy> copies, RetiredBuffer retire) {
-	if (copies.empty()) {
-		return;
-	}
+void BufferCache::QueueGarbageDownload(std::span<const DownloadCopy> copies, BufferId id,
+                                       uint64_t address, uint64_t size) {
+	EXIT_IF(copies.empty());
 	auto downloads = RecordDownloads(copies);
-	m_scheduler.DeferOperation(
-	    [this, downloads = std::move(downloads), retire = std::move(retire)]() mutable {
-		    PublishDownloads(downloads);
-		    for (const auto& range: downloads) {
-			    m_gpu_modified_ranges.Subtract(range.address, range.size);
-		    }
-		    // ForEachDownloadRange reports full tracker pages, and every exact GPU-owned
-		    // interval on those pages was downloaded and removed. Clearing the original
-		    // query therefore cannot orphan a dirty sibling on an edge page.
-		    m_memory_tracker.UnmarkRegionAsGpuModified(retire.address, retire.size);
-		    if (m_memory_tracker.IsRegionGpuModified(retire.address, retire.size) ||
-		        !m_gpu_modified_ranges.Intersections(retire.address, retire.size).empty()) {
-			    EXIT("BufferCache: asynchronous garbage collection retained GPU ownership\n");
-		    }
-		    m_memory_tracker.UntrackMemory(retire.address, retire.size);
-	    });
+	auto retired   = downloads;
+	m_scheduler.DeferPriorityOperation(
+	    [this, downloads = std::move(downloads)]() mutable { PublishDownloads(downloads); });
+	m_scheduler.DeferOperation([this, retired = std::move(retired), id, address, size]() mutable {
+		for (const auto& range: retired) {
+			m_gpu_modified_ranges.Subtract(range.address, range.size);
+		}
+		// ForEachDownloadRange reports full tracker pages, and every exact GPU-owned
+		// interval on those pages was downloaded and removed. Clearing the original
+		// query therefore cannot orphan a dirty sibling on an edge page.
+		m_memory_tracker.UnmarkRegionAsGpuModified(address, size);
+		if (m_memory_tracker.IsRegionGpuModified(address, size) ||
+		    !m_gpu_modified_ranges.Intersections(address, size).empty()) {
+			EXIT("BufferCache: asynchronous garbage collection retained GPU ownership\n");
+		}
+		m_memory_tracker.UntrackMemory(address, size);
+		EXIT_IF(!m_slot_buffers[id].is_deleted);
+		m_slot_buffers.erase(id);
+	});
 }
 
 BufferCache::BufferCache(GraphicContext& graphics, CommandScheduler& scheduler,
@@ -197,6 +187,10 @@ BufferCache::BufferCache(GraphicContext& graphics, CommandScheduler& scheduler,
       m_texture_cache(texture_cache) {
 	std::memset(m_gds_buffer.Mapped().data(), 0, static_cast<size_t>(m_gds_buffer.Size()));
 	m_gds_buffer.Flush(0, m_gds_buffer.Size());
+	const auto null_id =
+	    m_slot_buffers.insert(m_graphics, m_scheduler, MemoryUsage::DeviceLocal, 0, AllFlags, 16);
+	EXIT_IF(null_id != NULL_BUFFER_ID);
+	SetVulkanObjectNameF(m_graphics.device, GetBuffer(null_id).Handle(), "Kyty.NullBuffer");
 	if (!m_graphics.CanReportMemoryUsage()) {
 		return;
 	}
@@ -215,38 +209,16 @@ BufferCache::~BufferCache() {
 	if (!m_gpu_modified_ranges.Empty()) {
 		EXIT("BufferCache: destroyed with pending GPU-modified ranges\n");
 	}
-	for (const auto& [vaddr, cached]: m_buffers) {
+	for (const auto& [vaddr, id]: m_buffers) {
 		(void)vaddr;
-		if (m_memory_tracker.IsRegionGpuModified(cached->vaddr, cached->size)) {
+		const auto& buffer = m_slot_buffers[id];
+		if (m_memory_tracker.IsRegionGpuModified(buffer.CpuAddress(), buffer.Size())) {
 			EXIT("BufferCache: destroyed with GPU-modified buffer\n");
 		}
 	}
 	m_buffers.clear();
 }
 
-StreamBuffer& BufferCache::GetUtilityBuffer(MemoryUsage usage) noexcept {
-	switch (usage) {
-		case MemoryUsage::Upload: return m_staging_buffer;
-		case MemoryUsage::Stream: return m_stream_buffer;
-		case MemoryUsage::Download: return m_download_buffer;
-		case MemoryUsage::DeviceLocal: return m_device_buffer;
-	}
-	EXIT("BufferCache: invalid utility-buffer usage\n");
-}
-
-BufferBinding BufferCache::UploadTransient(const void* data, uint64_t size, uint64_t alignment) {
-	EXIT_IF(data == nullptr || size == 0);
-	if (auto [mapped, offset] = m_stream_buffer.Map(size, alignment, false); mapped != nullptr) {
-		std::memcpy(mapped, data, static_cast<size_t>(size));
-		m_stream_buffer.Commit();
-		return {{}, m_stream_buffer.Handle(), offset};
-	}
-	auto owner =
-	    std::make_shared<Buffer>(m_graphics, m_scheduler, MemoryUsage::Upload, 0, AllFlags, size);
-	owner->Write(0, data, size);
-	return {owner, owner->Handle(), 0};
-}
-
 void BufferCache::InvalidateMemory(uint64_t vaddr, uint64_t size) {
 	if (vaddr == 0 || size == 0 || vaddr >= TRACKER_ADDRESS_SIZE ||
 	    size > TRACKER_ADDRESS_SIZE - vaddr) {
@@ -257,11 +229,7 @@ void BufferCache::InvalidateMemory(uint64_t vaddr, uint64_t size) {
 }
 
 void BufferCache::ReadMemory(uint64_t vaddr, uint64_t size, bool is_write) {
-	if (Gpu::IsGpuThread()) {
-		ReadMemoryOnGpu(vaddr, size, is_write);
-		return;
-	}
-	if (CommandScheduler::InDeferredOperation()) {
+	if (!GuestGpu::IsGpuThread() && CommandScheduler::InDeferredOperation()) {
 		EXIT("unsupported buffer readback from an asynchronous GPU completion, "
 		     "addr=0x%016" PRIx64 " size=0x%016" PRIx64 "\n",
 		     vaddr, size);
@@ -291,14 +259,14 @@ void BufferCache::ReadMemoryOnGpu(uint64_t vaddr, uint64_t size, bool is_write)
 				    if (owner == m_buffers.begin()) {
 					    EXIT("BufferCache: invalidation readback has no buffer owner\n");
 				    }
-				    auto& cached = *std::prev(owner)->second;
-				    if (!cached.buffer->IsInBounds(copy_address, 1)) {
+				    auto& buffer = m_slot_buffers[std::prev(owner)->second];
+				    if (!buffer.IsInBounds(copy_address, 1)) {
 					    EXIT("BufferCache: invalidation readback is outside its buffer owner\n");
 				    }
-				    const auto copy_size =
-				        std::min(range.size - copied, cached.vaddr + cached.size - copy_address);
-				    copies.push_back({cached.buffer, cached.buffer->Offset(copy_address),
-				                      copy_address, copy_size});
+				    const auto copy_size = std::min(
+				        range.size - copied, buffer.CpuAddress() + buffer.Size() - copy_address);
+				    copies.push_back(
+				        {&buffer, buffer.Offset(copy_address), copy_address, copy_size});
 				    copied += copy_size;
 			    }
 		    }
@@ -333,38 +301,39 @@ void BufferCache::UnmapMemory(uint64_t vaddr, uint64_t size) {
 	std::vector<DownloadCopy>                  copies;
 	std::vector<std::pair<uint64_t, uint64_t>> modified_buffers;
 	std::vector<std::pair<uint64_t, uint64_t>> retired_buffers;
-	for (const auto& [begin, cached]: m_buffers) {
-		if (vaddr < begin + cached->size && begin < vaddr + size) {
-			retired_buffers.emplace_back(begin, cached->size);
+	for (const auto& [begin, id]: m_buffers) {
+		const auto& buffer = m_slot_buffers[id];
+		if (vaddr < begin + buffer.Size() && begin < vaddr + size) {
+			retired_buffers.emplace_back(begin, buffer.Size());
 		}
 	}
-	for (const auto& [begin, cached]: m_buffers) {
-		if (vaddr >= begin + cached->size || begin >= vaddr + size ||
-		    !m_memory_tracker.IsRegionGpuModified(begin, cached->size)) {
+	for (const auto& [begin, id]: m_buffers) {
+		const auto& buffer = m_slot_buffers[id];
+		if (vaddr >= begin + buffer.Size() || begin >= vaddr + size ||
+		    !m_memory_tracker.IsRegionGpuModified(begin, buffer.Size())) {
 			continue;
 		}
-		const auto dirty = m_gpu_modified_ranges.Intersections(begin, cached->size);
+		const auto dirty = m_gpu_modified_ranges.Intersections(begin, buffer.Size());
 		if (dirty.empty()) {
 			EXIT("BufferCache: GPU-modified buffer has no dirty ranges\n");
 		}
-		modified_buffers.emplace_back(begin, cached->size);
+		modified_buffers.emplace_back(begin, buffer.Size());
 	}
 	for (const auto& [begin, bytes]: modified_buffers) {
 		auto owner = m_buffers.find(begin);
-		if (owner == m_buffers.end() || owner->second->size != bytes) {
+		if (owner == m_buffers.end() || m_slot_buffers[owner->second].Size() != bytes) {
 			EXIT("BufferCache: unmap owner changed during collection\n");
 		}
-		auto& cached = *owner->second;
+		auto& buffer = m_slot_buffers[owner->second];
 		m_memory_tracker.ForEachDownloadRange<false>(
-		    begin, cached.size,
+		    begin, buffer.Size(),
 		    [&](uint64_t address, uint64_t bytes) noexcept {
 			    m_memory_tracker.ValidateGpuDirtyPages(m_gpu_modified_ranges, address, bytes,
 			                                           "unmap");
 		    },
 		    [&](uint64_t address, uint64_t bytes) noexcept {
 			    for (const auto& range: m_gpu_modified_ranges.Intersections(address, bytes)) {
-				    copies.push_back(
-				        {cached.buffer, range.address - begin, range.address, range.size});
+				    copies.push_back({&buffer, range.address - begin, range.address, range.size});
 			    }
 		    });
 	}
@@ -388,167 +357,146 @@ void BufferCache::UnmapMemory(uint64_t vaddr, uint64_t size) {
 		EXIT("BufferCache: unmap retained dirty byte ranges\n");
 	}
 	m_memory_tracker.UntrackMemory(vaddr, size);
-	for (auto it = m_buffers.begin(); it != m_buffers.end();) {
-		if (vaddr < it->first + it->second->size && it->first < vaddr + size) {
-			if (it->second->size > m_total_used_memory) {
-				EXIT("BufferCache: allocation accounting underflow\n");
-			}
-			m_total_used_memory -= it->second->size;
-			it = m_buffers.erase(it);
-		} else {
-			++it;
+	for (const auto& [begin, bytes]: retired_buffers) {
+		(void)bytes;
+		const auto found = m_buffers.find(begin);
+		if (found != m_buffers.end()) {
+			DeleteBuffer(found->second);
 		}
 	}
 }
 
-BufferCache::CachedBuffer& BufferCache::GetOrCreateBuffer(CommandBuffer& command, uint64_t vaddr,
-                                                          uint64_t size) {
-	const auto begin = AlignDown(vaddr);
-	const auto end   = AlignUp(vaddr + size);
-	auto       it    = m_buffers.upper_bound(vaddr);
-	if (it != m_buffers.begin()) {
-		auto previous = std::prev(it);
-		if (previous->second->buffer->IsInBounds(vaddr, size)) {
-			it = previous;
+BufferId BufferCache::FindBuffer(uint64_t vaddr, uint64_t size) {
+	if (vaddr == 0) {
+		return NULL_BUFFER_ID;
+	}
+	if (size == 0 || vaddr >= TRACKER_ADDRESS_SIZE || size > TRACKER_ADDRESS_SIZE - vaddr) {
+		EXIT("BufferCache: invalid buffer discovery request\n");
+	}
+	auto& command = m_scheduler.Current();
+	EXIT_IF(command.IsInvalid() || command.IsExecute());
+	auto       begin = vaddr & ~(CACHING_PAGE_SIZE - 1);
+	auto       end   = (vaddr + size + CACHING_PAGE_SIZE - 1) & ~(CACHING_PAGE_SIZE - 1);
+	const auto next  = m_buffers.upper_bound(vaddr);
+	if (next != m_buffers.begin()) {
+		const auto owner  = std::prev(next);
+		auto&      buffer = m_slot_buffers[owner->second];
+		if (buffer.IsInBounds(vaddr, size)) {
+			buffer.tick_accessed_last = m_gc_tick;
+			return owner->second;
 		}
 	}
-	if (it != m_buffers.end() && it->second->buffer->IsInBounds(vaddr, size)) {
-		it->second->tick_accessed_last = m_gc_tick;
-		return *it->second;
-	}
 
-	CacheRange merged {.address = begin, .size = end - begin};
-	using Iterator = decltype(m_buffers.begin());
-	std::vector<Iterator> overlaps;
+	auto merge = [&](const Buffer& buffer) {
+		const auto buffer_begin = buffer.CpuAddress();
+		const auto buffer_end   = buffer_begin + buffer.Size();
+		if (begin >= buffer_end || buffer_begin >= end) {
+			return false;
+		}
+		begin = std::min(begin, buffer_begin);
+		end   = std::max(end, buffer_end);
+		return true;
+	};
+	std::vector<BufferId> overlaps;
 	auto                  first = m_buffers.lower_bound(begin);
 	if (first != m_buffers.begin()) {
-		auto previous = std::prev(first);
-		if (ResolveOverlap(merged, {previous->second->vaddr, previous->second->size})) {
+		const auto previous = std::prev(first);
+		if (merge(m_slot_buffers[previous->second])) {
 			first = previous;
 		}
 	}
 	for (auto candidate = first; candidate != m_buffers.end(); ++candidate) {
-		if (candidate->first >= merged.address + merged.size) {
+		if (candidate->first >= end) {
 			break;
 		}
-		if (ResolveOverlap(merged, {candidate->second->vaddr, candidate->second->size})) {
-			overlaps.push_back(candidate);
+		if (merge(m_slot_buffers[candidate->second])) {
+			overlaps.push_back(candidate->second);
 		}
 	}
-	for (const auto overlap: overlaps) {
-		auto&                                      old = *overlap->second;
-		std::vector<std::pair<uint64_t, uint64_t>> uploads;
-		m_memory_tracker.ForEachUploadRange(
-		    old.vaddr, old.size, false,
-		    [&](uint64_t address, uint64_t bytes) noexcept {
-			    uploads.emplace_back(address, bytes);
-		    },
-		    [&]() noexcept {
-			    for (const auto& [address, bytes]: uploads) {
-				    Upload(command, *old.buffer, old.buffer->Offset(address),
-				           reinterpret_cast<const void*>(address), bytes);
-			    }
-		    });
-	}
 
-	auto cached                = std::make_unique<CachedBuffer>();
-	cached->vaddr              = merged.address;
-	cached->size               = merged.size;
-	cached->tick_accessed_last = m_gc_tick;
-	cached->buffer = std::make_shared<Buffer>(m_graphics, m_scheduler, MemoryUsage::DeviceLocal,
-	                                          merged.address, AllFlags, merged.size);
+	const auto id = m_slot_buffers.insert(m_graphics, m_scheduler, MemoryUsage::DeviceLocal, begin,
+	                                      AllFlags, end - begin);
+	auto&      buffer         = m_slot_buffers[id];
+	buffer.tick_accessed_last = m_gc_tick;
+	SetVulkanObjectNameF(m_graphics.device, buffer.Handle(),
+	                     "Kyty.GameBuffer[guest=0x{:016x} size=0x{:x}]", begin, end - begin);
 	for (const auto overlap: overlaps) {
-		const auto& old = *overlap->second;
-		cached->buffer->CopyFrom(command, *old.buffer, 0, old.vaddr - cached->vaddr, old.size);
-		command.RetainResourceUntilFence(old.buffer);
+		const auto& old = m_slot_buffers[overlap];
+		buffer.CopyFrom(command, old, 0, old.CpuAddress() - begin, old.Size());
 	}
 	for (const auto overlap: overlaps) {
-		if (overlap->second->size > m_total_used_memory) {
-			EXIT("BufferCache: allocation accounting underflow\n");
-		}
-		m_total_used_memory -= overlap->second->size;
-		m_buffers.erase(overlap);
+		DeleteBuffer(overlap);
 	}
-	m_total_used_memory += cached->size;
-	return *m_buffers.emplace(cached->vaddr, std::move(cached)).first->second;
+	m_total_used_memory += buffer.Size();
+	const auto [it, inserted] = m_buffers.emplace(begin, id);
+	(void)it;
+	EXIT_IF(!inserted);
+	return id;
 }
 
-BufferBinding BufferCache::ObtainBuffer(CommandBuffer& command, uint64_t vaddr, uint64_t size,
-                                        bool is_written, bool is_read, bool is_formatted) {
-	if (command.IsInvalid() || command.IsExecute()) {
-		EXIT("BufferCache: buffer request requires a recording command buffer\n");
-	}
-
-	if (is_read && !is_written && size <= CACHING_PAGE_SIZE &&
-	    !m_memory_tracker.IsRegionGpuModified(vaddr, size) &&
-	    m_memory_tracker.IsRegionCpuModified(vaddr, size)) {
-		const auto alignment = std::max<uint64_t>(
-		    m_graphics.physical_device_properties.limits.minUniformBufferOffsetAlignment, 1);
-		if (auto [mapped, offset] = m_stream_buffer.Map(size, alignment, false);
-		    mapped != nullptr) {
-			if (Libs::LibKernel::Memory::TryReadBacking(vaddr, mapped, size)) {
-				m_stream_buffer.Commit();
-				return {{}, m_stream_buffer.Handle(), offset};
-			}
-		} else {
-			auto owner = std::make_shared<Buffer>(m_graphics, m_scheduler, MemoryUsage::Upload, 0,
-			                                      AllFlags, size);
-			if (Libs::LibKernel::Memory::TryReadBacking(vaddr, owner->Mapped().data(), size)) {
-				owner->Flush(0, size);
-				return {owner, owner->Handle(), 0};
-			}
-		}
-	}
-
-	if (is_formatted && is_written) {
-		(void)m_texture_cache.InvalidateMemoryFromGPU(vaddr, size, true);
-	}
-
-	auto&                                      cached = GetOrCreateBuffer(command, vaddr, size);
+bool BufferCache::SynchronizeBuffer(Buffer& buffer, uint64_t vaddr, uint64_t size, bool is_written,
+                                    bool is_texel_buffer) {
 	std::vector<std::pair<uint64_t, uint64_t>> uploads;
 	m_memory_tracker.ForEachUploadRange(
 	    vaddr, size, is_written,
 	    [&](uint64_t address, uint64_t bytes) noexcept { uploads.emplace_back(address, bytes); },
 	    [&]() noexcept {
 		    for (const auto& [address, bytes]: uploads) {
-			    Upload(command, *cached.buffer, cached.buffer->Offset(address),
-			           reinterpret_cast<const void*>(address), bytes);
+			    WriteDataBuffer(buffer, address, reinterpret_cast<const void*>(address), bytes);
 		    }
 	    });
+	if (is_texel_buffer && !is_written) {
+		return SynchronizeBufferFromImage(buffer, vaddr, size);
+	}
+	return false;
+}
+
+std::pair<Buffer*, uint64_t> BufferCache::ObtainBuffer(uint64_t vaddr, uint64_t size,
+                                                       bool is_written, bool is_texel_buffer,
+                                                       BufferId id) {
+	auto& command = m_scheduler.Current();
+	if (command.IsInvalid() || command.IsExecute() || vaddr == 0 || size == 0 ||
+	    vaddr >= TRACKER_ADDRESS_SIZE || size > TRACKER_ADDRESS_SIZE - vaddr) {
+		EXIT("BufferCache: buffer request requires a recording command buffer\n");
+	}
+
+	if (!is_written && size <= CACHING_PAGE_SIZE &&
+	    !m_memory_tracker.IsRegionGpuModified(vaddr, size) &&
+	    m_memory_tracker.IsRegionCpuModified(vaddr, size)) {
+		const auto alignment = std::max<uint64_t>(
+		    m_graphics.physical_device_properties.limits.minUniformBufferOffsetAlignment, 1);
+		auto [mapped, offset] = m_stream_buffer.Map(size, alignment);
+		if (mapped != nullptr && Libs::LibKernel::Memory::TryReadBacking(vaddr, mapped, size)) {
+			m_stream_buffer.Commit();
+			return {&m_stream_buffer, offset};
+		}
+	}
+
+	auto* buffer = m_slot_buffers.try_get(id);
+	if (buffer == nullptr || buffer->is_deleted || !buffer->IsInBounds(vaddr, size)) {
+		id     = FindBuffer(vaddr, size);
+		buffer = &m_slot_buffers[id];
+	}
+	(void)SynchronizeBuffer(*buffer, vaddr, size, is_written, is_texel_buffer);
 	if (is_written) {
 		m_gpu_modified_ranges.Add(vaddr, size);
 	}
-	auto     owner  = cached.buffer;
-	uint64_t offset = cached.buffer->Offset(vaddr);
-	if (is_formatted && is_read && !is_written) {
-		(void)SynchronizeBufferFromImage(*owner, vaddr, size);
-	}
-	return {owner, owner->Handle(), offset};
+	return {buffer, buffer->Offset(vaddr)};
 }
 
-std::shared_ptr<Buffer> BufferCache::ObtainNullBuffer() {
-	if (m_null_buffer != nullptr) {
-		return m_null_buffer;
-	}
-	m_null_buffer = std::make_shared<Buffer>(m_graphics, m_scheduler, MemoryUsage::DeviceLocal, 0,
-	                                         AllFlags, 16);
-	const std::array<uint8_t, 16> zeros {};
-	Upload(m_scheduler.Current(), *m_null_buffer, 0, zeros.data(), zeros.size());
-	return m_null_buffer;
-}
-
-ImageBufferSource BufferCache::ObtainBufferForImage(uint64_t vaddr, uint64_t size) {
+std::pair<Buffer*, uint64_t> BufferCache::ObtainBufferForImage(uint64_t vaddr, uint64_t size) {
 	if (vaddr == 0 || size == 0 || vaddr >= TRACKER_ADDRESS_SIZE ||
 	    size > TRACKER_ADDRESS_SIZE - vaddr) {
 		EXIT("BufferCache: invalid image source\n");
 	}
-	auto find_owner = [&]() {
+	auto find_owner = [&]() -> Buffer* {
 		auto owner = m_buffers.upper_bound(vaddr);
 		if (owner == m_buffers.begin()) {
-			return m_buffers.end();
+			return nullptr;
 		}
 		--owner;
-		return owner->second->buffer->IsInBounds(vaddr, size) ? owner : m_buffers.end();
+		auto* buffer = m_slot_buffers.try_get(owner->second);
+		return buffer != nullptr && buffer->IsInBounds(vaddr, size) ? buffer : nullptr;
 	};
 
 	{
@@ -559,86 +507,48 @@ ImageBufferSource BufferCache::ObtainBufferForImage(uint64_t vaddr, uint64_t siz
 		m_memory_tracker.ValidateGpuDirtyOwnership(m_gpu_modified_ranges, vaddr, size,
 		                                           "image source");
 
-		auto owner = find_owner();
-		if (has_dirty_buffer_source && owner == m_buffers.end()) {
-			CacheRange merged {.address = AlignDown(vaddr),
-			                   .size    = AlignUp(vaddr + size) - AlignDown(vaddr)};
-			using Iterator = decltype(m_buffers.begin());
-			std::vector<Iterator> overlaps;
-			auto                  first = m_buffers.lower_bound(merged.address);
-			if (first != m_buffers.begin()) {
-				auto previous = std::prev(first);
-				if (ResolveOverlap(merged, {previous->second->vaddr, previous->second->size})) {
-					first = previous;
-				}
-			}
-			for (auto candidate = first; candidate != m_buffers.end(); ++candidate) {
-				if (candidate->first >= merged.address + merged.size) {
-					break;
-				}
-				if (ResolveOverlap(merged, {candidate->second->vaddr, candidate->second->size})) {
-					overlaps.push_back(candidate);
-				}
-			}
-			if (overlaps.empty()) {
+		auto* owner = find_owner();
+		if (has_dirty_buffer_source && owner == nullptr) {
+			if (!IsRegionRegistered(vaddr, size)) {
 				EXIT("BufferCache: GPU-dirty image source has no native buffer\n");
 			}
-
-			auto cached                = std::make_unique<CachedBuffer>();
-			cached->vaddr              = merged.address;
-			cached->size               = merged.size;
-			cached->tick_accessed_last = m_gc_tick;
-			cached->buffer =
-			    std::make_shared<Buffer>(m_graphics, m_scheduler, MemoryUsage::DeviceLocal,
-			                             merged.address, AllFlags, merged.size);
-			for (const auto overlap: overlaps) {
-				const auto& old = *overlap->second;
-				cached->buffer->CopyFrom(m_scheduler.Current(), *old.buffer, 0,
-				                         old.vaddr - cached->vaddr, old.size);
-				m_scheduler.Current().RetainResourceUntilFence(old.buffer);
-			}
-			for (const auto overlap: overlaps) {
-				if (overlap->second->size > m_total_used_memory) {
-					EXIT("BufferCache: allocation accounting underflow\n");
-				}
-				m_total_used_memory -= overlap->second->size;
-				m_buffers.erase(overlap);
-			}
-			m_total_used_memory += cached->size;
-			owner = m_buffers.emplace(cached->vaddr, std::move(cached)).first;
-			if (!owner->second->buffer->IsInBounds(vaddr, size)) {
+			const auto id       = FindBuffer(vaddr, size);
+			auto&      buffer   = m_slot_buffers[id];
+			owner               = &buffer;
+			const auto resolved = m_buffers.find(buffer.CpuAddress());
+			if (resolved == m_buffers.end() || resolved->second != id ||
+			    !buffer.IsInBounds(vaddr, size)) {
 				EXIT("BufferCache: merged image source does not contain the requested range\n");
 			}
 		}
-		if (owner != m_buffers.end() && !cpu_modified &&
-		    (!gpu_modified || has_dirty_buffer_source)) {
-			owner->second->tick_accessed_last = m_gc_tick;
-			return {owner->second->buffer.get(), owner->second->buffer->Offset(vaddr)};
+		if (owner != nullptr && !cpu_modified && (!gpu_modified || has_dirty_buffer_source)) {
+			owner->tick_accessed_last = m_gc_tick;
+			return {owner, owner->Offset(vaddr)};
 		}
-		if (has_dirty_buffer_source && owner == m_buffers.end()) {
+		if (has_dirty_buffer_source && owner == nullptr) {
 			EXIT("BufferCache: GPU-dirty image source could not resolve its native owner\n");
 		}
 	}
 
 	auto [staging, stage_offset] = m_staging_buffer.Map(size, 16);
-	if (staging == nullptr || !Libs::LibKernel::Memory::TryReadBacking(vaddr, staging, size)) {
+	if (staging == nullptr || (!Libs::LibKernel::Memory::TryReadBacking(vaddr, staging, size) &&
+	                           !Libs::LibKernel::Memory::TryReadPrtBacking(vaddr, staging, size))) {
 		EXIT("BufferCache: failed to read mapped guest image backing\n");
 	}
 	m_staging_buffer.Commit();
 
 	const auto dirty                   = m_gpu_modified_ranges.Intersections(vaddr, size);
 	const bool has_dirty_buffer_source = !dirty.empty();
-	auto       owner                   = find_owner();
-	if (has_dirty_buffer_source && owner == m_buffers.end()) {
+	auto*      owner                   = find_owner();
+	if (has_dirty_buffer_source && owner == nullptr) {
 		EXIT("BufferCache: GPU-dirty image source lost its native owner\n");
 	}
-	if (owner == m_buffers.end() ||
+	if (owner == nullptr ||
 	    (m_memory_tracker.IsRegionGpuModified(vaddr, size) && !has_dirty_buffer_source)) {
 		return {&m_staging_buffer, stage_offset};
 	}
 
-	auto& cached              = *owner->second;
-	cached.tick_accessed_last = m_gc_tick;
+	owner->tick_accessed_last = m_gc_tick;
 	std::vector<std::pair<uint64_t, uint64_t>> uploads;
 	m_memory_tracker.ForEachUploadRange(
 	    vaddr, size, false,
@@ -647,12 +557,12 @@ ImageBufferSource BufferCache::ObtainBufferForImage(uint64_t vaddr, uint64_t siz
 	    },
 	    [&]() noexcept {
 		    for (const auto& [address, upload_size]: uploads) {
-			    cached.buffer->CopyFrom(
-			        m_scheduler.Current(), m_staging_buffer, stage_offset + address - vaddr,
-			        cached.buffer->Offset(address), upload_size, vk::AccessFlagBits::eHostWrite);
+			    owner->CopyFrom(m_scheduler.Current(), m_staging_buffer,
+			                    stage_offset + address - vaddr, owner->Offset(address), upload_size,
+			                    vk::AccessFlagBits::eHostWrite);
 		    }
 	    });
-	return {cached.buffer.get(), cached.buffer->Offset(vaddr)};
+	return {owner, owner->Offset(vaddr)};
 }
 
 void BufferCache::WriteHostMemory(uint64_t vaddr, std::span<const uint8_t> data) {
@@ -662,16 +572,16 @@ void BufferCache::WriteHostMemory(uint64_t vaddr, std::span<const uint8_t> data)
 	Libs::LibKernel::Memory::WriteBacking(vaddr, data.data(), data.size());
 
 	const auto end = vaddr + data.size();
-	for (auto& [address, cached]: m_buffers) {
-		const auto cached_end = address + cached->size;
+	for (const auto& [address, id]: m_buffers) {
+		auto&      buffer     = m_slot_buffers[id];
+		const auto buffer_end = address + buffer.Size();
 		const auto begin      = std::max(vaddr, address);
-		const auto range_end  = std::min(end, cached_end);
+		const auto range_end  = std::min(end, buffer_end);
 		if (begin >= range_end) {
 			continue;
 		}
-		Upload(m_scheduler.Current(), *cached->buffer, cached->buffer->Offset(begin),
-		       data.data() + begin - vaddr, range_end - begin);
-		cached->tick_accessed_last = m_gc_tick;
+		WriteDataBuffer(buffer, begin, data.data() + begin - vaddr, range_end - begin);
+		buffer.tick_accessed_last = m_gc_tick;
 	}
 }
 
@@ -709,12 +619,11 @@ void BufferCache::FillBuffer(uint64_t vaddr, uint64_t size, uint32_t value, bool
 		}
 	}
 
-	auto& command = m_scheduler.Current();
-	auto  dst     = ObtainBuffer(command, vaddr, size, true, false, true);
-	EXIT_IF(dst.buffer == nullptr || dst.owner == nullptr);
-	command.RetainResourceUntilFence(dst.owner);
-	auto owner = std::static_pointer_cast<Buffer>(dst.owner);
-	owner->Fill(dst.offset, size, value);
+	m_texture_cache.InvalidateMemoryFromGPU(vaddr, size);
+	const auto id          = FindBuffer(vaddr, size);
+	auto [dst, dst_offset] = ObtainBuffer(vaddr, size, true, true, id);
+	EXIT_IF(dst == nullptr);
+	dst->Fill(dst_offset, size, value);
 }
 
 void BufferCache::CopyBuffer(uint64_t dst_vaddr, uint64_t src_vaddr, uint64_t size, bool dst_gds,
@@ -724,10 +633,11 @@ void BufferCache::CopyBuffer(uint64_t dst_vaddr, uint64_t src_vaddr, uint64_t si
 	if ((dst_memory && dst_vaddr == 0) || (src_memory && src_vaddr == 0) || size == 0 ||
 	    ((dst_gds || src_gds) && ((dst_vaddr | src_vaddr | size) & 3u) != 0) ||
 	    size > UINT64_MAX - dst_vaddr || size > UINT64_MAX - src_vaddr || (dst_gds && src_gds) ||
-	    (dst_gds == src_gds && src_vaddr < dst_vaddr + size && dst_vaddr < src_vaddr + size) ||
 	    (dst_gds && (dst_vaddr > m_gds_buffer.Size() || size > m_gds_buffer.Size() - dst_vaddr)) ||
 	    (src_gds && (src_vaddr > m_gds_buffer.Size() || size > m_gds_buffer.Size() - src_vaddr))) {
-		EXIT("BufferCache: invalid or overlapping copy range\n");
+		EXIT("BufferCache: invalid copy range, src=0x%016" PRIx64 " dst=0x%016" PRIx64
+		     " size=0x%016" PRIx64 " src_gds=%d dst_gds=%d\n",
+		     src_vaddr, dst_vaddr, size, static_cast<int>(src_gds), static_cast<int>(dst_gds));
 	}
 	if (src_memory || dst_memory) {
 		const auto src_region =
@@ -755,30 +665,20 @@ void BufferCache::CopyBuffer(uint64_t dst_vaddr, uint64_t src_vaddr, uint64_t si
 	}
 
 	auto& command = m_scheduler.Current();
-	auto  src = src_memory ? ObtainBuffer(command, src_vaddr, size, false, true, true)
-	                       : BufferBinding {.buffer = m_gds_buffer.Handle(), .offset = src_vaddr};
-	auto  dst = dst_memory ? ObtainBuffer(command, dst_vaddr, size, true, false, true)
-	                       : BufferBinding {.buffer = m_gds_buffer.Handle(), .offset = dst_vaddr};
-	EXIT_IF(src.buffer == nullptr || dst.buffer == nullptr || (dst_memory && dst.owner == nullptr));
-	if (src.owner != nullptr) {
-		command.RetainResourceUntilFence(src.owner);
+	if (dst_memory) {
+		m_texture_cache.InvalidateMemoryFromGPU(dst_vaddr, size);
 	}
-	if (dst.owner != nullptr) {
-		command.RetainResourceUntilFence(dst.owner);
-	}
-	if (src.buffer == dst.buffer && src.offset < dst.offset + size &&
-	    dst.offset < src.offset + size) {
+	const auto src_id      = src_memory ? FindBuffer(src_vaddr, size) : BufferId {};
+	const auto dst_id      = dst_memory ? FindBuffer(dst_vaddr, size) : BufferId {};
+	auto [src, src_offset] = src_memory ? ObtainBuffer(src_vaddr, size, false, true, src_id)
+	                                    : std::pair {&m_gds_buffer, src_vaddr};
+	auto [dst, dst_offset] = dst_memory ? ObtainBuffer(dst_vaddr, size, true, true, dst_id)
+	                                    : std::pair {&m_gds_buffer, dst_vaddr};
+	EXIT_IF(src == nullptr || dst == nullptr);
+	if (src == dst && src_offset < dst_offset + size && dst_offset < src_offset + size) {
 		EXIT("BufferCache: resolved Vulkan copy ranges overlap\n");
 	}
-	auto& source = src.owner != nullptr ? *std::static_pointer_cast<Buffer>(src.owner)
-	               : src_gds            ? m_gds_buffer
-	                                    : m_stream_buffer;
-	auto& destination =
-	    dst.owner != nullptr ? *std::static_pointer_cast<Buffer>(dst.owner) : m_gds_buffer;
-	if (source.Handle() != src.buffer || destination.Handle() != dst.buffer) {
-		EXIT("BufferCache: resolved copy owner does not match its Vulkan handle\n");
-	}
-	destination.CopyFrom(command, source, src.offset, dst.offset, size);
+	dst->CopyFrom(command, *src, src_offset, dst_offset, size);
 }
 
 bool BufferCache::IsRegionRegistered(uint64_t vaddr, uint64_t size) {
@@ -792,8 +692,8 @@ bool BufferCache::IsRegionRegistered(uint64_t vaddr, uint64_t size) {
 	if (candidate == m_buffers.begin()) {
 		return false;
 	}
-	const auto& [address, cached] = *std::prev(candidate);
-	return address + cached->size > vaddr;
+	const auto& [address, id] = *std::prev(candidate);
+	return address + m_slot_buffers[id].Size() > vaddr;
 }
 
 bool BufferCache::IsRegionGpuModified(uint64_t vaddr, uint64_t size) {
@@ -821,70 +721,56 @@ void BufferCache::RunGarbageCollector() {
 	const uint64_t age        = std::min<uint64_t>(aggressive ? 80 : 160, tick);
 	const size_t   limit      = aggressive ? 64 : 32;
 
-	std::vector<RetiredBuffer>                                       retires;
-	std::vector<std::pair<RetiredBuffer, std::vector<DownloadCopy>>> dirty_retires;
-	{
-		std::vector<uint64_t> candidates;
-		for (const auto& [address, owner]: m_buffers) {
-			const auto& cached = *owner;
-			if (tick - std::min(tick, cached.tick_accessed_last) < age) {
-				continue;
-			}
+	std::vector<uint64_t> candidates;
+	for (const auto& [address, id]: m_buffers) {
+		const auto& buffer = m_slot_buffers[id];
+		if (tick - std::min(tick, buffer.tick_accessed_last) >= age) {
 			candidates.push_back(address);
 		}
-		std::ranges::sort(candidates, [&](uint64_t left, uint64_t right) {
-			return m_buffers.at(left)->tick_accessed_last < m_buffers.at(right)->tick_accessed_last;
-		});
-		if (candidates.size() > limit) {
-			candidates.resize(limit);
-		}
-		for (const auto address: candidates) {
-			auto& cached = *m_buffers.at(address);
-			m_memory_tracker.ValidateGpuDirtyOwnership(m_gpu_modified_ranges, cached.vaddr,
-			                                           cached.size, "garbage collection");
-			retires.push_back({address, cached.size, cached.buffer});
-			// GC runs immediately before submission. Preserve every source referenced by commands
-			// already recorded in the active batch.
-			m_scheduler.Current().RetainResourceUntilFence(cached.buffer);
-		}
-		for (const auto& retire: retires) {
-			if (!m_memory_tracker.IsRegionGpuModified(retire.address, retire.size)) {
-				continue;
-			}
-			auto& copies = dirty_retires.emplace_back(retire, std::vector<DownloadCopy> {}).second;
+	}
+	std::ranges::sort(candidates, [&](uint64_t left, uint64_t right) {
+		return m_slot_buffers[m_buffers.at(left)].tick_accessed_last <
+		       m_slot_buffers[m_buffers.at(right)].tick_accessed_last;
+	});
+	if (candidates.size() > limit) {
+		candidates.resize(limit);
+	}
+
+	for (const auto address: candidates) {
+		auto found = m_buffers.find(address);
+		EXIT_IF(found == m_buffers.end());
+		const auto id     = found->second;
+		auto&      buffer = m_slot_buffers[id];
+		m_memory_tracker.ValidateGpuDirtyOwnership(m_gpu_modified_ranges, buffer.CpuAddress(),
+		                                           buffer.Size(), "garbage collection");
+		const bool dirty = m_memory_tracker.IsRegionGpuModified(buffer.CpuAddress(), buffer.Size());
+		std::vector<DownloadCopy> copies;
+		if (dirty) {
 			m_memory_tracker.ForEachDownloadRange<false>(
-			    retire.address, retire.size,
-			    [&](uint64_t address, uint64_t size) noexcept {
-				    m_memory_tracker.ValidateGpuDirtyPages(m_gpu_modified_ranges, address, size,
-				                                           "garbage collection");
+			    buffer.CpuAddress(), buffer.Size(),
+			    [&](uint64_t dirty_address, uint64_t dirty_size) noexcept {
+				    m_memory_tracker.ValidateGpuDirtyPages(m_gpu_modified_ranges, dirty_address,
+				                                           dirty_size, "garbage collection");
 			    },
-			    [&](uint64_t address, uint64_t size) noexcept {
-				    for (const auto range: m_gpu_modified_ranges.Intersections(address, size)) {
-					    copies.push_back({retire.owner, range.address - retire.address,
+			    [&](uint64_t dirty_address, uint64_t dirty_size) noexcept {
+				    for (const auto range:
+				         m_gpu_modified_ranges.Intersections(dirty_address, dirty_size)) {
+					    copies.push_back({&buffer, range.address - buffer.CpuAddress(),
 					                      range.address, range.size});
 				    }
 			    });
 		}
-	}
 
-	for (auto& [retire, copies]: dirty_retires) {
-		QueueGarbageDownload(copies, std::move(retire));
-	}
-
-	for (const auto& retire: retires) {
-		auto found = m_buffers.find(retire.address);
-		if (found == m_buffers.end() || found->second->size != retire.size ||
-		    found->second->buffer != retire.owner) {
-			EXIT("BufferCache: garbage-collection owner changed during download\n");
+		const auto retire_address = buffer.CpuAddress();
+		const auto retire_size    = buffer.Size();
+		if (dirty) {
+			EXIT_IF(copies.empty());
+			Unregister(id);
+			QueueGarbageDownload(copies, id, retire_address, retire_size);
+		} else {
+			m_memory_tracker.UntrackMemory(retire_address, retire_size);
+			DeleteBuffer(id);
 		}
-		if (!m_memory_tracker.IsRegionGpuModified(retire.address, retire.size)) {
-			m_memory_tracker.UntrackMemory(retire.address, retire.size);
-		}
-		if (retire.size > m_total_used_memory) {
-			EXIT("BufferCache: allocation accounting underflow\n");
-		}
-		m_total_used_memory -= retire.size;
-		m_buffers.erase(found);
 	}
 }
 

src/graphics/host_gpu/renderer/cache/bufferCache.h

@@ -3,12 +3,12 @@
 
 #include "common/abi.h"
 #include "common/common.h"
+#include "common/slotVector.h"
 #include "graphics/host_gpu/memoryTracker.h"
 #include "graphics/host_gpu/rangeSet.h"
 #include "graphics/host_gpu/renderer/cache/streamBuffer.h"
 
 #include <map>
-#include <memory>
 #include <span>
 #include <utility>
 #include <vector>
@@ -16,49 +16,44 @@
 namespace Libs::Graphics {
 
 struct GraphicContext;
-class CommandBuffer;
 class CommandScheduler;
 class TextureCache;
 
-struct BufferBinding {
-	std::shared_ptr<void> owner;
-	vk::Buffer            buffer = nullptr;
-	uint64_t              offset = 0;
-};
-
-struct ImageBufferSource {
-	Buffer*  buffer = nullptr;
-	uint64_t offset = 0;
-};
+using BufferId = Common::SlotId;
+inline constexpr BufferId NULL_BUFFER_ID {0};
 
 class BufferCache {
 public:
 	static constexpr uint64_t CACHING_PAGE_SIZE = 16ull * 1024ull;
-	static constexpr uint64_t GetBufferOffset(uint64_t vaddr) {
-		return vaddr & (CACHING_PAGE_SIZE - 1);
-	}
 
 	BufferCache(GraphicContext& graphics, CommandScheduler& scheduler, PageManager& page_manager,
 	            TextureCache& texture_cache);
 	~BufferCache();
 	KYTY_CLASS_NO_COPY(BufferCache);
 
-	void                        InvalidateMemory(uint64_t vaddr, uint64_t size);
-	void                        ReadMemory(uint64_t vaddr, uint64_t size, bool is_write = false);
-	void                        UnmapMemory(uint64_t vaddr, uint64_t size);
-	[[nodiscard]] BufferBinding ObtainBuffer(CommandBuffer& command, uint64_t vaddr, uint64_t size,
-	                                         bool is_written = false, bool is_read = true,
-	                                         bool is_formatted = false);
-	[[nodiscard]] StreamBuffer& GetUtilityBuffer(MemoryUsage usage) noexcept;
-	[[nodiscard]] Buffer&       GetGdsBuffer() noexcept { return m_gds_buffer; }
-	[[nodiscard]] const Buffer& GetGdsBuffer() const noexcept { return m_gds_buffer; }
-	[[nodiscard]] BufferBinding UploadTransient(const void* data, uint64_t size,
-	                                            uint64_t alignment);
-	[[nodiscard]] std::shared_ptr<Buffer> ObtainNullBuffer();
-	[[nodiscard]] ImageBufferSource       ObtainBufferForImage(uint64_t vaddr, uint64_t size);
-	void FillBuffer(uint64_t vaddr, uint64_t size, uint32_t value, bool is_gds = false);
-	void CopyBuffer(uint64_t dst_vaddr, uint64_t src_vaddr, uint64_t size, bool dst_gds = false,
-	                bool src_gds = false);
+	void                   InvalidateMemory(uint64_t vaddr, uint64_t size);
+	void                   ReadMemory(uint64_t vaddr, uint64_t size, bool is_write = false);
+	void                   UnmapMemory(uint64_t vaddr, uint64_t size);
+	[[nodiscard]] Buffer&  GetBuffer(BufferId id) { return m_slot_buffers[id]; }
+	[[nodiscard]] BufferId FindBuffer(uint64_t vaddr, uint64_t size);
+	[[nodiscard]] std::pair<Buffer*, uint64_t> ObtainBuffer(uint64_t vaddr, uint64_t size,
+	                                                        bool     is_written,
+	                                                        bool     is_texel_buffer = false,
+	                                                        BufferId id              = {});
+	[[nodiscard]] StreamBuffer&                GetUtilityBuffer(MemoryUsage usage) noexcept {
+		switch (usage) {
+			case MemoryUsage::Upload: return m_staging_buffer;
+			case MemoryUsage::Stream: return m_stream_buffer;
+			case MemoryUsage::Download: return m_download_buffer;
+			case MemoryUsage::DeviceLocal: return m_device_buffer;
+		}
+		EXIT("BufferCache: invalid utility-buffer usage\n");
+	}
+	[[nodiscard]] const Buffer* GetGdsBuffer() const noexcept { return &m_gds_buffer; }
+	[[nodiscard]] std::pair<Buffer*, uint64_t> ObtainBufferForImage(uint64_t vaddr, uint64_t size);
+	void FillBuffer(uint64_t vaddr, uint64_t size, uint32_t value, bool is_gds);
+	void CopyBuffer(uint64_t dst_vaddr, uint64_t src_vaddr, uint64_t size, bool dst_gds,
+	                bool src_gds);
 	// Cache-index and exact dirty-range queries require GPU-thread serialization.
 	[[nodiscard]] bool IsRegionRegistered(uint64_t vaddr, uint64_t size);
 	[[nodiscard]] bool HasGpuDirtyBytes(uint64_t vaddr, uint64_t size);
@@ -69,49 +64,42 @@ public:
 private:
 	friend struct BufferCacheTestAccess;
 
-	struct CacheRange {
-		uint64_t address = 0;
-		uint64_t size    = 0;
-	};
-	struct CachedBuffer;
 	struct DownloadCopy;
 	struct DownloadRange;
-	struct RetiredBuffer;
 	static constexpr uint64_t               DOWNLOAD_ALIGNMENT = 64;
-	[[nodiscard]] static uint64_t           AlignDown(uint64_t value) noexcept;
-	[[nodiscard]] static uint64_t           AlignUp(uint64_t value);
 	[[nodiscard]] static constexpr uint64_t AlignDownload(uint64_t size) noexcept {
 		return (size + DOWNLOAD_ALIGNMENT - 1) & ~(DOWNLOAD_ALIGNMENT - 1);
 	}
 	[[nodiscard]] static std::pair<uint64_t, uint64_t> DownloadEnvelope(const DownloadCopy& copy);
-	[[nodiscard]] static bool ResolveOverlap(CacheRange& merged, CacheRange candidate) noexcept;
-	void Upload(CommandBuffer& command, Buffer& destination, uint64_t destination_offset,
-	            const void* source, uint64_t size);
-	[[nodiscard]] CachedBuffer& GetOrCreateBuffer(CommandBuffer& command, uint64_t vaddr,
-	                                              uint64_t size);
+	void WriteDataBuffer(Buffer& buffer, uint64_t address, const void* source, uint64_t size);
+	void Unregister(BufferId id);
+	void DeleteBuffer(BufferId id);
+	[[nodiscard]] bool SynchronizeBuffer(Buffer& buffer, uint64_t vaddr, uint64_t size,
+	                                     bool is_written, bool is_texel_buffer);
 	[[nodiscard]] bool SynchronizeBufferFromImage(Buffer& buffer, uint64_t vaddr, uint64_t size);
 	[[nodiscard]] std::vector<DownloadRange> RecordDownloads(std::span<const DownloadCopy> copies);
 	void PublishDownloads(std::span<const DownloadRange> downloads);
-	void QueueGarbageDownload(std::span<const DownloadCopy> copies, RetiredBuffer retire);
+	void QueueGarbageDownload(std::span<const DownloadCopy> copies, BufferId id, uint64_t address,
+	                          uint64_t size);
 	void WriteHostMemory(uint64_t vaddr, std::span<const uint8_t> data);
 	void ReadMemoryOnGpu(uint64_t vaddr, uint64_t size, bool is_write);
 
-	GraphicContext&                                   m_graphics;
-	CommandScheduler&                                 m_scheduler;
-	Buffer                                            m_gds_buffer;
-	std::shared_ptr<Buffer>                           m_null_buffer;
-	std::map<uint64_t, std::unique_ptr<CachedBuffer>> m_buffers;
-	RangeSet                                          m_gpu_modified_ranges;
-	MemoryTracker                                     m_memory_tracker;
-	StreamBuffer                                      m_staging_buffer;
-	StreamBuffer                                      m_stream_buffer;
-	StreamBuffer                                      m_download_buffer;
-	StreamBuffer                                      m_device_buffer;
-	TextureCache&                                     m_texture_cache;
-	uint64_t                                          m_total_used_memory = 0;
-	uint64_t m_trigger_gc_memory  = 1ull * 1024 * 1024 * 1024;
-	uint64_t m_critical_gc_memory = 2ull * 1024 * 1024 * 1024;
-	uint64_t m_gc_tick            = 0;
+	GraphicContext&              m_graphics;
+	CommandScheduler&            m_scheduler;
+	Buffer                       m_gds_buffer;
+	Common::SlotVector<Buffer>   m_slot_buffers;
+	std::map<uint64_t, BufferId> m_buffers;
+	RangeSet                     m_gpu_modified_ranges;
+	MemoryTracker                m_memory_tracker;
+	StreamBuffer                 m_staging_buffer;
+	StreamBuffer                 m_stream_buffer;
+	StreamBuffer                 m_download_buffer;
+	StreamBuffer                 m_device_buffer;
+	TextureCache&                m_texture_cache;
+	uint64_t                     m_total_used_memory  = 0;
+	uint64_t                     m_trigger_gc_memory  = 1ull * 1024 * 1024 * 1024;
+	uint64_t                     m_critical_gc_memory = 2ull * 1024 * 1024 * 1024;
+	uint64_t                     m_gc_tick            = 0;
 };
 
 } // namespace Libs::Graphics

src/graphics/host_gpu/renderer/cache/gpuResourceManager.h

@@ -13,7 +13,7 @@
 namespace Libs::Graphics {
 
 class CommandScheduler;
-class Gpu;
+class GuestGpu;
 
 class GpuResourceManager {
 public:
@@ -23,7 +23,7 @@ public:
 
 	[[nodiscard]] BufferCache&  GetBufferCache() { return m_buffer_cache; }
 	[[nodiscard]] TextureCache& GetTextureCache() { return m_texture_cache; }
-	void                        SetGpu(Gpu* gpu) noexcept { m_gpu = gpu; }
+	void                        SetGpu(GuestGpu* gpu) noexcept { m_gpu = gpu; }
 
 	[[nodiscard]] bool HandleFault(PageFaultAccess access, uint64_t fault_vaddr) noexcept;
 	[[nodiscard]] bool InvalidateMemory(uint64_t vaddr, uint64_t size);
@@ -39,7 +39,7 @@ private:
 	TextureCache              m_texture_cache;
 	mutable std::shared_mutex m_mapped_ranges_mutex;
 	RangeSet                  m_mapped_ranges;
-	Gpu*                      m_gpu = nullptr;
+	GuestGpu*                 m_gpu = nullptr;
 };
 
 } // namespace Libs::Graphics

src/graphics/host_gpu/renderer/cache/streamBuffer.h

@@ -65,6 +65,10 @@ public:
 	                                                   vk::AccessFlagBits::eMemoryWrite);
 	void Fill(uint64_t offset, uint64_t size, uint32_t value);
 
+	// BufferCache state lives directly on the resource.
+	bool     is_deleted         = false;
+	uint64_t tick_accessed_last = 0;
+
 protected:
 	[[nodiscard]] GraphicContext&   Graphics() const noexcept { return *m_graphics; }
 	[[nodiscard]] CommandScheduler& Scheduler() const noexcept { return *m_scheduler; }

src/graphics/host_gpu/renderer/cache/textureCache.h

@@ -4,19 +4,17 @@
 #include "common/abi.h"
 #include "common/common.h"
 #include "common/lruCache.h"
+#include "common/slotVector.h"
 #include "graphics/host_gpu/pageManager.h"
 #include "graphics/host_gpu/regionManager.h"
 #include "graphics/host_gpu/renderer/cache/multiLevelPageTable.h"
 #include "graphics/host_gpu/renderer/image/blitHelper.h"
 #include "graphics/host_gpu/renderer/image/image.h"
+#include "graphics/host_gpu/renderer/image/tiler.h"
 
-#include <compare>
 #include <map>
-#include <memory>
-#include <optional>
-#include <set>
-#include <span>
-#include <utility>
+#include <unordered_map>
+#include <unordered_set>
 #include <vector>
 
 namespace Libs::Graphics {
@@ -27,8 +25,6 @@ class BufferCache;
 class CommandBuffer;
 class CommandScheduler;
 class RenderExecutor;
-class StreamBuffer;
-class TileManager;
 struct TextureCacheTestAccess;
 
 class TextureCache {
@@ -59,20 +55,26 @@ public:
 	[[nodiscard]] vk::ImageView FindTexture(ImageId id, const ImageDesc& desc);
 	[[nodiscard]] vk::ImageView FindRenderTarget(ImageId id, const ImageDesc& desc);
 	[[nodiscard]] vk::ImageView FindDepthTarget(ImageId id, const ImageDesc& desc);
-	[[nodiscard]] Image&        GetImage(ImageId id);
-	[[nodiscard]] const Image&  GetImage(ImageId id) const;
-	void                        MarkGpuWritten(ImageId id);
+	[[nodiscard]] Image&        GetImage(ImageId id) {
+		auto& image = m_slot_images[id];
+		TouchImage(image);
+		return image;
+	}
+	void MarkGpuWritten(ImageId id);
 
 	[[nodiscard]] bool ClearImageFromBuffer(CommandBuffer& command, uint64_t address, uint64_t size,
 	                                        uint32_t packed_clear);
 	void               InvalidateMemory(uint64_t address, uint64_t size);
-	[[nodiscard]] bool InvalidateMemoryFromGPU(uint64_t address, uint64_t size,
-	                                           bool formatted_buffer_write = false);
+	void               InvalidateMemoryFromGPU(uint64_t address, uint64_t size);
 	[[nodiscard]] RegionInfo QueryRegion(uint64_t address, uint64_t size);
 
 	[[nodiscard]] bool IsMeta(uint64_t address);
-	[[nodiscard]] bool IsMetaCleared(uint64_t address, uint32_t slice);
+	[[nodiscard]] bool IsMetaCleared(uint64_t address, uint32_t slice,
+	                                 uint32_t* fill_value = nullptr);
 	[[nodiscard]] bool ClearMeta(uint64_t address);
+	// Returns true when registered DCC absorbed the fill and the caller may skip the dispatch.
+	// False may still record PendingDcc state, but the guest dispatch must execute.
+	[[nodiscard]] bool TryConsumeDccFill(uint64_t address, uint64_t size, uint32_t fill_value);
 	[[nodiscard]] bool TouchMeta(uint64_t address, uint32_t slice, bool is_clear);
 
 	void UnmapMemory(uint64_t address, uint64_t size);
@@ -84,13 +86,18 @@ private:
 	struct ColorTransferPlan;
 	struct DownloadPlan;
 
-	struct Slot {
-		std::shared_ptr<Image> image;
-		uint32_t               generation = 1;
-	};
-
 	struct MetaDataInfo {
+		// A guest metadata-fill dispatch may initialize DCC before its render target is bound.
+		// PendingDcc retains that exact fill until FindRenderTarget classifies the address,
+		// without exposing an unconfirmed buffer address to the normal metadata heuristics.
+		// Keep all surface metadata in one entry so CMask/FMask can be
+		// registered beside HTile and DCC without introducing parallel tracking paths.
+		enum class Type : uint8_t { PendingDcc, CMask, FMask, HTile, Dcc };
+
+		Type     type       = Type::PendingDcc;
 		uint32_t clear_mask = 0;
+		uint32_t fill_value = 0xffffffffu;
+		uint64_t fill_size  = 0;
 	};
 
 	struct OverlapResult {
@@ -102,25 +109,21 @@ private:
 	using ImageIds       = InlinePageOwnerList<ImageId, 16>;
 	using ImagePageTable = MultiLevelPageTable<ImageIds, 20, 40, 10>;
 
-	[[nodiscard]] Image&                 ResolveImage(ImageId id);
-	[[nodiscard]] const Image&           ResolveImage(ImageId id) const;
-	[[nodiscard]] std::shared_ptr<Image> ResolveOwner(ImageId id) const;
-	[[nodiscard]] ImageId                InsertImage(const ImageInfo& info);
-	[[nodiscard]] ImageId                GetNullImage(const ImageDesc& desc);
-	void                                 RegisterImage(ImageId id);
-	void                                 UnregisterImage(ImageId id);
-	void                                 DeleteImage(ImageId id);
-	void DeleteImages(std::span<const ImageId> ids, std::optional<ImageId> native_source = {});
-	void RetainImage(CommandBuffer& command, ImageId id);
-	void TouchImage(Image& image);
-	void TrackImage(ImageId id);
-	void TrackImageHead(ImageId id);
-	void TrackImageTail(ImageId id);
-	void UntrackImage(ImageId id);
-	void UntrackImageHead(ImageId id);
-	void UntrackImageTail(ImageId id);
-	void TrackImageDownload(ImageId id);
-	void TrackImageDownloadLocked(ImageId id, Image& image);
+	[[nodiscard]] ImageId     InsertImage(const ImageInfo& info);
+	[[nodiscard]] ImageId     GetNullImage(const ImageDesc& desc);
+	void                      RegisterImage(ImageId id);
+	void                      UnregisterImage(ImageId id);
+	void                      DeleteImage(ImageId id);
+	void                      FreeImage(ImageId id);
+	void                      TouchImage(Image& image);
+	void                      TrackImage(ImageId id);
+	void                      TrackImageHead(ImageId id);
+	void                      TrackImageTail(ImageId id);
+	void                      UntrackImage(ImageId id);
+	void                      UntrackImageHead(ImageId id);
+	void                      UntrackImageTail(ImageId id);
+	void                      MarkAsMaybeDirty(ImageId id, Image& image);
+	void                      TrackImageDownload(ImageId id, Image& image);
 	[[nodiscard]] static bool SameBacking(const ImageInfo& cached, const ImageInfo& requested,
 	                                      bool exact_format);
 	[[nodiscard]] static BindingType UploadBinding(const Image& image);
@@ -149,31 +152,24 @@ private:
 	[[nodiscard]] bool CopyD16(Image& destination, Image& source);
 	void               CopyImage(ImageId destination, ImageId source);
 	void               AssociateStencil(ImageId depth, GuestRange stencil);
-	void               AssociateStencilLocked(ImageId depth, GuestRange stencil);
 	void CopyImageMip(ImageId destination, ImageId source, uint32_t mip, uint32_t layer);
 	void ValidateImageDesc(const ImageDesc& desc) const;
 
-	void InvalidateCpuAliases(uint64_t address, uint64_t size);
-	void ClearGpuModified(ImageId id);
-
-	void                                        DownloadImage(ImageId id);
-	[[nodiscard]] bool                          TryDownloadImage(ImageId id);
-	[[nodiscard]] std::pair<uint8_t*, uint64_t> MapDownload(uint64_t size, uint64_t alignment);
-	void QueueDownload(GuestRange range, StreamBuffer& download, uint8_t* mapped, uint64_t offset);
+	void               InvalidateCpuAliases(uint64_t address, uint64_t size);
+	[[nodiscard]] bool TryDownloadImage(ImageId id);
 
 	GraphicContext&                                   m_graphics;
 	CommandScheduler&                                 m_scheduler;
 	TrackingSpinLock                                  m_lock;
 	PageManager&                                      m_page_manager;
 	BlitHelper                                        m_blit_helper;
-	std::unique_ptr<TileManager>                      m_tiler;
+	TileManager                                       m_tiler;
 	BufferCache&                                      m_buffer_cache;
-	std::vector<Slot>                                 m_slots;
-	std::vector<uint32_t>                             m_free_slots;
+	Common::SlotVector<Image>                         m_slot_images;
 	ImagePageTable                                    m_image_page_table;
-	std::map<vk::Format, ImageId>                     m_null_images;
+	std::unordered_map<vk::Format, ImageId>           m_null_images;
 	Common::LeastRecentlyUsedCache<ImageId, uint64_t> m_lru_cache;
-	std::set<ImageId>                                 m_download_images;
+	std::unordered_set<ImageId>                       m_download_images;
 	std::map<uint64_t, MetaDataInfo>                  m_surface_metas;
 	uint64_t                                          m_total_used_memory  = 0;
 	uint64_t                                          m_trigger_gc_memory  = 0;

src/graphics/host_gpu/renderer/colorRenderTarget.cpp

@@ -9,7 +9,6 @@
 #include "graphics/host_gpu/graphicContext.h"
 #include "graphics/host_gpu/renderer/debug.h"
 #include "graphics/host_gpu/renderer/image/textureCommon.h"
-#include "graphics/host_gpu/renderer/pipeline/descriptorCache.h"
 #include "graphics/host_gpu/renderer/render.h"
 #include "graphics/host_gpu/renderer/renderContext.h"
 #include "graphics/host_gpu/vulkanCommon.h"
@@ -21,6 +20,28 @@ namespace Libs::Graphics {
 
 static std::atomic<uint32_t> g_render_color_log_count = 0;
 
+static void ResolveDccClearInfo(RenderColorInfo& info, vk::Format format, bool has_dcc,
+                                uint32_t packed_clear) {
+	// Register-backed DCC clears use the target's packed clear value. Decode one-word guest
+	// formats here; unsupported encodings remain tracked without unsafe materialization.
+	info.metadata_clear_supported =
+	    has_dcc && DecodePackedColorClear(format, packed_clear, info.color_clear_value);
+	switch (format) {
+		case vk::Format::eR8G8B8A8Unorm:
+		case vk::Format::eR8G8B8A8Srgb:
+		case vk::Format::eB8G8R8A8Unorm:
+		case vk::Format::eB8G8R8A8Srgb:
+		case vk::Format::eA2B10G10R10UnormPack32:
+		case vk::Format::eA2R10G10B10UnormPack32:
+			info.metadata_fixed_clear_supported = has_dcc;
+			break;
+		default: info.metadata_fixed_clear_supported = false; break;
+	}
+	if (!info.metadata_clear_supported) {
+		info.color_clear_value = {};
+	}
+}
+
 // NOLINTNEXTLINE(readability-function-cognitive-complexity)
 void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, RenderCommandBuffer& buffer,
                                               RenderColorInfo& r,
@@ -55,20 +76,22 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, RenderCommandB
 		}
 
 		// No color output
-		r.type               = RenderColorType::NoColorOutput;
-		r.desc               = {};
-		r.base_addr          = 0;
-		r.image_id           = {};
-		r.image_view         = nullptr;
-		r.format             = vk::Format::eUndefined;
-		r.extent             = {};
-		r.base_mip_level     = 0;
-		r.base_array_layer   = 0;
-		r.buffer_size        = 0;
-		r.samples            = 1;
-		r.export_mapping     = {};
-		r.color_clear_enable = false;
-		r.color_clear_value  = {};
+		r.type                           = RenderColorType::NoColorOutput;
+		r.desc                           = {};
+		r.base_addr                      = 0;
+		r.image_id                       = {};
+		r.image_view                     = nullptr;
+		r.format                         = vk::Format::eUndefined;
+		r.extent                         = {};
+		r.base_mip_level                 = 0;
+		r.base_array_layer               = 0;
+		r.buffer_size                    = 0;
+		r.samples                        = 1;
+		r.export_mapping                 = {};
+		r.color_clear_enable             = false;
+		r.metadata_clear_supported       = false;
+		r.metadata_fixed_clear_supported = false;
+		r.color_clear_value              = {};
 		return;
 	}
 	const auto samples = render_sample_count(rt.attrib.num_fragments);
@@ -101,14 +124,15 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, RenderCommandB
 			     " attrib3_tile=0x%08" PRIx32 " attrib3_dim=0x%08" PRIx32 " fmt=0x%08" PRIx32
 			     " nfmt=0x%08" PRIx32 " order=0x%08" PRIx32 "\n",
 			     rt_slot, rt.base.addr, mask, rt.attrib2.width, rt.attrib2.height,
-			     rt.attrib3.tile_mode, rt.attrib3.dimension, rt.info.format, rt.info.channel_type,
-			     rt.info.channel_order);
+			     static_cast<uint32_t>(rt.attrib3.tile_mode), rt.attrib3.dimension,
+			     static_cast<uint32_t>(rt.info.format), static_cast<uint32_t>(rt.info.channel_type),
+			     static_cast<uint32_t>(rt.info.channel_order));
 		}
 	}
 
-	// CB_COLOR_CONTROL describes the color-buffer operation / ROP3 logic op.
-	// ROP3 Copy is the normal color write path, not a render-target clear.
-	// SRGB clear words are still encoded as normalized component values.
+	// Color-control state selects the color-buffer operation and logical blend operation.
+	// The normal copy operation is a regular color write, not an attachment clear.
+	// Nonlinear clear values are still stored as normalized components.
 	// Fast color clears are metadata driven and must be handled explicitly when
 	// that metadata path is implemented; render-pass load must preserve contents.
 	r.color_clear_enable = false;
@@ -129,17 +153,21 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, RenderCommandB
 	if (volume && samples != 1) {
 		EXIT("multisampled 3D render targets are unsupported\n");
 	}
-	const uint32_t depth = volume ? rt.attrib3.depth + 1u : 1u;
-	const bool     standard64 =
-	    rt.attrib3.tile_mode == Prospero::GpuEnumValue(Prospero::TileMode::kStandard64KB);
+	const uint32_t depth        = volume ? rt.attrib3.depth + 1u : 1u;
+	const bool     standard4    = rt.attrib3.tile_mode == Prospero::TileMode::kStandard4KB;
+	const bool     standard64   = rt.attrib3.tile_mode == Prospero::TileMode::kStandard64KB;
+	const bool     depth_tile   = rt.attrib3.tile_mode == Prospero::TileMode::kDepth;
+	const bool     texture_tile = standard4 || standard64 || depth_tile;
 
 	switch (rt.attrib3.tile_mode) {
-		case Prospero::GpuEnumValue(Prospero::TileMode::kLinear):
-		case Prospero::GpuEnumValue(Prospero::TileMode::kStandard64KB):
-		case Prospero::GpuEnumValue(Prospero::TileMode::kRenderTarget):
+		case Prospero::TileMode::kLinear:
+		case Prospero::TileMode::kStandard4KB:
+		case Prospero::TileMode::kStandard64KB:
+		case Prospero::TileMode::kDepth:
+		case Prospero::TileMode::kRenderTarget:
 			tile = !RenderIsColorTileModeLinear(rt.attrib3.tile_mode);
 			break;
-		default: EXIT("unknown tile mode: %u\n", rt.attrib3.tile_mode);
+		default: EXIT("unknown tile mode: %u\n", static_cast<uint32_t>(rt.attrib3.tile_mode));
 	}
 	if (!tile && levels > 1) {
 		EXIT("linear mipmapped render targets are unsupported\n");
@@ -147,6 +175,9 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, RenderCommandB
 	if (samples > 1 && (!tile || levels != 1)) {
 		EXIT("multisampled render targets require a single-mip tiled surface\n");
 	}
+	if (texture_tile && samples != 1) {
+		EXIT("texture-tiled color render targets do not support multisampling\n");
+	}
 
 	width  = rt.attrib2.width + 1;
 	height = rt.attrib2.height + 1;
@@ -157,31 +188,38 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, RenderCommandB
 		EXIT("render-target format has no valid element size\n");
 	}
 	const auto transfer_format = ImageOps::RenderTargetTransferFormat(bytes_per_element);
-	if (standard64 &&
-	    (rt.attrib3.dimension != 1 || rt.attrib3.depth != 0 || levels != 1 ||
-	     rt.view.current_mip_level != 0 || view.base_layer != 0 || view.image_layers != 1 ||
-	     samples != 1 || bytes_per_element != 4 || rt.pitch.pitch_div8_minus1 != 0 ||
-	     (rt.base.addr & 0xffffu) != 0 || rt.info.fmask_compression_enable ||
-	     rt.info.fmask_data_compression_disable || rt.info.fmask_one_frag_mode ||
-	     rt.info.cmask_fast_clear_enable || rt.info.dcc_compression_enable ||
-	     rt.info.cmask_is_linear != 0 || rt.info.cmask_addr_type != 0 || rt.info.alt_tile_mode ||
-	     rt.cmask.addr != 0 || rt.fmask.addr != 0 || rt.dcc_addr.addr != 0 ||
-	     rt.dcc.data_write_on_dcc_clear_to_reg)) {
-		EXIT("unsupported Standard64KB render target: addr=0x%016" PRIx64
+	TileTextureBlockLayout texture_tile_layout {};
+	if (texture_tile &&
+	    (!TileGetTextureBlockLayout(transfer_format, rt.attrib3.tile_mode, volume,
+	                                texture_tile_layout) ||
+	     rt.pitch.pitch_div8_minus1 != 0 ||
+	     (rt.base.addr & (texture_tile_layout.block.block_size - 1u)) != 0 ||
+	     rt.info.fmask_compression_enable || rt.info.fmask_data_compression_disable ||
+	     rt.info.fmask_one_frag_mode || rt.info.cmask_fast_clear_enable ||
+	     rt.info.dcc_compression_enable || rt.info.cmask_is_linear != 0 ||
+	     rt.info.cmask_addr_type != 0 || rt.info.alt_tile_mode || rt.cmask.addr != 0 ||
+	     rt.fmask.addr != 0 || rt.dcc_addr.addr != 0 || rt.dcc.data_write_on_dcc_clear_to_reg)) {
+		EXIT("unsupported texture-tiled render target: addr=0x%016" PRIx64 " tile=%u"
 		     " dimension=%u depth=%u levels=%u layer=%u/%u samples=%u fragments=%u bpe=%u"
 		     " cmask=0x%016" PRIx64 " fmask=0x%016" PRIx64 " dcc=0x%016" PRIx64 "\n",
-		     rt.base.addr, rt.attrib3.dimension, rt.attrib3.depth, levels, view.base_layer,
-		     view.image_layers, rt.attrib.num_samples, rt.attrib.num_fragments, bytes_per_element,
-		     rt.cmask.addr, rt.fmask.addr, rt.dcc_addr.addr);
+		     rt.base.addr, static_cast<uint32_t>(rt.attrib3.tile_mode), rt.attrib3.dimension,
+		     rt.attrib3.depth, levels, view.base_layer, view.image_layers, rt.attrib.num_samples,
+		     rt.attrib.num_fragments, bytes_per_element, rt.cmask.addr, rt.fmask.addr,
+		     rt.dcc_addr.addr);
+	}
+	if ((standard64 || depth_tile) && (rt.attrib3.dimension != 1 || rt.attrib3.depth != 0 ||
+	                                   view.base_layer != 0 || view.image_layers != 1)) {
+		EXIT("unsupported 64KB texture-tiled render-target view: dimension=%u depth=%u"
+		     " layer=%u/%u\n",
+		     rt.attrib3.dimension, rt.attrib3.depth, view.base_layer, view.image_layers);
 	}
 	if (rt.pitch.pitch_div8_minus1 != 0) {
 		pitch = (rt.pitch.pitch_div8_minus1 + 1u) << 3u;
 	} else if (tile) {
 		if (volume) {
 			pitch = TileGetTexturePitch(transfer_format, width, rt.attrib3.tile_mode);
-		} else if (standard64) {
-			pitch = TileGetTexturePitch(Prospero::GpuEnumValue(Prospero::BufferFormat::k32Float),
-			                            width, rt.attrib3.tile_mode);
+		} else if (texture_tile) {
+			pitch = TileGetTexturePitch(transfer_format, width, rt.attrib3.tile_mode);
 		} else {
 			pitch = TileGetRenderTargetPitch(width, bytes_per_element, rt.attrib.num_fragments);
 		}
@@ -207,18 +245,17 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, RenderCommandB
 		                                          1};
 		if (!tile || !TileGetTiledTextureLayout(description, volume_layout)) {
 			EXIT("unsupported 3D render-target layout: %ux%ux%u levels=%u tile=%u\n", width, height,
-			     depth, levels, rt.attrib3.tile_mode);
+			     depth, levels, static_cast<uint32_t>(rt.attrib3.tile_mode));
 		}
 		size         = volume_layout.block_slice_size;
 		backing_size = volume_layout.total_size;
 	} else if (tile) {
 		TileSizeAlign layout {};
 		bool          valid_layout = false;
-		if (standard64) {
-			TileGetTextureSize(Prospero::GpuEnumValue(Prospero::BufferFormat::k32Float), width,
-			                   height, levels, rt.attrib3.tile_mode, &layout, mip_sizes,
-			                   mip_padded);
-			valid_layout = layout.size != 0 && layout.align == 65536;
+		if (texture_tile) {
+			TileGetTextureSize(transfer_format, width, height, levels, rt.attrib3.tile_mode,
+			                   &layout, mip_sizes, mip_padded);
+			valid_layout = layout.size != 0 && layout.align == texture_tile_layout.block.block_size;
 		} else {
 			valid_layout =
 			    levels == 1 ? TileGetRenderTargetSize(width, height, pitch, bytes_per_element,
@@ -278,8 +315,9 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, RenderCommandB
 		     " extent=%ux%ux%u view_mip=%u view_extent=%ux%u levels=%u pitch=%u"
 		     " fmt=0x%08" PRIx32 " nfmt=0x%08" PRIx32 " order=0x%08" PRIx32 " samples=%u tile=%s\n",
 		     rt_slot, rt.base.addr, backing_size, width, height, depth, rt.view.current_mip_level,
-		     view_extent.width, view_extent.height, levels, pitch, rt.info.format,
-		     rt.info.channel_type, rt.info.channel_order, samples, tile ? "tiled" : "linear");
+		     view_extent.width, view_extent.height, levels, pitch,
+		     static_cast<uint32_t>(rt.info.format), static_cast<uint32_t>(rt.info.channel_type),
+		     static_cast<uint32_t>(rt.info.channel_order), samples, tile ? "tiled" : "linear");
 	}
 
 	TextureCache::ImageDesc desc {};
@@ -294,6 +332,13 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, RenderCommandB
 	desc.info.bytes_per_block = bytes_per_element;
 	desc.info.samples         = samples;
 	desc.info.tile_mode       = rt.attrib3.tile_mode;
+	const bool has_dcc        = rt.info.dcc_compression_enable && rt.dcc_addr.addr != 0;
+	if (has_dcc) {
+		// DCC uses a separate metadata allocation. Carry its address through ImageInfo so
+		// TextureCache can associate metadata fills observed before target registration.
+		desc.info.metadata.kind          = ImageMetadataKind::Dcc;
+		desc.info.metadata.range.address = rt.dcc_addr.addr;
+	}
 	for (uint32_t level = 0; level < levels; level++) {
 		if (volume) {
 			const auto& mip             = volume_layout.mips[level];
@@ -340,7 +385,7 @@ void RenderExecutor::ResolveRenderColorTarget(uint64_t submit_id, RenderCommandB
 	r.samples                  = samples;
 	r.export_mapping           = target_format.export_mapping;
 	r.color_clear_enable       = false;
-	r.color_clear_value        = {};
+	ResolveDccClearInfo(r, target_format.format, has_dcc, rt.clear_word0.word0);
 	BindRenderTarget(r.image_id);
 }
 

src/graphics/host_gpu/renderer/colorRenderTarget.h

@@ -31,7 +31,9 @@ struct RenderColorInfo {
 	uint32_t                        target_slot      = 0;
 	uint32_t                        samples          = 1;
 	Prospero::ColorComponentMapping export_mapping;
-	bool                            color_clear_enable = false;
+	bool                            color_clear_enable             = false;
+	bool                            metadata_clear_supported       = false;
+	bool                            metadata_fixed_clear_supported = false;
 	vk::ClearColorValue             color_clear_value {};
 };
 

src/graphics/host_gpu/renderer/commandScheduler.cpp

@@ -226,12 +226,17 @@ void CommandScheduler::Wait(uint64_t tick) {
 	EXIT_IF(tick > CurrentTick());
 	if (tick >= CurrentTick()) {
 		// A stream-buffer wrap can wait while a draw is being prepared through a reference to
-		// Current(). Recycle the same command object so that reference remains valid.
-		FinishCurrent();
+		// Current(). Recycle the same command object so that reference remains valid. Deferred
+		// resources are released only at the next GPU operation boundary.
+		SubmitInfo submit;
+		auto&      submitted = SubmitCurrent(submit);
+		submitted.WaitForFenceAndReset();
+		m_master.Refresh();
+		submitted.Begin();
+		m_recording = true;
 		return;
 	}
 	m_master.Wait(tick);
-	PopPendingOperations();
 }
 
 void CommandScheduler::PopPendingOperations() {
@@ -243,17 +248,18 @@ void CommandScheduler::PopPendingOperations(bool refresh_gpu_tick) {
 		m_master.Refresh();
 	}
 	for (;;) {
-		Common::UniqueFunction<void> callback;
+		PendingOperation operation;
 		{
 			std::lock_guard lock(m_operation_mutex);
 			if (m_pending_operations.empty() ||
 			    !m_master.IsFree(m_pending_operations.front().tick)) {
 				return;
 			}
-			callback = std::move(m_pending_operations.front().callback);
+			operation = std::move(m_pending_operations.front());
 			m_pending_operations.pop();
 		}
-		RunOperation(std::move(callback));
+		WaitPriorityOperations(operation.tick);
+		RunOperation(std::move(operation.callback));
 	}
 }
 
@@ -423,10 +429,8 @@ void CommandScheduler::BeginNext() {
 	EXIT_IF(m_recording);
 
 	auto candidate = FindReusableBuffer(m_master.KnownGpuTick());
-	bool refreshed = false;
 	if (candidate < 0) {
 		m_master.Refresh();
-		refreshed = true;
 		candidate = FindReusableBuffer(m_master.KnownGpuTick());
 	}
 	if (candidate < 0) {
@@ -435,7 +439,6 @@ void CommandScheduler::BeginNext() {
 
 	m_current = candidate;
 	Current().WaitForFenceAndReset();
-	PopPendingOperations(!refreshed);
 	BindCurrent();
 	Current().Begin();
 	m_recording = true;

src/graphics/host_gpu/renderer/commandScheduler.h

@@ -55,13 +55,14 @@ public:
 	void                      DeferPriorityOperation(Common::UniqueFunction<void>&& operation);
 	[[nodiscard]] static bool InDeferredOperation() noexcept;
 
-	[[nodiscard]] bool            Active() const noexcept { return m_current >= 0; }
-	void                          CheckActive() const;
-	RenderCommandBuffer&          Current() const;
-	[[nodiscard]] uint64_t        CurrentTick() const noexcept { return m_master.CurrentTick(); }
-	[[nodiscard]] bool            IsFree(uint64_t tick);
-	[[nodiscard]] RenderContext&  Context() const noexcept { return m_context; }
-	[[nodiscard]] GraphicContext& Graphics() const noexcept { return m_graphics; }
+	[[nodiscard]] bool             Active() const noexcept { return m_current >= 0; }
+	void                           CheckActive() const;
+	RenderCommandBuffer&           Current() const;
+	[[nodiscard]] uint64_t         CurrentTick() const noexcept { return m_master.CurrentTick(); }
+	[[nodiscard]] bool             IsFree(uint64_t tick);
+	[[nodiscard]] MasterSemaphore& GetMasterSemaphore() noexcept { return m_master; }
+	[[nodiscard]] RenderContext&   Context() const noexcept { return m_context; }
+	[[nodiscard]] GraphicContext&  Graphics() const noexcept { return m_graphics; }
 
 private:
 	static constexpr size_t CommandBufferGrowStep = 4;

src/graphics/host_gpu/renderer/context.cpp

@@ -9,7 +9,6 @@
 #include "graphics/host_gpu/renderer/debug.h"
 #include "graphics/host_gpu/renderer/depthRenderTarget.h"
 #include "graphics/host_gpu/renderer/image/imageView.h"
-#include "graphics/host_gpu/renderer/pipeline/descriptorCache.h"
 #include "graphics/host_gpu/renderer/render.h"
 #include "graphics/host_gpu/renderer/renderContext.h"
 #include "graphics/host_gpu/vma.h"
@@ -17,30 +16,27 @@
 
 #include <algorithm>
 #include <bit>
+#include <cstdio>
 #include <cstring>
-#include <memory>
 namespace Libs::Graphics {
 
-FenceResourceRetainer::~FenceResourceRetainer() {
-	if (!m_resources.empty()) {
-		EXIT("fence resource retainer destroyed before release\n");
-	}
+namespace {
+
+void ReportVulkanFatal(const char* what, vk::Result result, uint32_t slot, uint64_t submit_seq,
+                       uint32_t debug_op, uint64_t debug_submit, uint32_t arg0, uint32_t arg1,
+                       uint32_t arg2, uint32_t arg3, uint64_t arg4) {
+	LOGF("%s failed: %s (%d), slot=%u submit_seq=%" PRIu64 " debug_op=%u debug_submit=%" PRIu64
+	     " args=%u,%u,%u,%u,0x%016" PRIx64 "\n",
+	     what, VulkanToString(result).c_str(), static_cast<int>(result), slot, submit_seq, debug_op,
+	     debug_submit, arg0, arg1, arg2, arg3, arg4);
+	std::printf("%s failed: %s (%d), slot=%u submit_seq=%" PRIu64
+	            " debug_op=%u debug_submit=%" PRIu64 " args=%u,%u,%u,%u,0x%016" PRIx64 "\n",
+	            what, VulkanToString(result).c_str(), static_cast<int>(result), slot, submit_seq,
+	            debug_op, debug_submit, arg0, arg1, arg2, arg3, arg4);
+	std::fflush(stdout);
 }
 
-void FenceResourceRetainer::Retain(std::shared_ptr<void> resource) {
-	if (resource == nullptr) {
-		EXIT("cannot retain a null fence resource\n");
-	}
-	if (std::ranges::none_of(m_resources, [&resource](const auto& retained) {
-		    return retained.get() == resource.get();
-	    })) {
-		m_resources.push_back(std::move(resource));
-	}
-}
-
-void FenceResourceRetainer::ReleaseAfterFence() noexcept {
-	m_resources.clear();
-}
+} // namespace
 
 CommandBuffer::CommandBuffer(CommandScheduler& scheduler)
     : m_context(scheduler.Context()), m_scheduler(scheduler), m_graphics(scheduler.Graphics()),
@@ -71,37 +67,11 @@ void CommandBuffer::Release() {
 
 	m_slot->busy = false;
 	m_slot->Reset();
-	ReleaseResourcesAfterFence();
 	m_slot = nullptr;
 
 	EXIT_NOT_IMPLEMENTED(!IsInvalid());
 }
 
-void CommandBuffer::RetireBufferAfterFence(std::unique_ptr<VulkanBuffer> buffer) {
-	if (IsInvalid() || m_execute || buffer == nullptr || buffer->buffer == nullptr) {
-		EXIT("cannot retire a buffer on an invalid or submitted command buffer\n");
-	}
-	m_retired_buffers.push_back(std::move(buffer));
-}
-
-void CommandBuffer::RetainResourceUntilFence(std::shared_ptr<void> resource) {
-	if (IsInvalid() || m_execute) {
-		EXIT("cannot retain a resource on an invalid or submitted command buffer\n");
-	}
-	m_fence_resources.Retain(std::move(resource));
-}
-
-void CommandBuffer::RecycleDescriptorAfterFence(VulkanDescriptorSet& set) {
-	m_descriptor_sets_after_fence.push_back(&set);
-}
-
-void CommandBuffer::RecycleDescriptorsAfterFence() {
-	for (auto* set: m_descriptor_sets_after_fence) {
-		m_context.GetDescriptorCache().Recycle(*set);
-	}
-	m_descriptor_sets_after_fence.clear();
-}
-
 void CommandBuffer::Begin() const {
 	EXIT_IF(m_rendering);
 	auto buffer = Handle();
@@ -169,8 +139,9 @@ void CommandBuffer::Execute(const SubmitInfo& submit) {
 
 	auto result = graphics.device.resetFences(1, &fence);
 	if (result != vk::Result::eSuccess) {
-		LOGF("vkResetFences failed before submit: %s (%d)\n", VulkanToString(result).c_str(),
-		     static_cast<int>(result));
+		ReportVulkanFatal("vkResetFences (before submit)", result, m_slot->id, m_submit_seq,
+		                  m_debug_op, m_debug_submit_id, m_debug_arg0, m_debug_arg1, m_debug_arg2,
+		                  m_debug_arg3, m_debug_arg4);
 	}
 	EXIT_NOT_IMPLEMENTED(result != vk::Result::eSuccess);
 
@@ -192,11 +163,9 @@ void CommandBuffer::Execute(const SubmitInfo& submit) {
 	m_fence_waited = false;
 
 	if (result != vk::Result::eSuccess) {
-		LOGF("vkQueueSubmit failed: %s (%d), slot=%u submit_seq=%" PRIu64
-		     " debug_op=%u debug_submit=%" PRIu64 " args=%u,%u,%u,%u,0x%016" PRIx64 "\n",
-		     VulkanToString(result).c_str(), static_cast<int>(result), m_slot->id, m_submit_seq,
-		     m_debug_op, m_debug_submit_id, m_debug_arg0, m_debug_arg1, m_debug_arg2, m_debug_arg3,
-		     m_debug_arg4);
+		ReportVulkanFatal("vkQueueSubmit", result, m_slot->id, m_submit_seq, m_debug_op,
+		                  m_debug_submit_id, m_debug_arg0, m_debug_arg1, m_debug_arg2, m_debug_arg3,
+		                  m_debug_arg4);
 	}
 	EXIT_NOT_IMPLEMENTED(result != vk::Result::eSuccess);
 }
@@ -213,11 +182,9 @@ void CommandBuffer::WaitForFenceOnly() {
 	auto device = m_graphics.device;
 	auto result = device.waitForFences(1, &m_slot->fence, VK_TRUE, UINT64_MAX);
 	if (result != vk::Result::eSuccess) {
-		LOGF("vkWaitForFences failed: %s (%d), slot=%u submit_seq=%" PRIu64
-		     " debug_op=%u debug_submit=%" PRIu64 " args=%u,%u,%u,%u,0x%016" PRIx64 "\n",
-		     VulkanToString(result).c_str(), static_cast<int>(result), m_slot->id, m_submit_seq,
-		     m_debug_op, m_debug_submit_id, m_debug_arg0, m_debug_arg1, m_debug_arg2, m_debug_arg3,
-		     m_debug_arg4);
+		ReportVulkanFatal("vkWaitForFences", result, m_slot->id, m_submit_seq, m_debug_op,
+		                  m_debug_submit_id, m_debug_arg0, m_debug_arg1, m_debug_arg2, m_debug_arg3,
+		                  m_debug_arg4);
 	}
 	EXIT_NOT_IMPLEMENTED(result != vk::Result::eSuccess);
 	m_fence_waited = true;
@@ -238,22 +205,6 @@ void CommandBuffer::FinalizeFence(bool reset_recording) {
 			m_slot->Reset();
 		}
 	}
-	if (was_executed) {
-		ReleaseResourcesAfterFence();
-	}
-	DeleteBuffersAfterFence();
-}
-
-void CommandBuffer::ReleaseResourcesAfterFence() {
-	RecycleDescriptorsAfterFence();
-	m_fence_resources.ReleaseAfterFence();
-}
-
-void CommandBuffer::DeleteBuffersAfterFence() {
-	for (const auto& buffer: m_retired_buffers) {
-		m_graphics.DeleteBuffer(*buffer);
-	}
-	m_retired_buffers.clear();
 }
 
 void CommandBuffer::BeginRendering(const RenderState& state) const {

src/graphics/host_gpu/renderer/debug.cpp

@@ -71,14 +71,16 @@ void uc_print(const char* func, const HW::UserConfig& uc) {
 	LOGF("\t GetPrimType()         = 0x%08" PRIx32 "\n"
 	     "\t GetIndexOffset()      = 0x%08" PRIx32 "\n"
 	     "\t GetObjectId()         = 0x%08" PRIx32 "\n"
+	     "\t primitive_reset       = 0x%08" PRIx32 "\n"
 	     "\t primitive_group_size  = 0x%04" PRIx16 "\n"
 	     "\t vertex_group_size     = 0x%04" PRIx16 "\n"
 	     "\t en_user_vgpr1         = %s\n"
 	     "\t en_user_vgpr2         = %s\n"
 	     "\t en_user_vgpr3         = %s\n",
-	     uc.GetPrimType(), uc.GetIndexOffset(), uc.GetObjectId(), ge_cntl.primitive_group_size,
-	     ge_cntl.vertex_group_size, user_en.vgpr1 ? "true" : "false",
-	     user_en.vgpr2 ? "true" : "false", user_en.vgpr3 ? "true" : "false");
+	     static_cast<uint32_t>(uc.GetPrimType()), uc.GetIndexOffset(), uc.GetObjectId(),
+	     uc.GetPrimitiveResetControl(), ge_cntl.primitive_group_size, ge_cntl.vertex_group_size,
+	     user_en.vgpr1 ? "true" : "false", user_en.vgpr2 ? "true" : "false",
+	     user_en.vgpr3 ? "true" : "false");
 }
 
 void uc_check(const HW::UserConfig& uc) {
@@ -141,11 +143,12 @@ std::vector<std::string> rt_print(const char* func, const HW::RenderTarget& rt)
 	    fmt::format("\t info.cmask_addr_type            = 0x{:08x}\n", rt.info.cmask_addr_type));
 	dst.push_back(fmt::format("\t info.alt_tile_mode              = {}\n",
 	                          rt.info.alt_tile_mode ? "true" : "false"));
-	dst.push_back(fmt::format("\t info.format                     = 0x{:08x}\n", rt.info.format));
-	dst.push_back(
-	    fmt::format("\t info.channel_type               = 0x{:08x}\n", rt.info.channel_type));
-	dst.push_back(
-	    fmt::format("\t info.channel_order              = 0x{:08x}\n", rt.info.channel_order));
+	dst.push_back(fmt::format("\t info.format                     = 0x{:08x}\n",
+	                          static_cast<uint32_t>(rt.info.format)));
+	dst.push_back(fmt::format("\t info.channel_type               = 0x{:08x}\n",
+	                          static_cast<uint32_t>(rt.info.channel_type)));
+	dst.push_back(fmt::format("\t info.channel_order              = 0x{:08x}\n",
+	                          static_cast<uint32_t>(rt.info.channel_order)));
 	dst.push_back(fmt::format("\t info.blend_bypa                 = {}\n",
 	                          rt.info.blend_bypass ? "true" : "false"));
 	dst.push_back(fmt::format("\t info.blend_clamp                = {}\n",
@@ -154,10 +157,10 @@ std::vector<std::string> rt_print(const char* func, const HW::RenderTarget& rt)
 	                          rt.info.round_mode ? "true" : "false"));
 	dst.push_back(fmt::format("\t attrib.force_dest_alpha_to_one  = {}\n",
 	                          rt.attrib.force_dest_alpha_to_one ? "true" : "false"));
-	dst.push_back(
-	    fmt::format("\t attrib.tile_mode                = 0x{:08x}\n", rt.attrib.tile_mode));
-	dst.push_back(
-	    fmt::format("\t attrib.fmask_tile_mode          = 0x{:08x}\n", rt.attrib.fmask_tile_mode));
+	dst.push_back(fmt::format("\t attrib.tile_mode                = 0x{:08x}\n",
+	                          static_cast<uint32_t>(rt.attrib.tile_mode)));
+	dst.push_back(fmt::format("\t attrib.fmask_tile_mode          = 0x{:08x}\n",
+	                          static_cast<uint32_t>(rt.attrib.fmask_tile_mode)));
 	dst.push_back(
 	    fmt::format("\t attrib.num_samples              = 0x{:08x}\n", rt.attrib.num_samples));
 	dst.push_back(
@@ -168,8 +171,8 @@ std::vector<std::string> rt_print(const char* func, const HW::RenderTarget& rt)
 	dst.push_back(
 	    fmt::format("\t attrib2.num_mip_levels          = 0x{:08x}\n", rt.attrib2.num_mip_levels));
 	dst.push_back(fmt::format("\t attrib3.depth                   = 0x{:08x}\n", rt.attrib3.depth));
-	dst.push_back(
-	    fmt::format("\t attrib3.tile_mode               = 0x{:08x}\n", rt.attrib3.tile_mode));
+	dst.push_back(fmt::format("\t attrib3.tile_mode               = 0x{:08x}\n",
+	                          static_cast<uint32_t>(rt.attrib3.tile_mode)));
 	dst.push_back(
 	    fmt::format("\t attrib3.dimension               = 0x{:08x}\n", rt.attrib3.dimension));
 	dst.push_back(fmt::format("\t attrib3.cmask_pipe_aligned      = {}\n",
@@ -212,22 +215,22 @@ std::vector<std::string> rt_print(const char* func, const HW::RenderTarget& rt)
 	return dst;
 }
 
-static bool RenderIsColorTileMode(uint32_t tile_mode) {
+static bool RenderIsColorTileMode(Prospero::TileMode tile_mode) {
 	// AGC CxRenderTarget::TileMode shifted by CB_COLOR*_ATTRIB3.COLOR_SW_MODE.
 	switch (tile_mode) {
-		case Prospero::GpuEnumValue(Prospero::TileMode::kLinear):
-		case Prospero::GpuEnumValue(Prospero::TileMode::kStandard256B):
-		case Prospero::GpuEnumValue(Prospero::TileMode::kStandard4KB):
-		case Prospero::GpuEnumValue(Prospero::TileMode::kStandard64KB):
-		case Prospero::GpuEnumValue(Prospero::TileMode::kPrt):
-		case Prospero::GpuEnumValue(Prospero::TileMode::kDepth):
-		case Prospero::GpuEnumValue(Prospero::TileMode::kRenderTarget): return true;
+		case Prospero::TileMode::kLinear:
+		case Prospero::TileMode::kStandard256B:
+		case Prospero::TileMode::kStandard4KB:
+		case Prospero::TileMode::kStandard64KB:
+		case Prospero::TileMode::kPrt:
+		case Prospero::TileMode::kDepth:
+		case Prospero::TileMode::kRenderTarget: return true;
 		default: return false;
 	}
 }
 
-bool RenderIsColorTileModeLinear(uint32_t tile_mode) {
-	return tile_mode == Prospero::GpuEnumValue(Prospero::TileMode::kLinear);
+bool RenderIsColorTileModeLinear(Prospero::TileMode tile_mode) {
+	return tile_mode == Prospero::TileMode::kLinear;
 }
 
 static bool RenderIsColorDimension(uint32_t dimension) {
@@ -263,20 +266,24 @@ static void RtCheck(const HW::RenderTarget& rt) {
 				logged = true;
 			}
 		}
+		const bool native_2x_one_frag = rt.info.fmask_compression_enable &&
+		                                !rt.info.fmask_data_compression_disable &&
+		                                rt.info.cmask_fast_clear_enable && rt.fmask.addr != 0 &&
+		                                rt.cmask.addr != 0 && rt.attrib.num_samples == 1 &&
+		                                rt.attrib.num_fragments == 1;
+		EXIT_NOT_IMPLEMENTED(rt.info.fmask_one_frag_mode && !native_2x_one_frag);
 		if (rt.info.fmask_compression_enable) {
 			EXIT_NOT_IMPLEMENTED(rt.attrib.num_samples == 0 && rt.attrib.num_fragments == 0);
 			static bool logged = false;
 			if (!logged) {
-				LOGF("RenderTarget: temporary: ignoring PS5 FMASK metadata for single-sample MSAA "
-				     "fallback, fmask=0x%016" PRIx64 "\n",
+				LOGF("RenderTarget: using native Vulkan MSAA without guest FMASK metadata, "
+				     "fmask=0x%016" PRIx64 "\n",
 				     rt.fmask.addr);
 				logged = true;
 			}
 		}
 
-		// EXIT_NOT_IMPLEMENTED(rt.info.fmask_compression_mode != 0x00000000);
 		EXIT_NOT_IMPLEMENTED(rt.info.fmask_data_compression_disable != false);
-		EXIT_NOT_IMPLEMENTED(rt.info.fmask_one_frag_mode != false);
 
 		if (rt.info.cmask_fast_clear_enable || rt.info.dcc_compression_enable) {
 			static bool logged = false;
@@ -361,7 +368,8 @@ static void RtCheck(const HW::RenderTarget& rt) {
 			}
 		}
 		if (!RenderIsColorTileMode(rt.attrib3.tile_mode)) {
-			EXIT("unknown PS5 render-target tile mode: 0x%08" PRIx32 "\n", rt.attrib3.tile_mode);
+			EXIT("unknown PS5 render-target tile mode: 0x%08" PRIx32 "\n",
+			     static_cast<uint32_t>(rt.attrib3.tile_mode));
 		}
 		if (!RenderIsColorDimension(rt.attrib3.dimension)) {
 			EXIT("unknown PS5 render-target dimension: 0x%08" PRIx32 "\n", rt.attrib3.dimension);
@@ -456,13 +464,14 @@ static void ZPrint(const char* func, const HW::DepthRenderTarget& z) {
 	     "\t height                                = 0x%08" PRIx32 "\n"
 	     "\t size.x_max                            = 0x%04" PRIx16 "\n"
 	     "\t size.y_max                            = 0x%04" PRIx16 "\n",
-	     z.z_info.format, z.z_info.num_samples,
-	     Prospero::GpuEnumValue(z.z_info.texture_compatibility),
+	     static_cast<uint32_t>(z.z_info.format), z.z_info.num_samples,
+	     static_cast<uint32_t>(z.z_info.texture_compatibility),
 	     z.z_info.htile_acceleration ? "true" : "false",
 	     z.z_info.expclear_enabled ? "true" : "false",
-	     Prospero::GpuEnumValue(z.z_info.z_compare_base),
+	     static_cast<uint32_t>(z.z_info.z_compare_base),
 	     z.z_info.partially_resident ? "true" : "false", z.z_info.max_mip_level,
-	     z.stencil_info.format, Prospero::GpuEnumValue(z.stencil_info.texture_compatibility),
+	     static_cast<uint32_t>(z.stencil_info.format),
+	     static_cast<uint32_t>(z.stencil_info.texture_compatibility),
 	     z.stencil_info.htile_stencil_disabled ? "true" : "false",
 	     z.stencil_info.expclear_enabled ? "true" : "false",
 	     z.stencil_info.partially_resident ? "true" : "false", z.depth_info.addr5_swizzle_mask,
@@ -491,15 +500,16 @@ static void ZCheck(const HW::DepthRenderTarget& z, const HW::DepthControl& dc,
 
 	EXIT_NOT_IMPLEMENTED(!z.z_info.HasValidTextureCompatibility());
 	EXIT_NOT_IMPLEMENTED(!z.stencil_info.HasValidTextureCompatibility());
-	if (z.z_info.format == 0) {
-		EXIT_NOT_IMPLEMENTED(z.z_info.format != 0);
+	if (z.z_info.format == Prospero::DepthFormat::kInvalid) {
+		EXIT_NOT_IMPLEMENTED(z.z_info.format != Prospero::DepthFormat::kInvalid);
 		EXIT_NOT_IMPLEMENTED(z.z_info.num_samples != 0);
 		EXIT_NOT_IMPLEMENTED(z.z_info.htile_acceleration != false);
 		EXIT_NOT_IMPLEMENTED(z.z_info.expclear_enabled != false);
 		EXIT_NOT_IMPLEMENTED(z.z_info.partially_resident != false);
 		EXIT_NOT_IMPLEMENTED(z.z_info.max_mip_level != 0);
 	} else {
-		EXIT_NOT_IMPLEMENTED(z.z_info.format != 0x00000001 && z.z_info.format != 0x00000003);
+		EXIT_NOT_IMPLEMENTED(z.z_info.format != Prospero::DepthFormat::kZ16 &&
+		                     z.z_info.format != Prospero::DepthFormat::kZ32F);
 		if (z.z_info.num_samples != 0x00000000) {
 			static bool logged = false;
 			if (!logged) {
@@ -513,16 +523,17 @@ static void ZCheck(const HW::DepthRenderTarget& z, const HW::DepthControl& dc,
 		EXIT_NOT_IMPLEMENTED(z.z_info.max_mip_level != 0);
 	}
 
-	if (z.stencil_info.format == 0) {
+	if (z.stencil_info.format == Prospero::StencilFormat::kInvalid) {
 		EXIT_NOT_IMPLEMENTED(z.stencil_info.expclear_enabled != false);
 		EXIT_NOT_IMPLEMENTED(z.stencil_info.partially_resident != false);
 	} else {
-		EXIT_NOT_IMPLEMENTED(z.stencil_info.format != 0x00000001);
+		EXIT_NOT_IMPLEMENTED(z.stencil_info.format != Prospero::StencilFormat::k8UInt);
 		EXIT_NOT_IMPLEMENTED(z.stencil_info.expclear_enabled != false);
 		EXIT_NOT_IMPLEMENTED(z.stencil_info.partially_resident != false);
 	}
 
-	if (z.z_info.format != 0 || z.stencil_info.format != 0) {
+	if (z.z_info.format != Prospero::DepthFormat::kInvalid ||
+	    z.stencil_info.format != Prospero::StencilFormat::kInvalid) {
 
 		EXIT_NOT_IMPLEMENTED(z.depth_info.addr5_swizzle_mask != 0x00000000);
 		EXIT_NOT_IMPLEMENTED(z.depth_info.array_mode != 0x00000000);
@@ -554,13 +565,11 @@ static void ZCheck(const HW::DepthRenderTarget& z, const HW::DepthControl& dc,
 				     z.depth_view.stencil_write_disable ? "true" : "false");
 			}
 		}
-		EXIT_NOT_IMPLEMENTED(z.z_read_base_addr != z.z_write_base_addr);
-		EXIT_NOT_IMPLEMENTED(z.stencil_read_base_addr != z.stencil_write_base_addr);
-		EXIT_NOT_IMPLEMENTED(z.z_write_base_addr == 0);
-		if (z.stencil_info.format != 0 && z.stencil_write_base_addr == 0) {
-			LOGF("\t warning: stencil format is set without a stencil base address, continuing "
-			     "without stencil attachment\n");
-		}
+		EXIT_NOT_IMPLEMENTED(!z.depth_view.depth_write_disable &&
+		                     z.z_read_base_addr != z.z_write_base_addr);
+		EXIT_NOT_IMPLEMENTED(!z.depth_view.stencil_write_disable &&
+		                     z.stencil_read_base_addr != z.stencil_write_base_addr);
+		EXIT_NOT_IMPLEMENTED(!z.depth_view.depth_write_disable && z.z_write_base_addr == 0);
 		// EXIT_NOT_IMPLEMENTED(z.pitch_div8_minus1 != 0x000000ff);
 		// EXIT_NOT_IMPLEMENTED(z.height_div8_minus1 != 0x0000008f);
 		// EXIT_NOT_IMPLEMENTED(z.slice_div64_minus1 != 0x00008fff);
@@ -609,8 +618,7 @@ static void ClipCheck(const HW::ClipControl& c) {
 	// dx_linear_attr_clip_enable preserves linear (noperspective) attributes at clip-generated
 	// vertices, which Vulkan provides as part of clipping and interpolation.
 	EXIT_NOT_IMPLEMENTED(c.user_clip_planes != 0 || c.user_clip_plane_mode != 0 ||
-	                     c.vertex_kill_any || !c.IsZClipModeRepresentable() ||
-	                     c.user_clip_plane_negate_y || c.clip_disable ||
+	                     c.vertex_kill_any || c.user_clip_plane_negate_y || c.clip_disable ||
 	                     c.user_clip_plane_cull_only || c.cull_on_clipping_error_disable ||
 	                     c.force_viewport_index_from_vs_enable);
 }
@@ -634,17 +642,6 @@ static void RcPrint(const char* func, const HW::RenderControl& c) {
 static void RcCheck(const HW::RenderControl& c) {
 	// EXIT_NOT_IMPLEMENTED(c.depth_clear_enable != false);
 	// EXIT_NOT_IMPLEMENTED(c.stencil_clear_enable != false);
-	if ((c.resummarize_enable || c.copy_centroid || c.copy_sample != 0)) {
-		static std::atomic<uint32_t> log_count {0};
-		if (log_count.fetch_add(1) < 16) {
-			LOGF("\t warning: unsupported PS5 render-control metadata state, continuing: "
-			     "resummarize=%s copy_centroid=%s copy_sample=%" PRIu8 "\n",
-			     c.resummarize_enable ? "true" : "false", c.copy_centroid ? "true" : "false",
-			     c.copy_sample);
-		}
-		return;
-	}
-	EXIT_NOT_IMPLEMENTED(c.resummarize_enable != false);
 	// EXIT_NOT_IMPLEMENTED(c.stencil_compress_disable != false);
 	// EXIT_NOT_IMPLEMENTED(c.depth_compress_disable != false);
 	EXIT_NOT_IMPLEMENTED(c.copy_centroid != false);
@@ -1218,8 +1215,9 @@ void hw_check(const RenderCommandBuffer& buffer) {
 			const auto& mrt = hw.GetRenderTarget(i);
 			LOGF("\t RT%u addr=0x%010" PRIx64 " mask=0x%x fmt=0x%08" PRIx32
 			     " width=%u height=%u tile=%u\n",
-			     i, mrt.base.addr, (hw.GetRenderTargetMask() >> (i * 4u)) & 0x0fu, mrt.info.format,
-			     mrt.attrib2.width + 1, mrt.attrib2.height + 1, mrt.attrib3.tile_mode);
+			     i, mrt.base.addr, (hw.GetRenderTargetMask() >> (i * 4u)) & 0x0fu,
+			     static_cast<uint32_t>(mrt.info.format), mrt.attrib2.width + 1,
+			     mrt.attrib2.height + 1, static_cast<uint32_t>(mrt.attrib3.tile_mode));
 		}
 	}
 	if (rc.depth_clear_enable && hw.GetDepthClearValue() != 0.0f &&
@@ -1258,9 +1256,10 @@ void hw_print(const RenderCommandBuffer& buffer) {
 		     "\t GetRenderTargetMask()   = 0x%08" PRIx32 "\n"
 		     "\t GetDepthClearValue()    = %f\n"
 		     "\t GetStencilClearValue()  = %" PRIu8 "\n"
-		     "\t GetLineWidth()          = %f\n",
+		     "\t GetLineWidth()          = %f\n"
+		     "\t primitive_reset_index   = 0x%08" PRIx32 "\n",
 		     hw.GetRenderTargetMask(), hw.GetDepthClearValue(), hw.GetStencilClearValue(),
-		     hw.GetLineWidth());
+		     hw.GetLineWidth(), hw.GetPrimitiveResetIndex());
 
 		LOGF("%s", Common::Concat(rt_print("RenderTraget:", rt), "").c_str());
 

src/graphics/host_gpu/renderer/debug.h

@@ -34,7 +34,7 @@ void                     uc_print(const char* func, const HW::UserConfig& uc);
 void                     uc_check(const HW::UserConfig& uc);
 void                     sh_print(const char* func, const HW::Shader& uc);
 std::vector<std::string> rt_print(const char* func, const HW::RenderTarget& rt);
-bool                     RenderIsColorTileModeLinear(uint32_t tile_mode);
+bool                     RenderIsColorTileModeLinear(Prospero::TileMode tile_mode);
 void                     hw_print(const RenderCommandBuffer& buffer);
 void                     hw_check(const RenderCommandBuffer& buffer);
 void                     LogDrawPhase(const char* draw_name, const char* phase);

src/graphics/host_gpu/renderer/depthRenderTarget.cpp

@@ -13,7 +13,6 @@
 #include "graphics/host_gpu/renderer/debug.h"
 #include "graphics/host_gpu/renderer/image/imageView.h"
 #include "graphics/host_gpu/renderer/image/textureCommon.h"
-#include "graphics/host_gpu/renderer/pipeline/descriptorCache.h"
 #include "graphics/host_gpu/renderer/render.h"
 #include "graphics/host_gpu/renderer/renderContext.h"
 #include "graphics/host_gpu/vulkanCommon.h"
@@ -67,9 +66,8 @@ static vk::StencilOp ConvertStencilOp(uint8_t value, uint8_t write_mask, uint8_t
 }
 
 static bool UsesStencilOpValue(uint8_t fail, uint8_t pass, uint8_t depth_fail) {
-	return fail == Prospero::GpuEnumValue(Prospero::StencilOp::kReplaceOp) ||
-	       pass == Prospero::GpuEnumValue(Prospero::StencilOp::kReplaceOp) ||
-	       depth_fail == Prospero::GpuEnumValue(Prospero::StencilOp::kReplaceOp);
+	constexpr auto replace_op = static_cast<uint8_t>(Prospero::StencilOp::kReplaceOp);
+	return fail == replace_op || pass == replace_op || depth_fail == replace_op;
 }
 
 [[nodiscard]] static vk::Format ResolveHostDepthAttachmentFormat(const RenderCommandBuffer& buffer,
@@ -103,15 +101,14 @@ void RenderExecutor::ResolveRenderDepthTarget(uint64_t submit_id, RenderCommandB
                                               RenderDepthInfo& r) {
 	KYTY_PROFILER_FUNCTION();
 	(void)submit_id;
-	const auto& hw = buffer.GetRegisters();
-	const auto& z  = hw.GetDepthRenderTarget();
-	const auto& rc = hw.GetRenderControl();
-	const auto& dc = hw.GetDepthControl();
-	const auto& sc = hw.GetStencilControl();
-	const auto& sm = hw.GetStencilMask();
-	const bool  has_stencil =
-	    z.stencil_info.format != Prospero::GpuEnumValue(Prospero::StencilFormat::kInvalid);
-	const bool depth_active =
+	const auto& hw          = buffer.GetRegisters();
+	const auto& z           = hw.GetDepthRenderTarget();
+	const auto& rc          = hw.GetRenderControl();
+	const auto& dc          = hw.GetDepthControl();
+	const auto& sc          = hw.GetStencilControl();
+	const auto& sm          = hw.GetStencilMask();
+	const bool  has_stencil = z.stencil_info.format != Prospero::StencilFormat::kInvalid;
+	const bool  depth_active =
 	    dc.z_enable || dc.z_write_enable || dc.depth_bounds_enable || rc.depth_clear_enable;
 	const bool stencil_active = has_stencil && (dc.stencil_enable || rc.stencil_clear_enable);
 	if (!depth_active && !stencil_active) {
@@ -122,9 +119,8 @@ void RenderExecutor::ResolveRenderDepthTarget(uint64_t submit_id, RenderCommandB
 		DepthFatal("invalid PS5 depth texture-compatibility encoding");
 	}
 	const bool attachment_unbound =
-	    z.z_info.format == Prospero::GpuEnumValue(Prospero::DepthFormat::kInvalid) &&
-	    z.stencil_info.format == Prospero::GpuEnumValue(Prospero::StencilFormat::kInvalid) &&
-	    z.z_info.num_samples == 0 &&
+	    z.z_info.format == Prospero::DepthFormat::kInvalid &&
+	    z.stencil_info.format == Prospero::StencilFormat::kInvalid && z.z_info.num_samples == 0 &&
 	    z.z_info.texture_compatibility == Prospero::TextureCompatiblePlaneCompression::kDisable &&
 	    !z.z_info.expclear_enabled && !z.z_info.partially_resident && z.z_info.max_mip_level == 0 &&
 	    z.stencil_info.texture_compatibility == Prospero::TextureCompatibleStencil::kDisable &&
@@ -169,27 +165,28 @@ void RenderExecutor::ResolveRenderDepthTarget(uint64_t submit_id, RenderCommandB
 			DepthFatal("invalid depth view: base=%u last=%u", z.depth_view.slice_start,
 			           z.depth_view.slice_max);
 	}
-	if (rc.resummarize_enable || rc.copy_centroid || rc.copy_sample != 0 ||
-	    z.z_info.expclear_enabled || z.stencil_info.expclear_enabled ||
-	    z.z_info.partially_resident || z.stencil_info.partially_resident ||
-	    z.z_info.max_mip_level != 0 || z.depth_view.current_mip_level != 0 ||
-	    z.depth_info.addr5_swizzle_mask != 0 || z.depth_info.array_mode != 0 ||
-	    z.depth_info.pipe_config != 0 || z.depth_info.bank_width != 0 ||
-	    z.depth_info.bank_height != 0 || z.depth_info.macro_tile_aspect != 0 ||
-	    z.depth_info.num_banks != 0 || z.htile_surface.linear != 0 ||
-	    z.htile_surface.full_cache != 0 || z.htile_surface.htile_uses_preload_win != 0 ||
-	    z.htile_surface.preload != 0 || z.htile_surface.prefetch_width != 0 ||
-	    z.htile_surface.prefetch_height != 0 || z.htile_surface.dst_outside_zero_to_one != 0 ||
-	    z.z_read_base_addr == 0 || z.z_write_base_addr != z.z_read_base_addr ||
+	if (rc.copy_centroid || rc.copy_sample != 0 || z.z_info.expclear_enabled ||
+	    z.stencil_info.expclear_enabled || z.z_info.partially_resident ||
+	    z.stencil_info.partially_resident || z.z_info.max_mip_level != 0 ||
+	    z.depth_view.current_mip_level != 0 || z.depth_info.addr5_swizzle_mask != 0 ||
+	    z.depth_info.array_mode != 0 || z.depth_info.pipe_config != 0 ||
+	    z.depth_info.bank_width != 0 || z.depth_info.bank_height != 0 ||
+	    z.depth_info.macro_tile_aspect != 0 || z.depth_info.num_banks != 0 ||
+	    z.htile_surface.linear != 0 || z.htile_surface.full_cache != 0 ||
+	    z.htile_surface.htile_uses_preload_win != 0 || z.htile_surface.preload != 0 ||
+	    z.htile_surface.prefetch_width != 0 || z.htile_surface.prefetch_height != 0 ||
+	    z.htile_surface.dst_outside_zero_to_one != 0 || z.z_read_base_addr == 0 ||
+	    (!z.depth_view.depth_write_disable && z.z_write_base_addr != z.z_read_base_addr) ||
 	    (z.z_read_base_addr & 0xffffu) != 0 ||
 	    dc.zfunc > static_cast<uint8_t>(vk::CompareOp::eAlways)) {
 		DepthFatal("unsupported depth register state");
 	}
 	if (has_stencil) {
-		if (z.stencil_info.format != Prospero::GpuEnumValue(Prospero::StencilFormat::k8UInt) ||
-		    !htile_stencil_compat || z.stencil_read_base_addr == 0 ||
-		    z.stencil_write_base_addr != z.stencil_read_base_addr ||
-		    (z.stencil_read_base_addr & 0xffffu) != 0 || z.depth_view.stencil_write_disable) {
+		if (z.stencil_info.format != Prospero::StencilFormat::k8UInt || !htile_stencil_compat ||
+		    z.stencil_read_base_addr == 0 ||
+		    (!z.depth_view.stencil_write_disable &&
+		     z.stencil_write_base_addr != z.stencil_read_base_addr) ||
+		    (z.stencil_read_base_addr & 0xffffu) != 0) {
 			DepthFatal("unsupported stencil attachment state");
 		}
 	} else if (z.stencil_read_base_addr != 0 || z.stencil_write_base_addr != 0 ||
@@ -229,7 +226,7 @@ void RenderExecutor::ResolveRenderDepthTarget(uint64_t submit_id, RenderCommandB
 		DepthFatal("no host depth/stencil format supports required usage for %s",
 		           VulkanToString(ideal_format).c_str());
 	}
-	const uint32_t guest_format = Prospero::GpuEnumValue(policy->guest_format);
+	const auto     guest_format = policy->guest_format;
 	const uint32_t bytes        = policy->bytes_per_element;
 	const auto     pitch        = TileGetDepthPitch(width, bytes, z.z_info.num_samples);
 	if (z.pitch_height_valid && ((static_cast<uint64_t>(z.pitch_div8_minus1) + 1u) * 8u != pitch ||
@@ -290,18 +287,22 @@ void RenderExecutor::ResolveRenderDepthTarget(uint64_t submit_id, RenderCommandB
 	r.depth_min_bounds         = hw.GetDepthBoundsMin();
 	r.depth_max_bounds         = hw.GetDepthBoundsMax();
 
-	r.stencil_clear_enable = has_stencil && rc.stencil_clear_enable;
-	r.stencil_clear_value  = hw.GetStencilClearValue();
-	r.stencil_test_enable  = has_stencil && dc.stencil_enable;
+	r.stencil_clear_enable =
+	    has_stencil && rc.stencil_clear_enable && !z.depth_view.stencil_write_disable;
+	r.stencil_clear_value = hw.GetStencilClearValue();
+	r.stencil_test_enable = has_stencil && dc.stencil_enable;
 	if (r.stencil_test_enable) {
-		const uint8_t front_write_mask = rc.stencil_clear_enable ? 0 : sm.stencil_writemask;
-		const uint8_t back_write_mask  = rc.stencil_clear_enable ? 0 : sm.stencil_writemask_bf;
+		const bool stencil_ops_disabled =
+		    rc.stencil_clear_enable || z.depth_view.stencil_write_disable;
+		const uint8_t front_write_mask = stencil_ops_disabled ? 0 : sm.stencil_writemask;
+		const uint8_t back_write_mask  = stencil_ops_disabled ? 0 : sm.stencil_writemask_bf;
 		if (dc.stencilfunc > static_cast<uint8_t>(vk::CompareOp::eAlways) ||
 		    (dc.backface_enable &&
 		     dc.stencilfunc_bf > static_cast<uint8_t>(vk::CompareOp::eAlways)) ||
-		    (UsesStencilOpValue(sc.stencil_fail, sc.stencil_zpass, sc.stencil_zfail) &&
+		    (front_write_mask != 0 &&
+		     UsesStencilOpValue(sc.stencil_fail, sc.stencil_zpass, sc.stencil_zfail) &&
 		     sm.stencil_opval != sm.stencil_testval) ||
-		    (dc.backface_enable &&
+		    (dc.backface_enable && back_write_mask != 0 &&
 		     UsesStencilOpValue(sc.stencil_fail_bf, sc.stencil_zpass_bf, sc.stencil_zfail_bf) &&
 		     sm.stencil_opval_bf != sm.stencil_testval_bf)) {
 			DepthFatal("unsupported stencil compare or replacement state");
@@ -343,7 +344,7 @@ void RenderExecutor::ResolveRenderDepthTarget(uint64_t submit_id, RenderCommandB
 	desc.info.pitch           = pitch;
 	desc.info.bytes_per_block = bytes;
 	desc.info.samples         = samples;
-	desc.info.tile_mode       = Prospero::GpuEnumValue(Prospero::TileMode::kDepth);
+	desc.info.tile_mode       = Prospero::TileMode::kDepth;
 	desc.info.mip_layout[0]   = {0, r.depth_buffer_size, pitch, height};
 	desc.info.metadata.range  = {r.htile_buffer_vaddr, r.htile_buffer_size};
 	desc.info.metadata.kind   = has_htile ? ImageMetadataKind::Htile : ImageMetadataKind::None;

src/graphics/host_gpu/renderer/image/image.cpp

@@ -646,13 +646,13 @@ void Validate(const ImageInfo& info) {
 	}
 }
 
-uint32_t RenderTargetTransferFormat(uint32_t bytes_per_element) {
+Prospero::BufferFormat RenderTargetTransferFormat(uint32_t bytes_per_element) {
 	switch (bytes_per_element) {
-		case 1: return Prospero::GpuEnumValue(Prospero::BufferFormat::k8UNorm);
-		case 2: return Prospero::GpuEnumValue(Prospero::BufferFormat::k16UNorm);
-		case 4: return Prospero::GpuEnumValue(Prospero::BufferFormat::k32Float);
-		case 8: return Prospero::GpuEnumValue(Prospero::BufferFormat::k16_16_16_16Float);
-		case 16: return Prospero::GpuEnumValue(Prospero::BufferFormat::k32_32_32_32Float);
+		case 1: return Prospero::BufferFormat::k8UNorm;
+		case 2: return Prospero::BufferFormat::k16UNorm;
+		case 4: return Prospero::BufferFormat::k32Float;
+		case 8: return Prospero::BufferFormat::k16_16_16_16Float;
+		case 16: return Prospero::BufferFormat::k32_32_32_32Float;
 		default: EXIT("unsupported render-target element size: %u\n", bytes_per_element);
 	}
 }

src/graphics/host_gpu/renderer/image/image.h

@@ -2,6 +2,7 @@
 #define EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_IMAGE_H_
 
 #include "common/assert.h"
+#include "common/slotVector.h"
 #include "graphics/host_gpu/graphicContext.h"
 #include "graphics/host_gpu/renderer/image/imageInfo.h"
 
@@ -19,15 +20,7 @@ class Buffer;
 class CommandScheduler;
 struct ImageTestAccess;
 
-struct ImageId {
-	uint32_t index      = std::numeric_limits<uint32_t>::max();
-	uint32_t generation = 0;
-
-	[[nodiscard]] explicit operator bool() const noexcept {
-		return index != std::numeric_limits<uint32_t>::max();
-	}
-	auto operator<=>(const ImageId&) const = default;
-};
+using ImageId = Common::SlotId;
 
 struct CachedImageView {
 	ImageViewInfo info;
@@ -55,6 +48,7 @@ struct ImageBinding {
 	bool is_target     = false;
 	bool needs_rebind  = false;
 	bool force_general = false;
+	bool shader_write  = false;
 };
 
 class Image final {
@@ -188,8 +182,8 @@ private:
 
 namespace ImageOps {
 
-void                   Validate(const ImageInfo& info);
-[[nodiscard]] uint32_t RenderTargetTransferFormat(uint32_t bytes_per_element);
+void                                 Validate(const ImageInfo& info);
+[[nodiscard]] Prospero::BufferFormat RenderTargetTransferFormat(uint32_t bytes_per_element);
 
 } // namespace ImageOps
 

src/graphics/host_gpu/renderer/image/imageInfo.h

@@ -67,14 +67,14 @@ struct ImageInfo {
 	ImageMetadataInfo            metadata;
 	uint32_t                     htile_clear_mask = UINT32_MAX;
 	vk::Format                   pixel_format     = vk::Format::eUndefined;
-	uint32_t                     guest_format     = 0;
+	Prospero::BufferFormat       guest_format     = Prospero::BufferFormat::kInvalid;
 	Prospero::ImageType          type             = Prospero::ImageType::kColor2D;
 	vk::Extent3D                 extent           = {1, 1, 1};
 	ImageSubresources            resources;
 	uint32_t                     pitch           = 0;
 	uint32_t                     bytes_per_block = 0;
 	uint32_t                     samples         = 1;
-	uint32_t                     tile_mode       = 0;
+	Prospero::TileMode           tile_mode       = Prospero::TileMode::kLinear;
 	bool                         bgra16          = false;
 	std::array<ImageMipInfo, 16> mip_layout {};
 
@@ -86,9 +86,7 @@ struct ImageInfo {
 	[[nodiscard]] bool IsBlock() const noexcept {
 		return Prospero::BlockCompressedBytesPerBlock(guest_format) != 0;
 	}
-	[[nodiscard]] bool IsTiled() const noexcept {
-		return tile_mode != Prospero::GpuEnumValue(Prospero::TileMode::kLinear);
-	}
+	[[nodiscard]] bool IsTiled() const noexcept { return tile_mode != Prospero::TileMode::kLinear; }
 	[[nodiscard]] constexpr bool IsVolume() const noexcept {
 		return type == Prospero::ImageType::kColor3D;
 	}
@@ -214,7 +212,6 @@ struct DepthFormatPolicy {
 	Prospero::DepthFormat     depth_format;
 	Prospero::BufferFormat    guest_format;
 	uint32_t                  bytes_per_element;
-	vk::Format                sampled_view_format;
 	vk::Format                depth_attachment_format;
 	std::array<vk::Format, 3> stencil_attachment_formats;
 };
@@ -223,21 +220,19 @@ inline constexpr std::array<DepthFormatPolicy, 2> DEPTH_FORMAT_POLICIES {{
     {Prospero::DepthFormat::kZ16,
      Prospero::BufferFormat::k16UNorm,
      2,
-     vk::Format::eR16Unorm,
      vk::Format::eD16Unorm,
      {vk::Format::eD16UnormS8Uint, vk::Format::eD24UnormS8Uint, vk::Format::eD32SfloatS8Uint}},
     {Prospero::DepthFormat::kZ32F,
      Prospero::BufferFormat::k32Float,
      4,
-     vk::Format::eR32Sfloat,
      vk::Format::eD32Sfloat,
      {vk::Format::eD32SfloatS8Uint, vk::Format::eUndefined, vk::Format::eUndefined}},
 }};
 
 [[nodiscard]] inline constexpr const DepthFormatPolicy*
-FindDepthFormatPolicy(uint32_t depth_format) noexcept {
+FindDepthFormatPolicy(Prospero::DepthFormat depth_format) noexcept {
 	for (const auto& policy: DEPTH_FORMAT_POLICIES) {
-		if (Prospero::GpuEnumValue(policy.depth_format) == depth_format) {
+		if (policy.depth_format == depth_format) {
 			return &policy;
 		}
 	}
@@ -245,9 +240,9 @@ FindDepthFormatPolicy(uint32_t depth_format) noexcept {
 }
 
 [[nodiscard]] inline constexpr const DepthFormatPolicy*
-FindGuestDepthFormatPolicy(uint32_t guest_format) noexcept {
+FindGuestDepthFormatPolicy(Prospero::BufferFormat guest_format) noexcept {
 	for (const auto& policy: DEPTH_FORMAT_POLICIES) {
-		if (Prospero::GpuEnumValue(policy.guest_format) == guest_format) {
+		if (policy.guest_format == guest_format) {
 			return &policy;
 		}
 	}
@@ -269,10 +264,11 @@ FindGuestDepthFormatPolicy(uint32_t guest_format) noexcept {
 	return has_stencil ? policy.stencil_attachment_formats.front() : policy.depth_attachment_format;
 }
 
-[[nodiscard]] inline constexpr vk::Format DepthAttachmentFormat(uint32_t depth_format,
-                                                                uint32_t stencil_format) noexcept {
+[[nodiscard]] inline constexpr vk::Format
+DepthAttachmentFormat(Prospero::DepthFormat   depth_format,
+                      Prospero::StencilFormat stencil_format) noexcept {
 	bool has_stencil = false;
-	switch (static_cast<Prospero::StencilFormat>(stencil_format)) {
+	switch (stencil_format) {
 		case Prospero::StencilFormat::kInvalid: break;
 		case Prospero::StencilFormat::k8UInt: has_stencil = true; break;
 		default: return vk::Format::eUndefined;
@@ -318,26 +314,6 @@ inline bool ImageInfo::IsDepth() const noexcept {
 	return std::bit_cast<uint32_t>(static_cast<float>(value) / 65535.0f);
 }
 
-[[nodiscard]] inline constexpr bool IsSupportedSampledDepthFormat(vk::Format image_format,
-                                                                  uint32_t   guest_format,
-                                                                  vk::Format view_format) noexcept {
-	const auto* policy = FindGuestDepthFormatPolicy(guest_format);
-	return policy != nullptr && view_format == policy->sampled_view_format &&
-	       (image_format == policy->depth_attachment_format ||
-	        IsStencilAttachmentFormat(*policy, image_format));
-}
-
-[[nodiscard]] inline constexpr bool IsSupportedSampledDepthFormat(vk::Format image_format,
-                                                                  vk::Format view_format) noexcept {
-	for (const auto& policy: DEPTH_FORMAT_POLICIES) {
-		if (IsSupportedSampledDepthFormat(image_format, Prospero::GpuEnumValue(policy.guest_format),
-		                                  view_format)) {
-			return true;
-		}
-	}
-	return false;
-}
-
 [[nodiscard]] inline constexpr bool IsSupportedDepthTargetFormat(const ImageInfo& info) {
 	const auto* policy = FindGuestDepthFormatPolicy(info.guest_format);
 	return policy != nullptr && info.bytes_per_block == policy->bytes_per_element &&
@@ -383,10 +359,10 @@ CanUseVideoOutNativeWithoutUpload(VideoOutCompression compression, bool render_t
 }
 
 struct VideoOutPixelFormatInfo {
-	vk::Format format            = vk::Format::eUndefined;
-	uint32_t   guest_format      = 0;
-	uint32_t   bytes_per_element = 0;
-	bool       bgra16            = false;
+	vk::Format             format            = vk::Format::eUndefined;
+	Prospero::BufferFormat guest_format      = Prospero::BufferFormat::kInvalid;
+	uint32_t               bytes_per_element = 0;
+	bool                   bgra16            = false;
 };
 
 struct VideoOutFormatPolicy {
@@ -396,23 +372,17 @@ struct VideoOutFormatPolicy {
 
 inline constexpr std::array<VideoOutFormatPolicy, 6> VIDEO_OUT_FORMAT_POLICIES {{
     {0x8000000022000000ull,
-     {vk::Format::eR8G8B8A8Srgb, Prospero::GpuEnumValue(Prospero::BufferFormat::k8_8_8_8Srgb), 4,
-      false}},
+     {vk::Format::eR8G8B8A8Srgb, Prospero::BufferFormat::k8_8_8_8Srgb, 4, false}},
     {0x8000000000000000ull,
-     {vk::Format::eB8G8R8A8Srgb, Prospero::GpuEnumValue(Prospero::BufferFormat::k8_8_8_8Srgb), 4,
-      false}},
+     {vk::Format::eB8G8R8A8Srgb, Prospero::BufferFormat::k8_8_8_8Srgb, 4, false}},
     {0x8100000022000000ull,
-     {vk::Format::eA2B10G10R10UnormPack32,
-      Prospero::GpuEnumValue(Prospero::BufferFormat::k10_10_10_2UNorm), 4, false}},
+     {vk::Format::eA2B10G10R10UnormPack32, Prospero::BufferFormat::k10_10_10_2UNorm, 4, false}},
     {0x8100000000000000ull,
-     {vk::Format::eA2R10G10B10UnormPack32,
-      Prospero::GpuEnumValue(Prospero::BufferFormat::k10_10_10_2UNorm), 4, false}},
+     {vk::Format::eA2R10G10B10UnormPack32, Prospero::BufferFormat::k10_10_10_2UNorm, 4, false}},
     {0xc001000622000000ull,
-     {vk::Format::eR16G16B16A16Sfloat,
-      Prospero::GpuEnumValue(Prospero::BufferFormat::k16_16_16_16Float), 8, false}},
+     {vk::Format::eR16G16B16A16Sfloat, Prospero::BufferFormat::k16_16_16_16Float, 8, false}},
     {0xc001000600000000ull,
-     {vk::Format::eR16G16B16A16Sfloat,
-      Prospero::GpuEnumValue(Prospero::BufferFormat::k16_16_16_16Float), 8, true}},
+     {vk::Format::eR16G16B16A16Sfloat, Prospero::BufferFormat::k16_16_16_16Float, 8, true}},
 }};
 
 [[nodiscard]] inline bool DecodeVideoOutPixelFormat(uint64_t                 pixel_format,
@@ -439,15 +409,15 @@ inline constexpr std::array<VideoOutFormatPolicy, 6> VIDEO_OUT_FORMAT_POLICIES {
 }
 
 [[nodiscard]] inline constexpr bool
-IsSupportedDisplayRenderTargetTileMode(uint32_t tile_mode) noexcept {
-	return tile_mode == Prospero::GpuEnumValue(Prospero::TileMode::kRenderTarget);
+IsSupportedDisplayRenderTargetTileMode(Prospero::TileMode tile_mode) noexcept {
+	return tile_mode == Prospero::TileMode::kRenderTarget;
 }
 
 [[nodiscard]] inline constexpr bool IsSupportedStandard64RenderTarget(const ImageInfo& info) {
-	if (info.tile_mode != Prospero::GpuEnumValue(Prospero::TileMode::kStandard64KB) ||
-	    info.data.address == 0 || (info.data.address & 0xffffu) != 0 || info.extent.width == 0 ||
-	    info.extent.height == 0 || info.bytes_per_block != 4 || info.resources.levels != 1 ||
-	    info.resources.layers != 1 || info.samples != 1) {
+	if (info.tile_mode != Prospero::TileMode::kStandard64KB || info.data.address == 0 ||
+	    (info.data.address & 0xffffu) != 0 || info.extent.width == 0 || info.extent.height == 0 ||
+	    info.bytes_per_block != 4 || info.resources.levels != 1 || info.resources.layers != 1 ||
+	    info.samples != 1) {
 		return false;
 	}
 	const auto expected_pitch =
@@ -460,7 +430,7 @@ IsSupportedDisplayRenderTargetTileMode(uint32_t tile_mode) noexcept {
 }
 
 [[nodiscard]] inline constexpr bool IsTiledRenderTarget(const ImageInfo& info) noexcept {
-	return info.tile_mode == Prospero::GpuEnumValue(Prospero::TileMode::kRenderTarget) ||
+	return info.tile_mode == Prospero::TileMode::kRenderTarget ||
 	       IsSupportedStandard64RenderTarget(info);
 }
 

src/graphics/host_gpu/renderer/image/imageView.cpp

@@ -36,17 +36,6 @@ namespace {
 	}
 }
 
-[[nodiscard]] bool IsDepthViewFormat(vk::Format format) {
-	switch (format) {
-		case vk::Format::eD16Unorm:
-		case vk::Format::eR16Unorm:
-		case vk::Format::eD32Sfloat:
-		case vk::Format::eR32Sfloat:
-		case vk::Format::eR32Uint: return true;
-		default: return false;
-	}
-}
-
 [[nodiscard]] bool IsValidViewType(const VulkanImage& image, const ImageViewInfo& info) {
 	switch (image.image_type) {
 		case vk::ImageType::e1D:
@@ -327,13 +316,13 @@ vk::ImageView Image::FindView(const ImageViewInfo& view_info) {
 	const auto& image      = backing;
 	auto        normalized = view_info;
 	const bool  is_storage = static_cast<bool>(normalized.usage & vk::ImageUsageFlagBits::eStorage);
-	normalized.aspect      = FullAspectMask(image.format);
-	if (normalized.aspect & vk::ImageAspectFlagBits::eDepth &&
-	    IsDepthViewFormat(normalized.format)) {
+	const auto  image_aspect = FullAspectMask(image.format);
+	if (image_aspect & vk::ImageAspectFlagBits::eDepth &&
+	    ImageViewOps::IsFormatDepthCompatible(normalized.format)) {
 		normalized.format = image.format;
 		normalized.aspect = vk::ImageAspectFlagBits::eDepth;
 	}
-	if (normalized.aspect & vk::ImageAspectFlagBits::eStencil &&
+	if (image_aspect & vk::ImageAspectFlagBits::eStencil &&
 	    IsStencilViewFormat(normalized.format)) {
 		normalized.format = image.format;
 		normalized.aspect = vk::ImageAspectFlagBits::eStencil;

src/graphics/host_gpu/renderer/image/imageView.h

@@ -12,8 +12,25 @@ namespace ImageViewOps {
 
 [[nodiscard]] vk::ImageAspectFlags DepthAspectMask(vk::Format format);
 [[nodiscard]] bool                 FormatsCompatible(vk::Format base, vk::Format view) noexcept;
+
+[[nodiscard]] inline bool IsFormatDepthCompatible(vk::Format format) noexcept {
+	switch (format) {
+		case vk::Format::eD32Sfloat:
+		case vk::Format::eR32Sfloat:
+		case vk::Format::eR32Uint:
+		case vk::Format::eD16Unorm:
+		case vk::Format::eR16Unorm: return true;
+		default: return false;
+	}
+}
 } // namespace ImageViewOps
 
+[[nodiscard]] inline bool IsSupportedSampledDepthFormat(vk::Format image_format,
+                                                        vk::Format view_format) noexcept {
+	return DepthAspectTransferFormat(image_format) != vk::Format::eUndefined &&
+	       ImageViewOps::IsFormatDepthCompatible(view_format);
+}
+
 [[nodiscard]] inline bool IsValidImageSwizzle(uint32_t swizzle) noexcept {
 	if ((swizzle & ~0xfffu) != 0) {
 		return false;
@@ -86,8 +103,14 @@ SelectSampledDepthView(vk::Format image_format, vk::Format view_format, uint32_t
 
 [[nodiscard]] inline bool
 IsSupportedSampledDepthResource(const ShaderRecompiler::IR::ImageResource& resource) noexcept {
-	return resource.kind == ShaderRecompiler::IR::ResourceKind::Image &&
-	       (resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim2D ||
+	if (resource.kind != ShaderRecompiler::IR::ResourceKind::Image &&
+	    resource.kind != ShaderRecompiler::IR::ResourceKind::ImageUint) {
+		return false;
+	}
+	if (resource.kind == ShaderRecompiler::IR::ResourceKind::ImageUint && resource.depth_compare) {
+		return false;
+	}
+	return (resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim2D ||
 	        resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim2DArray ||
 	        resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim2DMsaa ||
 	        resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim2DMsaaArray) &&
@@ -95,14 +118,6 @@ IsSupportedSampledDepthResource(const ShaderRecompiler::IR::ImageResource& resou
 	       !resource.written && !resource.atomic;
 }
 
-[[nodiscard]] inline bool
-IsSupportedSampledDepthUintResource(const ShaderRecompiler::IR::ImageResource& resource) noexcept {
-	return resource.kind == ShaderRecompiler::IR::ResourceKind::ImageUint &&
-	       resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim2D &&
-	       resource.mip_mode == ShaderRecompiler::IR::ImageMipMode::None && resource.read &&
-	       !resource.written && !resource.atomic && !resource.depth_compare;
-}
-
 inline void ValidateStorageColorView(vk::Format image_format, vk::Format view_format,
                                      uint32_t swizzle) noexcept {
 	if (!ImageViewOps::FormatsCompatible(image_format, view_format) ||
@@ -120,7 +135,11 @@ IsSupportedStorageImageResource(const ShaderRecompiler::IR::ImageResource& resou
 	        resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim2D ||
 	        resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim3D ||
 	        resource.dimension == ShaderRecompiler::Decoder::ImageDimension::Dim2DArray) &&
-	       resource.mip_mode == ShaderRecompiler::IR::ImageMipMode::None && resource.written &&
+	       ((resource.mip_mode == ShaderRecompiler::IR::ImageMipMode::None &&
+	         resource.mip_count == 1u) ||
+	        (resource.mip_mode == ShaderRecompiler::IR::ImageMipMode::DynamicStorage &&
+	         resource.mip_count != 0u)) &&
+	       resource.written &&
 	       (!resource.atomic ||
 	        (resource.kind == ShaderRecompiler::IR::ResourceKind::StorageImageUint &&
 	         resource.read)) &&

src/graphics/host_gpu/renderer/image/textureCommon.cpp

@@ -14,156 +14,67 @@
 namespace Libs::Graphics {
 namespace {
 
-struct RenderTargetFormatMapping {
-	Prospero::ChannelLayout layout;
-	Prospero::ChannelType   type;
-	Prospero::ChannelOrder  order;
-	RenderTargetFormatInfo  info;
+// Rows are channel order; columns are the number of physical components minus one. Unused
+// selectors complete each entry to a permutation so logical write masks can be inverted.
+constexpr Prospero::ColorComponentMapping kRenderTargetColorMappings[4][4] = {
+    {Prospero::ColorMappingRgba, Prospero::ColorMappingRgba, Prospero::ColorMappingRgba,
+     Prospero::ColorMappingRgba},
+    {Prospero::ColorMappingGr, Prospero::ColorMappingRabg, Prospero::ColorMappingRgab,
+     Prospero::ColorMappingBgra},
+    {Prospero::ColorMappingBgra, Prospero::ColorMappingGr, Prospero::ColorMappingBgra,
+     Prospero::ColorMappingAbgr},
+    {Prospero::ColorMappingAgba, Prospero::ColorMappingArbg, Prospero::ColorMappingAgbr,
+     Prospero::ColorMappingArgb},
 };
 
-constexpr RenderTargetFormatMapping kRenderTargetFormats[] = {
-    {Prospero::ChannelLayout::k8,
-     Prospero::ChannelType::kUInt,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR8Uint, 1}},
-    {Prospero::ChannelLayout::k8_8,
-     Prospero::ChannelType::kUNorm,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR8G8Unorm, 2}},
-    {Prospero::ChannelLayout::k8_8_8_8,
-     Prospero::ChannelType::kUNorm,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR8G8B8A8Unorm, 4}},
-    {Prospero::ChannelLayout::k8_8_8_8,
-     Prospero::ChannelType::kSNorm,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR8G8B8A8Snorm, 4}},
-    {Prospero::ChannelLayout::k8_8_8_8,
-     Prospero::ChannelType::kSrgb,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR8G8B8A8Srgb, 4}},
-    {Prospero::ChannelLayout::k8_8_8_8,
-     Prospero::ChannelType::kUNorm,
-     Prospero::ChannelOrder::kAlt,
-     {vk::Format::eB8G8R8A8Unorm, 4}},
-    {Prospero::ChannelLayout::k8_8_8_8,
-     Prospero::ChannelType::kSNorm,
-     Prospero::ChannelOrder::kAlt,
-     {vk::Format::eB8G8R8A8Snorm, 4}},
-    {Prospero::ChannelLayout::k8_8_8_8,
-     Prospero::ChannelType::kSrgb,
-     Prospero::ChannelOrder::kAlt,
-     {vk::Format::eB8G8R8A8Srgb, 4}},
-    {Prospero::ChannelLayout::k5_5_5_1,
-     Prospero::ChannelType::kUNorm,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR5G5B5A1UnormPack16, 2}},
-    {Prospero::ChannelLayout::k4_4_4_4,
-     Prospero::ChannelType::kUNorm,
-     Prospero::ChannelOrder::kReversed,
-     {vk::Format::eB4G4R4A4UnormPack16, 2}},
-    {Prospero::ChannelLayout::k10_10_10_2,
-     Prospero::ChannelType::kUNorm,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eA2B10G10R10UnormPack32, 4}},
-    {Prospero::ChannelLayout::k10_10_10_2,
-     Prospero::ChannelType::kUNorm,
-     Prospero::ChannelOrder::kAlt,
-     {vk::Format::eA2R10G10B10UnormPack32, 4}},
-    {Prospero::ChannelLayout::k11_11_10,
-     Prospero::ChannelType::kFloat,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eB10G11R11UfloatPack32, 4}},
-    {Prospero::ChannelLayout::k5_6_5,
-     Prospero::ChannelType::kUNorm,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eB5G6R5UnormPack16, 2}},
-    {Prospero::ChannelLayout::k16,
-     Prospero::ChannelType::kUNorm,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR16Unorm, 2}},
-    {Prospero::ChannelLayout::k16,
-     Prospero::ChannelType::kUInt,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR16Uint, 2}},
-    {Prospero::ChannelLayout::k16,
-     Prospero::ChannelType::kFloat,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR16Sfloat, 2}},
-    {Prospero::ChannelLayout::k16_16,
-     Prospero::ChannelType::kUNorm,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR16G16Unorm, 4}},
-    {Prospero::ChannelLayout::k16_16,
-     Prospero::ChannelType::kSNorm,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR16G16Snorm, 4}},
-    {Prospero::ChannelLayout::k16_16,
-     Prospero::ChannelType::kUInt,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR16G16Uint, 4}},
-    {Prospero::ChannelLayout::k16_16,
-     Prospero::ChannelType::kFloat,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR16G16Sfloat, 4}},
-    {Prospero::ChannelLayout::k16_16_16_16,
-     Prospero::ChannelType::kUNorm,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR16G16B16A16Unorm, 8}},
-    {Prospero::ChannelLayout::k16_16_16_16,
-     Prospero::ChannelType::kUInt,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR16G16B16A16Uint, 8}},
-    {Prospero::ChannelLayout::k16_16_16_16,
-     Prospero::ChannelType::kFloat,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR16G16B16A16Sfloat, 8}},
-    {Prospero::ChannelLayout::k16_16_16_16,
-     Prospero::ChannelType::kFloat,
-     Prospero::ChannelOrder::kAlt,
-     {vk::Format::eR16G16B16A16Sfloat, 8, Prospero::ColorMappingBgra}},
-    {Prospero::ChannelLayout::k16_16_16_16,
-     Prospero::ChannelType::kFloat,
-     Prospero::ChannelOrder::kReversed,
-     {vk::Format::eR16G16B16A16Sfloat, 8, Prospero::ColorMappingAbgr}},
-    {Prospero::ChannelLayout::k32,
-     Prospero::ChannelType::kFloat,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR32Sfloat, 4}},
-    {Prospero::ChannelLayout::k32_32,
-     Prospero::ChannelType::kUInt,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR32G32Uint, 8}},
-    {Prospero::ChannelLayout::k32_32,
-     Prospero::ChannelType::kFloat,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR32G32Sfloat, 8}},
-    {Prospero::ChannelLayout::k32_32_32_32,
-     Prospero::ChannelType::kFloat,
-     Prospero::ChannelOrder::kStandard,
-     {vk::Format::eR32G32B32A32Sfloat, 16}},
+struct RenderTargetHostFormatInfo {
+	vk::Format                      format = vk::Format::eUndefined;
+	Prospero::ColorComponentMapping host_to_storage;
 };
 
+RenderTargetHostFormatInfo ResolveRenderTargetHostFormat(Prospero::BufferFormat guest_format,
+                                                         Prospero::ChannelOrder order) {
+	if (order == Prospero::ChannelOrder::kAlt) {
+		switch (guest_format) {
+			case Prospero::BufferFormat::k8_8_8_8UNorm:
+				return {vk::Format::eB8G8R8A8Unorm, Prospero::ColorMappingBgra};
+			case Prospero::BufferFormat::k8_8_8_8SNorm:
+				return {vk::Format::eB8G8R8A8Snorm, Prospero::ColorMappingBgra};
+			case Prospero::BufferFormat::k8_8_8_8Srgb:
+				return {vk::Format::eB8G8R8A8Srgb, Prospero::ColorMappingBgra};
+			case Prospero::BufferFormat::k10_10_10_2UNorm:
+				return {vk::Format::eA2R10G10B10UnormPack32, Prospero::ColorMappingBgra};
+			default: break;
+		}
+	}
+	switch (guest_format) {
+		case Prospero::BufferFormat::k5_5_5_1UNorm:
+			return {vk::Format::eA1R5G5B5UnormPack16, Prospero::ColorMappingBgra};
+		case Prospero::BufferFormat::k1_5_5_5UNorm:
+			return {vk::Format::eR5G5B5A1UnormPack16, Prospero::ColorMappingAbgr};
+		case Prospero::BufferFormat::k4_4_4_4UNorm:
+			return {vk::Format::eR4G4B4A4UnormPack16, Prospero::ColorMappingAbgr};
+		default: return {VulkanFormat(guest_format), {}};
+	}
+}
+
 } // namespace
 
-// TODO: cleanup!
-RenderTargetFormatInfo TextureGetRenderTargetFormat(uint32_t raw_layout, uint32_t raw_type,
-                                                    uint32_t raw_order) {
-	const auto layout = static_cast<Prospero::ChannelLayout>(raw_layout);
-	const auto type   = static_cast<Prospero::ChannelType>(raw_type);
-	const auto order  = static_cast<Prospero::ChannelOrder>(raw_order);
-
-	if (layout == Prospero::ChannelLayout::k8 && type == Prospero::ChannelType::kUNorm &&
-	    raw_order <= Prospero::GpuEnumValue(Prospero::ChannelOrder::kAltReversed)) {
-		return {vk::Format::eR8Unorm, 1};
-	}
-	for (const auto& mapping: kRenderTargetFormats) {
-		if (mapping.layout == layout && mapping.type == type && mapping.order == order) {
-			return mapping.info;
+RenderTargetFormatInfo TextureGetRenderTargetFormat(Prospero::ChannelLayout layout,
+                                                    Prospero::ChannelType   type,
+                                                    Prospero::ChannelOrder  order) {
+	const auto encoding = Prospero::ResolveRenderTargetFormat(layout, type);
+	if (encoding.IsValid() && encoding.SupportsOrder(order)) {
+		const auto host  = ResolveRenderTargetHostFormat(encoding.buffer_format, order);
+		const auto bytes = Prospero::RenderTargetBytesPerElement(encoding.buffer_format);
+		if (host.format != vk::Format::eUndefined && bytes != 0) {
+			const auto order_mapping =
+			    kRenderTargetColorMappings[static_cast<size_t>(order)][encoding.components - 1u];
+			return {host.format, bytes, host.host_to_storage.Then(order_mapping)};
 		}
 	}
-	EXIT("unsupported render-target format combination: layout=%u type=%u order=%u\n", raw_layout,
-	     raw_type, raw_order);
+	EXIT("unsupported render-target format combination: layout=%u type=%u order=%u\n",
+	     static_cast<uint32_t>(layout), static_cast<uint32_t>(type), static_cast<uint32_t>(order));
 }
 
 namespace {
@@ -208,13 +119,16 @@ vk::ComponentMapping TextureGetComponentMapping(uint32_t swizzle) {
 	return components;
 }
 
-vk::Format TextureGetFormat(uint32_t fmt) {
-	const auto vk_format = VulkanFormat(fmt);
+SurfaceFormatInfo TextureGetSurfaceFormatInfo(Prospero::BufferFormat format) {
+	const auto backing_format    = Prospero::RemapTextureFormat(format);
+	const auto vk_format         = VulkanFormat(backing_format);
+	const auto conversion_format =
+	    backing_format != format ? format : Prospero::BufferFormat::kInvalid;
 	if (vk_format != vk::Format::eUndefined) {
-		return vk_format;
+		return SurfaceFormatInfo(vk_format, conversion_format);
 	}
-	EXIT("unknown format: fmt = %u\n", fmt);
-	return vk::Format::eUndefined;
+	EXIT("unknown format: fmt = %u\n", static_cast<uint32_t>(format));
+	return SurfaceFormatInfo(vk::Format::eUndefined, Prospero::BufferFormat::kInvalid);
 }
 
 namespace {
@@ -271,13 +185,14 @@ uint64_t SetLinearUploadLevels(TextureUploadMipLayout*         mips,
 
 } // namespace
 
-TextureUploadLayout TextureCalcUploadLayout(uint32_t fmt, uint32_t width, uint32_t height,
-                                            uint32_t levels, uint32_t depth, uint32_t tile,
-                                            uint64_t upload_size, bool allow_depth_tile,
-                                            bool volume_texture, const char* owner) {
+TextureUploadLayout TextureCalcUploadLayout(Prospero::BufferFormat format, uint32_t width,
+                                            uint32_t height, uint32_t levels, uint32_t depth,
+                                            Prospero::TileMode tile, uint64_t upload_size,
+                                            bool allow_depth_tile, bool volume_texture,
+                                            const char* owner) {
 	TextureUploadLayout layout {};
 	layout.surface.description = {
-	    fmt,
+	    format,
 	    tile,
 	    volume_texture ? TileSurfaceDimension::Dim3D : TileSurfaceDimension::Dim2D,
 	    width,
@@ -287,35 +202,39 @@ TextureUploadLayout TextureCalcUploadLayout(uint32_t fmt, uint32_t width, uint32
 	    volume_texture ? 1u : depth,
 	};
 	const auto&              description = layout.surface.description;
-	const auto               tile_mode   = static_cast<Prospero::TileMode>(description.tile_mode);
+	const auto               tile_mode   = description.tile_mode;
 	TileTextureElementLayout element {};
 
-	if (fmt == 0) {
+	if (format == Prospero::BufferFormat::kInvalid) {
 		EXIT("%s: legacy texture upload format unsupported: fmt=0 tile=%u size=%" PRIu64
 		     " extent=%ux%u levels=%u\n",
-		     owner, description.tile_mode, upload_size, width, height, levels);
+		     owner, static_cast<uint32_t>(description.tile_mode), upload_size, width, height,
+		     levels);
 	}
 
 	switch (tile_mode) {
 		case Prospero::TileMode::kLinear:
-			if (!TileGetTextureElementLayout(fmt, element)) {
-				EXIT("%s: unsupported linear texture format: fmt=%u\n", owner, fmt);
+			if (!TileGetTextureElementLayout(format, element)) {
+				EXIT("%s: unsupported linear texture format: fmt=%u\n", owner,
+				     static_cast<uint32_t>(format));
 			}
 			break;
 		case Prospero::TileMode::kDepth:
 			if (!allow_depth_tile || !TileGetTiledTextureLayout(description, layout.surface)) {
 				EXIT("%s: unsupported typed tiled upload: fmt=%u tile=%u "
 				     "size=%" PRIu64 " extent=%ux%u levels=%u\n",
-				     owner, static_cast<uint32_t>(fmt), description.tile_mode, upload_size, width,
-				     height, levels);
+				     owner, static_cast<uint32_t>(format),
+				     static_cast<uint32_t>(description.tile_mode), upload_size, width, height,
+				     levels);
 			}
 			break;
 		default:
 			if (!TileGetTiledTextureLayout(description, layout.surface)) {
 				EXIT("%s: unsupported typed tiled upload: fmt=%u tile=%u "
 				     "size=%" PRIu64 " extent=%ux%u levels=%u\n",
-				     owner, static_cast<uint32_t>(fmt), description.tile_mode, upload_size, width,
-				     height, levels);
+				     owner, static_cast<uint32_t>(format),
+				     static_cast<uint32_t>(description.tile_mode), upload_size, width, height,
+				     levels);
 			}
 			break;
 	}
@@ -324,12 +243,12 @@ TextureUploadLayout TextureCalcUploadLayout(uint32_t fmt, uint32_t width, uint32
 		element = {layout.surface.texture.block.bytes_per_element,
 		           layout.surface.texture.texel_width, layout.surface.texture.texel_height};
 	}
-	layout.pitch = TileGetTexturePitch(fmt, width, description.tile_mode);
+	layout.pitch = TileGetTexturePitch(format, width, description.tile_mode);
 	if (tile_mode == Prospero::TileMode::kLinear) {
 		TileSizeOffset level_sizes[16] {};
 		TilePaddedSize padded_sizes[16] {};
-		TileGetTextureSize(fmt, width, height, levels, description.tile_mode, nullptr, level_sizes,
-		                   padded_sizes);
+		TileGetTextureSize(format, width, height, levels, description.tile_mode, nullptr,
+		                   level_sizes, padded_sizes);
 		for (uint32_t level = 0; level < levels; ++level) {
 			layout.mips[level] = {level_sizes[level].offset, level_sizes[level].size,
 			                      padded_sizes[level].width, padded_sizes[level].height};
@@ -356,10 +275,9 @@ std::vector<vk::BufferImageCopy> TextureBuildImageCopies(const TextureUploadLayo
 	const auto  depth          = volume_texture ? description.depth : description.layers;
 	uint32_t    mip_width      = description.width;
 	uint32_t    mip_height     = description.height;
-	uint32_t mip_pitch = volume_texture && static_cast<Prospero::TileMode>(description.tile_mode) !=
-	                                           Prospero::TileMode::kLinear
-	                         ? description.width
-	                         : layout.pitch;
+	uint32_t    mip_pitch = volume_texture && description.tile_mode != Prospero::TileMode::kLinear
+	                            ? description.width
+	                            : layout.pitch;
 
 	std::vector<vk::BufferImageCopy> regions;
 	regions.reserve(GetTextureRegionCount(depth, description.levels, volume_texture));
@@ -376,8 +294,7 @@ std::vector<vk::BufferImageCopy> TextureBuildImageCopies(const TextureUploadLayo
 			                           1};
 			region.imageOffset.z    = volume_texture ? static_cast<int>(z) : 0;
 			region.imageExtent      = {mip_width, mip_height, 1};
-			const bool linear       = static_cast<Prospero::TileMode>(description.tile_mode) ==
-			                          Prospero::TileMode::kLinear;
+			const bool linear       = description.tile_mode == Prospero::TileMode::kLinear;
 			if (linear) {
 				region.bufferRowLength   = layout.mips[i].row_length;
 				region.bufferImageHeight = layout.mips[i].image_height;

src/graphics/host_gpu/renderer/image/textureCommon.h

@@ -3,7 +3,7 @@
 
 #include "common/abi.h"
 #include "common/common.h"
-#include "graphics/guest_gpu/gpu_defs.h"
+#include "graphics/guest_gpu/gpu_format.h"
 #include "graphics/guest_gpu/tile.h"
 #include "graphics/host_gpu/vulkanCommon.h"
 
@@ -19,6 +19,15 @@ struct RenderTargetFormatInfo {
 	Prospero::ColorComponentMapping export_mapping;
 };
 
+struct SurfaceFormatInfo {
+	explicit constexpr SurfaceFormatInfo(vk::Format format,
+	                                     Prospero::BufferFormat conversion_format)
+	    : vk_format(format), conversion_format(conversion_format) {}
+
+	vk::Format             vk_format;
+	Prospero::BufferFormat conversion_format;
+};
+
 struct TextureUploadMipLayout {
 	uint64_t offset       = 0;
 	uint64_t size         = 0;
@@ -35,12 +44,15 @@ struct TextureUploadLayout {
 };
 
 vk::ComponentMapping   TextureGetComponentMapping(uint32_t swizzle);
-vk::Format             TextureGetFormat(uint32_t fmt);
-RenderTargetFormatInfo TextureGetRenderTargetFormat(uint32_t layout, uint32_t type, uint32_t order);
-TextureUploadLayout    TextureCalcUploadLayout(uint32_t fmt, uint32_t width, uint32_t height,
-                                               uint32_t levels, uint32_t depth, uint32_t tile,
-                                               uint64_t upload_size, bool allow_depth_tile,
-                                               bool volume_texture, const char* owner);
+SurfaceFormatInfo      TextureGetSurfaceFormatInfo(Prospero::BufferFormat format);
+RenderTargetFormatInfo TextureGetRenderTargetFormat(Prospero::ChannelLayout layout,
+                                                    Prospero::ChannelType   type,
+                                                    Prospero::ChannelOrder  order);
+TextureUploadLayout    TextureCalcUploadLayout(Prospero::BufferFormat format, uint32_t width,
+                                               uint32_t height, uint32_t levels, uint32_t depth,
+                                               Prospero::TileMode tile, uint64_t upload_size,
+                                               bool allow_depth_tile, bool volume_texture,
+                                               const char* owner);
 std::vector<vk::BufferImageCopy> TextureBuildImageCopies(const TextureUploadLayout& layout);
 bool TextureBuildGpuTileInfos(uint64_t tiled_size, const std::vector<vk::BufferImageCopy>& regions,
                               const TextureUploadLayout& layout, uint32_t levels,

src/graphics/host_gpu/renderer/pipeline/descriptorCache.cpp

@@ -1,300 +0,0 @@
-#include "graphics/host_gpu/renderer/pipeline/descriptorCache.h"
-
-#include "common/assert.h"
-#include "common/profiler.h"
-#include "graphics/shader/recompiler/ir/ShaderIR.h"
-
-#include <array>
-
-namespace Libs::Graphics {
-namespace {
-
-using BindingKind = ShaderRecompiler::IR::DescriptorBindingKind;
-
-vk::ShaderStageFlags StageFlags(DescriptorCache::Stage stage) {
-	switch (stage) {
-		case DescriptorCache::Stage::Vertex: return vk::ShaderStageFlagBits::eVertex;
-		case DescriptorCache::Stage::Pixel: return vk::ShaderStageFlagBits::eFragment;
-		case DescriptorCache::Stage::Compute: return vk::ShaderStageFlagBits::eCompute;
-		default: EXIT("unknown descriptor stage\n");
-	}
-}
-
-bool IsSampledImage(BindingKind kind) {
-	switch (kind) {
-		case BindingKind::Sampled1D:
-		case BindingKind::Sampled1DArray:
-		case BindingKind::Sampled2D:
-		case BindingKind::Sampled2DArray:
-		case BindingKind::Sampled2DMsaa:
-		case BindingKind::Sampled2DMsaaArray:
-		case BindingKind::Sampled3D:
-		case BindingKind::SampledUint1D:
-		case BindingKind::SampledUint1DArray:
-		case BindingKind::SampledUint2D:
-		case BindingKind::SampledUint2DArray:
-		case BindingKind::SampledUint2DMsaa:
-		case BindingKind::SampledUint2DMsaaArray:
-		case BindingKind::SampledUint3D: return true;
-		default: return false;
-	}
-}
-
-bool IsStorageImage(BindingKind kind) {
-	switch (kind) {
-		case BindingKind::Storage1D:
-		case BindingKind::Storage1DArray:
-		case BindingKind::Storage2D:
-		case BindingKind::Storage2DArray:
-		case BindingKind::Storage3D:
-		case BindingKind::StorageUint1D:
-		case BindingKind::StorageUint1DArray:
-		case BindingKind::StorageUint2D:
-		case BindingKind::StorageUint2DArray:
-		case BindingKind::StorageUint3D: return true;
-		default: return false;
-	}
-}
-
-vk::DescriptorType DescriptorType(BindingKind kind) {
-	if (kind == BindingKind::Samplers) {
-		return vk::DescriptorType::eSampler;
-	}
-	if (IsSampledImage(kind)) {
-		return vk::DescriptorType::eSampledImage;
-	}
-	if (IsStorageImage(kind)) {
-		return vk::DescriptorType::eStorageImage;
-	}
-	return vk::DescriptorType::eStorageBuffer;
-}
-
-uint32_t DescriptorCount(const ShaderRecompiler::IR::DescriptorBinding& binding) {
-	return binding.resources.empty() ? 1u : static_cast<uint32_t>(binding.resources.size());
-}
-
-std::vector<uint32_t> LayoutKey(DescriptorCache::Stage               stage,
-                                const ShaderRecompiler::IR::Program& program) {
-	std::vector<uint32_t> key;
-	key.reserve(1u + program.bindings.descriptors.size() * 4u);
-	key.push_back(static_cast<uint32_t>(stage));
-	for (const auto& binding: program.bindings.descriptors) {
-		key.push_back(static_cast<uint32_t>(binding.kind));
-		key.push_back(binding.binding);
-		key.push_back(DescriptorCount(binding));
-		key.push_back(static_cast<uint32_t>(DescriptorType(binding.kind)));
-	}
-	return key;
-}
-
-vk::DescriptorBufferInfo BufferInfo(const BufferView& view) {
-	EXIT_IF(view.buffer == nullptr);
-	return {view.buffer, view.offset, view.range};
-}
-
-} // namespace
-
-vk::DescriptorImageInfo DescriptorCache::MakeImageInfo(const TextureBinding& texture) {
-	EXIT_IF(!texture.image_id || texture.image_view == nullptr ||
-	        texture.layout == vk::ImageLayout::eUndefined);
-	return {nullptr, texture.image_view, texture.layout};
-}
-
-DescriptorCache::~DescriptorCache() {
-	for (auto& [layout, sets]: m_free_sets_by_layout) {
-		(void)layout;
-		for (auto* set: sets) {
-			delete set;
-		}
-	}
-	for (const auto& [key, layout]: m_descriptor_set_layouts) {
-		(void)key;
-		m_graphics.device.destroyDescriptorSetLayout(layout, nullptr);
-	}
-	for (const auto& pool: m_pools) {
-		m_graphics.device.destroyDescriptorPool(pool.pool, nullptr);
-	}
-}
-
-vk::DescriptorSetLayout
-DescriptorCache::GetDescriptorSetLayoutInternal(Stage                                stage,
-                                                const ShaderRecompiler::IR::Program& program) {
-	const auto key = LayoutKey(stage, program);
-	if (const auto found = m_descriptor_set_layouts.find(key);
-	    found != m_descriptor_set_layouts.end()) {
-		return found->second;
-	}
-
-	std::vector<vk::DescriptorSetLayoutBinding> bindings;
-	bindings.reserve(program.bindings.descriptors.size());
-	for (const auto& descriptor: program.bindings.descriptors) {
-		bindings.push_back({descriptor.binding, DescriptorType(descriptor.kind),
-		                    DescriptorCount(descriptor), StageFlags(stage), nullptr});
-	}
-	if (bindings.empty()) {
-		return nullptr;
-	}
-
-	vk::DescriptorSetLayoutCreateInfo info {};
-	info.sType                     = vk::StructureType::eDescriptorSetLayoutCreateInfo;
-	info.bindingCount              = static_cast<uint32_t>(bindings.size());
-	info.pBindings                 = bindings.data();
-	vk::DescriptorSetLayout layout = nullptr;
-	const auto result = m_graphics.device.createDescriptorSetLayout(&info, nullptr, &layout);
-	EXIT_NOT_IMPLEMENTED(result != vk::Result::eSuccess || layout == nullptr);
-	m_descriptor_set_layouts.emplace(key, layout);
-	return layout;
-}
-
-void DescriptorCache::CreatePool() {
-	KYTY_PROFILER_FUNCTION();
-	constexpr uint32_t           MaxSets = 512;
-	const vk::DescriptorPoolSize sizes[] = {
-	    {vk::DescriptorType::eStorageBuffer,
-	     MaxSets * (ShaderRecompiler::IR::ShaderInfo::MaxBuffers +
-	                ShaderRecompiler::IR::ShaderInfo::MaxAddresses + 3u)},
-	    {vk::DescriptorType::eSampledImage, MaxSets * ShaderRecompiler::IR::ShaderInfo::MaxImages},
-	    {vk::DescriptorType::eStorageImage, MaxSets * ShaderRecompiler::IR::ShaderInfo::MaxImages},
-	    {vk::DescriptorType::eSampler, MaxSets * ShaderRecompiler::IR::ShaderInfo::MaxSamplers},
-	};
-	vk::DescriptorPoolCreateInfo info {};
-	info.sType         = vk::StructureType::eDescriptorPoolCreateInfo;
-	info.poolSizeCount = static_cast<uint32_t>(std::size(sizes));
-	info.pPoolSizes    = sizes;
-	info.maxSets       = MaxSets;
-	const auto pool_id = static_cast<int>(m_pools.size());
-	auto&      pool    = m_pools.emplace_back();
-	const auto result  = m_graphics.device.createDescriptorPool(&info, nullptr, &pool.pool);
-	EXIT_NOT_IMPLEMENTED(result != vk::Result::eSuccess || pool.pool == nullptr);
-	pool.next_free_pool = m_first_free_pool;
-	m_first_free_pool   = pool_id;
-}
-
-VulkanDescriptorSet* DescriptorCache::Allocate(Stage                                stage,
-                                               const ShaderRecompiler::IR::Program& program) {
-	KYTY_PROFILER_FUNCTION();
-	Common::LockGuard lock(m_mutex);
-	const auto        layout = GetDescriptorSetLayoutInternal(stage, program);
-	EXIT_IF(layout == nullptr);
-	auto& free_sets = m_free_sets_by_layout[layout];
-	if (!free_sets.empty()) {
-		auto* result = free_sets.back();
-		free_sets.pop_back();
-		return result;
-	}
-
-	constexpr uint32_t Batch = 32;
-	for (int attempt = 0; attempt < 2; attempt++) {
-		for (int pool_id = m_first_free_pool; pool_id != -1;
-		     pool_id     = m_pools[pool_id].next_free_pool) {
-			auto&                                      pool = m_pools[pool_id];
-			std::array<vk::DescriptorSetLayout, Batch> layouts;
-			layouts.fill(layout);
-			std::array<vk::DescriptorSet, Batch> sets {};
-			vk::DescriptorSetAllocateInfo        info {};
-			info.sType              = vk::StructureType::eDescriptorSetAllocateInfo;
-			info.descriptorPool     = pool.pool;
-			info.descriptorSetCount = Batch;
-			info.pSetLayouts        = layouts.data();
-			if (m_graphics.device.allocateDescriptorSets(&info, sets.data()) ==
-			    vk::Result::eSuccess) {
-				free_sets.reserve(free_sets.size() + Batch - 1u);
-				for (uint32_t i = 0; i + 1u < Batch; i++) {
-					free_sets.push_back(new VulkanDescriptorSet {sets[i], layout, pool_id});
-				}
-				return new VulkanDescriptorSet {sets.back(), layout, pool_id};
-			}
-			m_first_free_pool = pool.next_free_pool;
-			break;
-		}
-		CreatePool();
-	}
-	return nullptr;
-}
-
-void DescriptorCache::Recycle(VulkanDescriptorSet& set) {
-	EXIT_IF(set.set == nullptr || set.layout == nullptr);
-	Common::LockGuard lock(m_mutex);
-	m_free_sets_by_layout[set.layout].push_back(&set);
-}
-
-VulkanDescriptorSet& DescriptorCache::GetDescriptor(Stage                                stage,
-                                                    const ShaderRecompiler::IR::Program& program,
-                                                    const NativeDescriptors&             data) {
-	KYTY_PROFILER_FUNCTION();
-	EXIT_IF(data.buffers.size() != program.info.buffers.size() ||
-	        data.images.size() != program.info.images.size() ||
-	        data.samplers.size() != program.info.samplers.size() ||
-	        data.addresses.size() != program.info.addresses.size());
-	auto* set = Allocate(stage, program);
-	EXIT_NOT_IMPLEMENTED(set == nullptr);
-
-	const auto descriptor_count = program.info.buffers.size() + program.info.images.size() +
-	                              program.info.samplers.size() + program.info.addresses.size() + 3u;
-	std::vector<vk::DescriptorBufferInfo> buffer_infos;
-	std::vector<vk::DescriptorImageInfo>  image_infos;
-	std::vector<vk::WriteDescriptorSet>   writes;
-	buffer_infos.reserve(descriptor_count);
-	image_infos.reserve(descriptor_count);
-	writes.reserve(program.bindings.descriptors.size());
-
-	for (const auto& binding: program.bindings.descriptors) {
-		vk::WriteDescriptorSet write {};
-		write.sType             = vk::StructureType::eWriteDescriptorSet;
-		write.dstSet            = set->set;
-		write.dstBinding        = binding.binding;
-		write.descriptorType    = DescriptorType(binding.kind);
-		write.descriptorCount   = DescriptorCount(binding);
-		const auto buffer_start = buffer_infos.size();
-		const auto image_start  = image_infos.size();
-		switch (binding.kind) {
-			case BindingKind::Buffers:
-				for (const auto resource: binding.resources) {
-					buffer_infos.push_back(BufferInfo(data.buffers.at(resource)));
-				}
-				break;
-			case BindingKind::AddressMemory:
-				for (const auto resource: binding.resources) {
-					buffer_infos.push_back(BufferInfo(data.addresses.at(resource)));
-				}
-				break;
-			case BindingKind::FlattenedSrt:
-				buffer_infos.push_back(BufferInfo(data.flattened_srt));
-				break;
-			case BindingKind::UserData: buffer_infos.push_back(BufferInfo(data.user_data)); break;
-			case BindingKind::Gds: buffer_infos.push_back(BufferInfo(data.gds)); break;
-			case BindingKind::Samplers:
-				for (const auto resource: binding.resources) {
-					const auto sampler = data.samplers.at(resource);
-					EXIT_IF(sampler == nullptr);
-					image_infos.push_back({sampler, nullptr, vk::ImageLayout::eUndefined});
-				}
-				break;
-			default: {
-				for (const auto resource: binding.resources) {
-					const auto& texture = data.images.at(resource);
-					image_infos.push_back(MakeImageInfo(texture));
-				}
-				break;
-			}
-		}
-		if (buffer_infos.size() != buffer_start) {
-			write.pBufferInfo = buffer_infos.data() + buffer_start;
-		}
-		if (image_infos.size() != image_start) {
-			write.pImageInfo = image_infos.data() + image_start;
-		}
-		writes.push_back(write);
-	}
-	m_graphics.device.updateDescriptorSets(static_cast<uint32_t>(writes.size()), writes.data(), 0,
-	                                       nullptr);
-	return *set;
-}
-
-vk::DescriptorSetLayout
-DescriptorCache::GetDescriptorSetLayout(Stage stage, const ShaderRecompiler::IR::Program& program) {
-	Common::LockGuard lock(m_mutex);
-	return GetDescriptorSetLayoutInternal(stage, program);
-}
-
-} // namespace Libs::Graphics

src/graphics/host_gpu/renderer/pipeline/descriptorCache.h

@@ -1,114 +0,0 @@
-#ifndef EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_DESCRIPTORCACHE_H_
-#define EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_DESCRIPTORCACHE_H_
-
-#include "common/abi.h"
-#include "common/assert.h"
-#include "common/common.h"
-#include "common/threads.h"
-#include "graphics/host_gpu/graphicContext.h"
-#include "graphics/host_gpu/renderer/cache/textureCache.h"
-#include "graphics/host_gpu/vulkanCommon.h"
-#include "graphics/shader/shaderBindings.h"
-
-#include <map>
-#include <memory>
-#include <span>
-#include <unordered_map>
-#include <vector>
-
-namespace Libs::Graphics {
-
-namespace ShaderRecompiler::IR {
-struct DescriptorValue;
-struct ImageResource;
-struct Program;
-struct ResourceSnapshot;
-} // namespace ShaderRecompiler::IR
-
-class CommandBuffer;
-struct DescriptorCacheTestAccess;
-struct ShaderStageRuntime;
-
-struct VulkanDescriptorSet {
-	vk::DescriptorSet       set     = nullptr;
-	vk::DescriptorSetLayout layout  = nullptr;
-	int                     pool_id = -1;
-};
-
-struct BufferView {
-	std::shared_ptr<void> owner;
-	vk::Buffer            buffer = nullptr;
-	vk::DeviceSize        offset = 0;
-	vk::DeviceSize        range  = VK_WHOLE_SIZE;
-};
-
-class DescriptorCache {
-public:
-	enum class Stage { Unknown, Vertex, Pixel, Compute };
-
-	struct TextureBinding {
-		ImageId                 image_id;
-		vk::ImageView           image_view = nullptr;
-		TextureCache::ImageDesc desc;
-		vk::ImageLayout         layout = vk::ImageLayout::eUndefined;
-	};
-
-	struct NativeDescriptors {
-		std::vector<BufferView>     buffers;
-		std::vector<TextureBinding> images;
-		std::vector<vk::Sampler>    samplers;
-		std::vector<BufferView>     addresses;
-		BufferView                  gds;
-		BufferView                  flattened_srt;
-		BufferView                  user_data;
-	};
-
-	struct PreparedBindings {
-		std::shared_ptr<const ShaderRecompiler::IR::Program>          program;
-		std::shared_ptr<const ShaderRecompiler::IR::ResourceSnapshot> snapshot;
-		NativeDescriptors                                             resources;
-		std::vector<uint32_t>                                         flattened_srt;
-		std::vector<uint32_t>                                         user_data;
-		vk::ShaderStageFlags                                          shader_stage;
-		Stage                                                         stage     = Stage::Unknown;
-		bool                                                          committed = false;
-	};
-
-	explicit DescriptorCache(GraphicContext& graphics): m_graphics(graphics) {
-		EXIT_NOT_IMPLEMENTED(!Common::Thread::IsMainThread());
-	}
-	~DescriptorCache();
-	KYTY_CLASS_NO_COPY(DescriptorCache);
-
-	vk::DescriptorSetLayout GetDescriptorSetLayout(Stage                                stage,
-	                                               const ShaderRecompiler::IR::Program& program);
-	void                    Recycle(VulkanDescriptorSet& set);
-	VulkanDescriptorSet&    GetDescriptor(Stage stage, const ShaderRecompiler::IR::Program& program,
-	                                      const NativeDescriptors& descriptors);
-
-private:
-	friend struct DescriptorCacheTestAccess;
-
-	struct Pool {
-		vk::DescriptorPool pool           = nullptr;
-		int                next_free_pool = -1;
-	};
-
-	static vk::DescriptorImageInfo MakeImageInfo(const TextureBinding& texture);
-	void                           CreatePool();
-	VulkanDescriptorSet* Allocate(Stage stage, const ShaderRecompiler::IR::Program& program);
-	vk::DescriptorSetLayout
-	GetDescriptorSetLayoutInternal(Stage stage, const ShaderRecompiler::IR::Program& program);
-
-	GraphicContext&   m_graphics;
-	Common::Mutex     m_mutex;
-	std::vector<Pool> m_pools;
-	int               m_first_free_pool = -1;
-	std::unordered_map<vk::DescriptorSetLayout, std::vector<VulkanDescriptorSet*>>
-	                                                         m_free_sets_by_layout;
-	std::map<std::vector<uint32_t>, vk::DescriptorSetLayout> m_descriptor_set_layouts;
-};
-
-} // namespace Libs::Graphics
-
-#endif // EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_DESCRIPTORCACHE_H_

src/graphics/host_gpu/renderer/pipeline/descriptorHeap.cpp

@@ -0,0 +1,97 @@
+#include "graphics/host_gpu/renderer/pipeline/descriptorHeap.h"
+
+#include "common/assert.h"
+#include "common/profiler.h"
+#include "graphics/host_gpu/graphicContext.h"
+#include "graphics/host_gpu/renderer/masterSemaphore.h"
+
+namespace Libs::Graphics {
+namespace {
+
+constexpr uint32_t   DescriptorHeapCount = 1024;
+constexpr std::array DescriptorPoolSizes = {
+    vk::DescriptorPoolSize {vk::DescriptorType::eStorageBuffer, 8192},
+    vk::DescriptorPoolSize {vk::DescriptorType::eSampledImage, 8192},
+    vk::DescriptorPoolSize {vk::DescriptorType::eStorageImage, 1024},
+    vk::DescriptorPoolSize {vk::DescriptorType::eSampler, 1024},
+};
+
+} // namespace
+
+DescriptorHeap::DescriptorHeap(GraphicContext& graphics, MasterSemaphore& master_semaphore)
+    : m_graphics(graphics), m_master_semaphore(master_semaphore) {
+	CreatePool();
+}
+
+DescriptorHeap::~DescriptorHeap() {
+	m_graphics.device.destroyDescriptorPool(m_current_pool, nullptr);
+	for (const auto& [pool, tick]: m_pending_pools) {
+		m_master_semaphore.Wait(tick);
+		m_graphics.device.destroyDescriptorPool(pool, nullptr);
+	}
+}
+
+vk::DescriptorSet DescriptorHeap::Commit(vk::DescriptorSetLayout layout) {
+	KYTY_PROFILER_FUNCTION();
+	EXIT_IF(layout == nullptr);
+
+	auto& batch = m_sets[layout];
+	if (batch.size != 0) {
+		return batch.sets[--batch.size];
+	}
+	if (Allocate(layout, batch)) {
+		return batch.sets[--batch.size];
+	}
+
+	m_pending_pools.emplace_back(m_current_pool, m_master_semaphore.CurrentTick());
+	if (const auto& [pool, tick] = m_pending_pools.front(); m_master_semaphore.IsFree(tick)) {
+		m_current_pool = pool;
+		m_pending_pools.pop_front();
+		EXIT_IF(m_graphics.device.resetDescriptorPool(m_current_pool, {}) != vk::Result::eSuccess);
+	} else {
+		CreatePool();
+	}
+
+	m_sets.clear();
+	auto& fresh_batch = m_sets[layout];
+	EXIT_IF(!Allocate(layout, fresh_batch));
+	return fresh_batch.sets[--fresh_batch.size];
+}
+
+bool DescriptorHeap::Allocate(vk::DescriptorSetLayout layout, Batch& batch) {
+	std::array<vk::DescriptorSetLayout, DescriptorSetBatch> layouts;
+	layouts.fill(layout);
+
+	vk::DescriptorSetAllocateInfo allocate {};
+	allocate.sType          = vk::StructureType::eDescriptorSetAllocateInfo;
+	allocate.descriptorPool = m_current_pool;
+	allocate.pSetLayouts    = layouts.data();
+
+	for (;;) {
+		allocate.descriptorSetCount = batch.allocation;
+		const auto result = m_graphics.device.allocateDescriptorSets(&allocate, batch.sets.data());
+		if (result == vk::Result::eSuccess) {
+			batch.size = batch.allocation;
+			return true;
+		}
+		EXIT_IF(result != vk::Result::eErrorOutOfPoolMemory &&
+		        result != vk::Result::eErrorFragmentedPool);
+		if (batch.allocation == 1) {
+			return false;
+		}
+		batch.allocation /= 2;
+	}
+}
+
+void DescriptorHeap::CreatePool() {
+	vk::DescriptorPoolCreateInfo create {};
+	create.sType         = vk::StructureType::eDescriptorPoolCreateInfo;
+	create.flags         = {};
+	create.maxSets       = DescriptorHeapCount;
+	create.poolSizeCount = static_cast<uint32_t>(DescriptorPoolSizes.size());
+	create.pPoolSizes    = DescriptorPoolSizes.data();
+	EXIT_IF(m_graphics.device.createDescriptorPool(&create, nullptr, &m_current_pool) !=
+	        vk::Result::eSuccess);
+}
+
+} // namespace Libs::Graphics

src/graphics/host_gpu/renderer/pipeline/descriptorHeap.h

@@ -0,0 +1,45 @@
+#ifndef EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_DESCRIPTORHEAP_H_
+#define EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_DESCRIPTORHEAP_H_
+
+#include "common/common.h"
+#include "graphics/host_gpu/vulkanCommon.h"
+
+#include <array>
+#include <deque>
+#include <unordered_map>
+
+namespace Libs::Graphics {
+
+struct GraphicContext;
+class MasterSemaphore;
+
+class DescriptorHeap {
+public:
+	DescriptorHeap(GraphicContext& graphics, MasterSemaphore& master_semaphore);
+	~DescriptorHeap();
+	KYTY_CLASS_NO_COPY(DescriptorHeap);
+
+	[[nodiscard]] vk::DescriptorSet Commit(vk::DescriptorSetLayout layout);
+
+private:
+	static constexpr uint32_t DescriptorSetBatch = 32;
+
+	struct Batch {
+		std::array<vk::DescriptorSet, DescriptorSetBatch> sets {};
+		uint32_t                                          size       = 0;
+		uint32_t                                          allocation = DescriptorSetBatch;
+	};
+
+	[[nodiscard]] bool Allocate(vk::DescriptorSetLayout layout, Batch& batch);
+	void               CreatePool();
+
+	GraphicContext&                                     m_graphics;
+	MasterSemaphore&                                    m_master_semaphore;
+	vk::DescriptorPool                                  m_current_pool = nullptr;
+	std::deque<std::pair<vk::DescriptorPool, uint64_t>> m_pending_pools;
+	std::unordered_map<vk::DescriptorSetLayout, Batch>  m_sets;
+};
+
+} // namespace Libs::Graphics
+
+#endif // EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_DESCRIPTORHEAP_H_

src/graphics/host_gpu/renderer/pipeline/descriptors.h

@@ -2,6 +2,8 @@
 #define EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_DESCRIPTORS_H_
 
 #include "common/assert.h"
+#include "graphics/host_gpu/renderer/cache/bufferCache.h"
+#include "graphics/host_gpu/renderer/cache/textureCache.h"
 #include "graphics/host_gpu/renderer/image/image.h"
 #include "graphics/host_gpu/vulkanCommon.h"
 #include "graphics/shader/recompiler/ir/ShaderIR.h"
@@ -10,9 +12,56 @@
 #include <cstdint>
 #include <cstring>
 #include <type_traits>
+#include <vector>
 
 namespace Libs::Graphics {
 
+namespace ShaderRecompiler::IR {
+struct ResourceSnapshot;
+}
+
+struct BufferView {
+	vk::Buffer     buffer = nullptr;
+	vk::DeviceSize offset = 0;
+	vk::DeviceSize range  = VK_WHOLE_SIZE;
+};
+
+struct TextureBinding {
+	ImageId                    image_id;
+	vk::ImageView              image_view = nullptr;
+	TextureCache::ImageDesc    desc;
+	vk::ImageLayout            layout = vk::ImageLayout::eUndefined;
+	std::vector<vk::ImageView> mip_views;
+};
+
+struct NativeDescriptors {
+	std::vector<BufferView>     buffers;
+	std::vector<TextureBinding> images;
+	std::vector<vk::Sampler>    samplers;
+	std::vector<BufferView>     addresses;
+	BufferView                  gds;
+	BufferView                  flattened_srt;
+	BufferView                  user_data;
+};
+
+struct PreparedBindings {
+	const ShaderRecompiler::IR::Program*          program  = nullptr;
+	const ShaderRecompiler::IR::ResourceSnapshot* snapshot = nullptr;
+	NativeDescriptors                             resources;
+	std::vector<BufferId>                         buffer_ids;
+	std::vector<BufferId>                         address_ids;
+	std::vector<uint32_t>                         flattened_srt;
+	std::vector<uint32_t>                         user_data;
+	bool                                          committed = false;
+};
+
+[[nodiscard]] vk::DescriptorType
+NativeDescriptorType(ShaderRecompiler::IR::DescriptorBindingKind kind);
+[[nodiscard]] uint32_t
+NativeDescriptorCount(const ShaderRecompiler::IR::DescriptorBinding& binding);
+[[nodiscard]] vk::DescriptorImageInfo MakeImageInfo(const TextureBinding& texture,
+                                                    uint32_t              element = 0);
+
 template <typename T>
 [[nodiscard]] T DecodeNativeDescriptor(const ShaderRecompiler::IR::DescriptorValue& value) {
 	static_assert(std::is_trivially_copyable_v<T>);
@@ -34,9 +83,9 @@ ResolveTargetTextureView(const ShaderRecompiler::IR::ImageResource& resource,
                          Prospero::ImageType type, uint32_t base_layer, uint32_t image_layers);
 
 [[nodiscard]] bool IsSupportedDepthTargetDescriptor(const ShaderTextureResource& descriptor,
-                                                    const Image&                 image);
+                                                    const Image& image, bool r128 = false);
 [[nodiscard]] bool IsSupportedDepthTextureEncoding(const ShaderTextureResource& descriptor,
-                                                   const Image&                 image);
+                                                   const Image& image, bool r128 = false);
 [[nodiscard]] bool
 IsSupportedSampledVideoOutView(const ShaderRecompiler::IR::ImageResource& resource,
                                const ShaderTextureResource& descriptor, const Image& image);

src/graphics/host_gpu/renderer/pipeline/pipelineCache.cpp

@@ -42,6 +42,7 @@ PipelineCache::~PipelineCache() {
 			(void)key;
 			m_graphics.device.destroyPipeline(pipeline->pipeline, nullptr);
 			m_graphics.device.destroyPipelineLayout(pipeline->pipeline_layout, nullptr);
+			m_graphics.device.destroyDescriptorSetLayout(pipeline->descriptor_set_layout, nullptr);
 		}
 	};
 	destroy(m_graphics_pipelines);
@@ -55,8 +56,9 @@ bool PipelineStaticParameters::operator==(const PipelineStaticParameters& other)
 PipelineCache::GraphicsPipeline& PipelineCache::CreateGraphicsPipeline(
     RenderColorInfo* colors, uint32_t color_count, RenderDepthInfo& depth,
     ShaderVertexInputInfo& vs_input_info, RenderCommandBuffer& command,
-    ShaderPixelInputInfo* ps_input_info, vk::PrimitiveTopology topology, bool ps_active,
-    std::span<const uint32_t> vs_spirv, std::span<const uint32_t> ps_spirv) {
+    ShaderPixelInputInfo* ps_input_info, vk::PrimitiveTopology topology,
+    bool primitive_restart_enable, bool ps_active, std::span<const uint32_t> vs_spirv,
+    std::span<const uint32_t> ps_spirv) {
 	KYTY_PROFILER_BLOCK("PipelineCache::CreatePipeline(Gfx)", profiler::colors::DeepOrangeA200);
 
 	EXIT_IF(colors == nullptr);
@@ -120,6 +122,12 @@ PipelineCache::GraphicsPipeline& PipelineCache::CreateGraphicsPipeline(
 			     depth.samples);
 		}
 	}
+	if (color_count == 0 && !with_depth) {
+		attachment_samples = render_sample_count(ctx.GetAaConfig().msaa_num_samples);
+		EXIT_IF(!static_cast<bool>(
+		    m_graphics.GetPhysicalDeviceProperties().limits.framebufferNoAttachmentsSampleCounts &
+		    vulkan_sample_count(attachment_samples)));
+	}
 	EXIT_IF(attachment_samples == 0 ||
 	        vulkan_sample_count(attachment_samples) == vk::SampleCountFlagBits {});
 
@@ -130,12 +138,12 @@ PipelineCache::GraphicsPipeline& PipelineCache::CreateGraphicsPipeline(
 		}
 	}
 
-	const auto& clip_control = ctx.GetClipControl();
-	EXIT_NOT_IMPLEMENTED(!clip_control.IsZClipModeRepresentable());
-	static_params.negative_one_to_one = !clip_control.dx_clip_space;
-	static_params.depth_clip_enable   = clip_control.IsZClipEnabled();
-	static_params.topology            = topology;
-	static_params.samples             = attachment_samples;
+	const auto& clip_control               = ctx.GetClipControl();
+	static_params.negative_one_to_one      = !clip_control.dx_clip_space;
+	static_params.depth_clip_enable        = clip_control.IsZClipEnabled();
+	static_params.topology                 = topology;
+	static_params.primitive_restart_enable = primitive_restart_enable;
+	static_params.samples                  = attachment_samples;
 	static_params.sample_shading_enable =
 	    ps_active && attachment_samples > 1 && ps_input_info->ps_sample_shading;
 	if (static_params.sample_shading_enable && !m_graphics.sample_rate_shading_enabled) {
@@ -203,9 +211,9 @@ PipelineCache::GraphicsPipeline& PipelineCache::CreateGraphicsPipeline(
 	auto cached = std::make_unique<GraphicsPipeline>(p);
 	LogPipelineTrace("CreatePipelineInternal begin", vs_id.hash0, vs_id.crc32, ps_id.hash0,
 	                 ps_id.crc32);
-	CreatePipelineInternal(m_graphics, m_descriptor_cache, *cached, rendering, vs_input_info,
-	                       vs_spirv, ps_input_info, ps_spirv, static_params, vs_id.hash0,
-	                       vs_id.crc32, ps_id.hash0, ps_id.crc32, ps_active);
+	CreatePipelineInternal(m_graphics, *cached, rendering, vs_input_info, vs_spirv, ps_input_info,
+	                       ps_spirv, static_params, vs_id.hash0, vs_id.crc32, ps_id.hash0,
+	                       ps_id.crc32, ps_active);
 	LogPipelineTrace("CreatePipelineInternal done", vs_id.hash0, vs_id.crc32, ps_id.hash0,
 	                 ps_id.crc32);
 
@@ -245,7 +253,7 @@ PipelineCache::CreateComputePipeline(ShaderComputeInputInfo&      input_info,
 	}
 
 	auto cached = std::make_unique<ComputePipeline>(p);
-	CreatePipelineInternal(m_graphics, m_descriptor_cache, *cached, input_info, cs_spirv);
+	CreatePipelineInternal(m_graphics, *cached, input_info, cs_spirv);
 
 	EXIT_NOT_IMPLEMENTED(cached->pipeline == nullptr);
 	EXIT_NOT_IMPLEMENTED(cached->pipeline_layout == nullptr);

src/graphics/host_gpu/renderer/pipeline/pipelineCache.h

@@ -21,7 +21,6 @@ struct GraphicContext;
 struct RenderColorInfo;
 struct RenderDepthInfo;
 class RenderCommandBuffer;
-class DescriptorCache;
 
 namespace HW {
 class Context;
@@ -38,6 +37,7 @@ struct PipelineStaticParameters {
 	bool                       depth_clip_enable        = true;
 	int                        scissor_ltrb[4]          = {0};
 	vk::PrimitiveTopology      topology                 = vk::PrimitiveTopology::ePointList;
+	bool                       primitive_restart_enable = false;
 	uint32_t                   samples                  = 1;
 	bool                       sample_shading_enable    = false;
 	bool                       with_depth               = false;
@@ -79,9 +79,9 @@ static_assert(std::is_standard_layout_v<PipelineStaticParameters>);
 static_assert(alignof(PipelineStaticParameters) == 1);
 static_assert(sizeof(PipelineStaticParameters) ==
               sizeof(float[3]) + sizeof(float[3]) + sizeof(bool) * 2 + sizeof(int[4]) +
-                  sizeof(vk::PrimitiveTopology) + sizeof(uint32_t) + sizeof(bool) * 4 +
-                  sizeof(vk::CompareOp) + sizeof(bool) + sizeof(float) * 2 + sizeof(bool) +
-                  sizeof(PipelineStencilStaticState) * 2 + sizeof(uint32_t) +
+                  sizeof(vk::PrimitiveTopology) + sizeof(bool) + sizeof(uint32_t) +
+                  sizeof(bool) * 4 + sizeof(vk::CompareOp) + sizeof(bool) + sizeof(float) * 2 +
+                  sizeof(bool) + sizeof(PipelineStencilStaticState) * 2 + sizeof(uint32_t) +
                   sizeof(uint32_t[RENDER_COLOR_ATTACHMENTS_MAX]) + sizeof(bool) * 3 +
                   sizeof(uint8_t[RENDER_COLOR_ATTACHMENTS_MAX]) * 6 +
                   sizeof(bool[RENDER_COLOR_ATTACHMENTS_MAX]) * 3 + sizeof(float) * 4);
@@ -97,16 +97,17 @@ struct PipelineRenderingState {
 
 class PipelineCache {
 public:
-	PipelineCache(GraphicContext& graphics, DescriptorCache& descriptor_cache)
-	    : m_graphics(graphics), m_descriptor_cache(descriptor_cache) {
+	explicit PipelineCache(GraphicContext& graphics): m_graphics(graphics) {
 		EXIT_NOT_IMPLEMENTED(!Common::Thread::IsMainThread());
 	}
 	~PipelineCache();
 	KYTY_CLASS_NO_COPY(PipelineCache);
 
 	struct Pipeline {
-		vk::PipelineLayout pipeline_layout = nullptr;
-		vk::Pipeline       pipeline        = nullptr;
+		vk::PipelineLayout      pipeline_layout       = nullptr;
+		vk::Pipeline            pipeline              = nullptr;
+		vk::DescriptorSetLayout descriptor_set_layout = nullptr;
+		bool                    uses_push_descriptors = false;
 	};
 
 	struct GraphicsPipeline: Pipeline {
@@ -122,8 +123,8 @@ public:
 	CreateGraphicsPipeline(RenderColorInfo* colors, uint32_t color_count, RenderDepthInfo& depth,
 	                       ShaderVertexInputInfo& vs_input_info, RenderCommandBuffer& command,
 	                       ShaderPixelInputInfo* ps_input_info, vk::PrimitiveTopology topology,
-	                       bool ps_active, std::span<const uint32_t> vs_spirv,
-	                       std::span<const uint32_t> ps_spirv);
+	                       bool primitive_restart_enable, bool ps_active,
+	                       std::span<const uint32_t> vs_spirv, std::span<const uint32_t> ps_spirv);
 	ComputePipeline& CreateComputePipeline(ShaderComputeInputInfo&      input_info,
 	                                       const HW::ComputeShaderInfo& cs_regs,
 	                                       std::span<const uint32_t>    cs_spirv);
@@ -200,8 +201,7 @@ private:
 		}
 	};
 
-	GraphicContext&  m_graphics;
-	DescriptorCache& m_descriptor_cache;
+	GraphicContext& m_graphics;
 	std::unordered_map<GraphicsPipelineKey, std::unique_ptr<GraphicsPipeline>,
 	                   GraphicsPipelineKeyHash>
 	    m_graphics_pipelines;
@@ -213,16 +213,14 @@ private:
 void LogPipelineTrace(const char* phase, uint32_t vs_hash0, uint32_t vs_crc32, uint32_t ps_hash0,
                       uint32_t ps_crc32);
 void CreatePipelineInternal(
-    GraphicContext& graphics, DescriptorCache& descriptor_cache,
-    PipelineCache::GraphicsPipeline& pipeline, const PipelineRenderingState& rendering,
-    const ShaderVertexInputInfo& vs_input_info, std::span<const uint32_t> vs_shader,
-    const ShaderPixelInputInfo* ps_input_info, std::span<const uint32_t> ps_shader,
-    const PipelineStaticParameters& static_params, uint32_t vs_hash0, uint32_t vs_crc32,
-    uint32_t ps_hash0, uint32_t ps_crc32, bool ps_active);
-void CreatePipelineInternal(GraphicContext& graphics, DescriptorCache& descriptor_cache,
-                            PipelineCache::ComputePipeline& pipeline,
-                            const ShaderComputeInputInfo&   input_info,
-                            std::span<const uint32_t>       cs_shader);
+    GraphicContext& graphics, PipelineCache::GraphicsPipeline& pipeline,
+    const PipelineRenderingState& rendering, const ShaderVertexInputInfo& vs_input_info,
+    std::span<const uint32_t> vs_shader, const ShaderPixelInputInfo* ps_input_info,
+    std::span<const uint32_t> ps_shader, const PipelineStaticParameters& static_params,
+    uint32_t vs_hash0, uint32_t vs_crc32, uint32_t ps_hash0, uint32_t ps_crc32, bool ps_active);
+void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::ComputePipeline& pipeline,
+                            const ShaderComputeInputInfo& input_info,
+                            std::span<const uint32_t>     cs_shader);
 
 } // namespace Libs::Graphics
 

src/graphics/host_gpu/renderer/pipeline/shaderResourceBarrier.h

@@ -2,7 +2,7 @@
 #define EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_SHADERRESOURCEBARRIER_H_
 
 #include "graphics/host_gpu/graphicContext.h"
-#include "graphics/shader/recompiler/ir/ResourceMaterialization.h"
+#include "graphics/shader/recompiler/ir/passes/ResourceMaterialization.h"
 
 #include <vector>
 

src/graphics/host_gpu/renderer/pipeline/shaderSubgroup.cpp

@@ -1,60 +0,0 @@
-#include "graphics/host_gpu/renderer/pipeline/shaderSubgroup.h"
-
-#include "graphics/shader/recompiler/emitter/SpirvEmitter.h"
-
-namespace Libs::Graphics {
-
-ShaderSubgroupCapabilities::ShaderSubgroupCapabilities(const GraphicContext& graphics)
-    : subgroup_size(graphics.subgroup_size), min_subgroup_size(graphics.min_subgroup_size),
-      max_subgroup_size(graphics.max_subgroup_size),
-      required_subgroup_size_stages(graphics.required_subgroup_size_stages),
-      subgroup_size_control_enabled(graphics.subgroup_size_control_enabled) {}
-
-ShaderLaneMaskMode SelectGraphicsLaneMaskMode(uint32_t guest_wave_size) {
-	// A 64-wide guest wave can't assume the host GPU's lane layout matches
-	// GCN's. AMD RDNA also reports a 64-wide subgroup but uses a different
-	// lane/pixel layout, which corrupted rendering under "native wave" mode.
-	// Per-invocation emulation works regardless of the host's lane layout.
-	if (guest_wave_size == 64u) {
-		return ShaderLaneMaskMode::PerInvocation;
-	}
-	return ShaderLaneMaskMode::NativeWave;
-}
-
-ShaderSubgroupConfiguration ConfigureShaderSubgroup(const ShaderSubgroupCapabilities& capabilities,
-                                                    vk::ShaderStageFlagBits           stage,
-                                                    const ShaderRecompiler::IR::Program& program) {
-	const auto guest_wave_size = program.wave_size;
-	if (guest_wave_size != 32u && guest_wave_size != 64u) {
-		return {};
-	}
-	if (stage != vk::ShaderStageFlagBits::eCompute) {
-		const auto expected = SelectGraphicsLaneMaskMode(guest_wave_size);
-		if (program.lane_mask_mode != expected || (capabilities.subgroup_size != guest_wave_size &&
-		                                           expected != ShaderLaneMaskMode::PerInvocation)) {
-			return {};
-		}
-		return {expected == ShaderLaneMaskMode::NativeWave
-		            ? ShaderSubgroupMode::Natural
-		            : ShaderSubgroupMode::PerInvocationGraphics,
-		        0};
-	}
-	if (program.lane_mask_mode != ShaderLaneMaskMode::NativeWave) {
-		return {};
-	}
-	if (capabilities.subgroup_size == guest_wave_size) {
-		return {ShaderSubgroupMode::Natural, 0};
-	}
-	if (capabilities.subgroup_size_control_enabled &&
-	    (capabilities.required_subgroup_size_stages & stage) &&
-	    guest_wave_size >= capabilities.min_subgroup_size &&
-	    guest_wave_size <= capabilities.max_subgroup_size) {
-		return {ShaderSubgroupMode::Controlled, guest_wave_size};
-	}
-	if (!ShaderRecompiler::Spirv::ProgramRequiresExactSubgroupSize(program)) {
-		return {ShaderSubgroupMode::FlattenedMasks, 0};
-	}
-	return {};
-}
-
-} // namespace Libs::Graphics

src/graphics/host_gpu/renderer/pipeline/shaderSubgroup.h

@@ -1,40 +0,0 @@
-#ifndef EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_SHADERSUBGROUP_H_
-#define EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_SHADERSUBGROUP_H_
-
-#include "graphics/host_gpu/graphicContext.h"
-#include "graphics/shader/recompiler/ir/ShaderIR.h"
-
-namespace Libs::Graphics {
-
-enum class ShaderSubgroupMode {
-	Natural,
-	Controlled,
-	PerInvocationGraphics,
-	FlattenedMasks,
-	Unsupported
-};
-
-struct ShaderSubgroupConfiguration {
-	ShaderSubgroupMode mode          = ShaderSubgroupMode::Unsupported;
-	uint32_t           required_size = 0;
-};
-
-struct ShaderSubgroupCapabilities {
-	explicit ShaderSubgroupCapabilities(const GraphicContext& graphics);
-
-	uint32_t             subgroup_size     = 0;
-	uint32_t             min_subgroup_size = 0;
-	uint32_t             max_subgroup_size = 0;
-	vk::ShaderStageFlags required_subgroup_size_stages;
-	bool                 subgroup_size_control_enabled = false;
-};
-
-ShaderLaneMaskMode SelectGraphicsLaneMaskMode(uint32_t guest_wave_size);
-
-ShaderSubgroupConfiguration ConfigureShaderSubgroup(const ShaderSubgroupCapabilities& capabilities,
-                                                    vk::ShaderStageFlagBits           stage,
-                                                    const ShaderRecompiler::IR::Program& program);
-
-} // namespace Libs::Graphics
-
-#endif // EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_SHADERSUBGROUP_H_

src/graphics/host_gpu/renderer/pipeline/shaders.cpp

@@ -6,9 +6,8 @@
 #include "graphics/guest_gpu/gpu_defs.h"
 #include "graphics/host_gpu/graphicContext.h"
 #include "graphics/host_gpu/renderer/debug.h"
-#include "graphics/host_gpu/renderer/pipeline/descriptorCache.h"
+#include "graphics/host_gpu/renderer/pipeline/descriptors.h"
 #include "graphics/host_gpu/renderer/pipeline/pipelineCache.h"
-#include "graphics/host_gpu/renderer/pipeline/shaderSubgroup.h"
 #include "graphics/host_gpu/renderer/render.h"
 #include "graphics/host_gpu/renderer/renderContext.h"
 #include "graphics/host_gpu/renderer/renderTarget.h"
@@ -20,10 +19,13 @@
 #include <algorithm>
 #include <limits>
 #include <span>
+#include <vector>
 
 namespace Libs::Graphics {
 
 // IDK: maybe we can remove it?
+constexpr uint8_t kTemporaryVertexAttribFormat113 =
+    static_cast<uint8_t>(Prospero::VertexAttribFormat::k16_16SInt);
 constexpr uint32_t kTemporaryPs5BufferFormat121 = 121u;
 
 static bool NarrowInputFormat(vk::Format& format, uint32_t& size, uint32_t used_components) {
@@ -88,8 +90,9 @@ static bool NarrowInputFormat(vk::Format& format, uint32_t& size, uint32_t used_
 
 static void GetInputFormat(const ShaderBufferResource& res, vk::Format& format, uint32_t& size,
                            uint32_t used_components) {
-	auto fmt = res.Format();
-	if (fmt == Prospero::GpuEnumValue(Prospero::VertexAttribFormat::k16_16SInt)) {
+	const auto fmt        = res.Format();
+	const auto raw_format = res.RawFormat();
+	if (raw_format == kTemporaryVertexAttribFormat113) {
 		static bool logged_113 = false;
 		if (!logged_113) {
 			LOGF("InputFormat: temporary: accepting invalid PS5 buffer format 113 as "
@@ -99,12 +102,12 @@ static void GetInputFormat(const ShaderBufferResource& res, vk::Format& format,
 		format = vk::Format::eR32G32B32A32Sfloat;
 		size   = 4;
 		if (NarrowInputFormat(format, size, used_components)) {
-			LOGF("InputFormat: narrowing fmt=%u to %s for used_components=%u\n", fmt,
+			LOGF("InputFormat: narrowing fmt=%u to %s for used_components=%u\n", raw_format,
 			     VulkanToString(format).c_str(), used_components);
 		}
 		return;
 	}
-	if (fmt == kTemporaryPs5BufferFormat121) {
+	if (raw_format == kTemporaryPs5BufferFormat121) {
 		static bool logged_121 = false;
 		if (!logged_121) {
 			LOGF("InputFormat: accepting PS5 buffer format 121 as vk::Format::eR16G16Sfloat\n");
@@ -113,13 +116,13 @@ static void GetInputFormat(const ShaderBufferResource& res, vk::Format& format,
 		format = vk::Format::eR16G16Sfloat;
 		size   = 2;
 		if (NarrowInputFormat(format, size, used_components)) {
-			LOGF("InputFormat: narrowing fmt=%u to %s for used_components=%u\n", fmt,
+			LOGF("InputFormat: narrowing fmt=%u to %s for used_components=%u\n", raw_format,
 			     VulkanToString(format).c_str(), used_components);
 		}
 		return;
 	}
 
-	switch (static_cast<Prospero::BufferFormat>(fmt)) {
+	switch (fmt) {
 		case Prospero::BufferFormat::k32_32_32_32Float:
 			format = vk::Format::eR32G32B32A32Sfloat;
 			size   = 4;
@@ -329,7 +332,7 @@ static void GetInputFormat(const ShaderBufferResource& res, vk::Format& format,
 			size   = 1;
 			break;
 		default:
-			EXIT("unknown format: fmt = %u\n", fmt);
+			EXIT("unknown format: fmt = %u\n", raw_format);
 			format = vk::Format::eUndefined;
 			size   = 4;
 			break;
@@ -386,82 +389,42 @@ static vk::BlendOp GetBlendOp(uint32_t op) {
 	return vk::BlendOp::eAdd;
 }
 
-static void CreateLayout(DescriptorCache&                   descriptor_cache,
-                         std::span<vk::DescriptorSetLayout> set_layouts, uint32_t& set_layouts_num,
-                         std::span<vk::PushConstantRange>     push_constant_info,
-                         uint32_t&                            push_constant_info_num,
-                         const ShaderRecompiler::IR::Program& program,
-                         vk::ShaderStageFlags vk_stage, DescriptorCache::Stage stage) {
-	const auto& bindings        = program.bindings;
-	const bool  need_descriptor = !bindings.descriptors.empty();
-
-	if (bindings.push_constant_size != 0) {
-		auto index = push_constant_info_num;
-
-		push_constant_info[index].stageFlags = vk_stage;
-		push_constant_info[index].offset     = bindings.push_constant_offset;
-		push_constant_info[index].size       = bindings.push_constant_size;
-		push_constant_info_num++;
-	}
-
-	if (need_descriptor) {
-		EXIT_IF(bindings.descriptor_set != set_layouts_num);
-
-		set_layouts[set_layouts_num] = descriptor_cache.GetDescriptorSetLayout(stage, program);
-		set_layouts_num++;
+static void AddLayoutBindings(std::vector<vk::DescriptorSetLayoutBinding>& descriptor_bindings,
+                              const ShaderRecompiler::IR::Program& program,
+                              vk::ShaderStageFlagBits              stage) {
+	for (const auto& binding: program.bindings.descriptors) {
+		descriptor_bindings.push_back(
+		    {ShaderRecompiler::IR::NativeBinding(program.stage, binding.kind),
+		     NativeDescriptorType(binding.kind), NativeDescriptorCount(binding), stage, nullptr});
 	}
 }
 
-static void ConfigureSubgroupSize(const GraphicContext& graphics, vk::ShaderStageFlagBits vk_stage,
-                                  const ShaderRecompiler::IR::Program&                   program,
-                                  vk::PipelineShaderStageRequiredSubgroupSizeCreateInfo& required,
-                                  vk::PipelineShaderStageCreateInfo&                     stage) {
-	const auto config =
-	    ConfigureShaderSubgroup(ShaderSubgroupCapabilities {graphics}, vk_stage, program);
-	switch (config.mode) {
-		case ShaderSubgroupMode::Natural: return;
-		case ShaderSubgroupMode::PerInvocationGraphics: {
-			static std::atomic<uint32_t> log_count {0};
-			if (log_count.fetch_add(1, std::memory_order_relaxed) < 16) {
-				LOGF("Vulkan running %u-lane graphics shader stage 0x%08x with per-invocation "
-				     "EXEC/VCC on the host-native subgroup\n",
-				     program.wave_size, static_cast<uint32_t>(vk_stage));
-			}
-			return;
-		}
-		case ShaderSubgroupMode::FlattenedMasks: {
-			static std::atomic<uint32_t> log_count {0};
-			if (log_count.fetch_add(1, std::memory_order_relaxed) < 16) {
-				LOGF("Vulkan running %u-lane shader stage 0x%08x with flattened per-invocation "
-				     "EXEC/VCC\n",
-				     program.wave_size, static_cast<uint32_t>(vk_stage));
-			}
-			return;
-		}
-		case ShaderSubgroupMode::Controlled: break;
-		case ShaderSubgroupMode::Unsupported:
-		default:
-			EXIT("Vulkan cannot run %u-lane shader stage 0x%08x: default=%u min=%u max=%u "
-			     "controlled_stages=0x%08x\n",
-			     program.wave_size, static_cast<uint32_t>(vk_stage), graphics.subgroup_size,
-			     graphics.min_subgroup_size, graphics.max_subgroup_size,
-			     static_cast<vk::ShaderStageFlags::MaskType>(
-			         graphics.required_subgroup_size_stages));
+static void CreateDescriptorLayout(GraphicContext& graphics, PipelineCache::Pipeline& pipeline,
+                                   std::span<const vk::DescriptorSetLayoutBinding> bindings) {
+	uint32_t descriptor_count = 0;
+	for (const auto& binding: bindings) {
+		descriptor_count += binding.descriptorCount;
 	}
+	pipeline.uses_push_descriptors = descriptor_count <= graphics.max_push_descriptors;
 
-	required.sType = vk::StructureType::ePipelineShaderStageRequiredSubgroupSizeCreateInfo;
-	required.requiredSubgroupSize = config.required_size;
-	stage.pNext                   = &required;
+	vk::DescriptorSetLayoutCreateInfo create {};
+	create.sType        = vk::StructureType::eDescriptorSetLayoutCreateInfo;
+	create.flags        = pipeline.uses_push_descriptors
+	                          ? vk::DescriptorSetLayoutCreateFlagBits::ePushDescriptorKHR
+	                          : vk::DescriptorSetLayoutCreateFlags {};
+	create.bindingCount = static_cast<uint32_t>(bindings.size());
+	create.pBindings    = bindings.data();
+	EXIT_IF(graphics.device.createDescriptorSetLayout(
+	            &create, nullptr, &pipeline.descriptor_set_layout) != vk::Result::eSuccess);
 }
 
 // NOLINTNEXTLINE(readability-function-cognitive-complexity)
 void CreatePipelineInternal(
-    GraphicContext& graphics, DescriptorCache& descriptor_cache,
-    PipelineCache::GraphicsPipeline& pipeline, const PipelineRenderingState& rendering,
-    const ShaderVertexInputInfo& vs_input_info, std::span<const uint32_t> vs_shader,
-    const ShaderPixelInputInfo* ps_input_info, std::span<const uint32_t> ps_shader,
-    const PipelineStaticParameters& static_params, uint32_t vs_hash0, uint32_t vs_crc32,
-    uint32_t ps_hash0, uint32_t ps_crc32, bool ps_active) {
+    GraphicContext& graphics, PipelineCache::GraphicsPipeline& pipeline,
+    const PipelineRenderingState& rendering, const ShaderVertexInputInfo& vs_input_info,
+    std::span<const uint32_t> vs_shader, const ShaderPixelInputInfo* ps_input_info,
+    std::span<const uint32_t> ps_shader, const PipelineStaticParameters& static_params,
+    uint32_t vs_hash0, uint32_t vs_crc32, uint32_t ps_hash0, uint32_t ps_crc32, bool ps_active) {
 	EXIT_IF(ps_active && ps_input_info == nullptr);
 
 	const bool rect_list = static_params.topology == vk::PrimitiveTopology::ePatchList;
@@ -525,8 +488,7 @@ void CreatePipelineInternal(
 	    rect_list && (tess_control_shader_module == nullptr || tess_eval_shader_module == nullptr));
 	EXIT_NOT_IMPLEMENTED(ps_active && frag_shader_module == nullptr);
 
-	vk::PipelineShaderStageCreateInfo                     vert_shader_stage_info {};
-	vk::PipelineShaderStageRequiredSubgroupSizeCreateInfo vert_subgroup_size {};
+	vk::PipelineShaderStageCreateInfo vert_shader_stage_info {};
 
 	vert_shader_stage_info.sType               = vk::StructureType::ePipelineShaderStageCreateInfo;
 	vert_shader_stage_info.pNext               = nullptr;
@@ -535,12 +497,8 @@ void CreatePipelineInternal(
 	vert_shader_stage_info.module              = vert_shader_module;
 	vert_shader_stage_info.pName               = "main";
 	vert_shader_stage_info.pSpecializationInfo = nullptr;
-	EXIT_IF(!vs_input_info.stage);
-	ConfigureSubgroupSize(graphics, vk::ShaderStageFlagBits::eVertex, *vs_input_info.stage.program,
-	                      vert_subgroup_size, vert_shader_stage_info);
 
-	vk::PipelineShaderStageCreateInfo                     frag_shader_stage_info {};
-	vk::PipelineShaderStageRequiredSubgroupSizeCreateInfo frag_subgroup_size {};
+	vk::PipelineShaderStageCreateInfo frag_shader_stage_info {};
 	frag_shader_stage_info.sType               = vk::StructureType::ePipelineShaderStageCreateInfo;
 	frag_shader_stage_info.pNext               = nullptr;
 	frag_shader_stage_info.flags               = {};
@@ -548,12 +506,6 @@ void CreatePipelineInternal(
 	frag_shader_stage_info.module              = frag_shader_module;
 	frag_shader_stage_info.pName               = "main";
 	frag_shader_stage_info.pSpecializationInfo = nullptr;
-	if (ps_active) {
-		EXIT_IF(!ps_input_info->stage);
-		ConfigureSubgroupSize(graphics, vk::ShaderStageFlagBits::eFragment,
-		                      *ps_input_info->stage.program, frag_subgroup_size,
-		                      frag_shader_stage_info);
-	}
 
 	vk::PipelineShaderStageCreateInfo tess_control_shader_stage_info {};
 	tess_control_shader_stage_info.sType  = vk::StructureType::ePipelineShaderStageCreateInfo;
@@ -723,7 +675,8 @@ void CreatePipelineInternal(
 	input_assembly.pNext    = nullptr;
 	input_assembly.flags    = {};
 	input_assembly.topology = static_params.topology;
-	input_assembly.primitiveRestartEnable = VK_FALSE;
+	input_assembly.primitiveRestartEnable =
+	    static_params.primitive_restart_enable ? VK_TRUE : VK_FALSE;
 
 	vk::Viewport viewport {};
 	viewport.x        = static_params.viewport_offset[0] - static_params.viewport_scale[0];
@@ -873,40 +826,36 @@ void CreatePipelineInternal(
 	color_blending.blendConstants[2] = static_params.blend_color_blue;
 	color_blending.blendConstants[3] = static_params.blend_color_alpha;
 
-	vk::DescriptorSetLayout set_layouts[2]  = {};
-	uint32_t                set_layouts_num = 0;
-
-	vk::PushConstantRange push_constant_info[2];
-	uint32_t              push_constant_info_num = 0;
-
 	EXIT_IF(!vs_input_info.stage);
-	CreateLayout(descriptor_cache, set_layouts, set_layouts_num, push_constant_info,
-	             push_constant_info_num, *vs_input_info.stage.program,
-	             vk::ShaderStageFlagBits::eVertex, DescriptorCache::Stage::Vertex);
+	std::vector<vk::DescriptorSetLayoutBinding> descriptor_bindings;
+	AddLayoutBindings(descriptor_bindings, *vs_input_info.stage.program,
+	                  vk::ShaderStageFlagBits::eVertex);
 	if (ps_active) {
 		EXIT_IF(!ps_input_info->stage);
-		CreateLayout(descriptor_cache, set_layouts, set_layouts_num, push_constant_info,
-		             push_constant_info_num, *ps_input_info->stage.program,
-		             vk::ShaderStageFlagBits::eFragment, DescriptorCache::Stage::Pixel);
+		AddLayoutBindings(descriptor_bindings, *ps_input_info->stage.program,
+		                  vk::ShaderStageFlagBits::eFragment);
 	}
+	CreateDescriptorLayout(graphics, pipeline, descriptor_bindings);
+	constexpr auto GraphicsStages =
+	    vk::ShaderStageFlagBits::eVertex | vk::ShaderStageFlagBits::eFragment;
+	const vk::PushConstantRange push_constants {
+	    GraphicsStages, 0, ShaderRecompiler::IR::NativePushConstantSize};
 
 	vk::PipelineLayoutCreateInfo pipeline_layout_info {};
 	pipeline_layout_info.sType                  = vk::StructureType::ePipelineLayoutCreateInfo;
 	pipeline_layout_info.pNext                  = nullptr;
 	pipeline_layout_info.flags                  = {};
-	pipeline_layout_info.setLayoutCount         = set_layouts_num;
-	pipeline_layout_info.pSetLayouts            = (set_layouts_num > 0 ? set_layouts : nullptr);
-	pipeline_layout_info.pushConstantRangeCount = push_constant_info_num;
-	pipeline_layout_info.pPushConstantRanges =
-	    push_constant_info_num > 0 ? push_constant_info : nullptr;
+	pipeline_layout_info.setLayoutCount         = 1;
+	pipeline_layout_info.pSetLayouts            = &pipeline.descriptor_set_layout;
+	pipeline_layout_info.pushConstantRangeCount = 1;
+	pipeline_layout_info.pPushConstantRanges    = &push_constants;
 
 	EXIT_IF(pipeline.pipeline_layout != nullptr);
 
 	if (graphics_debug_dump_enabled()) {
 		LOGF("PipelineTrace: vkCreatePipelineLayout begin VS=0x%08" PRIx32 "/0x%08" PRIx32
-		     " PS=0x%08" PRIx32 "/0x%08" PRIx32 " set_layouts=%" PRIu32 " push_constants=%" PRIu32
-		     "\n",
-		     vs_hash0, vs_crc32, ps_hash0, ps_crc32, set_layouts_num, push_constant_info_num);
+		     " PS=0x%08" PRIx32 "/0x%08" PRIx32 " set_layouts=1 push_constants=%" PRIu32 "\n",
+		     vs_hash0, vs_crc32, ps_hash0, ps_crc32, 1u);
 	}
 	result = graphics.device.createPipelineLayout(&pipeline_layout_info, nullptr,
 	                                              &pipeline.pipeline_layout);
@@ -1038,10 +987,9 @@ void CreatePipelineInternal(
 }
 
 // NOLINTNEXTLINE(readability-function-cognitive-complexity)
-void CreatePipelineInternal(GraphicContext& graphics, DescriptorCache& descriptor_cache,
-                            PipelineCache::ComputePipeline& pipeline,
-                            const ShaderComputeInputInfo&   input_info,
-                            std::span<const uint32_t>       cs_shader) {
+void CreatePipelineInternal(GraphicContext& graphics, PipelineCache::ComputePipeline& pipeline,
+                            const ShaderComputeInputInfo& input_info,
+                            std::span<const uint32_t>     cs_shader) {
 	vk::ShaderModule comp_shader_module = nullptr;
 
 	vk::ShaderModuleCreateInfo create_info {};
@@ -1070,34 +1018,35 @@ void CreatePipelineInternal(GraphicContext& graphics, DescriptorCache& descripto
 	comp_shader_stage_info.pName               = "main";
 	comp_shader_stage_info.pSpecializationInfo = nullptr;
 	EXIT_IF(!input_info.stage);
-	ConfigureSubgroupSize(graphics, vk::ShaderStageFlagBits::eCompute, *input_info.stage.program,
-	                      comp_subgroup_size, comp_shader_stage_info);
+	const auto wave_size = input_info.stage.program->wave_size;
+	if (graphics.compute_subgroup_size_control_enabled &&
+	    wave_size >= graphics.min_subgroup_size && wave_size <= graphics.max_subgroup_size) {
+		comp_subgroup_size.sType =
+		    vk::StructureType::ePipelineShaderStageRequiredSubgroupSizeCreateInfo;
+		comp_subgroup_size.requiredSubgroupSize = wave_size;
+		comp_shader_stage_info.pNext            = &comp_subgroup_size;
+	}
 
-	vk::DescriptorSetLayout set_layouts[1]  = {};
-	uint32_t                set_layouts_num = 0;
-
-	vk::PushConstantRange push_constant_info[1];
-	uint32_t              push_constant_info_num = 0;
-
-	EXIT_IF(!input_info.stage);
-	CreateLayout(descriptor_cache, set_layouts, set_layouts_num, push_constant_info,
-	             push_constant_info_num, *input_info.stage.program,
-	             vk::ShaderStageFlagBits::eCompute, DescriptorCache::Stage::Compute);
+	std::vector<vk::DescriptorSetLayoutBinding> descriptor_bindings;
+	AddLayoutBindings(descriptor_bindings, *input_info.stage.program,
+	                  vk::ShaderStageFlagBits::eCompute);
+	CreateDescriptorLayout(graphics, pipeline, descriptor_bindings);
+	const vk::PushConstantRange push_constants {vk::ShaderStageFlagBits::eCompute, 0,
+	                                            ShaderRecompiler::IR::NativePushConstantSize};
 
 	vk::PipelineLayoutCreateInfo pipeline_layout_info {};
 	pipeline_layout_info.sType                  = vk::StructureType::ePipelineLayoutCreateInfo;
 	pipeline_layout_info.pNext                  = nullptr;
 	pipeline_layout_info.flags                  = {};
-	pipeline_layout_info.setLayoutCount         = set_layouts_num;
-	pipeline_layout_info.pSetLayouts            = (set_layouts_num > 0 ? set_layouts : nullptr);
-	pipeline_layout_info.pushConstantRangeCount = push_constant_info_num;
-	pipeline_layout_info.pPushConstantRanges =
-	    push_constant_info_num > 0 ? push_constant_info : nullptr;
+	pipeline_layout_info.setLayoutCount         = 1;
+	pipeline_layout_info.pSetLayouts            = &pipeline.descriptor_set_layout;
+	pipeline_layout_info.pushConstantRangeCount = 1;
+	pipeline_layout_info.pPushConstantRanges    = &push_constants;
 
 	EXIT_IF(pipeline.pipeline_layout != nullptr);
 
-	LOGF("PipelineTrace: vkCreatePipelineLayout CS begin set_layouts=%u push_constants=%u\n",
-	     set_layouts_num, push_constant_info_num);
+	LOGF("PipelineTrace: vkCreatePipelineLayout CS begin set_layouts=1 push_constants=%u\n",
+	     1u);
 	result = graphics.device.createPipelineLayout(&pipeline_layout_info, nullptr,
 	                                              &pipeline.pipeline_layout);
 	LOGF("PipelineTrace: vkCreatePipelineLayout CS done result=%s layout=%p\n",

src/graphics/host_gpu/renderer/render.h

@@ -4,12 +4,14 @@
 #include "common/abi.h"
 #include "common/assert.h"
 #include "common/common.h"
-#include "graphics/host_gpu/renderer/pipeline/descriptorCache.h"
+#include "graphics/host_gpu/renderer/pipeline/descriptors.h"
+#include "graphics/host_gpu/renderer/pipeline/pipelineCache.h"
 #include "graphics/host_gpu/renderer/renderTarget.h"
 #include "graphics/host_gpu/vulkanCommon.h"
 
 #include <array>
-#include <memory>
+#include <optional>
+#include <span>
 #include <vector>
 
 namespace Libs::Graphics {
@@ -24,8 +26,6 @@ struct GraphicContext;
 struct ShaderBufferResource;
 struct ShaderComputeInputInfo;
 struct CommandSlot;
-struct VulkanBuffer;
-struct VulkanDescriptorSet;
 struct RenderDepthInfo;
 struct RenderColorInfo;
 struct DrawCallInfo;
@@ -49,20 +49,6 @@ enum class CommandBufferDebugOp : uint32_t {
 	Unknown,
 };
 
-class FenceResourceRetainer {
-public:
-	FenceResourceRetainer() = default;
-	~FenceResourceRetainer();
-	KYTY_CLASS_NO_COPY(FenceResourceRetainer);
-
-	void               Retain(std::shared_ptr<void> resource);
-	void               ReleaseAfterFence() noexcept;
-	[[nodiscard]] bool Empty() const noexcept { return m_resources.empty(); }
-
-private:
-	std::vector<std::shared_ptr<void>> m_resources;
-};
-
 struct SubmitInfo {
 	static constexpr uint32_t MaxSemaphores = 3;
 
@@ -108,9 +94,6 @@ public:
 	void WaitForFenceOnly();
 	void WaitForFence();
 	void WaitForFenceAndReset();
-	void RetireBufferAfterFence(std::unique_ptr<VulkanBuffer> buffer);
-	void RetainResourceUntilFence(std::shared_ptr<void> resource);
-	void RecycleDescriptorAfterFence(VulkanDescriptorSet& set);
 
 	[[nodiscard]] vk::CommandBuffer Handle() const;
 	[[nodiscard]] GraphicContext&   GetGraphics() const noexcept { return m_graphics; }
@@ -120,29 +103,23 @@ public:
 private:
 	void Release();
 	void FinalizeFence(bool reset_recording);
-	void ReleaseResourcesAfterFence();
-	void DeleteBuffersAfterFence();
-	void RecycleDescriptorsAfterFence();
 
-	RenderContext&                             m_context;
-	CommandScheduler&                          m_scheduler;
-	GraphicContext&                            m_graphics;
-	CommandSlot*                               m_slot            = nullptr;
-	bool                                       m_execute         = false;
-	bool                                       m_fence_waited    = false;
-	uint64_t                                   m_submit_seq      = 0;
-	uint32_t                                   m_debug_op        = 0;
-	uint64_t                                   m_debug_submit_id = 0;
-	uint32_t                                   m_debug_arg0      = 0;
-	uint32_t                                   m_debug_arg1      = 0;
-	uint32_t                                   m_debug_arg2      = 0;
-	uint32_t                                   m_debug_arg3      = 0;
-	uint64_t                                   m_debug_arg4      = 0;
-	std::vector<std::unique_ptr<VulkanBuffer>> m_retired_buffers;
-	FenceResourceRetainer                      m_fence_resources;
-	std::vector<VulkanDescriptorSet*>          m_descriptor_sets_after_fence;
-	mutable RenderState                        m_render_state;
-	mutable bool                               m_rendering = false;
+	RenderContext&      m_context;
+	CommandScheduler&   m_scheduler;
+	GraphicContext&     m_graphics;
+	CommandSlot*        m_slot            = nullptr;
+	bool                m_execute         = false;
+	bool                m_fence_waited    = false;
+	uint64_t            m_submit_seq      = 0;
+	uint32_t            m_debug_op        = 0;
+	uint64_t            m_debug_submit_id = 0;
+	uint32_t            m_debug_arg0      = 0;
+	uint32_t            m_debug_arg1      = 0;
+	uint32_t            m_debug_arg2      = 0;
+	uint32_t            m_debug_arg3      = 0;
+	uint64_t            m_debug_arg4      = 0;
+	mutable RenderState m_render_state;
+	mutable bool        m_rendering = false;
 };
 
 class RenderCommandBuffer final: public CommandBuffer {
@@ -181,25 +158,23 @@ public:
 	void DispatchDirect(uint64_t submit_id, RenderCommandBuffer& buffer, uint32_t thread_group_x,
 	                    uint32_t thread_group_y, uint32_t thread_group_z, uint32_t mode);
 
-	[[nodiscard]] DescriptorCache::PreparedBindings
-	     PrepareBindings(CommandBuffer& buffer, const ShaderStageRuntime& runtime,
-	                     vk::ShaderStageFlags shader_stage, DescriptorCache::Stage stage);
-	void RebindBuffers(CommandBuffer& buffer, DescriptorCache::PreparedBindings& bindings);
-	void RebindImages(CommandBuffer& buffer, DescriptorCache::PreparedBindings& bindings);
+	[[nodiscard]] PreparedBindings PrepareBindings(const ShaderStageRuntime& runtime);
+	void                           FindBuffers(PreparedBindings& bindings);
+	void                           RebindBuffers(PreparedBindings& bindings);
+	void                           RebindImages(PreparedBindings& bindings);
 	void CommitBindings(CommandBuffer& buffer, vk::PipelineBindPoint pipeline_bind_point,
-	                    vk::PipelineLayout layout, DescriptorCache::PreparedBindings& bindings);
+	                    const PipelineCache::Pipeline&     pipeline,
+	                    std::span<PreparedBindings* const> bindings);
 
 private:
 	struct GraphicsBindings {
-		DescriptorCache::PreparedBindings                vertex;
-		std::optional<DescriptorCache::PreparedBindings> pixel;
+		PreparedBindings                vertex;
+		std::optional<PreparedBindings> pixel;
 	};
 
-	[[nodiscard]] DescriptorCache::TextureBinding
-	ResolveTexture(const ShaderRecompiler::IR::ImageResource&   resource,
-	               const ShaderRecompiler::IR::DescriptorValue& value);
-	[[nodiscard]] GraphicsBindings PrepareGraphicsBindings(CommandBuffer&            buffer,
-	                                                       const ShaderStageRuntime& vertex,
+	[[nodiscard]] TextureBinding ResolveTexture(const ShaderRecompiler::IR::ImageResource& resource,
+	                                            const ShaderRecompiler::IR::DescriptorValue& value);
+	[[nodiscard]] GraphicsBindings PrepareGraphicsBindings(const ShaderStageRuntime& vertex,
 	                                                       const ShaderStageRuntime& pixel,
 	                                                       bool                      pixel_active);
 	void ResolveRenderColorTarget(uint64_t submit_id, RenderCommandBuffer& buffer,
@@ -212,11 +187,12 @@ private:
 	                                          const DrawCallInfo& draw,
 	                                          uint32_t            render_target_slice_offset,
 	                                          bool log_setup_phases, DrawRenderState& state);
-	void ExecutePreparedDraw(uint64_t submit_id, RenderCommandBuffer& buffer,
-	                         const DrawCallInfo& draw, DrawRenderState& state,
-	                         vk::PrimitiveTopology topology, const DrawEmitInfo& emit,
-	                         const DrawIndexBufferSource& index_source, bool log_pipeline_phase,
-	                         bool set_bind_debug, bool set_auto_debug);
+	void               ExecutePreparedDraw(uint64_t submit_id, RenderCommandBuffer& buffer,
+	                                       const DrawCallInfo& draw, DrawRenderState& state,
+	                                       vk::PrimitiveTopology topology, const DrawEmitInfo& emit,
+	                                       const DrawIndexBufferSource& index_source,
+	                                       bool primitive_restart_enable, bool log_pipeline_phase,
+	                                       bool set_bind_debug, bool set_auto_debug);
 	[[nodiscard]] RenderState AcquireRenderTargets(CommandBuffer& buffer, RenderColorInfo* colors,
 	                                               uint32_t color_count, RenderDepthInfo& depth);
 	[[nodiscard]] bool        ResolveColorTargets(uint64_t submit_id, RenderCommandBuffer& buffer,
@@ -228,8 +204,14 @@ private:
 	[[nodiscard]] bool        TryConsumeComputeMetaClear(const ShaderComputeInputInfo& input,
 	                                                     const RenderCommandBuffer&    buffer);
 
-	RenderContext&                      m_context;
-	std::vector<std::shared_ptr<Image>> m_bound_images;
+	RenderContext&                        m_context;
+	std::vector<ImageId>                  m_bound_images;
+	std::vector<vk::DescriptorBufferInfo> m_descriptor_buffers;
+	std::vector<vk::DescriptorImageInfo>  m_descriptor_images;
+	std::vector<vk::WriteDescriptorSet>   m_descriptor_writes;
+	std::vector<uint32_t>                 m_image_occurrences;
+	std::array<uint32_t, ShaderRecompiler::IR::NativePushConstantSize / sizeof(uint32_t)>
+	    m_push_constants {};
 
 	friend struct RenderExecutorTestAccess;
 };

src/graphics/host_gpu/renderer/renderCompute.cpp

@@ -11,16 +11,14 @@
 #include "graphics/guest_gpu/hardwareContext.h"
 #include "graphics/host_gpu/graphicContext.h"
 #include "graphics/host_gpu/renderer/image/imageInfo.h"
-#include "graphics/host_gpu/renderer/pipeline/descriptorCache.h"
 #include "graphics/host_gpu/renderer/pipeline/descriptors.h"
 #include "graphics/host_gpu/renderer/pipeline/pipelineCache.h"
 #include "graphics/host_gpu/renderer/pipeline/shaderResourceBarrier.h"
-#include "graphics/host_gpu/renderer/pipeline/shaderSubgroup.h"
 #include "graphics/host_gpu/renderer/render.h"
 #include "graphics/host_gpu/renderer/renderContext.h"
 #include "graphics/host_gpu/vulkanCommon.h"
-#include "graphics/shader/recompiler/ir/ResourceMaterialization.h"
 #include "graphics/shader/recompiler/ir/ShaderIR.h"
+#include "graphics/shader/recompiler/ir/passes/ResourceMaterialization.h"
 #include "graphics/shader/shader.h"
 #include "kernel/eventQueue.h"
 #include "kernel/pthread.h"
@@ -93,7 +91,7 @@ bool ResolveComputeImageClear(const ShaderComputeInputInfo& input, uint32_t grou
 	const auto  descriptor = DecodeNativeDescriptor<ShaderBufferResource>(raw);
 	if (!resource.formatted || !resource.written || resource.read || resource.atomic ||
 	    resource.scalar || resource.max_byte_extent != 16 || descriptor.Stride() != 16 ||
-	    descriptor.Format() != Prospero::GpuEnumValue(Prospero::BufferFormat::k32_32_32_32UInt) ||
+	    descriptor.Format() != Prospero::BufferFormat::k32_32_32_32UInt ||
 	    descriptor.SwizzleEnabled() || descriptor.IndexStride() != 0 || descriptor.AddTid() ||
 	    resource.packed_stride != descriptor.PackedStride() || raw.dword_count != 4 ||
 	    program.user_data_base != 0 || resources.user_data.size() != 8) {
@@ -115,7 +113,7 @@ bool ResolveComputeImageClear(const ShaderComputeInputInfo& input, uint32_t grou
 	    group_z == 1 && input.dispatch_threads_num[0] == group_x &&
 	    input.dispatch_threads_num[1] == 1 && input.dispatch_threads_num[2] == 1 &&
 	    input.group_id[0] && !input.group_id[1] && !input.group_id[2] &&
-	    input.thread_ids_num == 1 && input.wave_size == 32 && !input.tg_size_en && mode == 0x61u &&
+	    input.thread_ids_num == 1 && input.wave_size == 64 && !input.tg_size_en && mode == 0x61u &&
 	    group_x % input.threads_num[0] == 0 && descriptor.NumRecords() == group_x;
 	const auto size = BufferDescriptorSize(descriptor);
 	if (!full_dispatch || size == 0) {
@@ -139,6 +137,21 @@ static bool TryConsumeComputeImageClear(const ShaderComputeInputInfo& input, Com
 	}
 	auto& cache = command.GetContext().GetTextureCache();
 	if (!cache.ClearImageFromBuffer(command, descriptor.Base48(), size, packed_clear)) {
+		// Recognized metadata-fill shaders access DCC as an ordinary storage buffer and may run
+		// before the render target is bound. TryConsumeDccFill either consumes registered state
+		// or retains a PendingDcc fill while allowing the dispatch to run.
+		const bool registered_metadata =
+		    cache.TryConsumeDccFill(descriptor.Base48(), size, packed_clear);
+		static std::atomic<uint32_t> logged_metadata_clears {0};
+		if (logged_metadata_clears.fetch_add(1, std::memory_order_relaxed) < 32) {
+			LOGF("GraphicsRenderDispatchDirect: %s metadata clear shader=0x%016" PRIx64
+			     " addr=0x%016" PRIx64 " size=0x%016" PRIx64 " value=0x%08" PRIx32 "\n",
+			     registered_metadata ? "tracked" : "deferred", input.stage.program->shader_hash,
+			     descriptor.Base48(), size, packed_clear);
+		}
+		if (registered_metadata) {
+			return true;
+		}
 		return false;
 	}
 	static std::atomic<uint32_t> logged_clears {0};
@@ -154,6 +167,7 @@ void RenderExecutor::DispatchDirect(uint64_t submit_id, RenderCommandBuffer& buf
                                     uint32_t thread_group_x, uint32_t thread_group_y,
                                     uint32_t thread_group_z, uint32_t mode) {
 	EXIT_IF(buffer.IsInvalid());
+	m_context.GetCommandScheduler().PopPendingOperations();
 	auto& ctx    = buffer.GetRegisters();
 	auto& sh_ctx = buffer.GetShaders();
 
@@ -196,7 +210,9 @@ void RenderExecutor::DispatchDirect(uint64_t submit_id, RenderCommandBuffer& buf
 
 	ShaderComputeInputInfo    input_info {};
 	std::span<const uint32_t> cs_shader;
-	if (!ShaderCompileInfoCS(cs_regs, sh_regs, input_info, cs_shader)) {
+	if (!ShaderCompileInfoCS(cs_regs, sh_regs,
+	                         !m_context.GetGraphics().compute_wave64_supported, input_info,
+	                         cs_shader)) {
 		EXIT("ShaderCompileInfoCS failed for dispatch with CS shader 0x%016" PRIx64 "\n",
 		     cs_regs.cs_regs.data_addr);
 	}
@@ -246,7 +262,7 @@ void RenderExecutor::DispatchDirect(uint64_t submit_id, RenderCommandBuffer& buf
 			LOGF("  CS buffer[%u]: source=%u usage=%s addr=0x%012" PRIx64
 			     " stride=%u records=%u format=%u\n",
 			     i, buffer.source, buffer.written ? "read-write" : "read-only", r.Base48(),
-			     r.Stride(), r.NumRecords(), r.Format());
+			     r.Stride(), r.NumRecords(), r.RawFormat());
 		}
 		for (uint32_t i = 0; i < program.info.images.size(); i++) {
 			const auto& image = program.info.images[i];
@@ -258,13 +274,14 @@ void RenderExecutor::DispatchDirect(uint64_t submit_id, RenderCommandBuffer& buf
 			      image.kind == ShaderRecompiler::IR::ResourceKind::ImageUint)
 			         ? "true"
 			         : "false",
-			     r.Base40(), static_cast<uint32_t>(r.Type()), r.Format(),
+			     r.Base40(), static_cast<uint32_t>(r.Type()), static_cast<uint32_t>(r.Format()),
 			     static_cast<uint32_t>(r.Width5()) + 1u, static_cast<uint32_t>(r.Height5()) + 1u,
 			     static_cast<uint32_t>(r.Depth()) + 1u,
-			     std::max<uint32_t>(static_cast<uint32_t>(r.LastLevel()),
-			                        static_cast<uint32_t>(r.MaxMip())) +
-			         1u,
-			     r.TileMode());
+			     r.Type() == Prospero::ImageType::kColor2DMsaa ||
+			             r.Type() == Prospero::ImageType::kColor2DMsaaArray
+			         ? 1u
+			         : static_cast<uint32_t>(image.r128 ? r.LastLevel() : r.MaxMip()) + 1u,
+			     static_cast<uint32_t>(r.TileMode()));
 		}
 		for (uint32_t i = 0; i < program.info.samplers.size(); i++) {
 			const auto r = DecodeNativeDescriptor<ShaderSamplerResource>(resources.samplers[i]);
@@ -318,13 +335,15 @@ void RenderExecutor::DispatchDirect(uint64_t submit_id, RenderCommandBuffer& buf
 	buffer.EndRendering();
 	auto& pipeline =
 	    m_context.GetPipelineCache().CreateComputePipeline(input_info, sh_ctx.GetCs(), cs_shader);
-	auto bindings = PrepareBindings(buffer, input_info.stage, vk::ShaderStageFlagBits::eCompute,
-	                                DescriptorCache::Stage::Compute);
-	RebindBuffers(buffer, bindings);
-	RebindImages(buffer, bindings);
+	auto bindings = PrepareBindings(input_info.stage);
+	FindBuffers(bindings);
+	RebindBuffers(bindings);
+	RebindImages(bindings);
 
-	auto vk_buffer = buffer.Handle();
-	CommitBindings(buffer, vk::PipelineBindPoint::eCompute, pipeline.pipeline_layout, bindings);
+	auto              vk_buffer        = buffer.Handle();
+	PreparedBindings* descriptor_stage = &bindings;
+	CommitBindings(buffer, vk::PipelineBindPoint::eCompute, pipeline,
+	               std::span {&descriptor_stage, 1u});
 	bool has_storage_writes = HasShaderBufferWrites(input_info.stage);
 	has_storage_writes =
 	    std::any_of(program.info.images.begin(), program.info.images.end(),

src/graphics/host_gpu/renderer/renderContext.cpp

@@ -13,8 +13,9 @@ namespace Libs::Graphics {
 
 RenderContext::RenderContext(GraphicContext& graphics)
     : m_graphics(graphics), m_render_executor(*this), m_command_scheduler(*this, graphics),
-      m_descriptor_cache(graphics), m_pipeline_cache(graphics, m_descriptor_cache),
-      m_sampler_cache(graphics), m_gpu_resources(graphics, m_command_scheduler) {
+      m_descriptor_heap(graphics, m_command_scheduler.GetMasterSemaphore()),
+      m_pipeline_cache(graphics), m_sampler_cache(graphics),
+      m_gpu_resources(graphics, m_command_scheduler) {
 	EXIT_NOT_IMPLEMENTED(!Common::Thread::IsMainThread());
 }
 
@@ -26,7 +27,7 @@ RenderContext::~RenderContext() {
 void RenderContext::InitializeGpu(VideoOut::VideoOutDriver* video_out) {
 	EXIT_IF(m_gpu != nullptr);
 	m_video_out = video_out;
-	m_gpu       = std::make_unique<Gpu>(*this);
+	m_gpu       = std::make_unique<GuestGpu>(*this);
 	m_gpu_resources.SetGpu(m_gpu.get());
 }
 
@@ -45,7 +46,7 @@ void RenderContext::ShutdownGpu() {
 	}
 }
 
-Gpu& RenderContext::GetGpu() const {
+GuestGpu& RenderContext::GetGpu() const {
 	EXIT_IF(m_gpu == nullptr);
 	return *m_gpu;
 }

src/graphics/host_gpu/renderer/renderContext.h

@@ -10,7 +10,7 @@
 #include "graphics/host_gpu/renderer/cache/samplerCache.h"
 #include "graphics/host_gpu/renderer/cache/textureCache.h"
 #include "graphics/host_gpu/renderer/commandScheduler.h"
-#include "graphics/host_gpu/renderer/pipeline/descriptorCache.h"
+#include "graphics/host_gpu/renderer/pipeline/descriptorHeap.h"
 #include "graphics/host_gpu/renderer/pipeline/pipelineCache.h"
 #include "kernel/eventQueue.h"
 
@@ -24,7 +24,7 @@ class VideoOutDriver;
 namespace Libs::Graphics {
 
 constexpr int AGC_USER_INTERRUPT_EVENT = 0x1800;
-class Gpu;
+class GuestGpu;
 
 class RenderContext {
 public:
@@ -35,13 +35,13 @@ public:
 	[[nodiscard]] GraphicContext&           GetGraphics() const noexcept { return m_graphics; }
 	void                                    InitializeGpu(VideoOut::VideoOutDriver* video_out);
 	void                                    ShutdownGpu();
-	[[nodiscard]] Gpu&                      GetGpu() const;
+	[[nodiscard]] GuestGpu&                 GetGpu() const;
 	[[nodiscard]] VideoOut::VideoOutDriver& GetVideoOut() const;
 
 	Common::Mutex&      GetMutex() { return m_mutex; }
 	CommandScheduler&   GetCommandScheduler() { return m_command_scheduler; }
 	PipelineCache&      GetPipelineCache() { return m_pipeline_cache; }
-	DescriptorCache&    GetDescriptorCache() { return m_descriptor_cache; }
+	DescriptorHeap&     GetDescriptorHeap() { return m_descriptor_heap; }
 	SamplerCache&       GetSamplerCache() { return m_sampler_cache; }
 	GpuResourceManager& GetGpuResources() { return m_gpu_resources; }
 	BufferCache&        GetBufferCache() { return m_gpu_resources.GetBufferCache(); }
@@ -63,11 +63,11 @@ private:
 	Common::Mutex             m_mutex;
 	RenderExecutor            m_render_executor;
 	CommandScheduler          m_command_scheduler;
-	DescriptorCache           m_descriptor_cache;
+	DescriptorHeap            m_descriptor_heap;
 	PipelineCache             m_pipeline_cache;
 	SamplerCache              m_sampler_cache;
 	GpuResourceManager        m_gpu_resources;
-	std::unique_ptr<Gpu>      m_gpu;
+	std::unique_ptr<GuestGpu> m_gpu;
 	VideoOut::VideoOutDriver* m_video_out = nullptr;
 
 	Common::Mutex                  m_eop_mutex;

src/graphics/host_gpu/renderer/renderDraw.cpp

@@ -15,15 +15,13 @@
 #include "graphics/host_gpu/renderer/colorRenderTarget.h"
 #include "graphics/host_gpu/renderer/debug.h"
 #include "graphics/host_gpu/renderer/depthRenderTarget.h"
-#include "graphics/host_gpu/renderer/pipeline/descriptorCache.h"
 #include "graphics/host_gpu/renderer/pipeline/pipelineCache.h"
 #include "graphics/host_gpu/renderer/pipeline/shaderResourceBarrier.h"
-#include "graphics/host_gpu/renderer/pipeline/shaderSubgroup.h"
 #include "graphics/host_gpu/renderer/render.h"
 #include "graphics/host_gpu/renderer/renderContext.h"
 #include "graphics/host_gpu/vulkanCommon.h"
-#include "graphics/shader/recompiler/ir/ResourceMaterialization.h"
 #include "graphics/shader/recompiler/ir/ShaderIR.h"
+#include "graphics/shader/recompiler/ir/passes/ResourceMaterialization.h"
 #include "graphics/shader/shader.h"
 #include "kernel/eventQueue.h"
 #include "kernel/memory.h"
@@ -35,9 +33,11 @@
 #include <atomic>
 #include <bit>
 #include <cmath>
+#include <cstdio>
 #include <cstring>
 #include <limits>
 #include <memory>
+#include <mutex>
 #include <optional>
 #include <span>
 #include <unordered_map>
@@ -123,7 +123,7 @@ static void LogFramebufferSkip(const char* draw_name, const RenderColorInfo& col
 	    log_id, draw_name, RenderColorTypeName(color.type), color.base_addr, color.buffer_size,
 	    color.image_id ? "yes" : "no", VulkanToString(depth.format).c_str(),
 	    depth.image_id ? "yes" : "no", depth.vaddr_num, ctx.GetRenderTargetMask(),
-	    ucfg.GetPrimType(), index_count, flags);
+	    static_cast<uint32_t>(ucfg.GetPrimType()), index_count, flags);
 }
 
 static void LogMrtState(const char* draw_name, const RenderCommandBuffer& buffer,
@@ -140,18 +140,6 @@ static void LogMrtState(const char* draw_name, const RenderCommandBuffer& buffer
 	                   (bc0.separate_alpha_blend && (IsDualSourceBlendFactor(bc0.alpha_srcblend) ||
 	                                                 IsDualSourceBlendFactor(bc0.alpha_destblend)));
 
-	for (uint32_t i = 1; i < 8; i++) {
-		const auto& rt = ctx.GetRenderTarget(i);
-		if (rt.base.addr != 0 || ps_input_info.target_output_mode[i] != 0 ||
-		    ((rt_mask >> (i * 4u)) & 0x0fu) != 0 || ((cb_shader_mask >> (i * 4u)) & 0x0fu) != 0) {
-			interesting = true;
-		}
-	}
-
-	if (!interesting) {
-		return;
-	}
-
 	auto log_id = g_mrt_state_log_count.fetch_add(1);
 	if (log_id >= 32) {
 		return;
@@ -179,10 +167,12 @@ static void LogMrtState(const char* draw_name, const RenderCommandBuffer& buffer
 		     " fmt=0x%08" PRIx32 " nfmt=0x%08" PRIx32 " order=0x%08" PRIx32
 		     " width=%u height=%u tile=%u"
 		     " blend=%s src=%u dst=%u alpha_src=%u alpha_dst=%u\n",
-		     log_id, i, rt.base.addr, ctm, csm, ps_input_info.target_output_mode[i], rt.info.format,
-		     rt.info.channel_type, rt.info.channel_order, rt.attrib2.width + 1,
-		     rt.attrib2.height + 1, rt.attrib3.tile_mode, bc.enable ? "true" : "false",
-		     bc.color_srcblend, bc.color_destblend, bc.alpha_srcblend, bc.alpha_destblend);
+		     log_id, i, rt.base.addr, ctm, csm, ps_input_info.target_output_mode[i],
+		     static_cast<uint32_t>(rt.info.format), static_cast<uint32_t>(rt.info.channel_type),
+		     static_cast<uint32_t>(rt.info.channel_order), rt.attrib2.width + 1,
+		     rt.attrib2.height + 1, static_cast<uint32_t>(rt.attrib3.tile_mode),
+		     bc.enable ? "true" : "false", bc.color_srcblend, bc.color_destblend, bc.alpha_srcblend,
+		     bc.alpha_destblend);
 	}
 }
 
@@ -225,7 +215,7 @@ static void LogDrawTargetState(const char* draw_name, const RenderColorInfo& col
 	    "scissor=(%d,%d)-(%d,%d)\n",
 	    log_id, buffer.GetContext().GetGpu().GetFrameNum(), draw_name,
 	    RenderColorTypeName(color.type), color.base_addr, extent.width, extent.height,
-	    ucfg.GetPrimType(), index_count, flags, ctx.GetRenderTargetMask(),
+	    static_cast<uint32_t>(ucfg.GetPrimType()), index_count, flags, ctx.GetRenderTargetMask(),
 	    color.color_clear_enable ? "true" : "false", color.color_clear_value.float32[0],
 	    color.color_clear_value.float32[1], color.color_clear_value.float32[2],
 	    color.color_clear_value.float32[3], cc.mode, cc.op, bc.enable ? "true" : "false",
@@ -289,8 +279,7 @@ static void LogDrawInputState(const RenderCommandBuffer& buffer, const RenderCol
 					const auto& rd        = vs_input_info.resources_dst[res_index];
 					const auto  offset    = b.attr_offsets[ai];
 					if (offset + 4u <= b.stride &&
-					    r.Format() ==
-					        Prospero::GpuEnumValue(Prospero::BufferFormat::k8_8_8_8UNorm)) {
+					    r.Format() == Prospero::BufferFormat::k8_8_8_8UNorm) {
 						uint32_t packed = 0;
 						std::memcpy(&packed, rec_bytes + offset, sizeof(packed));
 						const auto r8 = (packed >> 0u) & 0xffu;
@@ -320,89 +309,43 @@ static void LogDrawInputState(const RenderCommandBuffer& buffer, const RenderCol
 	}
 }
 
-static PipelineDynamicParameters BuildGraphicsDynamicParams(const RenderCommandBuffer& buffer,
-                                                            const RenderColorInfo*     colors,
-                                                            uint32_t                   color_count,
-                                                            const RenderDepthInfo&     depth) {
+static void SetGraphicsDynamicParams(const RenderCommandBuffer& buffer, vk::CommandBuffer vk_buffer,
+                                     const RenderColorInfo* colors, uint32_t color_count,
+                                     const RenderDepthInfo& depth) {
+	KYTY_PROFILER_FUNCTION();
+
 	EXIT_IF(colors == nullptr);
 	const auto& ctx = buffer.GetRegisters();
 
-	PipelineDynamicParameters ret {};
-	ret.color_write_count   = color_count;
-	ret.stencil_test_enable = depth.stencil_test_enable;
-
 	const auto&  vp = ctx.GetScreenViewport();
 	vk::Extent2D framebuffer_extent {};
 	if (color_count > 0 && colors[0].image_id) {
 		framebuffer_extent = colors[0].extent;
 	} else if (depth.image_id) {
 		framebuffer_extent = {depth.width, depth.height};
+	} else {
+		const auto& limits = buffer.GetGraphics().GetPhysicalDeviceProperties().limits;
+		framebuffer_extent = {limits.maxFramebufferWidth, limits.maxFramebufferHeight};
 	}
 
 	const auto final_scissor = calc_final_scissor(vp, ctx.GetScanModeControl(), framebuffer_extent);
 
-	ret.viewport_scale[0]  = vp.viewports[0].xscale;
-	ret.viewport_scale[1]  = vp.viewports[0].yscale;
-	ret.viewport_scale[2]  = vp.viewports[0].zscale;
-	ret.viewport_offset[0] = vp.viewports[0].xoffset;
-	ret.viewport_offset[1] = vp.viewports[0].yoffset;
-	ret.viewport_offset[2] = vp.viewports[0].zoffset;
-	ret.scissor_ltrb[0]    = final_scissor.left;
-	ret.scissor_ltrb[1]    = final_scissor.top;
-	ret.scissor_ltrb[2]    = final_scissor.right;
-	ret.scissor_ltrb[3]    = final_scissor.bottom;
-	ret.line_width         = ctx.GetLineWidth();
-	ret.stencil_front      = depth.stencil_dynamic_front;
-	ret.stencil_back       = depth.stencil_dynamic_back;
-
-	const auto& mode        = ctx.GetModeControl();
-	const auto& poly_offset = ctx.GetPolyOffset();
-	const bool  use_front   = mode.poly_offset_front_enable && !mode.cull_front;
-	const bool  use_back    = mode.poly_offset_back_enable && !mode.cull_back;
-	ret.depth_bias_enable   = use_front || use_back;
-	if (ret.depth_bias_enable) {
-		// Vulkan has one bias for both faces. Prefer a visible front face when both are enabled.
-		const float guest_constant_factor =
-		    use_front ? poly_offset.front_offset : poly_offset.back_offset;
-		ret.depth_bias_constant_factor =
-		    ConvertPolygonOffsetConstantFactor(guest_constant_factor, poly_offset, depth.format);
-		ret.depth_bias_clamp = poly_offset.clamp;
-		ret.depth_bias_slope_factor =
-		    (use_front ? poly_offset.front_scale : poly_offset.back_scale) / 16.0f;
-	}
-
-	// CB_COLOR_CONTROL.operation controls special CB operations, not the normal color component
-	// write mask. Use CB_TARGET_MASK here so scanout passes with mode=Disable do not go black.
-	for (uint32_t i = 0; i < RENDER_COLOR_ATTACHMENTS_MAX; i++) {
-		ret.color_write_enable[i] =
-		    (i < color_count &&
-		     render_target_mask_slot(ctx.GetRenderTargetMask(), colors[i].target_slot) != 0);
-	}
-
-	return ret;
-}
-
-static void SetDynamicParams(const RenderCommandBuffer& buffer, vk::CommandBuffer vk_buffer,
-                             const PipelineDynamicParameters& dynamic_params) {
-	KYTY_PROFILER_FUNCTION();
-
 	vk::Viewport viewport {};
-	viewport.x        = dynamic_params.viewport_offset[0] - dynamic_params.viewport_scale[0];
-	viewport.y        = dynamic_params.viewport_offset[1] - dynamic_params.viewport_scale[1];
-	viewport.width    = dynamic_params.viewport_scale[0] * 2.0f;
-	viewport.height   = dynamic_params.viewport_scale[1] * 2.0f;
-	viewport.minDepth = dynamic_params.viewport_offset[2];
-	viewport.maxDepth = dynamic_params.viewport_scale[2] + dynamic_params.viewport_offset[2];
+	viewport.x        = vp.viewports[0].xoffset - vp.viewports[0].xscale;
+	viewport.y        = vp.viewports[0].yoffset - vp.viewports[0].yscale;
+	viewport.width    = vp.viewports[0].xscale * 2.0f;
+	viewport.height   = vp.viewports[0].yscale * 2.0f;
+	viewport.minDepth = vp.viewports[0].zoffset;
+	viewport.maxDepth = vp.viewports[0].zscale + vp.viewports[0].zoffset;
 	vk_buffer.setViewport(0, 1, &viewport);
 
 	vk::Rect2D scissor {};
-	scissor.offset = {dynamic_params.scissor_ltrb[0], dynamic_params.scissor_ltrb[1]};
-	scissor.extent = {
-	    static_cast<uint32_t>(dynamic_params.scissor_ltrb[2] - dynamic_params.scissor_ltrb[0]),
-	    static_cast<uint32_t>(dynamic_params.scissor_ltrb[3] - dynamic_params.scissor_ltrb[1])};
+	scissor.offset = {final_scissor.left, final_scissor.top};
+	scissor.extent = {static_cast<uint32_t>(final_scissor.right - final_scissor.left),
+	                  static_cast<uint32_t>(final_scissor.bottom - final_scissor.top)};
 	vk_buffer.setScissor(0, 1, &scissor);
 
-	float line_width = dynamic_params.line_width;
+	float line_width = ctx.GetLineWidth();
 	if (line_width != 1.0f) {
 		static bool logged = false;
 		if (!logged) {
@@ -415,26 +358,36 @@ static void SetDynamicParams(const RenderCommandBuffer& buffer, vk::CommandBuffe
 	}
 	vk_buffer.setLineWidth(line_width);
 
-	vk_buffer.setDepthBiasEnable(dynamic_params.depth_bias_enable ? VK_TRUE : VK_FALSE);
-	if (dynamic_params.depth_bias_enable) {
-		vk_buffer.setDepthBias(dynamic_params.depth_bias_constant_factor,
-		                       dynamic_params.depth_bias_clamp,
-		                       dynamic_params.depth_bias_slope_factor);
+	const auto& mode              = ctx.GetModeControl();
+	const auto& poly_offset       = ctx.GetPolyOffset();
+	const bool  use_front         = mode.poly_offset_front_enable && !mode.cull_front;
+	const bool  use_back          = mode.poly_offset_back_enable && !mode.cull_back;
+	const bool  depth_bias_enable = use_front || use_back;
+	vk_buffer.setDepthBiasEnable(depth_bias_enable ? VK_TRUE : VK_FALSE);
+	if (depth_bias_enable) {
+		// Vulkan has one bias for both faces. Prefer a visible front face when both are enabled.
+		const float guest_constant_factor =
+		    use_front ? poly_offset.front_offset : poly_offset.back_offset;
+		const float constant_factor =
+		    ConvertPolygonOffsetConstantFactor(guest_constant_factor, poly_offset, depth.format);
+		const float slope_factor =
+		    (use_front ? poly_offset.front_scale : poly_offset.back_scale) / 16.0f;
+		vk_buffer.setDepthBias(constant_factor, poly_offset.clamp, slope_factor);
 	}
 
-	if (dynamic_params.stencil_test_enable) {
+	if (depth.stencil_test_enable) {
 		vk_buffer.setStencilCompareMask(vk::StencilFaceFlagBits::eFront,
-		                                dynamic_params.stencil_front.compareMask);
+		                                depth.stencil_dynamic_front.compareMask);
 		vk_buffer.setStencilCompareMask(vk::StencilFaceFlagBits::eBack,
-		                                dynamic_params.stencil_back.compareMask);
+		                                depth.stencil_dynamic_back.compareMask);
 		vk_buffer.setStencilWriteMask(vk::StencilFaceFlagBits::eFront,
-		                              dynamic_params.stencil_front.writeMask);
+		                              depth.stencil_dynamic_front.writeMask);
 		vk_buffer.setStencilWriteMask(vk::StencilFaceFlagBits::eBack,
-		                              dynamic_params.stencil_back.writeMask);
+		                              depth.stencil_dynamic_back.writeMask);
 		vk_buffer.setStencilReference(vk::StencilFaceFlagBits::eFront,
-		                              dynamic_params.stencil_front.reference);
+		                              depth.stencil_dynamic_front.reference);
 		vk_buffer.setStencilReference(vk::StencilFaceFlagBits::eBack,
-		                              dynamic_params.stencil_back.reference);
+		                              depth.stencil_dynamic_back.reference);
 	}
 
 #if defined(__APPLE__)
@@ -442,11 +395,15 @@ static void SetDynamicParams(const RenderCommandBuffer& buffer, vk::CommandBuffe
 	// eColorWriteEnableEXT dynamic state and relies on the static colorWriteMask instead.
 #else
 	vk::Bool32 enable[RENDER_COLOR_ATTACHMENTS_MAX] = {};
-	for (uint32_t i = 0; i < dynamic_params.color_write_count; i++) {
-		enable[i] = (dynamic_params.color_write_enable[i] ? VK_TRUE : VK_FALSE);
+	// Color-control operation selects special color-buffer paths, not the normal component write
+	// mask. Attachment availability therefore follows the target write mask.
+	for (uint32_t i = 0; i < color_count; i++) {
+		enable[i] = render_target_mask_slot(ctx.GetRenderTargetMask(), colors[i].target_slot) != 0
+		                ? VK_TRUE
+		                : VK_FALSE;
 	}
-	if (dynamic_params.color_write_count != 0) {
-		vk_buffer.setColorWriteEnableEXT(dynamic_params.color_write_count, enable);
+	if (color_count != 0) {
+		vk_buffer.setColorWriteEnableEXT(color_count, enable);
 	}
 #endif
 }
@@ -502,6 +459,12 @@ static bool ShouldSkipGeShader(const RenderCommandBuffer& buffer) {
 	    sh_regs.m_geMaxOutputPerSubgroup > 0x00000040;
 
 	if (unsupported_stage_mask || unsupported_gs_stage || ge_group_size || ge_shader_regs) {
+		static std::once_flag warning_once;
+		std::call_once(warning_once, [] {
+			std::printf("Warning: game uses unsupported graphics pipelines; some draw calls were "
+			            "skipped.\n");
+		});
+
 		const auto log_id = g_shader_stage_log_count.fetch_add(1);
 		if (log_id < 32) {
 			LOGF("Skipping unsupported GE shader draw: stages=0x%08" PRIx32
@@ -540,6 +503,70 @@ struct DrawCallInfo {
 	uint32_t             first_instance = 0;
 };
 
+static bool ResolveDccAttachmentClear(TextureCache& cache, const RenderColorInfo& target,
+                                      const ImageViewInfo& view, vk::ClearColorValue& clear_value) {
+	if (target.desc.info.metadata.kind != ImageMetadataKind::Dcc) {
+		return false;
+	}
+
+	// A DCC fast clear may update metadata only and leave the color allocation stale. Vulkan has
+	// no guest DCC state, so materialize the deferred value when the surface is next bound.
+	const auto metadata_address = target.desc.info.metadata.range.address;
+	uint32_t   metadata_value   = 0xffffffffu;
+	if (!cache.IsMetaCleared(metadata_address, view.base_layer, &metadata_value)) {
+		return false;
+	}
+
+	// Translate recognized constant and register-backed DCC states into a host clear value.
+	clear_value = {};
+	switch (static_cast<uint8_t>(metadata_value)) {
+		case 0x00: break;
+		case 0x20:
+			if (!target.metadata_clear_supported) {
+				return false;
+			}
+			clear_value = target.color_clear_value;
+			break;
+		case 0x40:
+			if (!target.metadata_fixed_clear_supported) {
+				return false;
+			}
+			clear_value.float32[3] = 1.0f;
+			break;
+		case 0x80:
+			if (!target.metadata_fixed_clear_supported) {
+				return false;
+			}
+			clear_value.float32[0] = 1.0f;
+			clear_value.float32[1] = 1.0f;
+			clear_value.float32[2] = 1.0f;
+			break;
+		case 0xc0:
+			if (!target.metadata_fixed_clear_supported) {
+				return false;
+			}
+			clear_value.float32[0] = 1.0f;
+			clear_value.float32[1] = 1.0f;
+			clear_value.float32[2] = 1.0f;
+			clear_value.float32[3] = 1.0f;
+			break;
+		default: return false;
+	}
+
+	for (uint32_t layer = 1; layer < view.layer_count; layer++) {
+		if (!cache.IsMetaCleared(metadata_address, view.base_layer + layer)) {
+			return false;
+		}
+	}
+	// Consume only after every layer in this Vulkan view can be materialized together.
+	for (uint32_t layer = 0; layer < view.layer_count; layer++) {
+		if (!cache.TouchMeta(metadata_address, view.base_layer + layer, false)) {
+			EXIT("failed to consume DCC clear state\n");
+		}
+	}
+	return true;
+}
+
 RenderState RenderExecutor::AcquireRenderTargets(CommandBuffer& buffer, RenderColorInfo* colors,
                                                  uint32_t color_count, RenderDepthInfo& depth) {
 	EXIT_IF(colors == nullptr || color_count > RENDER_COLOR_ATTACHMENTS_MAX);
@@ -553,7 +580,7 @@ RenderState RenderExecutor::AcquireRenderTargets(CommandBuffer& buffer, RenderCo
 	for (uint32_t i = 0; i < color_count; i++) {
 		auto& target = colors[i];
 		EXIT_IF(!target.image_id);
-		const auto old_image = cache.ResolveOwner(target.image_id);
+		const auto old_image = cache.m_slot_images.try_get(target.image_id);
 		if (old_image == nullptr || (!old_image->registered && !old_image->info.data.Empty()) ||
 		    old_image->binding.needs_rebind) {
 			if (old_image != nullptr) {
@@ -564,6 +591,15 @@ RenderState RenderExecutor::AcquireRenderTargets(CommandBuffer& buffer, RenderCo
 		}
 		target.image_view = cache.FindRenderTarget(target.image_id, target.desc);
 		auto& image       = cache.GetImage(target.image_id);
+		SetVulkanObjectNameF(m_context.GetGraphics().device, image.backing.image,
+		                     "Kyty.MRT{}.Image[guest=0x{:016x} size=0x{:x} format={}]",
+		                     target.target_slot, image.info.data.address, image.info.data.size,
+		                     static_cast<uint32_t>(image.info.pixel_format));
+		SetVulkanObjectNameF(m_context.GetGraphics().device, target.image_view,
+		                     "Kyty.MRT{}.View[guest=0x{:016x} mip={} layer={}+{}]",
+		                     target.target_slot, image.info.data.address,
+		                     target.desc.view_info.base_level, target.desc.view_info.base_layer,
+		                     target.desc.view_info.layer_count);
 		EXIT_IF(image.backing.samples != target.samples || target.image_view == nullptr);
 		if (attachment_samples == 0) {
 			attachment_samples = target.samples;
@@ -587,10 +623,16 @@ RenderState RenderExecutor::AcquireRenderTargets(CommandBuffer& buffer, RenderCo
 		attachment.image_view   = target.image_view;
 		attachment.image_layout = layout;
 		attachment.clear_value  = target.color_clear_value.uint32;
-		attachment.is_clear     = target.color_clear_enable;
+		vk::ClearColorValue metadata_clear_value {};
+		const bool          metadata_clear =
+		    ResolveDccAttachmentClear(cache, target, view, metadata_clear_value);
+		if (metadata_clear) {
+			attachment.clear_value = metadata_clear_value.uint32;
+		}
+		attachment.is_clear = target.color_clear_enable || metadata_clear;
 	}
 	if (depth.image_id) {
-		const auto owner = cache.ResolveOwner(depth.image_id);
+		const auto owner = cache.m_slot_images.try_get(depth.image_id);
 		if (owner == nullptr || !owner->registered || owner->binding.needs_rebind) {
 			EXIT("depth target changed after render-state discovery\n");
 		}
@@ -607,6 +649,14 @@ RenderState RenderExecutor::AcquireRenderTargets(CommandBuffer& buffer, RenderCo
 			EXIT("failed to consume HTile clear state\n");
 		}
 		auto& image = cache.GetImage(depth.image_id);
+		SetVulkanObjectNameF(m_context.GetGraphics().device, image.backing.image,
+		                     "Kyty.DepthTarget.Image[guest=0x{:016x} size=0x{:x} format={}]",
+		                     image.info.data.address, image.info.data.size,
+		                     static_cast<uint32_t>(image.info.pixel_format));
+		SetVulkanObjectNameF(m_context.GetGraphics().device, depth.image_view,
+		                     "Kyty.DepthTarget.View[guest=0x{:016x} layer={}+{}]",
+		                     image.info.data.address, depth.desc.view_info.base_layer,
+		                     depth.desc.view_info.layer_count);
 		EXIT_IF(depth.image_view == nullptr || image.backing.samples != depth.samples);
 		if (attachment_samples == 0) {
 			attachment_samples = depth.samples;
@@ -639,8 +689,12 @@ RenderState RenderExecutor::AcquireRenderTargets(CommandBuffer& buffer, RenderCo
 		attachment.has_stencil    = static_cast<bool>(aspects & vk::ImageAspectFlagBits::eStencil);
 		attachment.stencil_clear  = depth.stencil_clear_enable;
 	}
-	if (attachment_samples == 0 ||
-	    vulkan_sample_count(attachment_samples) == vk::SampleCountFlagBits {}) {
+	if (color_count == 0 && !depth.image_id) {
+		const auto& limits = buffer.GetGraphics().GetPhysicalDeviceProperties().limits;
+		state.width        = limits.maxFramebufferWidth;
+		state.height       = limits.maxFramebufferHeight;
+	} else if (attachment_samples == 0 ||
+	           vulkan_sample_count(attachment_samples) == vk::SampleCountFlagBits {}) {
 		EXIT("render state has no valid attachments\n");
 	}
 	if (state.num_layers == std::numeric_limits<uint32_t>::max()) {
@@ -652,22 +706,11 @@ RenderState RenderExecutor::AcquireRenderTargets(CommandBuffer& buffer, RenderCo
 	return state;
 }
 
-static bool DrawHasActivePixelShader(const RenderCommandBuffer& buffer,
-                                     const DrawRenderState& state, const DrawCallInfo& draw) {
-	EXIT_IF(draw.name == nullptr);
-	const auto& ctx    = buffer.GetRegisters();
-	const auto& sh_ctx = buffer.GetShaders();
-
-	const bool with_depth = (state.depth_info.format != vk::Format::eUndefined &&
-	                         static_cast<bool>(state.depth_info.image_id));
-	if (state.color_count != 0 || !with_depth) {
-		return true;
-	}
-
-	const auto& sh_regs = ctx.GetShaderRegisters();
-	const auto& ps      = sh_ctx.GetPs();
-	return ShaderAddressValid(ps.ps_regs.data_addr) &&
-	       PixelShaderHasDepthOrCoverageSideEffects(sh_regs);
+static bool DrawHasActivePixelShader(const RenderCommandBuffer& buffer) {
+	const auto& ctx              = buffer.GetRegisters();
+	const auto& sh_regs          = ctx.GetShaderRegisters();
+	const bool  has_color_output = (ctx.GetRenderTargetMask() & sh_regs.m_cbShaderMask) != 0;
+	return has_color_output || PixelShaderHasDepthOrCoverageSideEffects(sh_regs);
 }
 
 enum class CbColorMode : uint8_t {
@@ -682,10 +725,11 @@ enum class CbColorMode : uint8_t {
 static bool ConsumeMetadataColorOperation(const RenderCommandBuffer& buffer) {
 	const auto& ctx  = buffer.GetRegisters();
 	const auto  mode = ctx.GetColorControl().mode;
-	// These AGC CB modes run color-buffer metadata/decompression operations. The shader is a
-	// dummy vehicle for the CB, and its exported color must not be applied as a normal draw.
-	// Kyty currently stores host images as expanded Vulkan images and does not track CMASK/DCC
-	// metadata state, so the matching host operation is a no-op.
+	// These special modes run color-buffer metadata or decompression operations. The shader is a
+	// vehicle for that operation, and its exported color must not be applied as a normal draw.
+	// Kyty stores expanded Vulkan images rather than compressed guest surfaces, so no equivalent
+	// hardware pass is emitted. Tracked DCC clear state is materialized on attachment bind;
+	// future CMask/FMask support can consume its state through the same TextureCache path.
 	return mode == static_cast<uint8_t>(CbColorMode::EliminateFastClear) ||
 	       mode == static_cast<uint8_t>(CbColorMode::FmaskDecompress) ||
 	       mode == static_cast<uint8_t>(CbColorMode::DccDecompress);
@@ -706,11 +750,10 @@ struct DrawIndexBufferSource {
 };
 
 struct PreparedIndexBuffer {
-	std::shared_ptr<void> owner;
-	vk::Buffer            buffer = nullptr;
-	uint64_t              size   = 0;
-	vk::DeviceSize        offset = 0;
-	vk::IndexType         type   = vk::IndexType::eUint16;
+	vk::Buffer     buffer = nullptr;
+	uint64_t       size   = 0;
+	vk::DeviceSize offset = 0;
+	vk::IndexType  type   = vk::IndexType::eUint16;
 };
 
 static uint64_t VertexBufferDescriptorSize(const ShaderVertexInputBuffer& buffer) {
@@ -719,10 +762,10 @@ static uint64_t VertexBufferDescriptorSize(const ShaderVertexInputBuffer& buffer
 }
 
 struct VertexBufferRange {
-	uint64_t      base_address  = 0;
-	uint64_t      requested_end = 0;
-	uint64_t      acquired_end  = 0;
-	BufferBinding binding;
+	uint64_t                     base_address  = 0;
+	uint64_t                     requested_end = 0;
+	uint64_t                     acquired_end  = 0;
+	std::pair<Buffer*, uint64_t> binding;
 
 	[[nodiscard]] uint64_t RequestedSize() const { return requested_end - base_address; }
 };
@@ -730,11 +773,9 @@ struct VertexBufferRange {
 struct PreparedVertexBuffers {
 	static constexpr uint32_t MaxBuffers = ShaderVertexInputInfo::RES_MAX;
 
-	std::array<vk::Buffer, MaxBuffers>            buffers {};
-	std::array<vk::DeviceSize, MaxBuffers>        offsets {};
-	std::array<std::shared_ptr<void>, MaxBuffers> owners {};
-	uint32_t                                      count       = 0;
-	uint32_t                                      owner_count = 0;
+	std::array<vk::Buffer, MaxBuffers>     buffers {};
+	std::array<vk::DeviceSize, MaxBuffers> offsets {};
+	uint32_t                               count = 0;
 };
 
 static PreparedVertexBuffers AcquireVertexBuffers(RenderCommandBuffer&         buffer,
@@ -784,21 +825,22 @@ static PreparedVertexBuffers AcquireVertexBuffers(RenderCommandBuffer&         b
 		const auto size =
 		    Libs::LibKernel::Memory::ClampRangeSize(range.base_address, range.RequestedSize());
 		range.acquired_end = range.base_address + size;
-		range.binding      = cache.ObtainBuffer(buffer, range.base_address, size);
+		range.binding      = cache.ObtainBuffer(range.base_address, size, false);
+		SetVulkanObjectNameF(
+		    buffer.GetContext().GetGraphics().device, range.binding.first->Handle(),
+		    "Kyty.VertexBufferRange[guest=0x{:016x} size=0x{:x}]", range.base_address, size);
 	}
 
 	// Rebuild slot bindings, offsetting non-empty slots into their acquired merged range.
 	PreparedVertexBuffers prepared;
-	prepared.count = static_cast<uint32_t>(vs_input_info.buffers_num);
-	std::shared_ptr<Buffer> null_owner;
-	vk::Buffer              null_buffer = nullptr;
+	prepared.count         = static_cast<uint32_t>(vs_input_info.buffers_num);
+	vk::Buffer null_buffer = nullptr;
 	for (int i = 0; i < vs_input_info.buffers_num; i++) {
 		const auto& vertex = vs_input_info.buffers[i];
 		const auto  size   = VertexBufferDescriptorSize(vertex);
 		if (size == 0) {
-			if (null_owner == nullptr) {
-				null_owner  = cache.ObtainNullBuffer();
-				null_buffer = null_owner->Handle();
+			if (null_buffer == nullptr) {
+				null_buffer = cache.GetBuffer(NULL_BUFFER_ID).Handle();
 			}
 			prepared.buffers[i] = null_buffer;
 			prepared.offsets[i] = 0;
@@ -815,19 +857,14 @@ static PreparedVertexBuffers AcquireVertexBuffers(RenderCommandBuffer&         b
 			     vertex.addr);
 		}
 
-		prepared.buffers[i] = range->binding.buffer;
-		prepared.offsets[i] = range->binding.offset + vertex.addr - range->base_address;
+		prepared.buffers[i] = range->binding.first->Handle();
+		prepared.offsets[i] = range->binding.second + vertex.addr - range->base_address;
+		SetVulkanObjectNameF(
+		    buffer.GetContext().GetGraphics().device, prepared.buffers[i],
+		    "Kyty.VertexBuffer[slot={} guest=0x{:016x} size=0x{:x} stride={} records={}]", i,
+		    vertex.addr, size, vertex.stride, vertex.num_records);
 	}
 
-	if (null_owner != nullptr) {
-		prepared.owners[prepared.owner_count++] = std::move(null_owner);
-	}
-	for (uint32_t i = 0; i < merged_count; i++) {
-		if (merged_ranges[i].binding.owner != nullptr) {
-			EXIT_IF(prepared.owner_count >= PreparedVertexBuffers::MaxBuffers);
-			prepared.owners[prepared.owner_count++] = std::move(merged_ranges[i].binding.owner);
-		}
-	}
 	return prepared;
 }
 
@@ -844,7 +881,7 @@ static bool GetDrawTopology(const HW::UserConfig& ucfg, bool auto_draw,
 
 	topology = vk::PrimitiveTopology::ePointList;
 
-	switch (static_cast<Prospero::PrimitiveType>(ucfg.GetPrimType())) {
+	switch (ucfg.GetPrimType()) {
 		case Prospero::PrimitiveType::kNone: return false;
 		case Prospero::PrimitiveType::kPointList:
 			topology = vk::PrimitiveTopology::ePointList;
@@ -867,19 +904,54 @@ static bool GetDrawTopology(const HW::UserConfig& ucfg, bool auto_draw,
 			break;
 		case Prospero::PrimitiveType::kRectListLegacy:
 			if (!auto_draw) {
-				EXIT("unknown primitive type: %u\n", ucfg.GetPrimType());
+				EXIT("unknown primitive type: %u\n", static_cast<uint32_t>(ucfg.GetPrimType()));
 			}
 			topology = vk::PrimitiveTopology::eTriangleStrip;
 			break;
 		case Prospero::PrimitiveType::kQuadListLegacy:
 			topology = vk::PrimitiveTopology::eTriangleFan;
 			break;
-		default: EXIT("unknown primitive type: %u\n", ucfg.GetPrimType());
+		default: EXIT("unknown primitive type: %u\n", static_cast<uint32_t>(ucfg.GetPrimType()));
 	}
 
 	return true;
 }
 
+static bool ResolvePrimitiveRestart(const RenderCommandBuffer& buffer,
+                                    vk::PrimitiveTopology topology, uint32_t index_type_and_size) {
+	const auto control = buffer.GetUserConfig().GetPrimitiveResetControl();
+	EXIT_NOT_IMPLEMENTED((control & ~0x3u) != 0);
+	if ((control & 0x1u) == 0) {
+		return false;
+	}
+	switch (buffer.GetUserConfig().GetPrimType()) {
+		case Prospero::PrimitiveType::kLineStrip:
+		case Prospero::PrimitiveType::kTriFan:
+		case Prospero::PrimitiveType::kTriStrip: break;
+		default: return false;
+	}
+	if (topology != vk::PrimitiveTopology::eLineStrip &&
+	    topology != vk::PrimitiveTopology::eTriangleStrip &&
+	    topology != vk::PrimitiveTopology::eTriangleFan) {
+		return false;
+	}
+
+	uint32_t index_mask = 0;
+	switch (static_cast<Prospero::IndexType>(index_type_and_size)) {
+		case Prospero::IndexType::kIndex8: index_mask = 0xffu; break;
+		case Prospero::IndexType::kIndex16: index_mask = 0xffffu; break;
+		case Prospero::IndexType::kIndex32: index_mask = 0xffffffffu; break;
+		default: EXIT("unknown index_type_and_size: %u\n", index_type_and_size);
+	}
+
+	const auto reset_index = buffer.GetRegisters().GetPrimitiveResetIndex();
+	if ((control & 0x2u) != 0 && (reset_index & ~index_mask) != 0) {
+		return false;
+	}
+	EXIT_NOT_IMPLEMENTED((reset_index & index_mask) != index_mask);
+	return true;
+}
+
 bool RenderExecutor::PrepareDrawRenderState(uint64_t submit_id, RenderCommandBuffer& buffer,
                                             const DrawCallInfo& draw,
                                             uint32_t            render_target_slice_offset,
@@ -908,14 +980,14 @@ bool RenderExecutor::PrepareDrawRenderState(uint64_t submit_id, RenderCommandBuf
 	}
 	ResolveRenderDepthTarget(submit_id, buffer, state.depth_info);
 
+	state.ps_active       = DrawHasActivePixelShader(buffer);
 	const bool with_depth = (state.depth_info.format != vk::Format::eUndefined &&
 	                         static_cast<bool>(state.depth_info.image_id));
-	if (state.color_count == 0 && !with_depth) {
+	if (state.color_count == 0 && !with_depth && !state.ps_active) {
 		LogFramebufferSkip(draw.name, state.color_info[0], state.depth_info, buffer,
 		                   draw.index_count, draw.flags);
 		return false;
 	}
-	state.ps_active = DrawHasActivePixelShader(buffer, state, draw);
 
 	return true;
 }
@@ -938,12 +1010,10 @@ static void RefreshShaders(RenderCommandBuffer& buffer, const DrawCallInfo& draw
 	for (uint32_t i = 0; i < state.color_count; i++) {
 		target_export_mapping[state.color_info[i].target_slot] = state.color_info[i].export_mapping;
 	}
-	const auto lane_mask_mode = SelectGraphicsLaneMaskMode(64u);
-
 	if (log_phases) {
 		LogDrawPhase(draw.name, "ShaderCompileInfoVS");
 	}
-	if (!ShaderCompileInfoVS(vertex_shader_info, shader_regs, lane_mask_mode, state.vs_input_info,
+	if (!ShaderCompileInfoVS(vertex_shader_info, shader_regs, state.vs_input_info,
 	                         state.vs_shader)) {
 		EXIT("ShaderCompileInfoVS failed for draw %s\n", draw.name);
 	}
@@ -954,7 +1024,7 @@ static void RefreshShaders(RenderCommandBuffer& buffer, const DrawCallInfo& draw
 	if (log_phases) {
 		LogDrawPhase(draw.name, "ShaderCompileInfoPS");
 	}
-	if (!ShaderCompileInfoPS(pixel_shader_info, shader_regs, lane_mask_mode, state.vs_input_info,
+	if (!ShaderCompileInfoPS(pixel_shader_info, shader_regs, state.vs_input_info,
 	                         target_export_mapping, state.ps_input_info, state.ps_shader)) {
 		EXIT("ShaderCompileInfoPS failed for draw %s\n", draw.name);
 	}
@@ -980,28 +1050,29 @@ static PreparedIndexBuffer PrepareIndexBuffer(RenderCommandBuffer&         buffe
 	prepared.size = source.size;
 	prepared.type = source.type;
 	if (source.host_data != nullptr) {
-		auto binding =
-		    buffer.GetContext().GetBufferCache().UploadTransient(source.host_data, source.size, 16);
-		prepared.owner  = std::move(binding.owner);
-		prepared.buffer = binding.buffer;
-		prepared.offset = binding.offset;
+		auto& stream = buffer.GetContext().GetBufferCache().GetUtilityBuffer(MemoryUsage::Stream);
+		prepared.offset = stream.Copy(source.host_data, source.size, 16);
+		prepared.buffer = stream.Handle();
 	} else {
-		auto binding =
-		    buffer.GetContext().GetBufferCache().ObtainBuffer(buffer, source.address, source.size);
-		prepared.owner  = std::move(binding.owner);
-		prepared.buffer = binding.buffer;
-		prepared.offset = binding.offset;
+		auto [buffer_ptr, offset] =
+		    buffer.GetContext().GetBufferCache().ObtainBuffer(source.address, source.size, false);
+		prepared.buffer = buffer_ptr->Handle();
+		prepared.offset = offset;
+	}
+	if (source.host_data != nullptr) {
+		SetVulkanObjectNameF(buffer.GetContext().GetGraphics().device, prepared.buffer,
+		                     "Kyty.IndexBuffer[guest=transient size=0x{:x} type={}]", source.size,
+		                     static_cast<uint32_t>(source.type));
+	} else {
+		SetVulkanObjectNameF(buffer.GetContext().GetGraphics().device, prepared.buffer,
+		                     "Kyty.IndexBuffer[guest=0x{:016x} size=0x{:x} type={}]",
+		                     source.address, source.size, static_cast<uint32_t>(source.type));
 	}
 	return prepared;
 }
 
-static void CommitVertexBuffers(RenderCommandBuffer& buffer, vk::CommandBuffer vk_buffer,
-                                PreparedVertexBuffers& prepared) {
-	for (uint32_t i = 0; i < prepared.owner_count; i++) {
-		if (prepared.owners[i] != nullptr) {
-			buffer.RetainResourceUntilFence(std::move(prepared.owners[i]));
-		}
-	}
+static void CommitVertexBuffers(vk::CommandBuffer            vk_buffer,
+                                const PreparedVertexBuffers& prepared) {
 	for (uint32_t i = 0; i < prepared.count; i++) {
 		EXIT_IF(prepared.buffers[i] == nullptr);
 	}
@@ -1011,14 +1082,10 @@ static void CommitVertexBuffers(RenderCommandBuffer& buffer, vk::CommandBuffer v
 	}
 }
 
-static void CommitIndexBuffer(RenderCommandBuffer& buffer, vk::CommandBuffer vk_buffer,
-                              PreparedIndexBuffer& prepared) {
+static void CommitIndexBuffer(vk::CommandBuffer vk_buffer, const PreparedIndexBuffer& prepared) {
 	if (prepared.size == 0) {
 		return;
 	}
-	if (prepared.owner != nullptr) {
-		buffer.RetainResourceUntilFence(std::move(prepared.owner));
-	}
 	EXIT_IF(prepared.buffer == nullptr);
 	vk_buffer.bindIndexBuffer(prepared.buffer, prepared.offset, prepared.type);
 }
@@ -1051,7 +1118,7 @@ static void EmitDrawPrimitives(const HW::UserConfig& ucfg, vk::CommandBuffer vk_
                                const DrawEmitInfo& emit) {
 	EXIT_IF(draw.name == nullptr);
 
-	switch (static_cast<Prospero::PrimitiveType>(ucfg.GetPrimType())) {
+	switch (ucfg.GetPrimType()) {
 		case Prospero::PrimitiveType::kPointList:
 		case Prospero::PrimitiveType::kLineList:
 		case Prospero::PrimitiveType::kLineStrip:
@@ -1069,7 +1136,7 @@ static void EmitDrawPrimitives(const HW::UserConfig& ucfg, vk::CommandBuffer vk_
 			break;
 		case Prospero::PrimitiveType::kRectListLegacy:
 			if (emit.indexed) {
-				EXIT("unknown primitive type: %u\n", ucfg.GetPrimType());
+				EXIT("unknown primitive type: %u\n", static_cast<uint32_t>(ucfg.GetPrimType()));
 			}
 			// Sarah
 			EXIT_NOT_IMPLEMENTED(!(draw.index_count == 3 && vs_input_info.buffers_num == 0));
@@ -1087,7 +1154,7 @@ static void EmitDrawPrimitives(const HW::UserConfig& ucfg, vk::CommandBuffer vk_
 				}
 			}
 			break;
-		default: EXIT("unknown primitive type: %u\n", ucfg.GetPrimType());
+		default: EXIT("unknown primitive type: %u\n", static_cast<uint32_t>(ucfg.GetPrimType()));
 	}
 }
 
@@ -1095,14 +1162,14 @@ void RenderExecutor::ExecutePreparedDraw(uint64_t submit_id, RenderCommandBuffer
                                          const DrawCallInfo& draw, DrawRenderState& state,
                                          vk::PrimitiveTopology topology, const DrawEmitInfo& emit,
                                          const DrawIndexBufferSource& index_source,
-                                         bool log_pipeline_phase, bool set_bind_debug,
-                                         bool set_auto_debug) {
+                                         bool primitive_restart_enable, bool log_pipeline_phase,
+                                         bool set_bind_debug, bool set_auto_debug) {
 	EXIT_IF(draw.name == nullptr);
 	auto& ucfg = buffer.GetUserConfig();
 
 	LogDrawPhase(draw.name, "PrepareBindings");
-	auto bindings        = PrepareGraphicsBindings(buffer, state.vs_input_info.stage,
-	                                               state.ps_input_info.stage, state.ps_active);
+	auto bindings = PrepareGraphicsBindings(state.vs_input_info.stage, state.ps_input_info.stage,
+	                                        state.ps_active);
 	auto vertex_bindings = PrepareVertexBuffers(submit_id, buffer, draw, state.vs_input_info);
 	auto index_binding   = PrepareIndexBuffer(buffer, index_source);
 	state.rendering =
@@ -1113,7 +1180,8 @@ void RenderExecutor::ExecutePreparedDraw(uint64_t submit_id, RenderCommandBuffer
 	}
 	auto& pipeline = m_context.GetPipelineCache().CreateGraphicsPipeline(
 	    state.color_info, state.color_count, state.depth_info, state.vs_input_info, buffer,
-	    &state.ps_input_info, topology, state.ps_active, state.vs_shader, state.ps_shader);
+	    &state.ps_input_info, topology, primitive_restart_enable, state.ps_active, state.vs_shader,
+	    state.ps_shader);
 
 	// Resource preparation above may synchronously finish and restart the scheduler. From this
 	// point onward, every operation targets the current command buffer and cannot touch guest
@@ -1125,21 +1193,23 @@ void RenderExecutor::ExecutePreparedDraw(uint64_t submit_id, RenderCommandBuffer
 	if (set_auto_debug) {
 		SetDrawDebugPhase(buffer, submit_id, draw, 0x200u);
 	}
-	CommitVertexBuffers(buffer, vk_buffer, vertex_bindings);
-	CommitBindings(buffer, vk::PipelineBindPoint::eGraphics, pipeline.pipeline_layout,
-	               bindings.vertex);
+	CommitVertexBuffers(vk_buffer, vertex_bindings);
 	if (bindings.pixel.has_value()) {
 		if (set_auto_debug) {
 			SetDrawDebugPhase(buffer, submit_id, draw, 0x300u);
 		}
-		CommitBindings(buffer, vk::PipelineBindPoint::eGraphics, pipeline.pipeline_layout,
-		               *bindings.pixel);
 	}
-	CommitIndexBuffer(buffer, vk_buffer, index_binding);
+	std::array<PreparedBindings*, 2> descriptor_stages {&bindings.vertex, nullptr};
+	const size_t                     descriptor_stage_count = bindings.pixel.has_value() ? 2u : 1u;
+	if (bindings.pixel) {
+		descriptor_stages[1] = &*bindings.pixel;
+	}
+	CommitBindings(buffer, vk::PipelineBindPoint::eGraphics, pipeline,
+	               std::span {descriptor_stages.data(), descriptor_stage_count});
+	CommitIndexBuffer(vk_buffer, index_binding);
 
-	const auto dynamic_params =
-	    BuildGraphicsDynamicParams(buffer, state.color_info, state.color_count, state.depth_info);
-	SetDynamicParams(buffer, vk_buffer, dynamic_params);
+	SetGraphicsDynamicParams(buffer, vk_buffer, state.color_info, state.color_count,
+	                         state.depth_info);
 
 	LogDrawPhase(draw.name, "BeginRendering");
 	if (set_auto_debug) {
@@ -1180,6 +1250,7 @@ void RenderExecutor::DrawIndex(uint64_t submit_id, RenderCommandBuffer& buffer,
 	KYTY_PROFILER_FUNCTION();
 
 	EXIT_IF(buffer.IsInvalid());
+	m_context.GetCommandScheduler().PopPendingOperations();
 	auto& ucfg   = buffer.GetUserConfig();
 	auto& sh_ctx = buffer.GetShaders();
 
@@ -1237,6 +1308,7 @@ void RenderExecutor::DrawIndex(uint64_t submit_id, RenderCommandBuffer& buffer,
 	vk::IndexType index_type           = vk::IndexType::eUint16;
 	uint64_t      index_size           = 0;
 	bool          expand_index8_to_u16 = false;
+	const bool primitive_restart = ResolvePrimitiveRestart(buffer, topology, index_type_and_size);
 
 	switch (static_cast<Prospero::IndexType>(index_type_and_size)) {
 		case Prospero::IndexType::kIndex16:
@@ -1267,7 +1339,7 @@ void RenderExecutor::DrawIndex(uint64_t submit_id, RenderCommandBuffer& buffer,
 		const auto* src = static_cast<const uint8_t*>(index_addr);
 		expanded_indices.resize(index_count);
 		for (uint32_t i = 0; i < index_count; i++) {
-			expanded_indices[i] = src[i];
+			expanded_indices[i] = primitive_restart && src[i] == 0xffu ? 0xffffu : src[i];
 		}
 	}
 
@@ -1297,8 +1369,8 @@ void RenderExecutor::DrawIndex(uint64_t submit_id, RenderCommandBuffer& buffer,
 	emit.indexed       = true;
 	emit.vertex_offset = vertex_offset;
 
-	ExecutePreparedDraw(submit_id, buffer, draw, state, topology, emit, index_source, true, true,
-	                    false);
+	ExecutePreparedDraw(submit_id, buffer, draw, state, topology, emit, index_source,
+	                    primitive_restart, true, true, false);
 	ResetBindings();
 }
 
@@ -1310,6 +1382,7 @@ void RenderExecutor::DrawAuto(uint64_t submit_id, RenderCommandBuffer& buffer, u
 	KYTY_PROFILER_FUNCTION();
 
 	EXIT_IF(buffer.IsInvalid());
+	m_context.GetCommandScheduler().PopPendingOperations();
 	auto& ucfg   = buffer.GetUserConfig();
 	auto& sh_ctx = buffer.GetShaders();
 
@@ -1387,9 +1460,8 @@ void RenderExecutor::DrawAuto(uint64_t submit_id, RenderCommandBuffer& buffer, u
 	}
 
 	LogDrawStateIfNeeded(buffer, draw, state, false,
-	                     ucfg.GetPrimType() ==
-	                         Prospero::GpuEnumValue(Prospero::PrimitiveType::kRectListLegacy),
-	                     0, nullptr);
+	                     ucfg.GetPrimType() == Prospero::PrimitiveType::kRectListLegacy, 0,
+	                     nullptr);
 
 	const auto   vertex_offset = ResolveVertexOffset(ucfg.GetIndexOffset(), state.vs_input_info) +
 	                             static_cast<int32_t>(first_vertex);
@@ -1398,7 +1470,7 @@ void RenderExecutor::DrawAuto(uint64_t submit_id, RenderCommandBuffer& buffer, u
 
 	DrawIndexBufferSource index_source {};
 	ExecutePreparedDraw(submit_id, buffer, draw, state, topology, emit, index_source, false, false,
-	                    true);
+	                    false, true);
 	ResetBindings();
 }
 

src/graphics/host_gpu/renderer/renderDraw.h

@@ -1,43 +1,12 @@
 #ifndef EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_RENDERDRAW_H_
 #define EMULATOR_SRC_GRAPHICS_HOST_GPU_RENDERER_RENDERDRAW_H_
 
-#include "graphics/host_gpu/renderer/renderTarget.h"
-
 #include <cstdint>
-#include <type_traits>
 
 namespace Libs::Graphics {
 
-struct RenderColorInfo;
 struct ShaderVertexInputInfo;
 
-#pragma pack(push, 1)
-
-struct PipelineDynamicParameters {
-	bool stencil_test_enable = false;
-	bool depth_bias_enable   = false;
-
-	float viewport_scale[3]  = {};
-	float viewport_offset[3] = {};
-	int   scissor_ltrb[4]    = {0};
-
-	float    line_width                                       = 1.0f;
-	float    depth_bias_constant_factor                       = 0.0f;
-	float    depth_bias_clamp                                 = 0.0f;
-	float    depth_bias_slope_factor                          = 0.0f;
-	uint32_t color_write_count                                = 1;
-	bool     color_write_enable[RENDER_COLOR_ATTACHMENTS_MAX] = {true, true};
-
-	PipelineStencilDynamicState stencil_front;
-	PipelineStencilDynamicState stencil_back;
-};
-
-#pragma pack(pop)
-
-static_assert(std::is_trivially_copyable_v<PipelineDynamicParameters>);
-static_assert(std::is_standard_layout_v<PipelineDynamicParameters>);
-static_assert(alignof(PipelineDynamicParameters) == 1);
-
 [[nodiscard]] int32_t ResolveVertexOffset(uint32_t                     index_offset,
                                           const ShaderVertexInputInfo& vs_input_info);
 

src/graphics/host_gpu/renderer/sync.cpp

@@ -390,7 +390,7 @@ int DeleteEqEvent(LibKernel::EventQueue::KernelEqueue eq, int id) {
 	return result;
 }
 
-void ReadGds(Buffer& gds, uint32_t* dst, uint32_t dw_offset, uint32_t dw_size) {
+void ReadGds(const Buffer& gds, uint32_t* dst, uint32_t dw_offset, uint32_t dw_size) {
 	const auto offset = uint64_t {dw_offset} * sizeof(uint32_t);
 	const auto size   = uint64_t {dw_size} * sizeof(uint32_t);
 	EXIT_IF(dst == nullptr || offset > gds.Size() || size > gds.Size() - offset ||

src/graphics/host_gpu/renderer/sync.h

@@ -62,7 +62,7 @@ void WriteAtEndOfPipeWithInterruptWriteBackFlip32(uint64_t submit_id, CommandBuf
 int  AddEqEvent(RenderContext& renderer, LibKernel::EventQueue::KernelEqueue eq, int id,
                 void* udata);
 int  DeleteEqEvent(LibKernel::EventQueue::KernelEqueue eq, int id);
-void ReadGds(Buffer& gds, uint32_t* dst, uint32_t dw_offset, uint32_t dw_size);
+void ReadGds(const Buffer& gds, uint32_t* dst, uint32_t dw_offset, uint32_t dw_size);
 
 } // namespace Sync
 } // namespace Libs::Graphics

src/graphics/host_gpu/vma.cpp

@@ -199,20 +199,6 @@ void GraphicContext::CreateBuffer(uint64_t size, VulkanBuffer& buffer) {
 	VulkanTrackAllocation(buffer.memory);
 }
 
-void GraphicContext::DeleteBuffer(VulkanBuffer& buffer) {
-	KYTY_PROFILER_FUNCTION();
-	EXIT_IF(allocator == nullptr || buffer.buffer == nullptr ||
-	        buffer.memory.allocation == nullptr);
-
-	VulkanUntrackAllocation(buffer.memory);
-	vmaDestroyBuffer(allocator, buffer.buffer, buffer.memory.allocation);
-	buffer.buffer                 = nullptr;
-	buffer.memory.memory          = nullptr;
-	buffer.memory.allocation      = nullptr;
-	buffer.memory.allocation_info = {};
-	buffer.memory.offset          = 0;
-}
-
 bool GraphicContext::CreateImage(const vk::ImageCreateInfo& image_info, VulkanImage& image) {
 	KYTY_PROFILER_FUNCTION();
 	EXIT_IF(allocator == nullptr || image.image != nullptr || image.memory.allocation != nullptr);

src/graphics/host_gpu/vulkanCommon.cpp

@@ -83,11 +83,11 @@ constexpr FormatMapping kFormatMappings[] = {
 };
 
 constexpr auto MakeFormatLookup() {
-	constexpr auto kMaxFormat = Prospero::GpuEnumValue(Prospero::BufferFormat::kBc7Srgb);
+	constexpr auto kMaxFormat = static_cast<size_t>(Prospero::BufferFormat::kBc7Srgb);
 	std::array<vk::Format, kMaxFormat + 1> lookup {};
 	lookup.fill(vk::Format::eUndefined);
 	for (const auto& mapping: kFormatMappings) {
-		lookup[Prospero::GpuEnumValue(mapping.guest)] = mapping.host;
+		lookup[static_cast<size_t>(mapping.guest)] = mapping.host;
 	}
 	return lookup;
 }
@@ -112,9 +112,9 @@ std::string VulkanToString(vk::QueueFlags value) {
 	return vk::to_string(value);
 }
 
-vk::Format VulkanFormat(uint32_t guest_format) {
-	return guest_format < kFormatLookup.size() ? kFormatLookup[guest_format]
-	                                           : vk::Format::eUndefined;
+vk::Format VulkanFormat(Prospero::BufferFormat guest_format) {
+	const auto index = static_cast<size_t>(guest_format);
+	return index < kFormatLookup.size() ? kFormatLookup[index] : vk::Format::eUndefined;
 }
 
 void RequireVulkanSuccess(vk::Result result, const char* operation) {

src/graphics/host_gpu/vulkanCommon.h

@@ -10,8 +10,13 @@
 #define VMA_STATIC_VULKAN_FUNCTIONS  0
 #define VMA_DYNAMIC_VULKAN_FUNCTIONS 1
 
+#include "common/emulatorConfig.h"
+#include "graphics/guest_gpu/gpu_defs.h"
+
 #include <cstdint>
+#include <fmt/format.h>
 #include <string>
+#include <utility>
 #include <vector>
 #include <vulkan/vulkan.hpp>
 #include <vulkan/vulkan_hash.hpp>
@@ -20,18 +25,31 @@ namespace Libs::Graphics {
 
 using VulkanMemoryBarrier = vk::MemoryBarrier;
 
-template <typename Handle>
-[[nodiscard]] void* VulkanHandleToPointer(Handle handle) {
-	return reinterpret_cast<void*>(static_cast<typename Handle::CType>(handle));
-}
-
 std::string VulkanToString(vk::Result value);
 std::string VulkanToString(vk::Format value);
 std::string VulkanToString(vk::ImageLayout value);
 std::string VulkanToString(vk::QueueFlags value);
-vk::Format  VulkanFormat(uint32_t guest_format);
+vk::Format  VulkanFormat(Prospero::BufferFormat guest_format);
 void        RequireVulkanSuccess(vk::Result result, const char* operation);
 
+template <typename Handle, typename... Args>
+void SetVulkanObjectNameF(vk::Device device, Handle handle, fmt::format_string<Args...> format,
+                          Args&&... args) {
+	if (!Config::GraphicsDebugDumpEnabled() || device == nullptr || handle == nullptr ||
+	    VULKAN_HPP_DEFAULT_DISPATCHER.vkSetDebugUtilsObjectNameEXT == nullptr) {
+		return;
+	}
+
+	const auto                      name = fmt::format(format, std::forward<Args>(args)...);
+	vk::DebugUtilsObjectNameInfoEXT info {};
+	info.sType        = vk::StructureType::eDebugUtilsObjectNameInfoEXT;
+	info.objectType   = Handle::objectType;
+	info.objectHandle = static_cast<uint64_t>(
+	    reinterpret_cast<uintptr_t>(static_cast<typename Handle::CType>(handle)));
+	info.pObjectName = name.c_str();
+	(void)device.setDebugUtilsObjectNameEXT(&info);
+}
+
 template <typename T, typename Enumerator>
 [[nodiscard]] std::vector<T> EnumerateVulkan(const char* operation, Enumerator&& enumerate) {
 	for (;;) {

src/graphics/host_gpu/vulkanInstance.h

@@ -25,11 +25,13 @@ struct VulkanInstance {
 	VmaAllocator                       allocator                         = nullptr;
 	bool                               memory_budget_ext_enabled         = false;
 	bool                               rt_extensions_enabled             = false;
-	bool                               subgroup_size_control_enabled     = false;
+	bool                               compute_subgroup_size_control_enabled = false;
+	bool                               compute_wave64_supported          = false;
 	bool                               sample_rate_shading_enabled       = false;
 	uint32_t                           subgroup_size                     = 0;
 	uint32_t                           min_subgroup_size                 = 0;
 	uint32_t                           max_subgroup_size                 = 0;
+	uint32_t                           max_push_descriptors              = 0;
 	vk::ShaderStageFlags               required_subgroup_size_stages     = {};
 	Common::Mutex                      queue_mutex;
 	uint32_t                           queue_family = static_cast<uint32_t>(-1);

src/graphics/presentation/imeOverlay.cpp

@@ -32,64 +32,21 @@ namespace DialogIme = Libs::Dialog::ImeDialog;
 
 namespace Ime {
 
-enum class Type : uint32_t { Default = 0, BasicLatin = 1, Url = 2, Mail = 3, Number = 4 };
-enum class EnterLabel : uint32_t { Default = 0, Send = 1, Search = 2, Go = 3 };
-enum class Alignment : uint32_t { Start = 0, Center = 1, End = 2 };
-enum class ExternalAction : uint8_t {
-	None,
-	Text,
-	Backspace,
-	MoveLeft,
-	MoveRight,
-	Cancel,
-	Accept,
-	Newline,
-};
+using Type           = ImeCommon::Type;
+using EnterLabel     = ImeCommon::EnterLabel;
+using Alignment      = ImeCommon::Alignment;
+using ExternalAction = ImeCommon::ExternalAction;
+using ExternalInput  = ImeCommon::ExternalInput;
+using HostSnapshot   = ImeCommon::HostSnapshot;
 
-struct Keycode {
-	uint16_t keycode     = 0;
-	char16_t character   = u'\0';
-	uint32_t status      = 0;
-	uint32_t type        = 0;
-	int32_t  user_id     = 0;
-	uint32_t resource_id = 0;
-	uint64_t timestamp   = 0;
-};
-
-struct ExternalInput {
-	Keycode        key;
-	ExternalAction action = ExternalAction::None;
-	std::u16string text;
-};
-
-struct HostSnapshot {
-	uint64_t       generation           = 0;
-	Type           type                 = Type::Default;
-	EnterLabel     enter_label          = EnterLabel::Default;
-	uint32_t       option               = 0;
-	uint32_t       max_text_length      = 0;
-	uint32_t       cursor               = 0;
-	uint32_t       disable_device       = 0;
-	bool           key_panel_visible    = true;
-	float          posx                 = 0.0f;
-	float          posy                 = 0.0f;
-	Alignment      horizontal_alignment = Alignment::Start;
-	Alignment      vertical_alignment   = Alignment::Start;
-	uint32_t       panel_width          = 0;
-	uint32_t       panel_height         = 0;
-	std::u16string text;
-	std::u16string title;
-	std::u16string placeholder;
-};
-
-constexpr uint32_t OPTION_MULTILINE            = 0x00000001;
-constexpr uint32_t OPTION_NO_AUTO_CAPITALIZE   = 0x00000002;
-constexpr uint32_t OPTION_PASSWORD             = 0x00000004;
-constexpr uint32_t OPTION_FIXED_POSITION       = 0x00000040;
-constexpr uint32_t OPTION_DISABLE_POSITION_ADJ = 0x00000800;
-constexpr uint32_t OPTION_USE_OVER_2K          = 0x00004000;
-constexpr uint32_t DISABLE_DEVICE_CONTROLLER   = 0x00000001;
-constexpr uint32_t DISABLE_DEVICE_EXT_KEYBOARD = 0x00000002;
+constexpr uint32_t OPTION_MULTILINE            = ImeCommon::OPTION_MULTILINE;
+constexpr uint32_t OPTION_NO_AUTO_CAPITALIZE   = ImeCommon::OPTION_NO_AUTO_CAPITALIZE;
+constexpr uint32_t OPTION_PASSWORD             = ImeCommon::OPTION_PASSWORD;
+constexpr uint32_t OPTION_FIXED_POSITION       = ImeCommon::OPTION_FIXED_POSITION;
+constexpr uint32_t OPTION_DISABLE_POSITION_ADJ = ImeCommon::OPTION_DISABLE_POSITION_ADJUST;
+constexpr uint32_t OPTION_USE_OVER_2K          = ImeCommon::OPTION_USE_OVER_2K;
+constexpr uint32_t DISABLE_DEVICE_CONTROLLER   = ImeCommon::DISABLE_DEVICE_CONTROLLER;
+constexpr uint32_t DISABLE_DEVICE_EXT_KEYBOARD = ImeCommon::DISABLE_DEVICE_EXT_KEYBOARD;
 constexpr uint64_t CORE_GENERATION_BIT         = uint64_t {1} << 63;
 
 uint64_t PackCoreGeneration(uint64_t generation) {
@@ -108,49 +65,11 @@ bool GetHostSnapshot(HostSnapshot* snapshot) {
 	if (snapshot == nullptr) {
 		return false;
 	}
-	CoreIme::HostSnapshot core;
-	if (CoreIme::GetHostSnapshot(&core)) {
-		snapshot->generation           = PackCoreGeneration(core.generation);
-		snapshot->type                 = static_cast<Type>(core.type);
-		snapshot->enter_label          = static_cast<EnterLabel>(core.enter_label);
-		snapshot->option               = core.option;
-		snapshot->max_text_length      = core.max_text_length;
-		snapshot->cursor               = core.cursor;
-		snapshot->disable_device       = core.disable_device;
-		snapshot->key_panel_visible    = core.key_panel_visible;
-		snapshot->posx                 = core.posx;
-		snapshot->posy                 = core.posy;
-		snapshot->horizontal_alignment = static_cast<Alignment>(core.horizontal_alignment);
-		snapshot->vertical_alignment   = static_cast<Alignment>(core.vertical_alignment);
-		snapshot->panel_width          = core.panel_width;
-		snapshot->panel_height         = core.panel_height;
-		snapshot->text                 = std::move(core.text);
-		snapshot->title.clear();
-		snapshot->placeholder.clear();
+	if (CoreIme::GetHostSnapshot(snapshot)) {
+		snapshot->generation = PackCoreGeneration(snapshot->generation);
 		return true;
 	}
-	DialogIme::HostSnapshot dialog;
-	if (!DialogIme::GetHostSnapshot(&dialog)) {
-		return false;
-	}
-	snapshot->generation           = dialog.generation;
-	snapshot->type                 = static_cast<Type>(dialog.type);
-	snapshot->enter_label          = static_cast<EnterLabel>(dialog.enter_label);
-	snapshot->option               = dialog.option;
-	snapshot->max_text_length      = dialog.max_text_length;
-	snapshot->cursor               = dialog.cursor;
-	snapshot->disable_device       = dialog.disable_device;
-	snapshot->key_panel_visible    = dialog.key_panel_visible;
-	snapshot->posx                 = dialog.posx;
-	snapshot->posy                 = dialog.posy;
-	snapshot->horizontal_alignment = static_cast<Alignment>(dialog.horizontal_alignment);
-	snapshot->vertical_alignment   = static_cast<Alignment>(dialog.vertical_alignment);
-	snapshot->panel_width          = dialog.panel_width;
-	snapshot->panel_height         = dialog.panel_height;
-	snapshot->text                 = std::move(dialog.text);
-	snapshot->title                = std::move(dialog.title);
-	snapshot->placeholder          = std::move(dialog.placeholder);
-	return true;
+	return DialogIme::GetHostSnapshot(snapshot);
 }
 
 bool HostInsertText(uint64_t generation, std::u16string_view text) {
@@ -175,30 +94,9 @@ bool HostCancel(uint64_t generation) {
 }
 
 bool HostQueueExternalInput(uint64_t generation, ExternalInput input) {
-	if (!IsCoreGeneration(generation)) {
-		DialogIme::ExternalInput dialog;
-		dialog.key.keycode     = input.key.keycode;
-		dialog.key.character   = input.key.character;
-		dialog.key.status      = input.key.status;
-		dialog.key.type        = input.key.type;
-		dialog.key.user_id     = input.key.user_id;
-		dialog.key.resource_id = input.key.resource_id;
-		dialog.key.timestamp   = input.key.timestamp;
-		dialog.action          = static_cast<DialogIme::ExternalAction>(input.action);
-		dialog.text            = std::move(input.text);
-		return DialogIme::HostQueueExternalInput(generation, std::move(dialog));
-	}
-	CoreIme::ExternalInput core;
-	core.key.keycode     = input.key.keycode;
-	core.key.character   = input.key.character;
-	core.key.status      = input.key.status;
-	core.key.type        = input.key.type;
-	core.key.user_id     = input.key.user_id;
-	core.key.resource_id = input.key.resource_id;
-	core.key.timestamp   = input.key.timestamp;
-	core.action          = static_cast<CoreIme::ExternalAction>(input.action);
-	core.text            = std::move(input.text);
-	return CoreIme::HostQueueExternalInput(UnpackGeneration(generation), std::move(core));
+	return IsCoreGeneration(generation)
+	           ? CoreIme::HostQueueExternalInput(UnpackGeneration(generation), std::move(input))
+	           : DialogIme::HostQueueExternalInput(generation, std::move(input));
 }
 
 } // namespace Ime

src/graphics/presentation/renderDoc.cpp

@@ -1,15 +1,12 @@
 #include "graphics/presentation/renderDoc.h"
 
-#include "SDL_syswm.h"
-#include "SDL_version.h"
-#include "SDL_video.h"
 #include "common/logging/log.h"
 
 #include <array>
 #include <atomic>
-#include <cstdint>
-#include <cstring>
+#include <chrono>
 #include <filesystem>
+#include <renderdoc_app.h>
 #include <string>
 
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
@@ -21,361 +18,175 @@
 #undef max
 #else
 #include <dlfcn.h>
-
-// RenderDoc uses Windows-style names in its cross-platform API.
-#define __cdecl
-using HMODULE = void*;
 #endif
 
 namespace Libs::Graphics {
 
-using RenderDocDevicePointer = void*;
-using RenderDocWindowHandle  = void*;
-
-enum RenderDocVersion {
-	eRENDERDOC_API_Version_1_4_2 = 10402,
-};
-
-enum RenderDocInputButton {
-	eRENDERDOC_Key_NonPrintable = 0x100,
-	eRENDERDOC_Key_Divide,
-	eRENDERDOC_Key_Multiply,
-	eRENDERDOC_Key_Subtract,
-	eRENDERDOC_Key_Plus,
-	eRENDERDOC_Key_F1,
-};
-
-using pRENDERDOC_SetCaptureKeys             = void(__cdecl*)(RenderDocInputButton* keys, int num);
-using pRENDERDOC_SetCaptureFilePathTemplate = void(__cdecl*)(const char* pathtemplate);
-using pRENDERDOC_GetCaptureFilePathTemplate = const char*(__cdecl*)();
-using pRENDERDOC_GetNumCaptures             = uint32_t(__cdecl*)();
-using pRENDERDOC_GetCapture = uint32_t(__cdecl*)(uint32_t idx, char* filename, uint32_t* pathlength,
-                                                 uint64_t* timestamp);
-using pRENDERDOC_UnloadCrashHandler = void(__cdecl*)();
-using pRENDERDOC_SetActiveWindow    = void(__cdecl*)(RenderDocDevicePointer device,
-                                                     RenderDocWindowHandle  wndHandle);
-using pRENDERDOC_StartFrameCapture  = void(__cdecl*)(RenderDocDevicePointer device,
-                                                     RenderDocWindowHandle  wndHandle);
-using pRENDERDOC_IsFrameCapturing   = uint32_t(__cdecl*)();
-using pRENDERDOC_EndFrameCapture    = uint32_t(__cdecl*)(RenderDocDevicePointer device,
-                                                         RenderDocWindowHandle  wndHandle);
-using pRENDERDOC_GetAPI = int(__cdecl*)(RenderDocVersion version, void** out_api_pointers);
-
-struct RenderDocApi {
-	void*                                 GetAPIVersion;
-	void*                                 SetCaptureOptionU32;
-	void*                                 SetCaptureOptionF32;
-	void*                                 GetCaptureOptionU32;
-	void*                                 GetCaptureOptionF32;
-	void*                                 SetFocusToggleKeys;
-	pRENDERDOC_SetCaptureKeys             SetCaptureKeys;
-	void*                                 GetOverlayBits;
-	void*                                 MaskOverlayBits;
-	void*                                 RemoveHooks;
-	pRENDERDOC_UnloadCrashHandler         UnloadCrashHandler;
-	pRENDERDOC_SetCaptureFilePathTemplate SetCaptureFilePathTemplate;
-	pRENDERDOC_GetCaptureFilePathTemplate GetCaptureFilePathTemplate;
-	pRENDERDOC_GetNumCaptures             GetNumCaptures;
-	pRENDERDOC_GetCapture                 GetCapture;
-	void*                                 TriggerCapture;
-	void*                                 IsTargetControlConnected;
-	void*                                 LaunchReplayUI;
-	pRENDERDOC_SetActiveWindow            SetActiveWindow;
-	pRENDERDOC_StartFrameCapture          StartFrameCapture;
-	pRENDERDOC_IsFrameCapturing           IsFrameCapturing;
-	pRENDERDOC_EndFrameCapture            EndFrameCapture;
-	void*                                 TriggerMultiFrameCapture;
-	void*                                 SetCaptureFileComments;
-	void*                                 DiscardFrameCapture;
-};
-
 enum class RenderDocState : uint32_t {
 	Idle,
 	Requested,
+	Starting,
 	Capturing,
 };
 
-static RenderDocApi*               g_api             = nullptr;
-static HMODULE                     g_module          = nullptr;
-static RenderDocDevicePointer      g_device          = nullptr;
-static RenderDocWindowHandle       g_window          = nullptr;
+static RENDERDOC_API_1_6_0*        g_api             = nullptr;
 static std::atomic<RenderDocState> g_state           = RenderDocState::Idle;
-static std::atomic_bool            g_init_done       = false;
+static std::atomic_uint32_t        g_captured_flips  = 0;
 static std::atomic_bool            g_unavailable_log = false;
 
-static RenderDocDevicePointer GetRenderDocDevicePointer(vk::Instance instance) {
-	if (instance == nullptr) {
-		return nullptr;
-	}
-
-	return VulkanHandleToPointer(instance);
-}
-
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
 
 static bool BindRenderDocApi(HMODULE module) {
 	auto* get_api = reinterpret_cast<pRENDERDOC_GetAPI>(GetProcAddress(module, "RENDERDOC_GetAPI"));
-	if (get_api == nullptr) {
-		return false;
-	}
-
-	void* api = nullptr;
-	if (get_api(eRENDERDOC_API_Version_1_4_2, &api) == 0 || api == nullptr) {
-		return false;
-	}
-
-	g_module = module;
-	g_api    = static_cast<RenderDocApi*>(api);
-
-	g_api->SetCaptureFilePathTemplate("_RenderDoc/kyty");
-	g_api->SetCaptureKeys(nullptr, 0);
-	g_api->UnloadCrashHandler();
-
-	char module_path[MAX_PATH] = {};
-	GetModuleFileNameA(module, module_path, sizeof(module_path));
-	LOGF("RenderDoc: bound API from %s\n", module_path);
-	return true;
-}
-
-static RenderDocWindowHandle GetRenderDocWindowHandle(SDL_Window* window) {
-	if (window == nullptr) {
-		return nullptr;
-	}
-
-	SDL_SysWMinfo info {};
-	SDL_VERSION(&info.version);
-
-	if (SDL_GetWindowWMInfo(window, &info) != SDL_TRUE || info.subsystem != SDL_SYSWM_WINDOWS) {
-		return nullptr;
-	}
-
-	return info.info.win.window;
-}
-
 #else
-
-static bool BindRenderDocApi(HMODULE module) {
+static bool BindRenderDocApi(void* module) {
 	auto* get_api = reinterpret_cast<pRENDERDOC_GetAPI>(::dlsym(module, "RENDERDOC_GetAPI"));
+#endif
 	if (get_api == nullptr) {
 		return false;
 	}
 
 	void* api = nullptr;
-	if (get_api(eRENDERDOC_API_Version_1_4_2, &api) == 0 || api == nullptr) {
+	if (get_api(eRENDERDOC_API_Version_1_6_0, &api) != 1 || api == nullptr) {
 		return false;
 	}
 
-	g_module = module;
-	g_api    = static_cast<RenderDocApi*>(api);
-
-	g_api->SetCaptureFilePathTemplate("_RenderDoc/kyty");
+	g_api = static_cast<RENDERDOC_API_1_6_0*>(api);
 	g_api->SetCaptureKeys(nullptr, 0);
 	g_api->UnloadCrashHandler();
-
-	Dl_info info {};
-	if (::dladdr(reinterpret_cast<void*>(get_api), &info) != 0 && info.dli_fname != nullptr) {
-		LOGF("RenderDoc: bound API from %s\n", info.dli_fname);
-	} else {
-		LOGF("RenderDoc: bound API\n");
-	}
+	LOGF("RenderDoc: API 1.6.0 bound\n");
 	return true;
 }
 
-static RenderDocWindowHandle GetRenderDocWindowHandle(SDL_Window* window) {
-	if (window == nullptr) {
-		return nullptr;
-	}
-
-	SDL_SysWMinfo info {};
-	SDL_VERSION(&info.version);
-
-	if (SDL_GetWindowWMInfo(window, &info) != SDL_TRUE) {
-		return nullptr;
-	}
-
-#if defined(SDL_VIDEO_DRIVER_X11)
-	if (info.subsystem == SDL_SYSWM_X11) {
-		// RenderDoc takes the raw xlib Window id in the pointer slot, not a Display*.
-		return reinterpret_cast<RenderDocWindowHandle>(
-		    static_cast<uintptr_t>(info.info.x11.window));
-	}
-#endif
-
-	// Wayland capture works without an active-window handle.
-	static std::atomic_bool logged = false;
-	if (!logged.exchange(true)) {
-		LOGF("RenderDoc: no native window handle for SDL subsystem %d (Wayland?); the in-app "
-		     "overlay is unavailable, but --rd captures still work\n",
-		     static_cast<int>(info.subsystem));
-	}
-	return nullptr;
-}
-
-#endif
-
-#if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
-
-static bool IsAvailable() {
-	return g_api != nullptr && g_device != nullptr && g_window != nullptr;
-}
-
-#else
-
-static bool IsAvailable() {
-	return g_api != nullptr && g_device != nullptr;
-}
-
-#endif
-
 #if KYTY_PLATFORM == KYTY_PLATFORM_WINDOWS
 
 void RenderDocInit() {
-	bool expected = false;
-	if (!g_init_done.compare_exchange_strong(expected, true)) {
+	if (g_api != nullptr) {
 		return;
 	}
 
-	HKEY h_reg_key;
-	LONG result = RegOpenKeyExW(HKEY_LOCAL_MACHINE,
-	                            L"SOFTWARE\\Classes\\RenderDoc.RDCCapture.1\\DefaultIcon\\", 0,
-	                            KEY_READ, &h_reg_key);
-	if (result != ERROR_SUCCESS) {
-		return;
-	}
-	std::array<wchar_t, MAX_PATH> key_str {};
-	DWORD                         str_sz_out {key_str.size()};
-	result = RegQueryValueExW(h_reg_key, L"", 0, NULL, (LPBYTE)key_str.data(), &str_sz_out);
-	RegCloseKey(h_reg_key);
-	if (result != ERROR_SUCCESS) {
-		return;
-	}
-
-	std::filesystem::path path {key_str.cbegin(), key_str.cend()};
-	path                   = path.parent_path().append("renderdoc.dll");
-	const auto path_to_lib = path.generic_string();
-	auto*      module      = LoadLibraryA(path_to_lib.c_str());
+	auto* module = GetModuleHandleA("renderdoc.dll");
 	if (module == nullptr) {
-		return;
+		HKEY key = nullptr;
+		if (RegOpenKeyExW(HKEY_LOCAL_MACHINE,
+		                  L"SOFTWARE\\Classes\\RenderDoc.RDCCapture.1\\DefaultIcon\\", 0, KEY_READ,
+		                  &key) != ERROR_SUCCESS) {
+			return;
+		}
+
+		std::array<wchar_t, MAX_PATH> path_buffer {};
+		DWORD      path_size = static_cast<DWORD>(path_buffer.size() * sizeof(wchar_t));
+		const auto result    = RegQueryValueExW(
+		    key, L"", nullptr, nullptr, reinterpret_cast<LPBYTE>(path_buffer.data()), &path_size);
+		RegCloseKey(key);
+		if (result != ERROR_SUCCESS) {
+			return;
+		}
+
+		auto path = std::filesystem::path(path_buffer.data()).parent_path() / "renderdoc.dll";
+		module    = LoadLibraryW(path.c_str());
+		if (module == nullptr) {
+			return;
+		}
 	}
 
 	if (!BindRenderDocApi(module)) {
-		LOGF("RenderDoc: API 1.4.2 is not available; in-app capture disabled\n");
-		FreeLibrary(module);
-		return;
+		LOGF("RenderDoc: API 1.6.0 is unavailable\n");
 	}
 }
 
 #else
 
 void RenderDocInit() {
-	bool expected = false;
-	if (!g_init_done.compare_exchange_strong(expected, true)) {
+	if (g_api != nullptr) {
 		return;
 	}
 
-	// Prefer an injected RenderDoc instance.
 	auto* module = ::dlopen("librenderdoc.so", RTLD_NOW | RTLD_NOLOAD);
 	if (module == nullptr) {
 		module = ::dlopen("librenderdoc.so", RTLD_NOW);
 	}
 	if (module == nullptr) {
-		LOGF("RenderDoc: librenderdoc.so was not found; in-app capture disabled\n");
 		return;
 	}
 
 	if (!BindRenderDocApi(module)) {
-		LOGF("RenderDoc: API 1.4.2 is not available; in-app capture disabled\n");
+		LOGF("RenderDoc: API 1.6.0 is unavailable\n");
 		::dlclose(module);
-		return;
 	}
 }
 
 #endif
 
-void RenderDocSetActiveWindow(vk::Instance instance, SDL_Window* window) {
-	if (g_api == nullptr) {
-		return;
-	}
-
-	g_device = GetRenderDocDevicePointer(instance);
-	g_window = GetRenderDocWindowHandle(window);
-
-	if (g_device == nullptr || g_window == nullptr) {
-		LOGF("RenderDoc: active Vulkan window was not registered\n");
-		return;
-	}
-
-	g_api->SetActiveWindow(g_device, g_window);
-	LOGF("RenderDoc: active Vulkan window registered\n");
-}
-
 void RenderDocRequestCapture() {
-	if (!IsAvailable()) {
+	if (g_api == nullptr) {
 		if (!g_unavailable_log.exchange(true)) {
-			LOGF("RenderDoc: capture requested, but RenderDoc is not available\n");
+			LOGF("RenderDoc: capture requested, but RenderDoc is unavailable\n");
 		}
 		return;
 	}
 
 	RenderDocState expected = RenderDocState::Idle;
 	if (g_state.compare_exchange_strong(expected, RenderDocState::Requested)) {
-		LOGF("RenderDoc: capture requested; next complete presented frame will be captured\n");
-	} else {
-		LOGF("RenderDoc: capture request ignored because a capture is already pending\n");
+		LOGF("RenderDoc: capture requested\n");
 	}
 }
 
-static void LogNewestCapture() {
-	const auto count = g_api->GetNumCaptures();
-	if (count == 0) {
-		return;
-	}
-
-	char     filename[4096] = {};
-	uint32_t path_length    = sizeof(filename);
-	uint64_t timestamp      = 0;
-
-	if (g_api->GetCapture(count - 1, filename, &path_length, &timestamp) != 0) {
-		filename[sizeof(filename) - 1] = '\0';
-		LOGF("RenderDoc: wrote capture %s\n", filename);
-	}
+bool RenderDocCaptureRequested() {
+	return g_state.load(std::memory_order_acquire) == RenderDocState::Requested;
 }
 
-void RenderDocOnPresent() {
-	if (!IsAvailable()) {
+bool RenderDocCaptureInProgress() {
+	return g_state.load(std::memory_order_acquire) == RenderDocState::Capturing;
+}
+
+void RenderDocStartCapture() {
+	RenderDocState expected = RenderDocState::Requested;
+	if (g_api == nullptr || !g_state.compare_exchange_strong(expected, RenderDocState::Starting,
+	                                                         std::memory_order_acq_rel)) {
 		return;
 	}
 
-	switch (g_state.load()) {
-		case RenderDocState::Idle: return;
-		case RenderDocState::Requested:
-			if (g_api->IsFrameCapturing() != 0) {
-				LOGF("RenderDoc: capture request ignored because RenderDoc is already capturing\n");
-				g_state.store(RenderDocState::Idle);
-				return;
-			}
+	if (g_api->IsFrameCapturing() != 0) {
+		g_state.store(RenderDocState::Idle, std::memory_order_release);
+		LOGF("RenderDoc: capture request ignored because a capture is already active\n");
+		return;
+	}
 
-			g_api->StartFrameCapture(nullptr, nullptr);
-			if (g_api->IsFrameCapturing() == 0) {
-				LOGF("RenderDoc: StartFrameCapture returned, but RenderDoc is not capturing\n");
-				g_state.store(RenderDocState::Idle);
-				return;
-			}
-			g_state.store(RenderDocState::Capturing);
-			LOGF("RenderDoc: capture started\n");
-			return;
-		case RenderDocState::Capturing: break;
+	const auto capture_id   = std::chrono::duration_cast<std::chrono::microseconds>(
+	                              std::chrono::system_clock::now().time_since_epoch())
+	                              .count();
+	const auto capture_path = "_RenderDoc/kyty_" + std::to_string(capture_id);
+	g_api->SetCaptureFilePathTemplate(capture_path.c_str());
+	g_api->StartFrameCapture(nullptr, nullptr);
+	if (g_api->IsFrameCapturing() == 0) {
+		g_state.store(RenderDocState::Idle, std::memory_order_release);
+		LOGF("RenderDoc: capture failed to start\n");
+		return;
+	}
+	g_captured_flips.store(0, std::memory_order_release);
+	g_state.store(RenderDocState::Capturing, std::memory_order_release);
+	LOGF("RenderDoc: capture started\n");
+}
+
+void RenderDocEndCapture() {
+	if (g_api == nullptr || !RenderDocCaptureInProgress()) {
+		return;
 	}
 
 	const auto ok = g_api->EndFrameCapture(nullptr, nullptr);
-	g_state.store(RenderDocState::Idle);
+	g_state.store(RenderDocState::Idle, std::memory_order_release);
+	LOGF(ok != 0 ? "RenderDoc: capture finished\n" : "RenderDoc: capture failed\n");
+}
 
-	if (ok != 0) {
-		const auto count = g_api->GetNumCaptures();
-		LOGF("RenderDoc: capture finished, count=%u\n", count);
-		LogNewestCapture();
-	} else {
-		LOGF("RenderDoc: capture failed\n");
+void RenderDocOnGuestFlip() {
+	if (!RenderDocCaptureInProgress()) {
+		return;
+	}
+	const auto flip = g_captured_flips.fetch_add(1, std::memory_order_acq_rel) + 1;
+	LOGF("RenderDoc: captured guest flip %u/2\n", flip);
+	if (flip >= 2) {
+		RenderDocEndCapture();
 	}
 }
 

src/graphics/presentation/renderDoc.h

@@ -1,17 +1,16 @@
 #ifndef EMULATOR_INCLUDE_EMULATOR_GRAPHICS_RENDERDOC_H_
 #define EMULATOR_INCLUDE_EMULATOR_GRAPHICS_RENDERDOC_H_
 
-#include "common/common.h"
-#include "graphics/host_gpu/vulkanCommon.h"
-
-struct SDL_Window;
-
 namespace Libs::Graphics {
 
 void RenderDocInit();
-void RenderDocSetActiveWindow(vk::Instance instance, SDL_Window* window);
 void RenderDocRequestCapture();
-void RenderDocOnPresent();
+void RenderDocStartCapture();
+void RenderDocEndCapture();
+void RenderDocOnGuestFlip();
+
+[[nodiscard]] bool RenderDocCaptureRequested();
+[[nodiscard]] bool RenderDocCaptureInProgress();
 
 } // namespace Libs::Graphics
 

src/graphics/presentation/videoOut.cpp

@@ -16,6 +16,7 @@
 #include "graphics/host_gpu/renderer/render.h"
 #include "graphics/host_gpu/renderer/renderContext.h"
 #include "graphics/presentation/presenter.h"
+#include "graphics/presentation/renderDoc.h"
 #include "kernel/pthread.h"
 #include "libs/errno.h"
 #include "libs/libs.h"
@@ -42,6 +43,9 @@ constexpr int      VIDEO_OUT_EVENT_PRE_VBLANK_START                     = 2;
 constexpr int      VIDEO_OUT_EVENT_SET_MODE                             = 8;
 constexpr int      VIDEO_OUT_TRUE                                       = 1;
 constexpr int      VIDEO_OUT_FALSE                                      = 0;
+constexpr int      VIDEO_OUT_BUS_TYPE_MAIN                              = 0;
+constexpr int      VIDEO_OUT_BUS_TYPE_OVERLAY                           = 1;
+constexpr int      VIDEO_OUT_BUS_TYPE_SUB                               = 2;
 constexpr int      VIDEO_OUT_FLIP_MODE_VSYNC                            = 1;
 constexpr int      VIDEO_OUT_FLIP_MODE_VSYNC_MULTI                      = 4;
 constexpr int      VIDEO_OUT_BUFFER_INDEX_BLACK                         = -2;
@@ -233,7 +237,7 @@ private:
 
 struct VideoOutDriver::Impl {
 public:
-	static constexpr int VIDEO_OUT_NUM_MAX = 2;
+	static constexpr int VIDEO_OUT_NUM_MAX = 4;
 
 	Impl(uint32_t width, uint32_t height, Graphics::Presenter& presenter)
 	    : m_renderer(presenter.Renderer()), m_presenter(presenter), m_flip_queue(presenter) {
@@ -244,7 +248,7 @@ public:
 	~Impl();
 	KYTY_CLASS_NO_COPY(Impl);
 
-	int             Open();
+	int             Open(int bus_type);
 	bool            Close(int handle);
 	VideoOutConfig* Get(int handle);
 	VideoOutConfig* Get(int handle, uint64_t& generation);
@@ -534,8 +538,7 @@ Graphics::ImageInfo BufferAttributeGroup::ImageInfo(const VideoOutBuffer& buffer
 	if (!Graphics::DecodeVideoOutPixelFormat(attribute.pixel_format, pixel_format)) {
 		EXIT("unsupported video-out pixel format: 0x%016" PRIx64 "\n", attribute.pixel_format);
 	}
-	const auto tile_mode =
-	    Graphics::Prospero::GpuEnumValue(Graphics::Prospero::TileMode::kRenderTarget);
+	const auto tile_mode = Graphics::Prospero::TileMode::kRenderTarget;
 	const auto pitch =
 	    Graphics::TileGetTexturePitch(pixel_format.guest_format, attribute.width, tile_mode);
 	Graphics::TileSizeAlign total {};
@@ -607,19 +610,11 @@ void VideoOutDriver::Impl::Init(uint32_t width, uint32_t height) {
 	}
 }
 
-int VideoOutDriver::Impl::Open() {
+int VideoOutDriver::Impl::Open(int bus_type) {
 	Common::LockGuard lock(m_mutex);
 
-	int handle = -1;
-
-	for (int i = 1; i < VIDEO_OUT_NUM_MAX; i++) {
-		if (!m_video_out_ctx[i].opened) {
-			handle = i;
-			break;
-		}
-	}
-
-	if (handle < 0) {
+	const int handle = bus_type + 1;
+	if (m_video_out_ctx[handle].opened) {
 		return -1;
 	}
 	auto&             config = m_video_out_ctx[handle];
@@ -1128,6 +1123,7 @@ bool FlipQueue::Flip(uint32_t micros) {
 	m_mutex.Unlock();
 
 	m_presenter.Present(*r.frame);
+	Graphics::RenderDocOnGuestFlip();
 
 	m_mutex.Lock();
 	if (m_requests.empty() || m_requests.front().id != r.id ||
@@ -1172,12 +1168,15 @@ KYTY_SYSV_ABI int VideoOutOpen(int user_id, int bus_type, int index, const void*
 	PRINT_NAME();
 
 	EXIT_NOT_IMPLEMENTED(user_id != 255 && user_id != 0);
-	EXIT_NOT_IMPLEMENTED(bus_type != 0);
+	if (bus_type != VIDEO_OUT_BUS_TYPE_MAIN && bus_type != VIDEO_OUT_BUS_TYPE_OVERLAY &&
+	    bus_type != VIDEO_OUT_BUS_TYPE_SUB) {
+		return VIDEO_OUT_ERROR_INVALID_VALUE;
+	}
 	EXIT_NOT_IMPLEMENTED(index != 0);
 
 	LOGF("\t param = 0x%016" PRIx64 "\n", reinterpret_cast<uint64_t>(param));
 
-	int handle = DriverState().Open();
+	int handle = DriverState().Open(bus_type);
 
 	if (handle < 0) {
 		return VIDEO_OUT_ERROR_RESOURCE_BUSY;

src/graphics/presentation/window/hostInput.cpp

@@ -119,7 +119,8 @@ public:
 			}
 
 			const bool reserved =
-			    binding.key == SDLK_ESCAPE || binding.key == SDLK_SPACE || binding.key == SDLK_F1;
+			    binding.key == SDLK_ESCAPE || binding.key == SDLK_SPACE || binding.key == SDLK_F1 ||
+			    binding.key == SDLK_F11;
 			if (binding.control == INVALID_CONTROL || reserved ||
 			    (binding.key == SDLK_UNKNOWN && binding.mouse_button == 0)) {
 				EXIT("Invalid input mapping: %s\n", value.c_str());

src/graphics/presentation/window/swapchain.cpp

@@ -22,7 +22,6 @@
 #include "common/logging/log.h"
 #include "common/profiler.h"
 #include "common/stringUtils.h"
-#include "common/subsystems.h"
 #include "common/systemInfo.h"
 #include "common/threads.h"
 #include "common/timer.h"
@@ -33,7 +32,6 @@
 #include "graphics/host_gpu/vulkanCommon.h"
 #include "graphics/presentation/imeOverlay.h"
 #include "graphics/presentation/presenter.h"
-#include "graphics/presentation/renderDoc.h"
 #include "graphics/presentation/videoOut.h"
 #include "graphics/presentation/window/windowInternal.h"
 #include "libs/controller.h"
@@ -867,7 +865,6 @@ void Presenter::Present(Frame& frame, bool reuse) {
 			continue;
 		}
 
-		RenderDocOnPresent();
 		m_impl->presented_ime_revision.store(ime_visual.revision, std::memory_order_release);
 		window.UpdateTitle();
 		m_impl->frames.Release(&frame, true);

src/graphics/presentation/window/vulkanWindow.cpp

@@ -22,7 +22,6 @@
 #include "common/logging/log.h"
 #include "common/profiler.h"
 #include "common/stringUtils.h"
-#include "common/subsystems.h"
 #include "common/systemInfo.h"
 #include "common/threads.h"
 #include "common/timer.h"
@@ -33,7 +32,6 @@
 #include "graphics/host_gpu/vulkanCommon.h"
 #include "graphics/presentation/imeOverlay.h"
 #include "graphics/presentation/presenter.h"
-#include "graphics/presentation/renderDoc.h"
 #include "graphics/presentation/videoOut.h"
 #include "graphics/presentation/window.h"
 #include "graphics/presentation/window/windowInternal.h"
@@ -201,7 +199,15 @@ static void VulkanFindPhysicalDevice(vk::Instance instance, vk::SurfaceKHR surfa
 
 		vk::PhysicalDeviceVulkan12Features features12 {};
 		features12.sType = vk::StructureType::ePhysicalDeviceVulkan12Features;
+#if defined(__APPLE__)
 		features12.pNext = &depth_clip_control;
+#else
+		vk::PhysicalDeviceFragmentShaderBarycentricFeaturesKHR fragment_barycentric {};
+		fragment_barycentric.sType =
+		    vk::StructureType::ePhysicalDeviceFragmentShaderBarycentricFeaturesKHR;
+		fragment_barycentric.pNext = &depth_clip_control;
+		features12.pNext           = &fragment_barycentric;
+#endif
 		features13.pNext = &features12;
 
 		device_features2.sType = vk::StructureType::ePhysicalDeviceFeatures2;
@@ -233,6 +239,12 @@ static void VulkanFindPhysicalDevice(vk::Instance instance, vk::SurfaceKHR surfa
 			skip_device = true;
 #endif
 		}
+#if !defined(__APPLE__)
+		if (fragment_barycentric.fragmentShaderBarycentric != VK_TRUE) {
+			LOGF("fragmentShaderBarycentric is not supported\n");
+			skip_device = true;
+		}
+#endif
 
 		if (features12.samplerMirrorClampToEdge != VK_TRUE) {
 			LOGF("samplerMirrorClampToEdge is not supported\n");
@@ -264,6 +276,22 @@ static void VulkanFindPhysicalDevice(vk::Instance instance, vk::SurfaceKHR surfa
 			LOGF("depthBiasClamp is not supported\n");
 			skip_device = true;
 		}
+		if (device_features2.features.shaderClipDistance != VK_TRUE) {
+			LOGF("shaderClipDistance is not supported\n");
+			skip_device = true;
+		}
+		if (device_features2.features.shaderCullDistance != VK_TRUE) {
+			LOGF("shaderCullDistance is not supported\n");
+			skip_device = true;
+		}
+		if (device_features2.features.largePoints != VK_TRUE) {
+			LOGF("largePoints is not supported\n");
+			skip_device = true;
+		}
+		if (features12.shaderOutputLayer != VK_TRUE) {
+			LOGF("shaderOutputLayer is not supported\n");
+			skip_device = true;
+		}
 
 		if (device_features2.features.fragmentStoresAndAtomics != VK_TRUE) {
 			LOGF("fragmentStoresAndAtomics is not supported\n");
@@ -288,10 +316,6 @@ static void VulkanFindPhysicalDevice(vk::Instance instance, vk::SurfaceKHR surfa
 			LOGF("shaderStorageImageWriteWithoutFormat is not supported\n");
 			skip_device = true;
 		}
-		if (device_features2.features.shaderStorageImageReadWithoutFormat != VK_TRUE) {
-			LOGF("shaderStorageImageReadWithoutFormat is not supported\n");
-			skip_device = true;
-		}
 
 		if (device_features2.features.shaderImageGatherExtended != VK_TRUE) {
 			LOGF("shaderImageGatherExtended is not supported\n");
@@ -469,14 +493,19 @@ static void VulkanInitSubgroupSizeControl(vk::PhysicalDevice physical_device,
 	graphics.min_subgroup_size             = subgroup_size_control.minSubgroupSize;
 	graphics.max_subgroup_size             = subgroup_size_control.maxSubgroupSize;
 	graphics.required_subgroup_size_stages = subgroup_size_control.requiredSubgroupSizeStages;
-	graphics.subgroup_size_control_enabled = features13.subgroupSizeControl == VK_TRUE &&
-	                                         subgroup_size_control.minSubgroupSize <= 64 &&
-	                                         subgroup_size_control.maxSubgroupSize >= 64;
+	graphics.compute_subgroup_size_control_enabled =
+	    features13.subgroupSizeControl == VK_TRUE &&
+	    (graphics.required_subgroup_size_stages & vk::ShaderStageFlagBits::eCompute) &&
+	    subgroup_size_control.minSubgroupSize <= 64 &&
+	    subgroup_size_control.maxSubgroupSize >= 64;
+	graphics.compute_wave64_supported =
+	    graphics.subgroup_size == 64u || graphics.compute_subgroup_size_control_enabled;
 
-	LOGF("Vulkan subgroup: default=%u min=%u max=%u stages=0x%08x size_control=%s\n",
+	LOGF("Vulkan subgroup: default=%u min=%u max=%u stages=0x%08x size_control=%s wave64=%s\n",
 	     graphics.subgroup_size, graphics.min_subgroup_size, graphics.max_subgroup_size,
 	     static_cast<vk::ShaderStageFlags::MaskType>(graphics.required_subgroup_size_stages),
-	     graphics.subgroup_size_control_enabled ? "true" : "false");
+	     graphics.compute_subgroup_size_control_enabled ? "true" : "false",
+	     graphics.compute_wave64_supported ? "true" : "false");
 }
 
 static vk::Device VulkanCreateDevice(vk::PhysicalDevice physical_device, const VulkanExtensions& r,
@@ -520,10 +549,6 @@ static vk::Device VulkanCreateDevice(vk::PhysicalDevice physical_device, const V
 	features12.pNext                    = &depth_clip_control;
 	features12.samplerMirrorClampToEdge = VK_TRUE;
 
-	vk::PhysicalDeviceSubgroupSizeControlFeatures subgroup_size_control {};
-	subgroup_size_control.sType = vk::StructureType::ePhysicalDeviceSubgroupSizeControlFeatures;
-	subgroup_size_control.pNext = &features12;
-
 	vk::PhysicalDeviceVulkan13Features supported_features13 {};
 	supported_features13.sType = vk::StructureType::ePhysicalDeviceVulkan13Features;
 
@@ -532,6 +557,14 @@ static vk::Device VulkanCreateDevice(vk::PhysicalDevice physical_device, const V
 	supported_features12.pNext = nullptr;
 	supported_features13.pNext = &supported_features12;
 
+#if !defined(__APPLE__)
+	vk::PhysicalDeviceFragmentShaderBarycentricFeaturesKHR supported_fragment_barycentric {};
+	supported_fragment_barycentric.sType =
+	    vk::StructureType::ePhysicalDeviceFragmentShaderBarycentricFeaturesKHR;
+	supported_fragment_barycentric.pNext = nullptr;
+	supported_features12.pNext           = &supported_fragment_barycentric;
+#endif
+
 	const auto robustness2_ext_enabled =
 	    HasExtension(device_extensions, VK_EXT_ROBUSTNESS_2_EXTENSION_NAME);
 
@@ -539,7 +572,11 @@ static vk::Device VulkanCreateDevice(vk::PhysicalDevice physical_device, const V
 	supported_robustness2.sType = vk::StructureType::ePhysicalDeviceRobustness2FeaturesEXT;
 	supported_robustness2.pNext = nullptr;
 	if (robustness2_ext_enabled) {
+#if defined(__APPLE__)
 		supported_features12.pNext = &supported_robustness2;
+#else
+		supported_fragment_barycentric.pNext = &supported_robustness2;
+#endif
 	}
 
 	vk::PhysicalDeviceFeatures2 supported_features2 {};
@@ -554,7 +591,23 @@ static vk::Device VulkanCreateDevice(vk::PhysicalDevice physical_device, const V
 	                     supported_features13.synchronization2 != VK_TRUE);
 	EXIT_NOT_IMPLEMENTED(supported_features2.features.sampleRateShading != VK_TRUE);
 	EXIT_NOT_IMPLEMENTED(supported_features2.features.depthBiasClamp != VK_TRUE);
+	EXIT_NOT_IMPLEMENTED(supported_features2.features.shaderClipDistance != VK_TRUE);
+	EXIT_NOT_IMPLEMENTED(supported_features2.features.shaderCullDistance != VK_TRUE);
+	EXIT_NOT_IMPLEMENTED(supported_features2.features.largePoints != VK_TRUE);
+	EXIT_NOT_IMPLEMENTED(supported_features12.shaderOutputLayer != VK_TRUE);
+#if !defined(__APPLE__)
+	EXIT_NOT_IMPLEMENTED(supported_fragment_barycentric.fragmentShaderBarycentric != VK_TRUE);
+#endif
 	features12.timelineSemaphore = VK_TRUE;
+	features12.shaderOutputLayer = VK_TRUE;
+
+	if (Config::GpuAssistedValidationEnabled()) {
+		if (supported_features12.bufferDeviceAddress != VK_TRUE) {
+			EXIT("--gpu-assisted-validation requires bufferDeviceAddress, unsupported by this "
+			     "device\n");
+		}
+		features12.bufferDeviceAddress = VK_TRUE;
+	}
 
 	vk::PhysicalDeviceFeatures device_features {};
 	device_features.fragmentStoresAndAtomics = VK_TRUE;
@@ -564,49 +617,64 @@ static vk::Device VulkanCreateDevice(vk::PhysicalDevice physical_device, const V
 	device_features.depthBounds = VK_TRUE; // unsupported by MoltenVK
 #endif
 	device_features.shaderStorageImageWriteWithoutFormat = VK_TRUE;
-	device_features.shaderStorageImageReadWithoutFormat  = VK_TRUE;
 	device_features.shaderImageGatherExtended            = VK_TRUE;
 	device_features.independentBlend                     = VK_TRUE;
 	device_features.tessellationShader                   = VK_TRUE;
 	device_features.sampleRateShading                    = VK_TRUE;
 	device_features.depthBiasClamp                       = VK_TRUE;
+	device_features.shaderClipDistance                   = VK_TRUE;
+	device_features.shaderCullDistance                   = VK_TRUE;
+	device_features.largePoints                          = VK_TRUE;
 	graphics.sample_rate_shading_enabled                 = true;
 	device_features.vertexPipelineStoresAndAtomics =
 	    supported_features2.features.vertexPipelineStoresAndAtomics;
-
-	if (graphics.subgroup_size_control_enabled &&
-	    supported_features13.subgroupSizeControl == VK_TRUE) {
-		subgroup_size_control.subgroupSizeControl = VK_TRUE;
+	if (Config::GpuAssistedValidationEnabled()) {
+		device_features.shaderInt64 = supported_features2.features.shaderInt64;
 	}
 
-	const auto* base_feature_chain = (subgroup_size_control.subgroupSizeControl == VK_TRUE
-	                                      ? static_cast<const void*>(&subgroup_size_control)
-	                                      : static_cast<const void*>(&features12));
-
 	vk::PhysicalDeviceRobustness2FeaturesEXT robustness2 {};
 	robustness2.sType = vk::StructureType::ePhysicalDeviceRobustness2FeaturesEXT;
-	robustness2.pNext = const_cast<void*>(base_feature_chain);
+#if defined(__APPLE__)
+	robustness2.pNext = &features12;
+#else
+	vk::PhysicalDeviceFragmentShaderBarycentricFeaturesKHR fragment_barycentric {};
+	fragment_barycentric.sType =
+	    vk::StructureType::ePhysicalDeviceFragmentShaderBarycentricFeaturesKHR;
+	fragment_barycentric.pNext                     = &features12;
+	fragment_barycentric.fragmentShaderBarycentric = VK_TRUE;
+	robustness2.pNext                              = &fragment_barycentric;
+#endif
 	if (robustness2_ext_enabled) {
 		robustness2.robustBufferAccess2 = supported_robustness2.robustBufferAccess2;
 		robustness2.robustImageAccess2  = supported_robustness2.robustImageAccess2;
 		robustness2.nullDescriptor      = supported_robustness2.nullDescriptor;
 	}
 
+	const bool subgroup_size_control_enabled =
+	    graphics.compute_subgroup_size_control_enabled &&
+	    supported_features13.subgroupSizeControl == VK_TRUE;
+
 	auto features13 = required_features13;
-	features13.pNext =
-	    robustness2_ext_enabled ? &robustness2 : const_cast<void*>(base_feature_chain);
-	features13.robustImageAccess = supported_features13.robustImageAccess;
+#if defined(__APPLE__)
+	features13.pNext = robustness2_ext_enabled ? static_cast<void*>(&robustness2)
+	                                           : static_cast<void*>(&features12);
+#else
+	features13.pNext = robustness2_ext_enabled ? static_cast<void*>(&robustness2)
+	                                           : static_cast<void*>(&fragment_barycentric);
+#endif
+	features13.robustImageAccess   = supported_features13.robustImageAccess;
+	features13.subgroupSizeControl = subgroup_size_control_enabled ? VK_TRUE : VK_FALSE;
 
 	LOGF("Vulkan robustness: robustImageAccess=%s robustImageAccess2=%s\n",
 	     features13.robustImageAccess == VK_TRUE ? "true" : "false",
 	     robustness2_ext_enabled && robustness2.robustImageAccess2 == VK_TRUE ? "true" : "false");
 
 	vk::DeviceCreateInfo create_info {};
-	create_info.sType                = vk::StructureType::eDeviceCreateInfo;
-	create_info.pNext                = &features13;
-	create_info.flags                = {};
-	create_info.pQueueCreateInfos    = &queue_create_info;
-	create_info.queueCreateInfoCount = 1;
+	create_info.sType                   = vk::StructureType::eDeviceCreateInfo;
+	create_info.pNext                   = &features13;
+	create_info.flags                   = {};
+	create_info.pQueueCreateInfos       = &queue_create_info;
+	create_info.queueCreateInfoCount    = 1;
 	create_info.enabledExtensionCount   = static_cast<uint32_t>(device_extensions.size());
 	create_info.ppEnabledExtensionNames = device_extensions.data();
 	create_info.pEnabledFeatures        = &device_features;
@@ -835,23 +903,26 @@ void WindowContext::CreateVulkan() {
 	app_info.engineVersion      = 1;
 	app_info.apiVersion         = VULKAN_TARGET_API_VERSION; // NOLINT
 
-	vk::ValidationFeatureEnableEXT enabled_features[] = {
+	if (Config::SpirvDebugPrintfEnabled() && Config::GpuAssistedValidationEnabled()) {
+		EXIT("--spirv-debug-printf and --gpu-assisted-validation are mutually exclusive\n");
+	}
+
+	vk::ValidationFeatureEnableEXT enabled_features[3]    = {};
+	uint32_t                       enabled_features_count = 0;
 #ifdef KYTY_ENABLE_BEST_PRACTICES
-	    vk::ValidationFeatureEnableEXT::eBestPractices,
+	enabled_features[enabled_features_count++] = vk::ValidationFeatureEnableEXT::eBestPractices;
 #endif
 #ifdef KYTY_ENABLE_DEBUG_PRINTF
-	    vk::ValidationFeatureEnableEXT::eDebugPrintf,
-#endif
-	};
-
-	uint32_t enabled_features_count =
-	    sizeof(enabled_features) / sizeof(vk::ValidationFeatureEnableEXT);
-
-#ifdef KYTY_ENABLE_DEBUG_PRINTF
-	if (!Config::SpirvDebugPrintfEnabled()) {
-		enabled_features_count--;
+	if (Config::SpirvDebugPrintfEnabled()) {
+		enabled_features[enabled_features_count++] = vk::ValidationFeatureEnableEXT::eDebugPrintf;
 	}
 #endif
+	if (Config::GpuAssistedValidationEnabled()) {
+		enabled_features[enabled_features_count++] = vk::ValidationFeatureEnableEXT::eGpuAssisted;
+		LOGF("Vulkan GPU-assisted validation is enabled; expect a large slowdown\n");
+		std::printf("Vulkan GPU-assisted validation is enabled; expect a large slowdown\n");
+		std::fflush(stdout);
+	}
 
 	vk::ValidationFeaturesEXT validation_features {};
 	validation_features.sType                          = vk::StructureType::eValidationFeaturesEXT;
@@ -924,6 +995,7 @@ void WindowContext::CreateVulkan() {
 #else
 	device_extensions.push_back(VK_EXT_DEPTH_CLIP_ENABLE_EXTENSION_NAME);
 	device_extensions.push_back(VK_EXT_COLOR_WRITE_ENABLE_EXTENSION_NAME);
+	device_extensions.push_back(VK_KHR_FRAGMENT_SHADER_BARYCENTRIC_EXTENSION_NAME);
 #endif
 
 #ifdef KYTY_ENABLE_DEBUG_PRINTF
@@ -941,7 +1013,15 @@ void WindowContext::CreateVulkan() {
 		EXIT("Could not find suitable device");
 	}
 
-	graphic_ctx.physical_device.getProperties(&graphic_ctx.physical_device_properties);
+	vk::PhysicalDevicePushDescriptorProperties push_descriptor_properties {};
+	vk::PhysicalDeviceProperties2              physical_device_properties {};
+	push_descriptor_properties.sType = vk::StructureType::ePhysicalDevicePushDescriptorProperties;
+	push_descriptor_properties.pNext = nullptr;
+	physical_device_properties.sType = vk::StructureType::ePhysicalDeviceProperties2;
+	physical_device_properties.pNext = &push_descriptor_properties;
+	graphic_ctx.physical_device.getProperties2(&physical_device_properties);
+	graphic_ctx.physical_device_properties = physical_device_properties.properties;
+	graphic_ctx.max_push_descriptors       = push_descriptor_properties.maxPushDescriptors;
 	graphic_ctx.physical_device.getMemoryProperties(&graphic_ctx.physical_device_memory_properties);
 	const auto& device_properties = graphic_ctx.GetPhysicalDeviceProperties();
 
@@ -987,7 +1067,6 @@ void WindowContext::CreateVulkan() {
 	render_context = std::make_unique<RenderContext>(graphic_ctx);
 	LibKernel::Memory::InstallGpuResources(&render_context->GetGpuResources());
 	presenter = std::make_unique<Presenter>(*this);
-	RenderDocSetActiveWindow(graphic_ctx.instance, window);
 }
 
 void WindowContext::RefreshSurfaceCapabilities() {

src/graphics/presentation/window/window.cpp

@@ -219,7 +219,24 @@ static void GameEventTerminate(WindowLoopState& game) {
 	game.need_exit = true;
 }
 
+static void ToggleDesktopFullscreen() {
+	if (g_window == nullptr || g_window->window == nullptr) {
+		return;
+	}
+
+	const auto flags = static_cast<uint32_t>(SDL_GetWindowFlags(g_window->window));
+	const bool fullscreen =
+	    (flags & static_cast<uint32_t>(SDL_WINDOW_FULLSCREEN_DESKTOP)) != 0u;
+	const auto mode =
+	    fullscreen ? 0u : static_cast<uint32_t>(SDL_WINDOW_FULLSCREEN_DESKTOP);
+	if (SDL_SetWindowFullscreen(g_window->window, mode) != 0) {
+		LOGF("Toggle fullscreen failed: %s\n", SDL_GetError());
+	}
+}
+
 static void GameEventKeyboard(WindowLoopState& game, const EventKeyboard& key) {
+	static SDL_Keycode fullscreen_key = SDLK_UNKNOWN;
+
 #ifdef KYTY_DBG_INPUT
 	LOGF("Key: time = %.04f, %s%s, %s%s, %s, scan = %d, key = %d, mod = %04" PRIx16 "\n",
 	     key.timestamp_seconds, (key.down ? "down" : ""), (key.up ? "up" : ""),
@@ -237,13 +254,30 @@ static void GameEventKeyboard(WindowLoopState& game, const EventKeyboard& key) {
 					RenderDocRequestCapture();
 				}
 				break;
+			case SDLK_F11:
+				if (!key.repeat) {
+					ToggleDesktopFullscreen();
+				}
+				break;
+			case SDLK_RETURN:
+			case SDLK_KP_ENTER:
+				if (!key.repeat && (key.mod & KMOD_ALT) != 0) {
+					ToggleDesktopFullscreen();
+					fullscreen_key = key.key_code;
+				}
+				break;
 			default: break;
 		}
 	}
 
 #endif
 
-	if ((key.down || key.up) && !key.repeat) {
+	const bool fullscreen_key_event =
+	    fullscreen_key != SDLK_UNKNOWN && key.key_code == fullscreen_key;
+	if (fullscreen_key_event && key.up) {
+		fullscreen_key = SDLK_UNKNOWN;
+	}
+	if ((key.down || key.up) && !key.repeat && !fullscreen_key_event) {
 		HostInputKey(key.key_code, key.down);
 	}
 }

src/graphics/shader/recompiler/ExecMask.cpp

@@ -1,51 +0,0 @@
-#include "graphics/shader/recompiler/ExecMask.h"
-
-namespace Libs::Graphics::ShaderRecompiler::Exec {
-
-Mask FullMask(uint32_t wave_size) {
-	return {0xffffffffu, wave_size == 64u ? 0xffffffffu : 0u};
-}
-
-Mask ApplyWaveSize(Mask mask, uint32_t wave_size) {
-	const auto full = FullMask(wave_size);
-	mask.low &= full.low;
-	mask.high &= full.high;
-	return mask;
-}
-
-Mask ReadExec(const State& state) {
-	return ApplyWaveSize(state.exec, state.wave_size);
-}
-
-void WriteExec(State& state, Mask mask) {
-	state.exec = ApplyWaveSize(mask, state.wave_size);
-}
-
-void AndExec(State& state, Mask mask) {
-	WriteExec(state, {state.exec.low & mask.low, state.exec.high & mask.high});
-}
-
-void OrExec(State& state, Mask mask) {
-	WriteExec(state, {state.exec.low | mask.low, state.exec.high | mask.high});
-}
-
-void XorExec(State& state, Mask mask) {
-	WriteExec(state, {state.exec.low ^ mask.low, state.exec.high ^ mask.high});
-}
-
-void InvertExec(State& state) {
-	const auto full = FullMask(state.wave_size);
-	WriteExec(state, {~state.exec.low & full.low, ~state.exec.high & full.high});
-}
-
-bool IsExecZero(const State& state) {
-	const auto exec = ReadExec(state);
-	return exec.low == 0u && exec.high == 0u;
-}
-
-bool IsVccZero(const State& state) {
-	const auto vcc = ApplyWaveSize(state.vcc, state.wave_size);
-	return vcc.low == 0u && vcc.high == 0u;
-}
-
-} // namespace Libs::Graphics::ShaderRecompiler::Exec

src/graphics/shader/recompiler/ExecMask.h

@@ -1,34 +0,0 @@
-#ifndef EMULATOR_INCLUDE_EMULATOR_GRAPHICS_SHADER_RECOMPILER_EXECMASK_H_
-#define EMULATOR_INCLUDE_EMULATOR_GRAPHICS_SHADER_RECOMPILER_EXECMASK_H_
-
-#include "common/common.h"
-
-namespace Libs::Graphics::ShaderRecompiler::Exec {
-
-struct Mask {
-	uint32_t low  = 0;
-	uint32_t high = 0;
-};
-
-struct State {
-	uint32_t wave_size = 64;
-	Mask     exec;
-	Mask     vcc;
-	bool     scc = false;
-};
-
-Mask FullMask(uint32_t wave_size);
-Mask ApplyWaveSize(Mask mask, uint32_t wave_size);
-
-Mask ReadExec(const State& state);
-void WriteExec(State& state, Mask mask);
-void AndExec(State& state, Mask mask);
-void OrExec(State& state, Mask mask);
-void XorExec(State& state, Mask mask);
-void InvertExec(State& state);
-bool IsExecZero(const State& state);
-bool IsVccZero(const State& state);
-
-} // namespace Libs::Graphics::ShaderRecompiler::Exec
-
-#endif /* EMULATOR_INCLUDE_EMULATOR_GRAPHICS_SHADER_RECOMPILER_EXECMASK_H_ */

src/graphics/shader/recompiler/ShaderRecompiler.cpp

@@ -2,18 +2,22 @@
 
 #include "common/assert.h"
 #include "common/logging/log.h"
-#include "graphics/shader/recompiler/cfg/ShaderCFG.h"
-#include "graphics/shader/recompiler/decompiler/ShaderDecoder.h"
-#include "graphics/shader/recompiler/emitter/SpirvEmitter.h"
-#include "graphics/shader/recompiler/ir/BindingLayout.h"
-#include "graphics/shader/recompiler/ir/ReadLaneElimination.h"
-#include "graphics/shader/recompiler/ir/ResourceMaterialization.h"
-#include "graphics/shader/recompiler/ir/ResourceTracking.h"
-#include "graphics/shader/recompiler/ir/ScalarProvenance.h"
+#include "graphics/shader/recompiler/backend/spirv/SpirvEmitter.h"
+#include "graphics/shader/recompiler/frontend/cfg/ShaderCFG.h"
+#include "graphics/shader/recompiler/frontend/decode/ShaderDecoder.h"
+#include "graphics/shader/recompiler/frontend/translate/Translate.h"
 #include "graphics/shader/recompiler/ir/ShaderIR.h"
-#include "graphics/shader/recompiler/ir/ShaderInfoCollection.h"
-#include "graphics/shader/recompiler/ir/SrtPatcher.h"
-#include "graphics/shader/recompiler/ir/SrtWalker.h"
+#include "graphics/shader/recompiler/ir/ValueProgram.h"
+#include "graphics/shader/recompiler/ir/passes/BindingLayout.h"
+#include "graphics/shader/recompiler/ir/passes/ConstantPropagation.h"
+#include "graphics/shader/recompiler/ir/passes/DeadCodeElimination.h"
+#include "graphics/shader/recompiler/ir/passes/ReadLaneElimination.h"
+#include "graphics/shader/recompiler/ir/passes/ResourceMaterialization.h"
+#include "graphics/shader/recompiler/ir/passes/ResourceTracking.h"
+#include "graphics/shader/recompiler/ir/passes/ShaderInfoCollection.h"
+#include "graphics/shader/recompiler/ir/passes/SharedMemoryBarrier.h"
+#include "graphics/shader/recompiler/ir/passes/SrtWalker.h"
+#include "graphics/shader/recompiler/ir/passes/SsaRewrite.h"
 
 #include <algorithm>
 #include <array>
@@ -43,7 +47,9 @@ std::string MakeIrDump(const CFG::Graph& cfg, const IR::Program& ir) {
 	std::string dump = "CFG:\n";
 	dump += CFG::GraphToString(cfg);
 	dump += "\nIR:\n";
-	dump += IR::ProgramToString(ir);
+	dump += fmt::format("mode={} scratch_dwords={}\n",
+	                    ir.dispatcher_fallback ? "dispatcher" : "structured", ir.scratch_dwords);
+	dump += ir.values != nullptr ? IR::ValueProgramToString(*ir.values) : IR::ProgramToString(ir);
 	return dump;
 }
 
@@ -56,76 +62,22 @@ const char* StageName(ShaderType stage) {
 	}
 }
 
-std::string FormatCfgFailure(const CFG::Graph& cfg, const CompileOptions& options,
-                             const std::string& reason) {
-	const auto block_id = cfg.failure_block != UINT32_MAX ? cfg.failure_block : cfg.entry_block;
-	return CFG::FormatBlockDiagnostic(cfg, block_id, StageName(options.stage), reason);
-}
-
-bool InstructionMaySplitSpirvBlock(const IR::Instruction& inst) {
-	switch (inst.op) {
-		case IR::Opcode::SLoadDword:
-		case IR::Opcode::SBufferLoadDword:
-		case IR::Opcode::BufferLoadUbyte:
-		case IR::Opcode::BufferLoadSbyte:
-		case IR::Opcode::BufferLoadUshort:
-		case IR::Opcode::BufferLoadSshort:
-		case IR::Opcode::BufferLoadDword:
-		case IR::Opcode::BufferStoreByte:
-		case IR::Opcode::BufferStoreShort:
-		case IR::Opcode::BufferStoreDword:
-		case IR::Opcode::AtomicSwapU32:
-		case IR::Opcode::AtomicAddU32:
-		case IR::Opcode::AtomicSubU32:
-		case IR::Opcode::AtomicSMinI32:
-		case IR::Opcode::AtomicUMinU32:
-		case IR::Opcode::AtomicSMaxI32:
-		case IR::Opcode::AtomicUMaxU32:
-		case IR::Opcode::AtomicAndU32:
-		case IR::Opcode::AtomicOrU32:
-		case IR::Opcode::AtomicXorU32:
-		case IR::Opcode::AtomicFMinF32:
-		case IR::Opcode::AtomicFMaxF32:
-		case IR::Opcode::FlatLoadUbyte:
-		case IR::Opcode::FlatLoadSbyte:
-		case IR::Opcode::FlatLoadUshort:
-		case IR::Opcode::FlatLoadSshort:
-		case IR::Opcode::FlatLoadDword:
-		case IR::Opcode::FlatStoreByte:
-		case IR::Opcode::FlatStoreShort:
-		case IR::Opcode::FlatStoreDword:
-		case IR::Opcode::DsMinF32:
-		case IR::Opcode::DsMaxF32:
-		case IR::Opcode::DsWriteByte:
-		case IR::Opcode::DsWriteShort:
-		case IR::Opcode::DsWriteB32:
-		case IR::Opcode::DsWriteAddtidB32:
-		case IR::Opcode::DsAppend:
-		case IR::Opcode::DsConsume:
-		case IR::Opcode::ImageStore:
-		case IR::Opcode::Export: return true;
-		default: return false;
-	}
-}
-
-bool NeedsDispatcherForStructuredLoopHeader(const IR::Program& ir, std::string* reason) {
-	for (const auto& block: ir.blocks) {
-		if (!block.terminator.loop_header) {
-			continue;
-		}
-		for (const auto& inst: block.instructions) {
-			if (!InstructionMaySplitSpirvBlock(inst)) {
-				continue;
-			}
-			if (reason != nullptr) {
-				*reason = fmt::format("loop header block {} contains an instruction whose SPIR-V "
-				                      "lowering emits internal control flow: {}",
-				                      block.id, IR::InstructionToString(inst).c_str());
-			}
-			return true;
-		}
-	}
-	return false;
+void LogDispatcherFallback(const CompileOptions& options, const CFG::Graph& cfg, const char* phase,
+                           const std::string& reason) {
+	const auto* block        = cfg.FindBlock(cfg.failure_block);
+	const auto  start        = block != nullptr ? block->start_pc : UINT32_MAX;
+	const auto  end          = block != nullptr ? block->end_pc : UINT32_MAX;
+	const auto  predecessors = block != nullptr ? block->predecessors.size() : 0u;
+	const auto  successors   = block != nullptr ? block->successors.size() : 0u;
+	LOGF("%s CFG dispatcher fallback: stage=%s hash=0x%016" PRIx64
+	     " phase=%s failure=%s block=%" PRIu32 " pc=0x%08" PRIx32 "..0x%08" PRIx32 " preds=%" PRIu64
+	     " succs=%" PRIu64 " blocks=%" PRIu64 " loops=%" PRIu64 " back_edges=%" PRIu64
+	     " reason=%s\n",
+	     GetDumpLabel(options), StageName(options.stage), options.shader_hash, phase,
+	     CFG::FailureKindToString(cfg.failure_kind).c_str(), cfg.failure_block, start, end,
+	     static_cast<uint64_t>(predecessors), static_cast<uint64_t>(successors),
+	     static_cast<uint64_t>(cfg.blocks.size()), static_cast<uint64_t>(cfg.natural_loops.size()),
+	     static_cast<uint64_t>(cfg.back_edges.size()), reason.c_str());
 }
 
 enum class EmbeddedFetchValueType {
@@ -199,19 +151,19 @@ uint32_t DecodedDstSize(const Decoder::Instruction& inst) {
 }
 
 uint32_t EmbeddedFetchDstSize(const Decoder::Instruction& inst) {
-	return inst.opcode == Decoder::Opcode::VMadU64U32 ? 2u : DecodedDstSize(inst);
+	return inst.opcode == Decoder::Opcode::V_MAD_U64_U32 ? 2u : DecodedDstSize(inst);
 }
 
 bool EmbeddedFetchHasBranch(Decoder::Opcode opcode) {
 	switch (opcode) {
-		case Decoder::Opcode::SSetpcB64:
-		case Decoder::Opcode::SBranch:
-		case Decoder::Opcode::SCbranchScc0:
-		case Decoder::Opcode::SCbranchScc1:
-		case Decoder::Opcode::SCbranchVccz:
-		case Decoder::Opcode::SCbranchVccnz:
-		case Decoder::Opcode::SCbranchExecz:
-		case Decoder::Opcode::SCbranchExecnz: return true;
+		case Decoder::Opcode::S_SETPC_B64:
+		case Decoder::Opcode::S_BRANCH:
+		case Decoder::Opcode::S_CBRANCH_SCC0:
+		case Decoder::Opcode::S_CBRANCH_SCC1:
+		case Decoder::Opcode::S_CBRANCH_VCCZ:
+		case Decoder::Opcode::S_CBRANCH_VCCNZ:
+		case Decoder::Opcode::S_CBRANCH_EXECZ:
+		case Decoder::Opcode::S_CBRANCH_EXECNZ: return true;
 		default: return false;
 	}
 }
@@ -260,32 +212,32 @@ bool TryDecodedSmemOffset(const std::array<EmbeddedFetchSgprInfo, 108>& sgprs,
 
 bool IsEmbeddedFetchSLoad(const Decoder::Instruction& inst) {
 	switch (inst.opcode) {
-		case Decoder::Opcode::SLoadDword:
-		case Decoder::Opcode::SLoadDwordx2:
-		case Decoder::Opcode::SLoadDwordx4:
-		case Decoder::Opcode::SLoadDwordx8:
-		case Decoder::Opcode::SLoadDwordx16: return true;
+		case Decoder::Opcode::S_LOAD_DWORD:
+		case Decoder::Opcode::S_LOAD_DWORDX2:
+		case Decoder::Opcode::S_LOAD_DWORDX4:
+		case Decoder::Opcode::S_LOAD_DWORDX8:
+		case Decoder::Opcode::S_LOAD_DWORDX16: return true;
 		default: return false;
 	}
 }
 
 bool IsEmbeddedFetchBufferLoad(const Decoder::Instruction& inst) {
 	switch (inst.opcode) {
-		case Decoder::Opcode::BufferLoadFormatX:
-		case Decoder::Opcode::BufferLoadFormatXy:
-		case Decoder::Opcode::BufferLoadFormatXyz:
-		case Decoder::Opcode::BufferLoadFormatXyzw: return true;
+		case Decoder::Opcode::BUFFER_LOAD_FORMAT_X:
+		case Decoder::Opcode::BUFFER_LOAD_FORMAT_XY:
+		case Decoder::Opcode::BUFFER_LOAD_FORMAT_XYZ:
+		case Decoder::Opcode::BUFFER_LOAD_FORMAT_XYZW: return true;
 		default: return false;
 	}
 }
 
 bool IsEmbeddedFetchAttribPropagationAlu(const Decoder::Instruction& inst) {
 	switch (inst.opcode) {
-		case Decoder::Opcode::SBfeU32:
-		case Decoder::Opcode::SAndB32:
-		case Decoder::Opcode::SAddI32:
-		case Decoder::Opcode::SAddU32:
-		case Decoder::Opcode::SLshlB32: return true;
+		case Decoder::Opcode::S_BFE_U32:
+		case Decoder::Opcode::S_AND_B32:
+		case Decoder::Opcode::S_ADD_I32:
+		case Decoder::Opcode::S_ADD_U32:
+		case Decoder::Opcode::S_LSHL_B32: return true;
 		default: return false;
 	}
 }
@@ -308,9 +260,6 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
                                             uint32_t user_data_base, uint32_t user_data_count,
                                             uint32_t wave_size) {
 	EmbeddedFetchData data;
-	if (input_info == nullptr || !input_info->fetch_embedded) {
-		return data;
-	}
 	data.loads.reserve(input_info->resources_num);
 	int32_t offset_candidate = -1;
 	bool    offset_conflict  = false;
@@ -348,10 +297,11 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 		uint32_t   sad_zero = 0;
 		const bool vertex_offset_add =
 		    vertex_index_accumulator && IsDecodedSgpr(inst.src0) &&
-		    ((inst.opcode == Decoder::Opcode::VAddI32 && IsDecodedVgpr(inst.src1) &&
+		    ((inst.opcode == Decoder::Opcode::V_ADD_I32 && IsDecodedVgpr(inst.src1) &&
 		      inst.src1.reg == inst.dst.reg) ||
-		     (user_data_base == 8 && inst.dst.reg == 5 && inst.opcode == Decoder::Opcode::VSadU32 &&
-		      IsDecodedVgpr(inst.src2) && inst.src2.reg == inst.dst.reg &&
+		     (user_data_base == 8 && inst.dst.reg == 5 &&
+		      inst.opcode == Decoder::Opcode::V_SAD_U32 && IsDecodedVgpr(inst.src2) &&
+		      inst.src2.reg == inst.dst.reg &&
 		      TryDecodedOperandConstant(sgprs, inst.src1, sad_zero) && sad_zero == 0));
 		if (data.loads.empty() && vertex_offset_add) {
 			const auto reg = DecodedSgprReg(inst.src0);
@@ -364,7 +314,7 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 			}
 		}
 		switch (inst.opcode) {
-			case Decoder::Opcode::VWritelaneB32: {
+			case Decoder::Opcode::V_WRITELANE_B32: {
 				uint32_t lane = 0;
 				if (IsDecodedVgpr(inst.dst) && inst.dst.reg < vgprs.size()) {
 					vgprs[inst.dst.reg] = {};
@@ -380,7 +330,7 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 				}
 				break;
 			}
-			case Decoder::Opcode::VReadlaneB32: {
+			case Decoder::Opcode::V_READLANE_B32: {
 				uint32_t lane = 0;
 				if (track_vector_lanes && IsDecodedSgpr(inst.dst) &&
 				    DecodedSgprReg(inst.dst) < sgprs.size() && IsDecodedVgpr(inst.src0) &&
@@ -394,7 +344,7 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 				}
 				break;
 			}
-			case Decoder::Opcode::SMovB32:
+			case Decoder::Opcode::S_MOV_B32:
 				if (IsDecodedSgpr(inst.dst) && IsDecodedSgpr(inst.src0) &&
 				    DecodedSgprReg(inst.src0) < sgprs.size()) {
 					sgprs[DecodedSgprReg(inst.dst)] = sgprs[DecodedSgprReg(inst.src0)];
@@ -410,7 +360,7 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 					}
 				}
 				break;
-			case Decoder::Opcode::SMovkI32:
+			case Decoder::Opcode::S_MOVK_I32:
 				if (IsDecodedSgpr(inst.dst)) {
 					auto& dst = sgprs[DecodedSgprReg(inst.dst)];
 					dst.type  = EmbeddedFetchValueType::Constant;
@@ -471,7 +421,7 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 					} else {
 						ClearEmbeddedFetchSgprs(sgprs, inst.dst, DecodedDstSize(inst));
 					}
-				} else if (inst.opcode == Decoder::Opcode::VCndmaskB32) {
+				} else if (inst.opcode == Decoder::Opcode::V_CNDMASK_B32) {
 					if (IsDecodedVgpr(inst.dst) && inst.dst.reg < vgprs.size()) {
 						ClearEmbeddedFetchVectorLanes(&vector_lanes, inst.dst.reg);
 					}
@@ -493,11 +443,11 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 							auto& dst = sgprs[DecodedSgprReg(inst.dst)];
 							dst.type  = EmbeddedFetchValueType::Constant;
 							switch (inst.opcode) {
-								case Decoder::Opcode::SAndB32: dst.value = src0 & src1; break;
-								case Decoder::Opcode::SLshlB32:
+								case Decoder::Opcode::S_AND_B32: dst.value = src0 & src1; break;
+								case Decoder::Opcode::S_LSHL_B32:
 									dst.value = src0 << (src1 & 31u);
 									break;
-								case Decoder::Opcode::SBfeU32:
+								case Decoder::Opcode::S_BFE_U32:
 									dst.value = src0 >> (src1 & 31u);
 									break;
 								default: dst.value = src0 + src1; break;
@@ -526,9 +476,9 @@ EmbeddedFetchData DetectEmbeddedVertexFetch(const Decoder::Program&      decoded
 				}
 				break;
 		}
-		if (inst.opcode == Decoder::Opcode::VMovreldB32) {
+		if (inst.opcode == Decoder::Opcode::V_MOVRELD_B32) {
 			vector_lanes.clear();
-		} else if (inst.opcode != Decoder::Opcode::VWritelaneB32 && IsDecodedVgpr(inst.dst)) {
+		} else if (inst.opcode != Decoder::Opcode::V_WRITELANE_B32 && IsDecodedVgpr(inst.dst)) {
 			for (uint32_t i = 0; i < EmbeddedFetchDstSize(inst) && inst.dst.reg + i < vgprs.size();
 			     i++) {
 				ClearEmbeddedFetchVectorLanes(&vector_lanes, inst.dst.reg + i);
@@ -559,9 +509,6 @@ bool IsIrFetchPrologLoad(const IR::Instruction& inst) {
 
 int ResolveEmbeddedFetchResource(const ShaderVertexInputInfo* input_info,
                                  const EmbeddedFetchLoad&     load) {
-	if (input_info == nullptr) {
-		return -1;
-	}
 	if (load.attrib_id >= 0 && load.attrib_id < input_info->resources_num &&
 	    input_info->resources_dst[load.attrib_id].attr_id == load.attrib_id) {
 		return load.attrib_id;
@@ -583,7 +530,7 @@ int ResolveEmbeddedFetchResource(const ShaderVertexInputInfo* input_info,
 
 uint32_t RewriteEmbeddedVertexFetches(IR::Program& ir, const ShaderVertexInputInfo* input_info,
                                       const std::vector<EmbeddedFetchLoad>& loads) {
-	if (input_info == nullptr || loads.empty()) {
+	if (loads.empty()) {
 		return 0;
 	}
 
@@ -609,7 +556,9 @@ uint32_t RewriteEmbeddedVertexFetches(IR::Program& ir, const ShaderVertexInputIn
 				continue;
 			}
 			const auto* load = FindEmbeddedFetchLoad(loads, inst.pc);
-			if (load == nullptr || inst.memory.component_index >= load->components) {
+			if (load == nullptr || inst.memory.data_dwords != load->components ||
+			    inst.dst.kind != IR::OperandKind::Register ||
+			    inst.dst.reg.file != IR::RegisterFile::Vector) {
 				continue;
 			}
 
@@ -621,16 +570,16 @@ uint32_t RewriteEmbeddedVertexFetches(IR::Program& ir, const ShaderVertexInputIn
 				continue;
 			}
 
-			inst.op              = IR::Opcode::LoadInputF32;
-			inst.input_info.attr = static_cast<uint32_t>(resource_id);
-			inst.input_info.chan = inst.memory.component_index;
-			inst.memory          = {};
-			inst.src_count       = 0;
-
+			inst.op                         = IR::Opcode::LoadInputF32;
+			inst.input_info.attr            = static_cast<uint32_t>(resource_id);
+			inst.input_info.chan            = 0;
+			inst.input_info.component_count = load->components;
+			inst.memory                     = {};
+			inst.src_count                  = 0;
 			mutable_input_info->resource_fetch_components[resource_id] =
 			    std::max(mutable_input_info->resource_fetch_components[resource_id],
-			             static_cast<int>(inst.input_info.chan + 1u));
-			rewritten++;
+			             static_cast<int>(load->components));
+			rewritten += load->components;
 		}
 	}
 
@@ -697,27 +646,25 @@ bool TryRecompile(std::span<const uint32_t> code, const CompileOptions& options,
 	bool        dispatcher_fallback = false;
 	std::string dispatcher_reason;
 	if (cfg.irreducible) {
-		dispatcher_fallback   = true;
-		dispatcher_reason     = cfg.unsupported_reason;
-		const auto diagnostic = FormatCfgFailure(cfg, options, cfg.unsupported_reason);
-		LOGF("%s irreducible CFG detected: %s\n", GetDumpLabel(options), diagnostic.c_str());
+		dispatcher_fallback = true;
+		dispatcher_reason   = cfg.unsupported_reason;
+		LogDispatcherFallback(options, cfg, "build", dispatcher_reason);
 	} else {
 		std::string structure_error;
 		const auto  unstructured_cfg = cfg;
 		LOGF("%s phase begin: stage=%s hash=0x%016" PRIx64 " CFG Structurize\n",
 		     GetDumpLabel(options), StageName(options.stage), options.shader_hash);
 		if (!CFG::Structurize(cfg, &structure_error)) {
-			const auto diagnostic = FormatCfgFailure(cfg, options, structure_error);
-			LOGF("%s structured CFG bug/failure: %s\n", GetDumpLabel(options), diagnostic.c_str());
 			dispatcher_fallback      = true;
 			dispatcher_reason        = structure_error;
 			const auto failure_kind  = cfg.failure_kind;
 			const auto failure_block = cfg.failure_block;
-			cfg                      = unstructured_cfg;
-			cfg.unsupported          = true;
-			cfg.failure_kind         = failure_kind;
-			cfg.failure_block        = failure_block;
-			cfg.unsupported_reason   = structure_error;
+			LogDispatcherFallback(options, cfg, "structurize", dispatcher_reason);
+			cfg                    = unstructured_cfg;
+			cfg.unsupported        = true;
+			cfg.failure_kind       = failure_kind;
+			cfg.failure_block      = failure_block;
+			cfg.unsupported_reason = structure_error;
 		} else {
 			LOGF("%s structured CFG success: blocks=%" PRIu64 "\n", GetDumpLabel(options),
 			     static_cast<uint64_t>(cfg.blocks.size()));
@@ -732,34 +679,77 @@ bool TryRecompile(std::span<const uint32_t> code, const CompileOptions& options,
 	IR::Program ir;
 	LOGF("%s phase begin: stage=%s hash=0x%016" PRIx64 " IR LowerProgram\n", GetDumpLabel(options),
 	     StageName(options.stage), options.shader_hash);
-	if (!IR::LowerProgram(decoded, cfg, options.stage, options.wave_size, ir, error)) {
+	const auto lower_program = [&](IR::Program& output) {
+		output = {};
+		if (!IR::LowerProgram(decoded, cfg, options.stage, options.wave_size, output, error)) {
+			return false;
+		}
+		output.shader_hash     = options.shader_hash;
+		output.user_data_base  = options.user_data_base;
+		output.user_data_count = options.user_data_count;
+		output.scratch_dwords  = options.scratch_dwords;
+		return true;
+	};
+	if (!lower_program(ir)) {
 		return false;
 	}
-	ir.lane_mask_mode  = options.lane_mask_mode;
-	ir.shader_hash     = options.shader_hash;
-	ir.user_data_base  = options.user_data_base;
-	ir.user_data_count = options.user_data_count;
+
 	LOGF("%s phase end: stage=%s hash=0x%016" PRIx64 " IR LowerProgram blocks=%" PRIu64
 	     " elapsed_ms=%" PRIu64 "\n",
 	     GetDumpLabel(options), StageName(options.stage), options.shader_hash,
 	     static_cast<uint64_t>(ir.blocks.size()), phase_ms());
+	const ShaderVertexInputInfo*  vertex  = nullptr;
+	const ShaderPixelInputInfo*   pixel   = nullptr;
+	const ShaderComputeInputInfo* compute = nullptr;
+	switch (options.stage) {
+		case ShaderType::Vertex: vertex = options.input_info.vertex; break;
+		case ShaderType::Pixel: pixel = options.input_info.pixel; break;
+		case ShaderType::Compute: compute = options.input_info.compute; break;
+		default: break;
+	}
 	EmbeddedFetchData embedded_fetch;
-	if (options.stage == ShaderType::Vertex && options.vertex_input_info != nullptr &&
-	    options.vertex_input_info->fetch_embedded) {
-		embedded_fetch =
-		    DetectEmbeddedVertexFetch(decoded, options.vertex_input_info, ir.user_data_base,
-		                              ir.user_data_count, options.wave_size);
-		auto rewritten =
-		    RewriteEmbeddedVertexFetches(ir, options.vertex_input_info, embedded_fetch.loads);
+	if (options.stage == ShaderType::Vertex && vertex->fetch_embedded) {
+		embedded_fetch = DetectEmbeddedVertexFetch(decoded, vertex, ir.user_data_base,
+		                                           ir.user_data_count, options.wave_size);
+		auto rewritten = RewriteEmbeddedVertexFetches(ir, vertex, embedded_fetch.loads);
 		if (rewritten > 0 || !embedded_fetch.loads.empty()) {
 			LOGF("%s embedded vertex fetch rewrite: detected=%" PRIu64 " rewritten=%" PRIu32 "\n",
 			     GetDumpLabel(options), static_cast<uint64_t>(embedded_fetch.loads.size()),
 			     rewritten);
 		}
 	}
-	if (!IR::BuildScalarProvenance(ir, error)) {
+	ir.dispatcher_fallback = dispatcher_fallback;
+	if (!dispatcher_reason.empty()) {
+		ir.fallback_reason = dispatcher_reason;
+	}
+
+	ir.values = std::make_shared<IR::ValueProgram>();
+	if (!Frontend::TranslateProgram(ir, *ir.values, vertex, pixel, compute, error)) {
 		return false;
 	}
+	ir.blocks.clear();
+	IR::RewriteToSsa(ir.values->blocks);
+	IR::ConstantPropagationPass(ir.values->blocks);
+	IR::ResolveControlFlowIdentities(*ir.values);
+	IR::RemoveIdentities(ir.values->blocks);
+	IR::EliminateDeadCode(ir.values->blocks);
+	const auto read_lane_stats = IR::EliminateReadLane(*ir.values, ir.wave_size);
+	if (read_lane_stats.rewritten_reads != 0) {
+		LOGF("%s read-lane elimination: reads=%" PRIu32 "\n", GetDumpLabel(options),
+		     read_lane_stats.rewritten_reads);
+		IR::ConstantPropagationPass(ir.values->blocks);
+		IR::ResolveControlFlowIdentities(*ir.values);
+		IR::RemoveIdentities(ir.values->blocks);
+		IR::EliminateDeadCode(ir.values->blocks);
+	}
+	if (options.stage == ShaderType::Compute) {
+		const auto lds_barriers =
+		    IR::InsertSharedMemoryBarriers(*ir.values, ir.wave_size, *compute);
+		if (lds_barriers.inserted_barriers != 0) {
+			LOGF("%s wave64 LDS synchronization: barriers=%" PRIu32 "\n", GetDumpLabel(options),
+			     lds_barriers.inserted_barriers);
+		}
+	}
 	std::string srt_error;
 	if (!IR::BuildSrtPlan(ir, &srt_error)) {
 		LOGF("%s SRT planning failed: %s\n", GetDumpLabel(options), srt_error.c_str());
@@ -768,29 +758,14 @@ bool TryRecompile(std::span<const uint32_t> code, const CompileOptions& options,
 		}
 		return false;
 	}
-	ir.dispatcher_fallback = dispatcher_fallback;
-	if (!dispatcher_reason.empty()) {
-		ir.fallback_reason = dispatcher_reason;
-	}
-
-	if (!IR::PatchSrtReads(ir, error) || !IR::TrackResources(ir, error)) {
+	IR::EliminateDeadCode(ir.values->blocks);
+	if (!IR::TrackResources(ir, error)) {
 		return false;
 	}
+	IR::EliminateDeadCode(ir.values->blocks);
 	if (options.stage == ShaderType::Vertex) {
 		ir.info.vertex_offset_sgpr = embedded_fetch.vertex_offset_sgpr;
 	}
-	if (!dispatcher_fallback) {
-		std::string emitter_reason;
-		if (NeedsDispatcherForStructuredLoopHeader(ir, &emitter_reason)) {
-			dispatcher_fallback    = true;
-			dispatcher_reason      = emitter_reason;
-			ir.dispatcher_fallback = true;
-			ir.cfg_failure_kind    = CFG::FailureKind::StructuredControlFlow;
-			ir.fallback_reason     = emitter_reason;
-			LOGF("%s structured CFG emitter fallback: %s\n", GetDumpLabel(options),
-			     emitter_reason.c_str());
-		}
-	}
 
 	IR::ResourceSnapshot resources;
 	if (options.resource_snapshot != nullptr) {
@@ -809,6 +784,7 @@ bool TryRecompile(std::span<const uint32_t> code, const CompileOptions& options,
 		runtime.shader_base = options.shader_base != 0 ? options.shader_base
 		                                               : reinterpret_cast<uint64_t>(code.data());
 		runtime.read_memory = options.read_memory;
+		runtime.read_specialization_memory = options.read_specialization_memory;
 		if (runtime.read_memory == nullptr && options.user_data == nullptr) {
 			runtime.read_memory = ReadZeroMemory;
 		}
@@ -822,29 +798,18 @@ bool TryRecompile(std::span<const uint32_t> code, const CompileOptions& options,
 		return false;
 	}
 
-	ShaderVertexInputInfo  default_vertex {};
-	ShaderPixelInputInfo   default_pixel {};
-	ShaderComputeInputInfo default_compute {};
-	IR::ShaderInfoOptions  info_options;
-	info_options.vertex =
-	    options.vertex_input_info != nullptr ? options.vertex_input_info : &default_vertex;
-	info_options.pixel =
-	    options.pixel_input_info != nullptr ? options.pixel_input_info : &default_pixel;
-	info_options.compute =
-	    options.compute_input_info != nullptr ? options.compute_input_info : &default_compute;
+	IR::ShaderInfoOptions info_options;
+	info_options.vertex  = vertex;
+	info_options.pixel   = pixel;
+	info_options.compute = compute;
 	if (!IR::CollectShaderInfo(ir, info_options, error)) {
 		return false;
 	}
-	IR::BindingLayoutOptions layout_options;
-	layout_options.descriptor_set       = options.descriptor_set;
-	layout_options.push_constant_offset = options.push_constant_offset;
-	if (!IR::AllocateBindings(ir, layout_options, error)) {
+	if (!IR::AllocateBindings(ir, options.push_constant_offset, error)) {
 		return false;
 	}
-	const auto read_lane_stats = IR::EliminateReadLane(ir);
-	if (read_lane_stats.rewritten_reads != 0) {
-		LOGF("%s read-lane elimination: reads=%" PRIu32 " shadow_writes=%" PRIu32 "\n",
-		     GetDumpLabel(options), read_lane_stats.rewritten_reads, read_lane_stats.shadow_writes);
+	if (!Spirv::AnalyzeProgramRequirements(ir, error)) {
+		return false;
 	}
 	std::string ir_dump;
 	if (options.dump_ir) {
@@ -858,8 +823,8 @@ bool TryRecompile(std::span<const uint32_t> code, const CompileOptions& options,
 	std::string           emit_error;
 	LOGF("%s phase begin: stage=%s hash=0x%016" PRIx64 " SPIR-V EmitProgram\n",
 	     GetDumpLabel(options), StageName(options.stage), options.shader_hash);
-	if (!Spirv::EmitProgram(ir, resources, options.vertex_input_info, options.pixel_input_info,
-	                        options.compute_input_info, spirv, &emit_error)) {
+	if (!Spirv::EmitProgram(ir, resources, options.input_info, spirv, &emit_error)) {
+		LOGF("%s typed SPIR-V emit failed: %s\n", GetDumpLabel(options), emit_error.c_str());
 		if (dispatcher_fallback && error != nullptr) {
 			*error = fmt::format("dispatcher fallback failed after {}: {}",
 			                     dispatcher_reason.c_str(), emit_error.c_str());
@@ -873,10 +838,6 @@ bool TryRecompile(std::span<const uint32_t> code, const CompileOptions& options,
 	     " elapsed_ms=%" PRIu64 "\n",
 	     GetDumpLabel(options), StageName(options.stage), options.shader_hash,
 	     static_cast<uint64_t>(spirv.size()), phase_ms());
-	if (dispatcher_fallback) {
-		LOGF("%s dispatcher fallback used: %s\n", GetDumpLabel(options), dispatcher_reason.c_str());
-	}
-
 	result.spirv     = std::move(spirv);
 	result.program   = std::move(ir);
 	result.resources = std::move(resources);

src/graphics/shader/recompiler/ShaderRecompiler.h

@@ -3,7 +3,7 @@
 
 #include "common/common.h"
 #include "common/stringUtils.h"
-#include "graphics/shader/recompiler/ir/ResourceMaterialization.h"
+#include "graphics/shader/recompiler/ir/passes/ResourceMaterialization.h"
 #include "graphics/shader/shader.h"
 
 #include <optional>
@@ -13,26 +13,24 @@
 namespace Libs::Graphics::ShaderRecompiler {
 
 struct CompileOptions {
-	ShaderType                    stage           = ShaderType::Compute;
-	ShaderLaneMaskMode            lane_mask_mode  = ShaderLaneMaskMode::NativeWave;
-	uint32_t                      wave_size       = 64;
-	uint32_t                      user_data_base  = 0;
-	uint32_t                      user_data_count = 64;
-	uint64_t                      shader_hash     = 0;
-	uint64_t                      shader_base     = 0;
-	std::optional<uint64_t>       flat_memory_base;
-	uint32_t                      descriptor_set       = 0;
-	uint32_t                      push_constant_offset = 0;
-	bool                          dump_ir              = true;
-	bool                          early_dump           = false;
-	const char*                   dump_label           = nullptr;
-	const uint32_t*               user_data            = nullptr;
-	IR::SrtMemoryReader           read_memory          = nullptr;
-	void*                         read_memory_data     = nullptr;
-	const IR::ResourceSnapshot*   resource_snapshot    = nullptr;
-	const ShaderVertexInputInfo*  vertex_input_info    = nullptr;
-	const ShaderPixelInputInfo*   pixel_input_info     = nullptr;
-	const ShaderComputeInputInfo* compute_input_info   = nullptr;
+	ShaderType                  stage           = ShaderType::Compute;
+	uint32_t                    wave_size       = 64;
+	uint32_t                    user_data_base  = 0;
+	uint32_t                    user_data_count = 64;
+	uint32_t                    scratch_dwords  = 0;
+	uint64_t                    shader_hash     = 0;
+	uint64_t                    shader_base     = 0;
+	std::optional<uint64_t>     flat_memory_base;
+	uint32_t                    push_constant_offset      = 0;
+	bool                        dump_ir                    = true;
+	bool                        early_dump                 = false;
+	const char*                 dump_label                 = nullptr;
+	const uint32_t*             user_data                  = nullptr;
+	IR::SrtMemoryReader         read_memory                = nullptr;
+	IR::SrtMemoryReader         read_specialization_memory = nullptr;
+	void*                       read_memory_data           = nullptr;
+	const IR::ResourceSnapshot* resource_snapshot          = nullptr;
+	ShaderStageInputInfo        input_info;
 };
 
 struct CompileResult {

src/graphics/shader/recompiler/backend/spirv/SpirvBuilder.cpp

@@ -0,0 +1,270 @@
+#include "graphics/shader/recompiler/backend/spirv/SpirvBuilder.h"
+
+#include "common/debug.h"
+
+#include <cstring>
+
+namespace Libs::Graphics::ShaderRecompiler::Spirv {
+
+static constexpr size_t InitialSpirvSectionReserve         = 4096;
+static constexpr size_t InitialSpirvFunctionSectionReserve = 32768;
+
+static void AppendInstructionWords(std::vector<uint32_t>& section, const uint32_t* words,
+                                   size_t words_num) {
+	if (words_num == 0) {
+		return;
+	}
+	const auto opcode     = words[0];
+	const auto word_count = static_cast<uint32_t>(words_num);
+	section.push_back((word_count << 16u) | opcode);
+	section.insert(section.end(), words + 1, words + words_num);
+}
+
+Builder::Builder(uint32_t version): m_version(version) {
+	m_debug.reserve(InitialSpirvSectionReserve);
+	m_annotations.reserve(InitialSpirvSectionReserve);
+	m_declarations.reserve(InitialSpirvSectionReserve);
+	m_functions.reserve(InitialSpirvFunctionSectionReserve);
+}
+
+uint32_t Builder::AllocateId() {
+	return m_next_id++;
+}
+
+void Builder::RequireCapability(uint32_t capability) {
+	if (m_required_capabilities.insert(capability).second) {
+		AddCapability({capability});
+	}
+}
+
+void Builder::RequireExtension(const char* name) {
+	if (m_required_extensions.emplace(name).second) {
+		AddExtension(name);
+	}
+}
+
+uint32_t Builder::Import(const char* name) {
+	if (const auto it = m_import_ids.find(name); it != m_import_ids.end()) {
+		return it->second;
+	}
+	const auto id = AllocateId();
+	m_import_ids.emplace(name, id);
+	AddExtInstImport(id, name);
+	return id;
+}
+
+uint32_t Builder::Type(uint32_t opcode, std::initializer_list<uint32_t> operands) {
+	return Type(opcode, std::vector<uint32_t>(operands));
+}
+
+uint32_t Builder::Type(uint32_t opcode, const std::vector<uint32_t>& operands) {
+	std::vector<uint32_t> key;
+	key.reserve(operands.size() + 2u);
+	key.push_back(opcode);
+	key.push_back(static_cast<uint32_t>(operands.size()));
+	key.insert(key.end(), operands.begin(), operands.end());
+	if (const auto it = m_declaration_ids.find(key); it != m_declaration_ids.end()) {
+		return it->second;
+	}
+	const auto id = AllocateId();
+	m_declaration_ids.emplace(std::move(key), id);
+	std::vector<uint32_t> words {opcode, id};
+	words.insert(words.end(), operands.begin(), operands.end());
+	AppendInstructionWords(m_declarations, words.data(), words.size());
+	return id;
+}
+
+uint32_t Builder::DecoratedType(uint32_t opcode, std::initializer_list<uint32_t> operands,
+                                std::initializer_list<TypeAnnotation> annotations) {
+	if (annotations.size() == 0) {
+		return Type(opcode, operands);
+	}
+	std::vector<uint32_t> key {opcode, static_cast<uint32_t>(operands.size())};
+	key.insert(key.end(), operands.begin(), operands.end());
+	key.push_back(static_cast<uint32_t>(annotations.size()));
+	for (const auto& annotation: annotations) {
+		key.push_back(annotation.opcode);
+		key.push_back(static_cast<uint32_t>(annotation.operands.size()));
+		key.insert(key.end(), annotation.operands.begin(), annotation.operands.end());
+	}
+	if (const auto it = m_declaration_ids.find(key); it != m_declaration_ids.end()) {
+		return it->second;
+	}
+	const auto id = AllocateId();
+	m_declaration_ids.emplace(std::move(key), id);
+	std::vector<uint32_t> words {opcode, id};
+	words.insert(words.end(), operands.begin(), operands.end());
+	AppendInstructionWords(m_declarations, words.data(), words.size());
+	for (const auto& annotation: annotations) {
+		words = {annotation.opcode, id};
+		words.insert(words.end(), annotation.operands.begin(), annotation.operands.end());
+		AppendInstructionWords(m_annotations, words.data(), words.size());
+	}
+	return id;
+}
+
+uint32_t Builder::Constant(uint32_t opcode, uint32_t type,
+                           std::initializer_list<uint32_t> operands) {
+	return Constant(opcode, type, std::vector<uint32_t>(operands));
+}
+
+uint32_t Builder::Constant(uint32_t opcode, uint32_t type, const std::vector<uint32_t>& operands) {
+	std::vector<uint32_t> key;
+	key.reserve(operands.size() + 2u);
+	key.push_back(opcode);
+	key.push_back(type);
+	key.insert(key.end(), operands.begin(), operands.end());
+	if (const auto it = m_declaration_ids.find(key); it != m_declaration_ids.end()) {
+		return it->second;
+	}
+	const auto id = AllocateId();
+	m_declaration_ids.emplace(std::move(key), id);
+	std::vector<uint32_t> words {opcode, type, id};
+	words.insert(words.end(), operands.begin(), operands.end());
+	AppendInstructionWords(m_declarations, words.data(), words.size());
+	return id;
+}
+
+uint32_t Builder::DefineGlobalVariable(uint32_t pointer_type, uint32_t storage_class) {
+	const auto id = AllocateId();
+	DefineGlobalVariable(id, pointer_type, storage_class);
+	return id;
+}
+
+void Builder::DefineGlobalVariable(uint32_t id, uint32_t pointer_type, uint32_t storage_class) {
+	AddType({59u, pointer_type, id, storage_class});
+}
+
+void Builder::AppendString(std::vector<uint32_t>& words, const char* text) {
+	const auto len        = text != nullptr ? std::strlen(text) : 0;
+	const auto word_count = (len + 1u + 3u) / 4u;
+	for (size_t i = 0; i < word_count; i++) {
+		uint32_t word = 0;
+		for (size_t byte = 0; byte < 4; byte++) {
+			const auto index = i * 4u + byte;
+			if (index < len) {
+				word |= static_cast<uint32_t>(static_cast<unsigned char>(text[index]))
+				        << (byte * 8u);
+			}
+		}
+		words.push_back(word);
+	}
+}
+
+void Builder::AppendInstruction(std::vector<uint32_t>& section, uint32_t opcode,
+                                const std::vector<uint32_t>& operands) {
+	const uint32_t word_count = static_cast<uint32_t>(operands.size() + 1u);
+	section.push_back((word_count << 16u) | opcode);
+	section.insert(section.end(), operands.begin(), operands.end());
+}
+
+void Builder::AppendInstruction(std::vector<uint32_t>& section, uint32_t opcode,
+                                std::initializer_list<uint32_t> operands) {
+	const uint32_t word_count = static_cast<uint32_t>(operands.size() + 1u);
+	section.push_back((word_count << 16u) | opcode);
+	section.insert(section.end(), operands.begin(), operands.end());
+}
+
+void Builder::AddCapability(std::initializer_list<uint32_t> operands) {
+	AppendInstruction(m_capabilities, 17u, operands);
+}
+
+void Builder::AddExtension(const char* name) {
+	std::vector<uint32_t> operands;
+	AppendString(operands, name);
+	AppendInstruction(m_extensions, 10u, operands);
+}
+
+void Builder::AddExtInstImport(uint32_t id, const char* name) {
+	std::vector<uint32_t> operands = {id};
+	AppendString(operands, name);
+	AppendInstruction(m_ext_inst_imports, 11u, operands);
+}
+
+void Builder::AddMemoryModel(std::initializer_list<uint32_t> operands) {
+	AppendInstruction(m_memory_model, 14u, operands);
+}
+
+void Builder::AddEntryPoint(uint32_t execution_model, uint32_t entry_point, const char* name,
+                            const std::vector<uint32_t>& interfaces) {
+	std::vector<uint32_t> operands = {execution_model, entry_point};
+	AppendString(operands, name);
+	operands.insert(operands.end(), interfaces.begin(), interfaces.end());
+	AppendInstruction(m_entry_points, 15u, operands);
+}
+
+void Builder::AddExecutionMode(std::initializer_list<uint32_t> operands) {
+	AppendInstruction(m_execution_modes, 16u, operands);
+}
+
+void Builder::AddName(uint32_t target, const char* name) {
+	std::vector<uint32_t> operands = {target};
+	AppendString(operands, name);
+	AppendInstruction(m_debug, 5u, operands);
+}
+
+void Builder::AddAnnotation(std::initializer_list<uint32_t> words) {
+	AppendInstructionWords(m_annotations, words.begin(), words.size());
+}
+
+void Builder::AddType(std::initializer_list<uint32_t> words) {
+	AppendInstructionWords(m_declarations, words.begin(), words.size());
+}
+
+void Builder::AddFunction(std::initializer_list<uint32_t> words) {
+	AppendInstructionWords(m_functions, words.begin(), words.size());
+}
+
+void Builder::AddFunction(const std::vector<uint32_t>& words) {
+	AppendInstructionWords(m_functions, words.data(), words.size());
+}
+
+DeferredPhi Builder::AddDeferredPhi(uint32_t type, uint32_t result, size_t incoming_count) {
+	std::vector<uint32_t> words {245u, type, result};
+	words.resize(words.size() + incoming_count * 2u);
+	const DeferredPhi phi {m_functions.size(), incoming_count};
+	AddFunction(words);
+	m_unpatched_phi_incomings += incoming_count;
+	return phi;
+}
+
+void Builder::PatchDeferredPhi(DeferredPhi phi, size_t incoming, uint32_t value, uint32_t parent) {
+	EXIT_IF(incoming >= phi.incoming_count || value == 0 || parent == 0);
+	const auto value_word  = phi.word_offset + 3u + incoming * 2u;
+	const auto parent_word = value_word + 1u;
+	EXIT_IF(m_functions.at(value_word) != 0 || m_functions.at(parent_word) != 0);
+	m_functions[value_word]  = value;
+	m_functions[parent_word] = parent;
+	m_unpatched_phi_incomings--;
+}
+
+std::vector<uint32_t> Builder::Build() const {
+	EXIT_IF(m_unpatched_phi_incomings != 0);
+
+	std::vector<uint32_t> module;
+	module.reserve(5u + m_capabilities.size() + m_extensions.size() + m_ext_inst_imports.size() +
+	               m_memory_model.size() + m_entry_points.size() + m_execution_modes.size() +
+	               m_debug.size() + m_annotations.size() + m_declarations.size() +
+	               m_functions.size());
+
+	module.push_back(0x07230203u);
+	module.push_back(m_version);
+	module.push_back(0u);
+	module.push_back(m_next_id);
+	module.push_back(0u);
+
+	module.insert(module.end(), m_capabilities.begin(), m_capabilities.end());
+	module.insert(module.end(), m_extensions.begin(), m_extensions.end());
+	module.insert(module.end(), m_ext_inst_imports.begin(), m_ext_inst_imports.end());
+	module.insert(module.end(), m_memory_model.begin(), m_memory_model.end());
+	module.insert(module.end(), m_entry_points.begin(), m_entry_points.end());
+	module.insert(module.end(), m_execution_modes.begin(), m_execution_modes.end());
+	module.insert(module.end(), m_debug.begin(), m_debug.end());
+	module.insert(module.end(), m_annotations.begin(), m_annotations.end());
+	module.insert(module.end(), m_declarations.begin(), m_declarations.end());
+	module.insert(module.end(), m_functions.begin(), m_functions.end());
+
+	return module;
+}
+
+} // namespace Libs::Graphics::ShaderRecompiler::Spirv

src/graphics/shader/recompiler/backend/spirv/SpirvBuilder.h

@@ -0,0 +1,89 @@
+#ifndef EMULATOR_INCLUDE_EMULATOR_GRAPHICS_SHADER_RECOMPILER_SPIRVBUILDER_H_
+#define EMULATOR_INCLUDE_EMULATOR_GRAPHICS_SHADER_RECOMPILER_SPIRVBUILDER_H_
+
+#include "common/common.h"
+
+#include <initializer_list>
+#include <map>
+#include <set>
+#include <string>
+#include <vector>
+
+namespace Libs::Graphics::ShaderRecompiler::Spirv {
+
+struct TypeAnnotation {
+	uint32_t              opcode = 0;
+	std::vector<uint32_t> operands;
+};
+
+struct DeferredPhi {
+	size_t word_offset    = 0;
+	size_t incoming_count = 0;
+};
+
+class Builder {
+public:
+	explicit Builder(uint32_t version = 0x00010300u);
+	~Builder() = default;
+	KYTY_CLASS_DEFAULT_COPY(Builder);
+
+	uint32_t AllocateId();
+	void     RequireCapability(uint32_t capability);
+	void     RequireExtension(const char* name);
+	uint32_t Import(const char* name);
+	uint32_t Type(uint32_t opcode, std::initializer_list<uint32_t> operands = {});
+	uint32_t Type(uint32_t opcode, const std::vector<uint32_t>& operands);
+	uint32_t DecoratedType(uint32_t opcode, std::initializer_list<uint32_t> operands,
+	                       std::initializer_list<TypeAnnotation> annotations);
+	uint32_t Constant(uint32_t opcode, uint32_t type,
+	                  std::initializer_list<uint32_t> operands = {});
+	uint32_t Constant(uint32_t opcode, uint32_t type, const std::vector<uint32_t>& operands);
+	uint32_t DefineGlobalVariable(uint32_t pointer_type, uint32_t storage_class);
+	void     DefineGlobalVariable(uint32_t id, uint32_t pointer_type, uint32_t storage_class);
+
+	void        AddMemoryModel(std::initializer_list<uint32_t> operands);
+	void        AddEntryPoint(uint32_t execution_model, uint32_t entry_point, const char* name,
+	                          const std::vector<uint32_t>& interfaces);
+	void        AddExecutionMode(std::initializer_list<uint32_t> operands);
+	void        AddName(uint32_t target, const char* name);
+	void        AddAnnotation(std::initializer_list<uint32_t> words);
+	void        AddFunction(std::initializer_list<uint32_t> words);
+	void        AddFunction(const std::vector<uint32_t>& words);
+	DeferredPhi AddDeferredPhi(uint32_t type, uint32_t result, size_t incoming_count);
+	void        PatchDeferredPhi(DeferredPhi phi, size_t incoming, uint32_t value, uint32_t parent);
+
+	[[nodiscard]] std::vector<uint32_t> Build() const;
+
+private:
+	static void AppendInstruction(std::vector<uint32_t>& section, uint32_t opcode,
+	                              const std::vector<uint32_t>& operands);
+	static void AppendInstruction(std::vector<uint32_t>& section, uint32_t opcode,
+	                              std::initializer_list<uint32_t> operands);
+	static void AppendString(std::vector<uint32_t>& words, const char* text);
+	void        AddCapability(std::initializer_list<uint32_t> operands);
+	void        AddExtension(const char* name);
+	void        AddExtInstImport(uint32_t id, const char* name);
+	void        AddType(std::initializer_list<uint32_t> words);
+
+	uint32_t                                  m_next_id = 1;
+	uint32_t                                  m_version = 0;
+	std::vector<uint32_t>                     m_capabilities;
+	std::vector<uint32_t>                     m_extensions;
+	std::vector<uint32_t>                     m_ext_inst_imports;
+	std::vector<uint32_t>                     m_memory_model;
+	std::vector<uint32_t>                     m_entry_points;
+	std::vector<uint32_t>                     m_execution_modes;
+	std::vector<uint32_t>                     m_debug;
+	std::vector<uint32_t>                     m_annotations;
+	std::vector<uint32_t>                     m_declarations;
+	std::vector<uint32_t>                     m_functions;
+	std::set<uint32_t>                        m_required_capabilities;
+	std::set<std::string>                     m_required_extensions;
+	std::map<std::string, uint32_t>           m_import_ids;
+	std::map<std::vector<uint32_t>, uint32_t> m_declaration_ids;
+	size_t                                    m_unpatched_phi_incomings = 0;
+};
+
+} // namespace Libs::Graphics::ShaderRecompiler::Spirv
+
+#endif /* EMULATOR_INCLUDE_EMULATOR_GRAPHICS_SHADER_RECOMPILER_SPIRVBUILDER_H_ */

src/graphics/shader/recompiler/backend/spirv/SpirvEmitter.cpp

@@ -0,0 +1,364 @@
+#include "graphics/shader/recompiler/backend/spirv/SpirvEmitter.h"
+
+#include "graphics/shader/recompiler/backend/spirv/spirvEmitterInternal.h"
+#include "graphics/shader/recompiler/ir/ValueProgram.h"
+#include "graphics/shader/shader.h"
+
+#include <algorithm>
+#include <array>
+
+namespace Libs::Graphics::ShaderRecompiler::Spirv {
+
+using ShaderError::Fail;
+
+namespace {
+
+bool ValidateNativeProgram(const IR::Program& program, std::string* error) {
+	using Kind                                             = IR::DescriptorBindingKind;
+	constexpr auto                               KindCount = static_cast<size_t>(Kind::Count);
+	std::array<std::vector<uint32_t>, KindCount> expected;
+	std::array<bool, KindCount>                  present {};
+	const auto                                   Dense = [](size_t size) {
+		std::vector<uint32_t> values(size);
+		for (uint32_t i = 0; i < values.size(); i++) {
+			values[i] = i;
+		}
+		return values;
+	};
+	auto Expect = [&](Kind kind, std::vector<uint32_t> resources = {}) {
+		const auto index = static_cast<size_t>(kind);
+		present[index]   = true;
+		expected[index]  = std::move(resources);
+	};
+	if (!program.info.buffers.empty()) {
+		Expect(Kind::Buffers, Dense(program.info.buffers.size()));
+	}
+	for (uint32_t i = 0; i < program.info.images.size(); i++) {
+		const auto kind = IR::DescriptorBindingForImage(program.info.images[i]);
+		if (!kind.has_value()) {
+			return Fail(error, "native shader plan has an invalid image class");
+		}
+		present[static_cast<size_t>(*kind)] = true;
+		const auto dynamic = program.info.images[i].mip_mode == IR::ImageMipMode::DynamicStorage;
+		const auto count   = dynamic ? program.info.images[i].mip_count : 1u;
+		if (count == 0u || (!dynamic && program.info.images[i].mip_count != 1u)) {
+			return Fail(error, "native shader plan has an invalid image mip descriptor count");
+		}
+		expected[static_cast<size_t>(*kind)].insert(expected[static_cast<size_t>(*kind)].end(),
+		                                            count, i);
+	}
+	if (!program.info.samplers.empty()) {
+		Expect(Kind::Samplers, Dense(program.info.samplers.size()));
+	}
+	bool uses_gds = false;
+	if (program.values != nullptr) {
+		for (const auto* block: program.values->blocks) {
+			for (const auto& inst: *block) {
+				if (IR::SharedAccessOf(inst.GetOpcode()) == IR::SharedAccess::None) {
+					continue;
+				}
+				const auto index = inst.Flags<IR::MemoryFlags>().index;
+				if (index >= program.values->memory_info.size()) {
+					return Fail(error, "shared operation has invalid memory metadata");
+				}
+				const auto kind = program.values->memory_info[index].kind;
+				if (kind != IR::ResourceKind::Lds && kind != IR::ResourceKind::Gds) {
+					return Fail(error, "shared operation has invalid resource kind");
+				}
+				uses_gds |= kind == IR::ResourceKind::Gds;
+			}
+		}
+	}
+	if (uses_gds) {
+		Expect(Kind::Gds);
+	}
+	if (!program.info.addresses.empty()) {
+		Expect(Kind::AddressMemory, Dense(program.info.addresses.size()));
+	}
+	const bool uses_flattened_runtime =
+	    program.values != nullptr &&
+	    (!program.values->srt_reads.empty() ||
+	     std::ranges::any_of(program.info.images, [](const IR::ImageResource& image) {
+		     return image.indirect_mapping_capacity != 0u;
+	     }));
+	if (uses_flattened_runtime) {
+		Expect(Kind::FlattenedSrt);
+	}
+	if (program.bindings.ShaderDataDwords() != 0 && program.bindings.push_constant_size == 0) {
+		Expect(Kind::UserData);
+	}
+
+	std::array<bool, KindCount> seen {};
+	for (const auto& binding: program.bindings.descriptors) {
+		const auto kind = static_cast<size_t>(binding.kind);
+		if (kind >= KindCount || seen[kind] || !present[kind] ||
+		    binding.resources != expected[kind]) {
+			return Fail(error, "native descriptor groups do not match shader topology");
+		}
+		seen[kind] = true;
+	}
+	for (size_t i = 0; i < KindCount; i++) {
+		if (present[i] != seen[i]) {
+			return Fail(error, "native shader plan is missing a required descriptor group");
+		}
+	}
+	const auto has_user_storage = present[static_cast<size_t>(Kind::UserData)];
+	if (program.bindings.push_constant_offset % sizeof(uint32_t) != 0 ||
+	    program.bindings.push_constant_offset + program.bindings.push_constant_size >
+	        IR::NativePushConstantSize ||
+	    program.bindings.memory_offset_dword != program.bindings.user_data_registers.size() ||
+	    program.bindings.memory_offset_count !=
+	        program.info.buffers.size() + program.info.addresses.size() ||
+	    (!has_user_storage &&
+	     program.bindings.push_constant_size != program.bindings.ShaderDataDwords() * 4u) ||
+	    (has_user_storage && program.bindings.push_constant_size != 0) ||
+	    !std::is_sorted(program.bindings.user_data_registers.begin(),
+	                    program.bindings.user_data_registers.end()) ||
+	    std::adjacent_find(program.bindings.user_data_registers.begin(),
+	                       program.bindings.user_data_registers.end()) !=
+	        program.bindings.user_data_registers.end()) {
+		return Fail(error, "native user-data layout is inconsistent");
+	}
+
+	if (program.values == nullptr) {
+		return Fail(error, "native shader plan has no typed SSA");
+	}
+	const auto planning_only_handle = [&](const IR::Inst& handle) {
+		return !handle.Uses().empty() &&
+		       std::ranges::all_of(handle.Uses(), [&](const IR::Use& use) {
+			       const auto op = use.user->GetOpcode();
+			       if (op != IR::ValueOpcode::LoadAddressU32 &&
+			           op != IR::ValueOpcode::ReadConstBuffer) {
+				       return false;
+			       }
+			       const auto index = use.user->Flags<IR::MemoryFlags>().index;
+			       return index < program.values->memory_info.size() &&
+			              program.values->memory_info[index].planning_only;
+		       });
+	};
+	for (const auto* block: program.values->blocks) {
+		for (const auto& inst: *block) {
+			const auto dense = inst.Flags<uint32_t>();
+			switch (inst.GetOpcode()) {
+				case IR::ValueOpcode::GetBufferResource:
+					if (planning_only_handle(inst)) {
+						break;
+					}
+					if (dense >= program.info.buffers.size()) {
+						return Fail(error, "typed buffer handle has an invalid dense resource");
+					}
+					break;
+				case IR::ValueOpcode::GetAddressResource:
+					if (planning_only_handle(inst)) {
+						break;
+					}
+					if (dense >= program.info.addresses.size()) {
+						return Fail(error, "typed address handle has an invalid dense resource");
+					}
+					break;
+				case IR::ValueOpcode::GetScratchResource:
+					if (inst.NumArgs() != 0 || program.scratch_dwords == 0) {
+						return Fail(error, "typed scratch handle has invalid shader metadata");
+					}
+					break;
+				case IR::ValueOpcode::GetImageResource:
+					if (dense >= program.info.images.size()) {
+						return Fail(error, "typed image handle has an invalid dense resource");
+					}
+					break;
+				case IR::ValueOpcode::GetSamplerResource:
+					if (dense >= program.info.samplers.size()) {
+						return Fail(error, "typed sampler handle has an invalid dense resource");
+					}
+					break;
+				case IR::ValueOpcode::ReadConst: {
+					const auto slot = inst.Arg(1).Resolve();
+					if (!slot.IsImmediate() || slot.GetType() != IR::Type::U32 ||
+					    slot.U32() >= program.values->srt_reads.size()) {
+						return Fail(error, "flattened SRT read has an invalid dense slot");
+					}
+					break;
+				}
+				default: break;
+			}
+		}
+	}
+	return true;
+}
+
+} // namespace
+
+bool AnalyzeProgramRequirements(IR::Program& program, std::string* error) {
+	program.spirv_requirements.reset();
+	IR::SpirvRequirements requirements {};
+	if (program.values == nullptr) {
+		program.spirv_requirements.emplace(requirements);
+		return true;
+	}
+	const auto MarkBallot = [&] { requirements.subgroup_ballot = true; };
+	for (const auto* block: program.values->blocks) {
+		for (const auto& inst: *block) {
+			if (IR::AddressOpcodeInfoOf(inst.GetOpcode()).access != IR::AddressAccess::None) {
+				const auto memory_index = inst.Flags<IR::MemoryFlags>().index;
+				if (memory_index >= program.values->memory_info.size()) {
+					return Fail(error, "address operation has invalid memory metadata");
+				}
+				if (program.values->memory_info[memory_index].kind == IR::ResourceKind::Scratch) {
+					if (program.scratch_dwords == 0) {
+						return Fail(error, "scratch operation has no per-thread storage");
+					}
+					requirements.function_scratch = true;
+				}
+			}
+			if (IR::BufferAccessOf(inst.GetOpcode()) != IR::BufferAccess::None) {
+				const auto memory_index = inst.Flags<IR::MemoryFlags>().index;
+				if (memory_index >= program.values->memory_info.size()) {
+					return Fail(error, "buffer operation has invalid memory metadata");
+				}
+				const auto& memory = program.values->memory_info[memory_index];
+				if (memory.kind == IR::ResourceKind::Buffer) {
+					if (memory.resource >= program.info.buffers.size()) {
+						return Fail(error, "buffer operation has invalid resource metadata");
+					}
+					if ((program.info.buffers[memory.resource].packed_stride & (1u << 20u)) != 0u) {
+						if (program.stage != ShaderType::Compute) {
+							return Fail(error, "buffer ADD_TID is only valid for compute shaders");
+						}
+						requirements.subgroup_local_invocation_id = true;
+					}
+				}
+			}
+			const auto shared_access = IR::SharedAccessOf(inst.GetOpcode());
+			if (shared_access != IR::SharedAccess::None) {
+				const auto index = inst.Flags<IR::MemoryFlags>().index;
+				if (index >= program.values->memory_info.size()) {
+					return Fail(error, "shared operation has invalid memory metadata");
+				}
+				const auto kind = program.values->memory_info[index].kind;
+				if (kind != IR::ResourceKind::Lds && kind != IR::ResourceKind::Gds) {
+					return Fail(error, "shared operation has invalid resource kind");
+				}
+				if (program.stage != ShaderType::Compute && kind == IR::ResourceKind::Lds) {
+					requirements.function_lds = true;
+				}
+				if (shared_access == IR::SharedAccess::Append ||
+				    shared_access == IR::SharedAccess::Consume) {
+					MarkBallot();
+					requirements.subgroup_shuffle             = true;
+					requirements.subgroup_local_invocation_id = true;
+				}
+			}
+			switch (inst.GetOpcode()) {
+				case IR::ValueOpcode::Ballot: MarkBallot(); break;
+				case IR::ValueOpcode::DppMoveU32:
+				case IR::ValueOpcode::ReadFirstLane:
+				case IR::ValueOpcode::ReadLane: {
+					MarkBallot();
+					requirements.subgroup_shuffle = true;
+					if (inst.GetOpcode() == IR::ValueOpcode::DppMoveU32) {
+						requirements.subgroup_local_invocation_id = true;
+					}
+					break;
+				}
+				case IR::ValueOpcode::DppUpdateU32:
+				case IR::ValueOpcode::WqmMask:
+				case IR::ValueOpcode::WriteLane: {
+					MarkBallot();
+					requirements.subgroup_local_invocation_id = true;
+					break;
+				}
+				case IR::ValueOpcode::Permlane16U32: {
+					MarkBallot();
+					requirements.subgroup_shuffle             = true;
+					requirements.subgroup_local_invocation_id = true;
+					break;
+				}
+				case IR::ValueOpcode::SwizzleU32: {
+					MarkBallot();
+					requirements.subgroup_shuffle             = true;
+					requirements.subgroup_local_invocation_id = true;
+					break;
+				}
+				case IR::ValueOpcode::LaneId:
+					requirements.subgroup_local_invocation_id = true;
+					break;
+				case IR::ValueOpcode::ImageQueryLod: requirements.compute_derivatives = true; break;
+				case IR::ValueOpcode::ImageGatherRaw:
+					requirements.image_gather_extended = true;
+					break;
+				case IR::ValueOpcode::SetAttribute: {
+					const auto index = inst.Flags<IR::ExportFlags>().index;
+					if (index >= program.values->export_info.size()) {
+						return Fail(error, "attribute export has invalid metadata");
+					}
+					if (program.stage == ShaderType::Pixel &&
+					    program.values->export_info[index].vm) {
+						requirements.pixel_valid_mask = true;
+					}
+					break;
+				}
+				default: break;
+			}
+		}
+	}
+	program.spirv_requirements.emplace(requirements);
+	return true;
+}
+
+bool EmitProgram(const IR::Program& program, const IR::ResourceSnapshot& resources,
+                 ShaderStageInputInfo input_info, std::vector<uint32_t>& spirv,
+                 std::string* error) {
+	using namespace Emitter;
+
+	if (program.stage != ShaderType::Compute && program.stage != ShaderType::Vertex &&
+	    program.stage != ShaderType::Pixel) {
+		SetError(error, "binary SPIR-V emitter supports compute, vertex, and pixel shaders");
+		return false;
+	}
+	if (!program.srt_plan_complete || !program.resource_tracking_complete ||
+	    !program.shader_info_complete || !program.binding_layout_complete ||
+	    !program.spirv_requirements.has_value()) {
+		SetError(error, "SPIR-V emitter requires a fully planned native shader program");
+		return false;
+	}
+	if (!IR::ValidateResourceSnapshot(program, resources, error)) {
+		return false;
+	}
+	if (!IR::ValidateResourceSpecialization(program, resources, error)) {
+		return false;
+	}
+	if (!ValidateNativeProgram(program, error)) {
+		return false;
+	}
+	if (program.values == nullptr) {
+		SetError(error, "SPIR-V emitter requires planned typed SSA");
+		return false;
+	}
+	const auto& value_program = *program.values;
+	if (!IR::ValidateValueProgram(value_program, true, error)) {
+		return false;
+	}
+	EmitterState state(program, resources, input_info);
+	state.stage     = program.stage;
+	state.wave_size = program.wave_size;
+	state.inputs.reserve(program.info.inputs.size());
+	state.outputs.reserve(program.info.outputs.size());
+	state.interface_variables.reserve(program.info.inputs.size() + program.info.outputs.size());
+	CopyProgramInputsAndOutputs(state, program);
+	AllocateInputVariables(state);
+	AllocateOutputVariables(state);
+	DefineModule(state);
+	if (!EmitValueProgram(state, value_program, error)) {
+		return false;
+	}
+
+	auto binary = state.builder.Build();
+	if (binary.empty()) {
+		SetError(error, "SPIR-V builder returned an empty module");
+		return false;
+	}
+	spirv = std::move(binary);
+	return true;
+}
+
+} // namespace Libs::Graphics::ShaderRecompiler::Spirv

src/graphics/shader/recompiler/backend/spirv/SpirvEmitter.h

@@ -3,19 +3,16 @@
 
 #include "common/common.h"
 #include "common/stringUtils.h"
-#include "graphics/shader/recompiler/ir/ResourceMaterialization.h"
+#include "graphics/shader/recompiler/ir/passes/ResourceMaterialization.h"
 
 #include <vector>
 
 namespace Libs::Graphics::ShaderRecompiler::Spirv {
 
-bool ProgramRequiresExactSubgroupSize(const IR::Program& program);
+bool AnalyzeProgramRequirements(IR::Program& program, std::string* error);
 
 bool EmitProgram(const IR::Program& program, const IR::ResourceSnapshot& resources,
-                 const ShaderVertexInputInfo*  vertex_input_info,
-                 const ShaderPixelInputInfo*   pixel_input_info,
-                 const ShaderComputeInputInfo* compute_input_info, std::vector<uint32_t>& spirv,
-                 std::string* error);
+                 ShaderStageInputInfo input_info, std::vector<uint32_t>& spirv, std::string* error);
 
 } // namespace Libs::Graphics::ShaderRecompiler::Spirv
 

src/graphics/shader/recompiler/backend/spirv/spirvEmitterAlu.cpp

@@ -0,0 +1,690 @@
+#include "graphics/shader/recompiler/backend/spirv/spirvEmitterInternal.h"
+
+#include <array>
+
+namespace Libs::Graphics::ShaderRecompiler::Spirv::Emitter {
+namespace {
+
+struct Pair {
+	uint32_t low  = 0;
+	uint32_t high = 0;
+};
+
+uint32_t NewUnary(EmitterState& state, uint32_t opcode, uint32_t type, uint32_t value) {
+	const auto result = state.builder.AllocateId();
+	state.builder.AddFunction({opcode, type, result, value});
+	return result;
+}
+
+uint32_t NewBinary(EmitterState& state, uint32_t opcode, uint32_t type, uint32_t lhs,
+                   uint32_t rhs) {
+	const auto result = state.builder.AllocateId();
+	state.builder.AddFunction({opcode, type, result, lhs, rhs});
+	return result;
+}
+
+uint32_t NewSelect(EmitterState& state, uint32_t type, uint32_t condition, uint32_t true_value,
+                   uint32_t false_value) {
+	const auto result = state.builder.AllocateId();
+	state.builder.AddFunction({OpSelect, type, result, condition, true_value, false_value});
+	return result;
+}
+
+Pair ExtractPair(EmitterState& state, uint32_t value) {
+	Pair result {state.builder.AllocateId(), state.builder.AllocateId()};
+	state.builder.AddFunction({OpCompositeExtract, TypeU32(state), result.low, value, 0});
+	state.builder.AddFunction({OpCompositeExtract, TypeU32(state), result.high, value, 1});
+	return result;
+}
+
+uint32_t MakePair(EmitterState& state, uint32_t low, uint32_t high) {
+	const auto result = state.builder.AllocateId();
+	state.builder.AddFunction({OpCompositeConstruct, TypeU64(state), result, low, high});
+	return result;
+}
+
+uint32_t CompareEqual64(EmitterState& state, uint32_t lhs_value, uint32_t rhs_value,
+                        bool not_equal) {
+	const auto compare = NewBinary(state, not_equal ? OpINotEqual : OpIEqual,
+	                               TypeBoolVector(state, 2), lhs_value, rhs_value);
+	return NewUnary(state, not_equal ? OpAny : OpAll, TypeBool(state), compare);
+}
+
+uint32_t CompareOrdered64(EmitterState& state, uint32_t lhs_value, uint32_t rhs_value,
+                          uint32_t high_compare, uint32_t low_compare) {
+	const auto lhs         = ExtractPair(state, lhs_value);
+	const auto rhs         = ExtractPair(state, rhs_value);
+	const auto high_equal  = NewBinary(state, OpIEqual, TypeBool(state), lhs.high, rhs.high);
+	const auto high_result = NewBinary(state, high_compare, TypeBool(state), lhs.high, rhs.high);
+	const auto low_result  = NewBinary(state, low_compare, TypeBool(state), lhs.low, rhs.low);
+	const auto low_path = NewBinary(state, OpLogicalAnd, TypeBool(state), high_equal, low_result);
+	return NewBinary(state, OpLogicalOr, TypeBool(state), high_result, low_path);
+}
+
+uint32_t EmitAdd64(EmitterState& state, uint32_t lhs_value, uint32_t rhs_value) {
+	const auto lhs      = ExtractPair(state, lhs_value);
+	const auto rhs      = ExtractPair(state, rhs_value);
+	const auto low_pair = state.builder.AllocateId();
+	const auto low      = state.builder.AllocateId();
+	const auto carry    = state.builder.AllocateId();
+	state.builder.AddFunction({OpIAddCarry, TypeU32Pair(state), low_pair, lhs.low, rhs.low});
+	state.builder.AddFunction({OpCompositeExtract, TypeU32(state), low, low_pair, 0});
+	state.builder.AddFunction({OpCompositeExtract, TypeU32(state), carry, low_pair, 1});
+	const auto high0 = NewBinary(state, OpIAdd, TypeU32(state), lhs.high, rhs.high);
+	const auto high  = NewBinary(state, OpIAdd, TypeU32(state), high0, carry);
+	return MakePair(state, low, high);
+}
+
+uint32_t EmitSub64(EmitterState& state, uint32_t lhs_value, uint32_t rhs_value) {
+	const auto lhs    = ExtractPair(state, lhs_value);
+	const auto rhs    = ExtractPair(state, rhs_value);
+	const auto low    = NewBinary(state, OpISub, TypeU32(state), lhs.low, rhs.low);
+	const auto borrow = NewBinary(state, OpULessThan, TypeBool(state), lhs.low, rhs.low);
+	const auto borrow_u32 =
+	    NewSelect(state, TypeU32(state), borrow, ConstantU32(state, 1), ConstantU32(state, 0));
+	const auto high0 = NewBinary(state, OpISub, TypeU32(state), lhs.high, rhs.high);
+	const auto high  = NewBinary(state, OpISub, TypeU32(state), high0, borrow_u32);
+	return MakePair(state, low, high);
+}
+
+uint32_t EmitMulHigh(EmitterState& state, uint32_t lhs, uint32_t rhs, bool signed_value) {
+	const auto operand_type = signed_value ? TypeI32(state) : TypeU32(state);
+	const auto pair_type    = signed_value ? TypeI32Pair(state) : TypeU32Pair(state);
+	uint32_t   lhs_operand  = lhs;
+	uint32_t   rhs_operand  = rhs;
+	if (signed_value) {
+		lhs_operand = NewUnary(state, OpBitcast, TypeI32(state), lhs);
+		rhs_operand = NewUnary(state, OpBitcast, TypeI32(state), rhs);
+	}
+	const auto extended = state.builder.AllocateId();
+	state.builder.AddFunction({signed_value ? OpSMulExtended : OpUMulExtended, pair_type, extended,
+	                           lhs_operand, rhs_operand});
+	const auto high = state.builder.AllocateId();
+	state.builder.AddFunction({OpCompositeExtract, operand_type, high, extended, 1});
+	return signed_value ? NewUnary(state, OpBitcast, TypeU32(state), high) : high;
+}
+
+uint32_t EmitMul64(EmitterState& state, uint32_t lhs_value, uint32_t rhs_value) {
+	const auto lhs   = ExtractPair(state, lhs_value);
+	const auto rhs   = ExtractPair(state, rhs_value);
+	const auto low   = NewBinary(state, OpIMul, TypeU32(state), lhs.low, rhs.low);
+	const auto high0 = EmitMulHigh(state, lhs.low, rhs.low, false);
+	const auto high1 = NewBinary(state, OpIMul, TypeU32(state), lhs.low, rhs.high);
+	const auto high2 = NewBinary(state, OpIMul, TypeU32(state), lhs.high, rhs.low);
+	return MakePair(state, low,
+	                NewBinary(state, OpIAdd, TypeU32(state),
+	                          NewBinary(state, OpIAdd, TypeU32(state), high0, high1), high2));
+}
+
+uint32_t EmitShift64(EmitterState& state, uint32_t opcode, uint32_t value, uint32_t shift) {
+	const auto pair = ExtractPair(state, value);
+	uint32_t   low  = 0;
+	uint32_t   high = 0;
+	if (opcode == OpShiftLeftLogical) {
+		EmitShiftLeftLogicalU64Values(state, pair.low, pair.high, shift, low, high);
+	} else if (opcode == OpShiftRightLogical) {
+		EmitShiftRightLogicalU64Values(state, pair.low, pair.high, shift, low, high);
+	} else {
+		const auto amount =
+		    NewBinary(state, OpBitwiseAnd, TypeU32(state), shift, ConstantU32(state, 63));
+		const auto word_shift =
+		    NewBinary(state, OpBitwiseAnd, TypeU32(state), amount, ConstantU32(state, 31));
+		const auto at_least_32 =
+		    NewBinary(state, OpUGreaterThanEqual, TypeBool(state), amount, ConstantU32(state, 32));
+		const auto nonzero =
+		    NewBinary(state, OpINotEqual, TypeBool(state), amount, ConstantU32(state, 0));
+		const auto high_shifted =
+		    NewBinary(state, OpShiftRightArithmetic, TypeU32(state), pair.high, word_shift);
+		const auto low_shifted =
+		    NewBinary(state, OpShiftRightLogical, TypeU32(state), pair.low, word_shift);
+		const auto carry_count =
+		    NewBinary(state, OpBitwiseAnd, TypeU32(state),
+		              NewBinary(state, OpISub, TypeU32(state), ConstantU32(state, 32), word_shift),
+		              ConstantU32(state, 31));
+		const auto carry =
+		    NewSelect(state, TypeU32(state), nonzero,
+		              NewBinary(state, OpShiftLeftLogical, TypeU32(state), pair.high, carry_count),
+		              ConstantU32(state, 0));
+		const auto low_below_32 = NewBinary(state, OpBitwiseOr, TypeU32(state), low_shifted, carry);
+		const auto negative =
+		    NewBinary(state, OpSLessThan, TypeBool(state), pair.high, ConstantU32(state, 0));
+		const auto sign_fill = NewSelect(state, TypeU32(state), negative,
+		                                 ConstantU32(state, 0xffffffffu), ConstantU32(state, 0));
+		return MakePair(state,
+		                NewSelect(state, TypeU32(state), at_least_32, high_shifted, low_below_32),
+		                NewSelect(state, TypeU32(state), at_least_32, sign_fill, high_shifted));
+	}
+	return MakePair(state, low, high);
+}
+
+uint32_t EmitConstantShift64(EmitterState& state, uint32_t opcode, uint32_t value, uint32_t shift) {
+	shift &= 63u;
+	if (shift == 0u) {
+		return value;
+	}
+	const auto pair = ExtractPair(state, value);
+	if (opcode == OpShiftLeftLogical) {
+		if (shift < 32u) {
+			const auto low  = NewBinary(state, OpShiftLeftLogical, TypeU32(state), pair.low,
+			                            ConstantU32(state, shift));
+			const auto high = NewBinary(state, OpBitwiseOr, TypeU32(state),
+			                            NewBinary(state, OpShiftLeftLogical, TypeU32(state),
+			                                      pair.high, ConstantU32(state, shift)),
+			                            NewBinary(state, OpShiftRightLogical, TypeU32(state),
+			                                      pair.low, ConstantU32(state, 32u - shift)));
+			return MakePair(state, low, high);
+		}
+		return MakePair(state, ConstantU32(state, 0),
+		                shift == 32u ? pair.low
+		                             : NewBinary(state, OpShiftLeftLogical, TypeU32(state),
+		                                         pair.low, ConstantU32(state, shift - 32u)));
+	}
+	if (shift < 32u) {
+		const auto low = NewBinary(state, OpBitwiseOr, TypeU32(state),
+		                           NewBinary(state, OpShiftRightLogical, TypeU32(state), pair.low,
+		                                     ConstantU32(state, shift)),
+		                           NewBinary(state, OpShiftLeftLogical, TypeU32(state), pair.high,
+		                                     ConstantU32(state, 32u - shift)));
+		const auto high =
+		    NewBinary(state, opcode, TypeU32(state), pair.high, ConstantU32(state, shift));
+		return MakePair(state, low, high);
+	}
+	const auto high = opcode == OpShiftRightArithmetic
+	                      ? NewBinary(state, OpShiftRightArithmetic, TypeU32(state), pair.high,
+	                                  ConstantU32(state, 31u))
+	                      : ConstantU32(state, 0);
+	const auto low  = shift == 32u ? pair.high
+	                               : NewBinary(state, opcode, TypeU32(state), pair.high,
+	                                           ConstantU32(state, shift - 32u));
+	return MakePair(state, low, high);
+}
+
+uint32_t EmitFindMsb64(EmitterState& state, uint32_t value) {
+	const auto pair   = ExtractPair(state, value);
+	const auto high_i = state.builder.AllocateId();
+	const auto low_i  = state.builder.AllocateId();
+	state.builder.AddFunction(
+	    {OpExtInst, TypeI32(state), high_i, GlslStd450(state), GlslFindUMsb, pair.high});
+	state.builder.AddFunction(
+	    {OpExtInst, TypeI32(state), low_i, GlslStd450(state), GlslFindUMsb, pair.low});
+	const auto high = NewUnary(state, OpBitcast, TypeU32(state), high_i);
+	const auto low  = NewUnary(state, OpBitcast, TypeU32(state), low_i);
+	const auto high_nonzero =
+	    NewBinary(state, OpINotEqual, TypeBool(state), pair.high, ConstantU32(state, 0));
+	return NewSelect(state, TypeU32(state), high_nonzero,
+	                 NewBinary(state, OpIAdd, TypeU32(state), high, ConstantU32(state, 32)), low);
+}
+
+uint32_t EmitMinMax3(EmitterState& state, uint32_t a, uint32_t b, uint32_t c, bool signed_value,
+                     bool max_value) {
+	const auto ab = signed_value ? EmitMinMaxI32Value(state, a, b, max_value)
+	                             : EmitMinMaxU32Value(state, a, b, max_value);
+	return signed_value ? EmitMinMaxI32Value(state, ab, c, max_value)
+	                    : EmitMinMaxU32Value(state, ab, c, max_value);
+}
+
+uint32_t EmitMed3(EmitterState& state, uint32_t a, uint32_t b, uint32_t c, bool signed_value) {
+	const auto minimum = EmitMinMax3(state, a, b, c, signed_value, false);
+	const auto maximum = EmitMinMax3(state, a, b, c, signed_value, true);
+	const auto ab      = NewBinary(state, OpIAdd, TypeU32(state), a, b);
+	const auto abc     = NewBinary(state, OpIAdd, TypeU32(state), ab, c);
+	return NewBinary(state, OpISub, TypeU32(state),
+	                 NewBinary(state, OpISub, TypeU32(state), abc, minimum), maximum);
+}
+
+uint32_t EmitFMinMax3(EmitterState& state, uint32_t a, uint32_t b, uint32_t c, bool max_value) {
+	return EmitMinMaxF32Value(state, EmitMinMaxF32Value(state, a, b, max_value), c, max_value);
+}
+
+uint32_t EmitFMed3(EmitterState& state, uint32_t a, uint32_t b, uint32_t c) {
+	const auto min_ab   = EmitMinMaxF32Value(state, a, b, false);
+	const auto min3     = EmitMinMaxF32Value(state, min_ab, c, false);
+	const auto max_ab   = EmitMinMaxF32Value(state, a, b, true);
+	const auto high_min = EmitMinMaxF32Value(state, max_ab, c, false);
+	const auto median   = EmitMinMaxF32Value(state, min_ab, high_min, true);
+	const auto nan_ab   = NewBinary(state, OpLogicalOr, TypeBool(state),
+	                                EmitClassifyF32(state, a).nan, EmitClassifyF32(state, b).nan);
+	const auto any_nan =
+	    NewBinary(state, OpLogicalOr, TypeBool(state), nan_ab, EmitClassifyF32(state, c).nan);
+	return NewSelect(state, TypeF32(state), any_nan, min3, median);
+}
+
+uint32_t EmitExt(EmitterState& state, uint32_t type, uint32_t opcode,
+                 std::initializer_list<uint32_t> args) {
+	const auto            result = state.builder.AllocateId();
+	std::vector<uint32_t> words {OpExtInst, type, result, GlslStd450(state), opcode};
+	words.insert(words.end(), args.begin(), args.end());
+	state.builder.AddFunction(words);
+	return result;
+}
+
+uint32_t EmitF32ToU32(EmitterState& state, uint32_t src, bool signed_value) {
+	const auto trunc         = EmitTruncF32Value(state, src);
+	const auto converted_raw = state.builder.AllocateId();
+	if (signed_value) {
+		const auto converted_i = state.builder.AllocateId();
+		state.builder.AddFunction({OpConvertFToS, TypeI32(state), converted_i, trunc});
+		state.builder.AddFunction({OpBitcast, TypeU32(state), converted_raw, converted_i});
+	} else {
+		state.builder.AddFunction({OpConvertFToU, TypeU32(state), converted_raw, trunc});
+	}
+	const auto nan = EmitClassifyF32(state, src).nan;
+	if (signed_value) {
+		const auto below = NewBinary(state, OpFOrdLessThanEqual, TypeBool(state), src,
+		                             ConstantF32(state, 0xcf000000u));
+		const auto above = NewBinary(state, OpFOrdGreaterThanEqual, TypeBool(state), src,
+		                             ConstantF32(state, 0x4f000000u));
+		const auto high =
+		    NewSelect(state, TypeU32(state), above, ConstantU32(state, 0x7fffffffu), converted_raw);
+		const auto low =
+		    NewSelect(state, TypeU32(state), below, ConstantU32(state, 0x80000000u), high);
+		return NewSelect(state, TypeU32(state), nan, ConstantU32(state, 0), low);
+	}
+	const auto below =
+	    NewBinary(state, OpFOrdLessThanEqual, TypeBool(state), src, ConstantF32(state, 0));
+	const auto above = NewBinary(state, OpFOrdGreaterThanEqual, TypeBool(state), src,
+	                             ConstantF32(state, 0x4f800000u));
+	const auto zero  = NewBinary(state, OpLogicalOr, TypeBool(state), nan, below);
+	const auto high =
+	    NewSelect(state, TypeU32(state), above, ConstantU32(state, 0xffffffffu), converted_raw);
+	return NewSelect(state, TypeU32(state), zero, ConstantU32(state, 0), high);
+}
+
+uint32_t EmitPackHalf(EmitterState& state, uint32_t src) {
+	return EmitExt(state, TypeU32(state), GlslPackHalf2x16, {src});
+}
+
+uint32_t EmitUnpackHalf(EmitterState& state, uint32_t bits) {
+	const auto pair   = EmitExt(state, TypeF32Vector(state, 2), GlslUnpackHalf2x16, {bits});
+	const auto result = state.builder.AllocateId();
+	state.builder.AddFunction({OpCompositeExtract, TypeF32(state), result, pair, 0});
+	return result;
+}
+
+} // namespace
+
+bool EmitValueAlu(ValueEmitContext& ctx, const IR::Inst& inst) {
+	auto&      state = ctx.state;
+	const auto op    = inst.GetOpcode();
+	const auto unary = [&](uint32_t spirv, IR::Type type) {
+		ctx.Emit(inst, spirv, type, {ctx.Arg(inst, 0)});
+		return true;
+	};
+	const auto binary = [&](uint32_t spirv, IR::Type type) {
+		ctx.Emit(inst, spirv, type, {ctx.Arg(inst, 0), ctx.Arg(inst, 1)});
+		return true;
+	};
+	const auto ext_unary = [&](uint32_t ext) {
+		ctx.Define(inst, EmitExt(state, TypeF32(state), ext, {ctx.Arg(inst, 0)}));
+		return true;
+	};
+	switch (op) {
+		case IR::ValueOpcode::BitCastU16F16:
+		case IR::ValueOpcode::BitCastF16U16:
+		case IR::ValueOpcode::ConvertU32U16:
+		case IR::ValueOpcode::ConvertU32U8:
+		case IR::ValueOpcode::BitCastU32F32: return unary(OpBitcast, IR::Type::U32);
+		case IR::ValueOpcode::BitCastF32U32: return unary(OpBitcast, IR::Type::F32);
+		case IR::ValueOpcode::ConvertU16U32:
+			ctx.Emit(inst, OpBitwiseAnd, IR::Type::U16,
+			         {ctx.Arg(inst, 0), ConstantU32(state, 0xffffu)});
+			return true;
+		case IR::ValueOpcode::ConvertU8U32:
+			ctx.Emit(inst, OpBitwiseAnd, IR::Type::U8,
+			         {ctx.Arg(inst, 0), ConstantU32(state, 0xffu)});
+			return true;
+		case IR::ValueOpcode::ConvertF16F32: {
+			const auto pair = state.builder.AllocateId();
+			state.builder.AddFunction({OpCompositeConstruct, TypeF32Vector(state, 2), pair,
+			                           ctx.Arg(inst, 0), ConstantF32(state, 0)});
+			ctx.Define(inst, EmitPackHalf(state, pair));
+			return true;
+		}
+		case IR::ValueOpcode::ConvertF32F16:
+			ctx.Define(inst, EmitUnpackHalf(state, ctx.Arg(inst, 0)));
+			return true;
+		case IR::ValueOpcode::ConvertS32F32:
+			ctx.Define(inst, EmitF32ToU32(state, ctx.Arg(inst, 0), true));
+			return true;
+		case IR::ValueOpcode::ConvertU32F32:
+			ctx.Define(inst, EmitF32ToU32(state, ctx.Arg(inst, 0), false));
+			return true;
+		case IR::ValueOpcode::ConvertF32S32: {
+			const auto signed_value = NewUnary(state, OpBitcast, TypeI32(state), ctx.Arg(inst, 0));
+			ctx.Emit(inst, OpConvertSToF, IR::Type::F32, {signed_value});
+			return true;
+		}
+		case IR::ValueOpcode::ConvertF32U32: return unary(OpConvertUToF, IR::Type::F32);
+		case IR::ValueOpcode::CompositeConstructU64:
+			ctx.Emit(inst, OpCompositeConstruct, IR::Type::U64,
+			         {ctx.Arg(inst, 0), ctx.Arg(inst, 1)});
+			return true;
+		case IR::ValueOpcode::CompositeConstructU32x2:
+			ctx.Emit(inst, OpCompositeConstruct, IR::Type::U32x2,
+			         {ctx.Arg(inst, 0), ctx.Arg(inst, 1)});
+			return true;
+		case IR::ValueOpcode::CompositeConstructU32x3:
+			ctx.Emit(inst, OpCompositeConstruct, IR::Type::U32x3,
+			         {ctx.Arg(inst, 0), ctx.Arg(inst, 1), ctx.Arg(inst, 2)});
+			return true;
+		case IR::ValueOpcode::CompositeConstructF32x2:
+			ctx.Emit(inst, OpCompositeConstruct, IR::Type::F32x2,
+			         {ctx.Arg(inst, 0), ctx.Arg(inst, 1)});
+			return true;
+		case IR::ValueOpcode::CompositeConstructU32x4:
+			ctx.Emit(inst, OpCompositeConstruct, IR::Type::U32x4,
+			         {ctx.Arg(inst, 0), ctx.Arg(inst, 1), ctx.Arg(inst, 2), ctx.Arg(inst, 3)});
+			return true;
+		case IR::ValueOpcode::CompositeExtractU64:
+		case IR::ValueOpcode::CompositeExtractU32x2:
+		case IR::ValueOpcode::CompositeExtractU32x3:
+		case IR::ValueOpcode::CompositeExtractU32x4:
+			ctx.Emit(inst, OpCompositeExtract, IR::Type::U32,
+			         {ctx.Arg(inst, 0), inst.Arg(1).U32()});
+			return true;
+		case IR::ValueOpcode::PackHalf2x16:
+			ctx.Define(inst, EmitExt(state, TypeU32(state), GlslPackHalf2x16, {ctx.Arg(inst, 0)}));
+			return true;
+		case IR::ValueOpcode::PackSnorm2x16:
+			ctx.Define(inst, EmitExt(state, TypeU32(state), GlslPackSnorm2x16, {ctx.Arg(inst, 0)}));
+			return true;
+		case IR::ValueOpcode::PackUnorm2x16:
+			ctx.Define(inst, EmitExt(state, TypeU32(state), GlslPackUnorm2x16, {ctx.Arg(inst, 0)}));
+			return true;
+		case IR::ValueOpcode::PackFloat2x16Rtz: {
+			const auto low = EmitF32ToF16RtzBits(state, ctx.Arg(inst, 0));
+			const auto high =
+			    NewBinary(state, OpShiftLeftLogical, TypeU32(state),
+			              EmitF32ToF16RtzBits(state, ctx.Arg(inst, 1)), ConstantU32(state, 16));
+			ctx.Define(inst, NewBinary(state, OpBitwiseOr, TypeU32(state), low, high));
+			return true;
+		}
+		case IR::ValueOpcode::FPAbs32:
+			ctx.Define(inst, EmitFAbsValue(state, ctx.Arg(inst, 0)));
+			return true;
+		case IR::ValueOpcode::FPNeg32:
+			ctx.Define(inst, EmitFNegateValue(state, ctx.Arg(inst, 0)));
+			return true;
+		case IR::ValueOpcode::FPSaturate32:
+			ctx.Define(inst, EmitExt(state, TypeF32(state), GlslFClamp,
+			                         {ctx.Arg(inst, 0), ConstantF32(state, 0),
+			                          ConstantF32(state, 0x3f800000u)}));
+			return true;
+		case IR::ValueOpcode::BitFieldInsert:
+			ctx.Emit(inst, OpBitFieldInsert, IR::Type::U32,
+			         {ctx.Arg(inst, 0), ctx.Arg(inst, 1), ctx.Arg(inst, 2), ctx.Arg(inst, 3)});
+			return true;
+		case IR::ValueOpcode::BitFieldUExtract:
+			ctx.Emit(inst, OpBitFieldUExtract, IR::Type::U32,
+			         {ctx.Arg(inst, 0), ctx.Arg(inst, 1), ctx.Arg(inst, 2)});
+			return true;
+		case IR::ValueOpcode::BitFieldSExtract:
+			ctx.Emit(inst, OpBitFieldSExtract, IR::Type::U32,
+			         {ctx.Arg(inst, 0), ctx.Arg(inst, 1), ctx.Arg(inst, 2)});
+			return true;
+		case IR::ValueOpcode::SelectU1:
+			ctx.Emit(inst, OpSelect, IR::Type::U1,
+			         {ctx.Arg(inst, 0), ctx.Arg(inst, 1), ctx.Arg(inst, 2)});
+			return true;
+		case IR::ValueOpcode::SelectU32:
+			ctx.Emit(inst, OpSelect, IR::Type::U32,
+			         {ctx.Arg(inst, 0), ctx.Arg(inst, 1), ctx.Arg(inst, 2)});
+			return true;
+		case IR::ValueOpcode::SelectF32:
+			ctx.Emit(inst, OpSelect, IR::Type::F32,
+			         {ctx.Arg(inst, 0), ctx.Arg(inst, 1), ctx.Arg(inst, 2)});
+			return true;
+		case IR::ValueOpcode::IAdd32: return binary(OpIAdd, IR::Type::U32);
+		case IR::ValueOpcode::ISub32: return binary(OpISub, IR::Type::U32);
+		case IR::ValueOpcode::IMul32: return binary(OpIMul, IR::Type::U32);
+		case IR::ValueOpcode::UDiv32: return binary(OpUDiv, IR::Type::U32);
+		case IR::ValueOpcode::IAdd64:
+			ctx.Define(inst, EmitAdd64(state, ctx.Arg(inst, 0), ctx.Arg(inst, 1)));
+			return true;
+		case IR::ValueOpcode::ISub64:
+			ctx.Define(inst, EmitSub64(state, ctx.Arg(inst, 0), ctx.Arg(inst, 1)));
+			return true;
+		case IR::ValueOpcode::IMul64:
+			ctx.Define(inst, EmitMul64(state, ctx.Arg(inst, 0), ctx.Arg(inst, 1)));
+			return true;
+		case IR::ValueOpcode::IAddCarry32:
+			ctx.Emit(inst, OpIAddCarry, IR::Type::U32x2, {ctx.Arg(inst, 0), ctx.Arg(inst, 1)});
+			return true;
+		case IR::ValueOpcode::SMulHi:
+			ctx.Define(inst, EmitMulHigh(state, ctx.Arg(inst, 0), ctx.Arg(inst, 1), true));
+			return true;
+		case IR::ValueOpcode::UMulHi:
+			ctx.Define(inst, EmitMulHigh(state, ctx.Arg(inst, 0), ctx.Arg(inst, 1), false));
+			return true;
+		case IR::ValueOpcode::IAbs32: {
+			const auto value = ctx.Arg(inst, 0);
+			const auto neg   = NewUnary(state, OpSNegate, TypeU32(state), value);
+			const auto negative =
+			    NewBinary(state, OpSLessThan, TypeBool(state), value, ConstantU32(state, 0));
+			ctx.Define(inst, NewSelect(state, TypeU32(state), negative, neg, value));
+			return true;
+		}
+		case IR::ValueOpcode::ShiftLeftLogical32: return binary(OpShiftLeftLogical, IR::Type::U32);
+		case IR::ValueOpcode::ShiftRightLogical32:
+			return binary(OpShiftRightLogical, IR::Type::U32);
+		case IR::ValueOpcode::ShiftRightArithmetic32:
+			return binary(OpShiftRightArithmetic, IR::Type::U32);
+		case IR::ValueOpcode::ShiftLeftLogical64:
+			ctx.Define(inst, inst.Arg(1).Resolve().IsImmediate()
+			                     ? EmitConstantShift64(state, OpShiftLeftLogical, ctx.Arg(inst, 0),
+			                                           inst.Arg(1).Resolve().U32())
+			                     : EmitShift64(state, OpShiftLeftLogical, ctx.Arg(inst, 0),
+			                                   ctx.Arg(inst, 1)));
+			return true;
+		case IR::ValueOpcode::ShiftRightLogical64:
+			ctx.Define(inst, inst.Arg(1).Resolve().IsImmediate()
+			                     ? EmitConstantShift64(state, OpShiftRightLogical, ctx.Arg(inst, 0),
+			                                           inst.Arg(1).Resolve().U32())
+			                     : EmitShift64(state, OpShiftRightLogical, ctx.Arg(inst, 0),
+			                                   ctx.Arg(inst, 1)));
+			return true;
+		case IR::ValueOpcode::ShiftRightArithmetic64:
+			ctx.Define(inst,
+			           inst.Arg(1).Resolve().IsImmediate()
+			               ? EmitConstantShift64(state, OpShiftRightArithmetic, ctx.Arg(inst, 0),
+			                                     inst.Arg(1).Resolve().U32())
+			               : EmitShift64(state, OpShiftRightArithmetic, ctx.Arg(inst, 0),
+			                             ctx.Arg(inst, 1)));
+			return true;
+		case IR::ValueOpcode::BitwiseAnd32: return binary(OpBitwiseAnd, IR::Type::U32);
+		case IR::ValueOpcode::BitwiseOr32: return binary(OpBitwiseOr, IR::Type::U32);
+		case IR::ValueOpcode::BitwiseXor32: return binary(OpBitwiseXor, IR::Type::U32);
+		case IR::ValueOpcode::BitwiseNot32: return unary(OpNot, IR::Type::U32);
+		case IR::ValueOpcode::BitwiseAnd64:
+			ctx.Define(inst, NewBinary(state, OpBitwiseAnd, TypeU64(state), ctx.Arg(inst, 0),
+			                           ctx.Arg(inst, 1)));
+			return true;
+		case IR::ValueOpcode::BitReverse32: return unary(OpBitReverse, IR::Type::U32);
+		case IR::ValueOpcode::BitCount32: return unary(OpBitCount, IR::Type::U32);
+		case IR::ValueOpcode::BitCount64: {
+			const auto pair =
+			    ExtractPair(state, NewUnary(state, OpBitCount, TypeU64(state), ctx.Arg(inst, 0)));
+			ctx.Define(inst, NewBinary(state, OpIAdd, TypeU32(state), pair.low, pair.high));
+			return true;
+		}
+		case IR::ValueOpcode::FindILsb32: {
+			const auto value = EmitExt(state, TypeI32(state), GlslFindILsb, {ctx.Arg(inst, 0)});
+			ctx.Define(inst, NewUnary(state, OpBitcast, TypeU32(state), value));
+			return true;
+		}
+		case IR::ValueOpcode::FindUMsb32: {
+			const auto value = EmitExt(state, TypeI32(state), GlslFindUMsb, {ctx.Arg(inst, 0)});
+			ctx.Define(inst, NewUnary(state, OpBitcast, TypeU32(state), value));
+			return true;
+		}
+		case IR::ValueOpcode::FindUMsb64:
+			ctx.Define(inst, EmitFindMsb64(state, ctx.Arg(inst, 0)));
+			return true;
+		case IR::ValueOpcode::SMin32:
+			ctx.Define(inst, EmitMinMaxI32Value(state, ctx.Arg(inst, 0), ctx.Arg(inst, 1), false));
+			return true;
+		case IR::ValueOpcode::SMax32:
+			ctx.Define(inst, EmitMinMaxI32Value(state, ctx.Arg(inst, 0), ctx.Arg(inst, 1), true));
+			return true;
+		case IR::ValueOpcode::UMin32:
+			ctx.Define(inst, EmitMinMaxU32Value(state, ctx.Arg(inst, 0), ctx.Arg(inst, 1), false));
+			return true;
+		case IR::ValueOpcode::UMax32:
+			ctx.Define(inst, EmitMinMaxU32Value(state, ctx.Arg(inst, 0), ctx.Arg(inst, 1), true));
+			return true;
+		case IR::ValueOpcode::SMinTri32:
+			ctx.Define(inst, EmitMinMax3(state, ctx.Arg(inst, 0), ctx.Arg(inst, 1),
+			                             ctx.Arg(inst, 2), true, false));
+			return true;
+		case IR::ValueOpcode::SMaxTri32:
+			ctx.Define(inst, EmitMinMax3(state, ctx.Arg(inst, 0), ctx.Arg(inst, 1),
+			                             ctx.Arg(inst, 2), true, true));
+			return true;
+		case IR::ValueOpcode::UMinTri32:
+			ctx.Define(inst, EmitMinMax3(state, ctx.Arg(inst, 0), ctx.Arg(inst, 1),
+			                             ctx.Arg(inst, 2), false, false));
+			return true;
+		case IR::ValueOpcode::UMaxTri32:
+			ctx.Define(inst, EmitMinMax3(state, ctx.Arg(inst, 0), ctx.Arg(inst, 1),
+			                             ctx.Arg(inst, 2), false, true));
+			return true;
+		case IR::ValueOpcode::SMedTri32:
+			ctx.Define(inst,
+			           EmitMed3(state, ctx.Arg(inst, 0), ctx.Arg(inst, 1), ctx.Arg(inst, 2), true));
+			return true;
+		case IR::ValueOpcode::UMedTri32:
+			ctx.Define(
+			    inst, EmitMed3(state, ctx.Arg(inst, 0), ctx.Arg(inst, 1), ctx.Arg(inst, 2), false));
+			return true;
+		case IR::ValueOpcode::SLessThan32: return binary(OpSLessThan, IR::Type::U1);
+		case IR::ValueOpcode::ULessThan32: return binary(OpULessThan, IR::Type::U1);
+		case IR::ValueOpcode::IEqual32: return binary(OpIEqual, IR::Type::U1);
+		case IR::ValueOpcode::SLessThanEqual32: return binary(OpSLessThanEqual, IR::Type::U1);
+		case IR::ValueOpcode::ULessThanEqual32: return binary(OpULessThanEqual, IR::Type::U1);
+		case IR::ValueOpcode::SGreaterThan32: return binary(OpSGreaterThan, IR::Type::U1);
+		case IR::ValueOpcode::UGreaterThan32: return binary(OpUGreaterThan, IR::Type::U1);
+		case IR::ValueOpcode::INotEqual32: return binary(OpINotEqual, IR::Type::U1);
+		case IR::ValueOpcode::SGreaterThanEqual32: return binary(OpSGreaterThanEqual, IR::Type::U1);
+		case IR::ValueOpcode::UGreaterThanEqual32: return binary(OpUGreaterThanEqual, IR::Type::U1);
+		case IR::ValueOpcode::IEqual64:
+			ctx.Define(inst, CompareEqual64(state, ctx.Arg(inst, 0), ctx.Arg(inst, 1), false));
+			return true;
+		case IR::ValueOpcode::INotEqual64:
+			ctx.Define(inst, CompareEqual64(state, ctx.Arg(inst, 0), ctx.Arg(inst, 1), true));
+			return true;
+		case IR::ValueOpcode::ULessThan64:
+			ctx.Define(inst, CompareOrdered64(state, ctx.Arg(inst, 0), ctx.Arg(inst, 1),
+			                                  OpULessThan, OpULessThan));
+			return true;
+		case IR::ValueOpcode::SLessThan64:
+			ctx.Define(inst, CompareOrdered64(state, ctx.Arg(inst, 0), ctx.Arg(inst, 1),
+			                                  OpSLessThan, OpULessThan));
+			return true;
+		case IR::ValueOpcode::UGreaterThan64:
+			ctx.Define(inst, CompareOrdered64(state, ctx.Arg(inst, 0), ctx.Arg(inst, 1),
+			                                  OpUGreaterThan, OpUGreaterThan));
+			return true;
+		case IR::ValueOpcode::LogicalOr: return binary(OpLogicalOr, IR::Type::U1);
+		case IR::ValueOpcode::LogicalAnd: return binary(OpLogicalAnd, IR::Type::U1);
+		case IR::ValueOpcode::LogicalXor: return binary(OpLogicalNotEqual, IR::Type::U1);
+		case IR::ValueOpcode::LogicalNot: return unary(OpLogicalNot, IR::Type::U1);
+		case IR::ValueOpcode::FPOrdEqual32: return binary(OpFOrdEqual, IR::Type::U1);
+		case IR::ValueOpcode::FPUnordEqual32: return binary(OpFUnordEqual, IR::Type::U1);
+		case IR::ValueOpcode::FPOrdNotEqual32: return binary(OpFOrdNotEqual, IR::Type::U1);
+		case IR::ValueOpcode::FPUnordNotEqual32: return binary(OpFUnordNotEqual, IR::Type::U1);
+		case IR::ValueOpcode::FPOrdLessThan32: return binary(OpFOrdLessThan, IR::Type::U1);
+		case IR::ValueOpcode::FPUnordLessThan32: return binary(OpFUnordLessThan, IR::Type::U1);
+		case IR::ValueOpcode::FPOrdGreaterThan32: return binary(OpFOrdGreaterThan, IR::Type::U1);
+		case IR::ValueOpcode::FPUnordGreaterThan32:
+			return binary(OpFUnordGreaterThan, IR::Type::U1);
+		case IR::ValueOpcode::FPOrdLessThanEqual32:
+			return binary(OpFOrdLessThanEqual, IR::Type::U1);
+		case IR::ValueOpcode::FPUnordLessThanEqual32:
+			return binary(OpFUnordLessThanEqual, IR::Type::U1);
+		case IR::ValueOpcode::FPOrdGreaterThanEqual32:
+			return binary(OpFOrdGreaterThanEqual, IR::Type::U1);
+		case IR::ValueOpcode::FPUnordGreaterThanEqual32:
+			return binary(OpFUnordGreaterThanEqual, IR::Type::U1);
+		case IR::ValueOpcode::FPIsNan32:
+			ctx.Emit(inst, OpFUnordNotEqual, IR::Type::U1, {ctx.Arg(inst, 0), ctx.Arg(inst, 0)});
+			return true;
+		case IR::ValueOpcode::FPCmpClass32:
+			ctx.Define(inst, EmitClassMaskF32(state, ctx.Arg(inst, 0), ctx.Arg(inst, 1)));
+			return true;
+		case IR::ValueOpcode::FPAdd32: return binary(OpFAdd, IR::Type::F32);
+		case IR::ValueOpcode::FPSub32: return binary(OpFSub, IR::Type::F32);
+		case IR::ValueOpcode::FPMul32: return binary(OpFMul, IR::Type::F32);
+		case IR::ValueOpcode::FPFma32:
+			ctx.Define(inst, EmitExt(state, TypeF32(state), GlslFma,
+			                         {ctx.Arg(inst, 0), ctx.Arg(inst, 1), ctx.Arg(inst, 2)}));
+			return true;
+		case IR::ValueOpcode::FPMin32:
+			ctx.Define(inst, EmitMinMaxF32Value(state, ctx.Arg(inst, 0), ctx.Arg(inst, 1), false));
+			return true;
+		case IR::ValueOpcode::FPMax32:
+			ctx.Define(inst, EmitMinMaxF32Value(state, ctx.Arg(inst, 0), ctx.Arg(inst, 1), true));
+			return true;
+		case IR::ValueOpcode::FPMinTri32:
+			ctx.Define(inst, EmitFMinMax3(state, ctx.Arg(inst, 0), ctx.Arg(inst, 1),
+			                              ctx.Arg(inst, 2), false));
+			return true;
+		case IR::ValueOpcode::FPMaxTri32:
+			ctx.Define(inst, EmitFMinMax3(state, ctx.Arg(inst, 0), ctx.Arg(inst, 1),
+			                              ctx.Arg(inst, 2), true));
+			return true;
+		case IR::ValueOpcode::FPMedTri32:
+			ctx.Define(inst,
+			           EmitFMed3(state, ctx.Arg(inst, 0), ctx.Arg(inst, 1), ctx.Arg(inst, 2)));
+			return true;
+		case IR::ValueOpcode::FPRecip32: {
+			const auto source = EmitFlushF32DenormToSignedZero(state, ctx.Arg(inst, 0));
+			ctx.Define(inst, NewBinary(state, OpFDiv, TypeF32(state),
+			                           ConstantF32(state, 0x3f800000u), source));
+			return true;
+		}
+		case IR::ValueOpcode::FPRecipIFlag32:
+			// Integer-to-float inputs used by IFLAG cannot be denormal.
+			ctx.Define(inst, NewBinary(state, OpFDiv, TypeF32(state),
+			                           ConstantF32(state, 0x3f800000u), ctx.Arg(inst, 0)));
+			return true;
+		case IR::ValueOpcode::FPRecipSqrt32:
+			ctx.Define(inst, EmitExt(state, TypeF32(state), GlslInverseSqrt,
+			                         {EmitFlushF32DenormToSignedZero(state, ctx.Arg(inst, 0))}));
+			return true;
+		case IR::ValueOpcode::FPSqrt:
+			ctx.Define(inst, EmitExt(state, TypeF32(state), GlslSqrt,
+			                         {EmitFlushF32DenormToSignedZero(state, ctx.Arg(inst, 0))}));
+			return true;
+		case IR::ValueOpcode::FPSin:
+		case IR::ValueOpcode::FPCos: {
+			auto source = EmitTrigCycleF32(state, ctx.Arg(inst, 0), op == IR::ValueOpcode::FPSin);
+			source =
+			    NewBinary(state, OpFMul, TypeF32(state), source, ConstantF32(state, 0x40c90fdbu));
+			ctx.Define(inst, EmitExt(state, TypeF32(state),
+			                         op == IR::ValueOpcode::FPSin ? GlslSin : GlslCos, {source}));
+			return true;
+		}
+		case IR::ValueOpcode::FPExp2:
+			ctx.Define(inst, EmitExt(state, TypeF32(state), GlslExp2,
+			                         {EmitFlushF32DenormToSignedZero(state, ctx.Arg(inst, 0))}));
+			return true;
+		case IR::ValueOpcode::FPLog2:
+			ctx.Define(inst, EmitExt(state, TypeF32(state), GlslLog2,
+			                         {EmitFlushF32DenormToSignedZero(state, ctx.Arg(inst, 0))}));
+			return true;
+		case IR::ValueOpcode::FPLdexp: {
+			const auto exponent = NewUnary(state, OpBitcast, TypeI32(state), ctx.Arg(inst, 1));
+			ctx.Define(inst,
+			           EmitExt(state, TypeF32(state), GlslLdexp, {ctx.Arg(inst, 0), exponent}));
+			return true;
+		}
+		case IR::ValueOpcode::FPRoundEven32: return ext_unary(GlslRoundEven);
+		case IR::ValueOpcode::FPFloor32: return ext_unary(GlslFloor);
+		case IR::ValueOpcode::FPCeil32: return ext_unary(GlslCeil);
+		case IR::ValueOpcode::FPTrunc32: return ext_unary(GlslTrunc);
+		case IR::ValueOpcode::FPFract32: return ext_unary(GlslFract);
+		default: return false;
+	}
+}
+
+} // namespace Libs::Graphics::ShaderRecompiler::Spirv::Emitter

src/graphics/shader/recompiler/backend/spirv/spirvEmitterAluHelpers.cpp

@@ -0,0 +1,393 @@
+#include "graphics/shader/recompiler/backend/spirv/spirvEmitterInternal.h"
+
+namespace Libs::Graphics::ShaderRecompiler::Spirv::Emitter {
+
+uint32_t EmitSelectValueU32(EmitterState& state, uint32_t cond, uint32_t true_value,
+                            uint32_t false_value) {
+	const auto ret = state.builder.AllocateId();
+	state.builder.AddFunction({OpSelect, TypeU32(state), ret, cond, true_value, false_value});
+	return ret;
+}
+
+uint32_t EmitAndConstant(EmitterState& state, uint32_t value, uint32_t mask);
+
+void EmitShiftLeftLogicalU64Values(EmitterState& state, uint32_t low, uint32_t high, uint32_t shift,
+                                   uint32_t& out_low, uint32_t& out_high) {
+	const auto amount           = EmitAndConstant(state, shift, 63u);
+	const auto dword_shift      = EmitAndConstant(state, amount, 31u);
+	const auto less32           = state.builder.AllocateId();
+	const auto non_zero         = state.builder.AllocateId();
+	const auto carry_count_base = state.builder.AllocateId();
+	const auto carry_count      = state.builder.AllocateId();
+	const auto low_part         = state.builder.AllocateId();
+	const auto high_part        = state.builder.AllocateId();
+	const auto carry_part       = state.builder.AllocateId();
+	const auto carry            = state.builder.AllocateId();
+	const auto high_less        = state.builder.AllocateId();
+	const auto high_ge          = state.builder.AllocateId();
+	state.builder.AddFunction(
+	    {OpULessThan, TypeBool(state), less32, amount, ConstantU32(state, 32)});
+	state.builder.AddFunction(
+	    {OpINotEqual, TypeBool(state), non_zero, amount, ConstantU32(state, 0)});
+	state.builder.AddFunction(
+	    {OpISub, TypeU32(state), carry_count_base, ConstantU32(state, 32), dword_shift});
+	state.builder.AddFunction(
+	    {OpBitwiseAnd, TypeU32(state), carry_count, carry_count_base, ConstantU32(state, 31)});
+	state.builder.AddFunction({OpShiftLeftLogical, TypeU32(state), low_part, low, dword_shift});
+	state.builder.AddFunction({OpShiftLeftLogical, TypeU32(state), high_part, high, dword_shift});
+	state.builder.AddFunction({OpShiftRightLogical, TypeU32(state), carry_part, low, carry_count});
+	state.builder.AddFunction(
+	    {OpSelect, TypeU32(state), carry, non_zero, carry_part, ConstantU32(state, 0)});
+	state.builder.AddFunction({OpBitwiseOr, TypeU32(state), high_less, high_part, carry});
+	state.builder.AddFunction({OpShiftLeftLogical, TypeU32(state), high_ge, low, dword_shift});
+	out_low  = EmitSelectValueU32(state, less32, low_part, ConstantU32(state, 0));
+	out_high = EmitSelectValueU32(state, less32, high_less, high_ge);
+}
+
+void EmitShiftRightLogicalU64Values(EmitterState& state, uint32_t low, uint32_t high,
+                                    uint32_t shift, uint32_t& out_low, uint32_t& out_high) {
+	const auto amount           = EmitAndConstant(state, shift, 63u);
+	const auto dword_shift      = EmitAndConstant(state, amount, 31u);
+	const auto less32           = state.builder.AllocateId();
+	const auto non_zero         = state.builder.AllocateId();
+	const auto carry_count_base = state.builder.AllocateId();
+	const auto carry_count      = state.builder.AllocateId();
+	const auto low_part         = state.builder.AllocateId();
+	const auto high_part        = state.builder.AllocateId();
+	const auto carry_part       = state.builder.AllocateId();
+	const auto carry            = state.builder.AllocateId();
+	const auto low_less         = state.builder.AllocateId();
+	const auto low_ge           = state.builder.AllocateId();
+	state.builder.AddFunction(
+	    {OpULessThan, TypeBool(state), less32, amount, ConstantU32(state, 32)});
+	state.builder.AddFunction(
+	    {OpINotEqual, TypeBool(state), non_zero, amount, ConstantU32(state, 0)});
+	state.builder.AddFunction(
+	    {OpISub, TypeU32(state), carry_count_base, ConstantU32(state, 32), dword_shift});
+	state.builder.AddFunction(
+	    {OpBitwiseAnd, TypeU32(state), carry_count, carry_count_base, ConstantU32(state, 31)});
+	state.builder.AddFunction({OpShiftRightLogical, TypeU32(state), low_part, low, dword_shift});
+	state.builder.AddFunction({OpShiftRightLogical, TypeU32(state), high_part, high, dword_shift});
+	state.builder.AddFunction({OpShiftLeftLogical, TypeU32(state), carry_part, high, carry_count});
+	state.builder.AddFunction(
+	    {OpSelect, TypeU32(state), carry, non_zero, carry_part, ConstantU32(state, 0)});
+	state.builder.AddFunction({OpBitwiseOr, TypeU32(state), low_less, low_part, carry});
+	state.builder.AddFunction({OpShiftRightLogical, TypeU32(state), low_ge, high, dword_shift});
+	out_low  = EmitSelectValueU32(state, less32, low_less, low_ge);
+	out_high = EmitSelectValueU32(state, less32, high_part, ConstantU32(state, 0));
+}
+
+uint32_t EmitAndConstant(EmitterState& state, uint32_t value, uint32_t mask) {
+	const auto ret = state.builder.AllocateId();
+	state.builder.AddFunction({OpBitwiseAnd, TypeU32(state), ret, value, ConstantU32(state, mask)});
+	return ret;
+}
+
+uint32_t EmitShiftRightConstant(EmitterState& state, uint32_t value, uint32_t shift) {
+	const auto ret = state.builder.AllocateId();
+	state.builder.AddFunction(
+	    {OpShiftRightLogical, TypeU32(state), ret, value, ConstantU32(state, shift)});
+	return ret;
+}
+
+uint32_t EmitOrU32(EmitterState& state, uint32_t lhs, uint32_t rhs) {
+	const auto ret = state.builder.AllocateId();
+	state.builder.AddFunction({OpBitwiseOr, TypeU32(state), ret, lhs, rhs});
+	return ret;
+}
+
+uint32_t EmitCompareU32Constant(EmitterState& state, uint32_t opcode, uint32_t value,
+                                uint32_t constant) {
+	const auto ret = state.builder.AllocateId();
+	state.builder.AddFunction({opcode, TypeBool(state), ret, value, ConstantU32(state, constant)});
+	return ret;
+}
+
+uint32_t EmitSubConstantMinusU32(EmitterState& state, uint32_t constant, uint32_t value) {
+	const auto ret = state.builder.AllocateId();
+	state.builder.AddFunction({OpISub, TypeU32(state), ret, ConstantU32(state, constant), value});
+	return ret;
+}
+
+uint32_t EmitF32ToF16RtzBits(EmitterState& state, uint32_t f32) {
+	const auto bits = state.builder.AllocateId();
+	state.builder.AddFunction({OpBitcast, TypeU32(state), bits, f32});
+
+	const auto sign = EmitAndConstant(state, EmitShiftRightConstant(state, bits, 16), 0x8000u);
+	const auto exp  = EmitAndConstant(state, EmitShiftRightConstant(state, bits, 23), 0xffu);
+	const auto mant = EmitAndConstant(state, bits, 0x007fffffu);
+
+	const auto half_exp       = state.builder.AllocateId();
+	const auto normal_exp     = state.builder.AllocateId();
+	const auto normal_mant    = EmitShiftRightConstant(state, mant, 13);
+	const auto normal_payload = state.builder.AllocateId();
+	const auto normal         = state.builder.AllocateId();
+	state.builder.AddFunction({OpISub, TypeU32(state), half_exp, exp, ConstantU32(state, 112)});
+	state.builder.AddFunction(
+	    {OpShiftLeftLogical, TypeU32(state), normal_exp, half_exp, ConstantU32(state, 10)});
+	state.builder.AddFunction(
+	    {OpBitwiseOr, TypeU32(state), normal_payload, normal_exp, normal_mant});
+	state.builder.AddFunction({OpBitwiseOr, TypeU32(state), normal, sign, normal_payload});
+
+	const auto mant_with_hidden = EmitOrU32(state, mant, ConstantU32(state, 0x00800000u));
+	const auto raw_sub_shift    = EmitSubConstantMinusU32(state, 126, exp);
+	const auto exp_lt_103       = EmitCompareU32Constant(state, OpULessThan, exp, 103);
+	const auto exp_gt_112       = EmitCompareU32Constant(state, OpUGreaterThan, exp, 112);
+	const auto sub_shift_low =
+	    EmitSelectValueU32(state, exp_lt_103, ConstantU32(state, 31), raw_sub_shift);
+	const auto sub_shift =
+	    EmitSelectValueU32(state, exp_gt_112, ConstantU32(state, 14), sub_shift_low);
+	const auto sub_mant  = state.builder.AllocateId();
+	const auto subnormal = state.builder.AllocateId();
+	state.builder.AddFunction(
+	    {OpShiftRightLogical, TypeU32(state), sub_mant, mant_with_hidden, sub_shift});
+	state.builder.AddFunction({OpBitwiseOr, TypeU32(state), subnormal, sign, sub_mant});
+
+	const auto nan_payload =
+	    EmitOrU32(state, EmitShiftRightConstant(state, mant, 13), ConstantU32(state, 0x0200u));
+	const auto nan =
+	    EmitOrU32(state, sign, EmitOrU32(state, ConstantU32(state, 0x7c00u), nan_payload));
+	const auto inf        = EmitOrU32(state, sign, ConstantU32(state, 0x7c00u));
+	const auto max_finite = EmitOrU32(state, sign, ConstantU32(state, 0x7bffu));
+	const auto mant_zero  = EmitCompareU32Constant(state, OpIEqual, mant, 0);
+	const auto special    = EmitSelectValueU32(state, mant_zero, inf, nan);
+
+	const auto exp_le_112 = EmitCompareU32Constant(state, OpULessThanEqual, exp, 112);
+	const auto exp_ge_143 = EmitCompareU32Constant(state, OpUGreaterThanEqual, exp, 143);
+	const auto exp_eq_255 = EmitCompareU32Constant(state, OpIEqual, exp, 255);
+	const auto finite0    = EmitSelectValueU32(state, exp_le_112, subnormal, normal);
+	const auto finite1    = EmitSelectValueU32(state, exp_lt_103, sign, finite0);
+	const auto finite2    = EmitSelectValueU32(state, exp_ge_143, max_finite, finite1);
+	return EmitAndConstant(state, EmitSelectValueU32(state, exp_eq_255, special, finite2), 0xffffu);
+}
+
+uint32_t EmitMinMaxU32Value(EmitterState& state, uint32_t lhs, uint32_t rhs, bool max_value) {
+	const auto cond = state.builder.AllocateId();
+	const auto ret  = state.builder.AllocateId();
+	state.builder.AddFunction(
+	    {max_value ? OpUGreaterThan : OpULessThan, TypeBool(state), cond, lhs, rhs});
+	state.builder.AddFunction({OpSelect, TypeU32(state), ret, cond, lhs, rhs});
+	return ret;
+}
+
+uint32_t EmitMinMaxI32Value(EmitterState& state, uint32_t lhs, uint32_t rhs, bool max_value) {
+	const auto cond = state.builder.AllocateId();
+	const auto ret  = state.builder.AllocateId();
+	state.builder.AddFunction(
+	    {max_value ? OpSGreaterThan : OpSLessThan, TypeBool(state), cond, lhs, rhs});
+	state.builder.AddFunction({OpSelect, TypeU32(state), ret, cond, lhs, rhs});
+	return ret;
+}
+
+uint32_t EmitBitcastF32ToU32(EmitterState& state, uint32_t value) {
+	const auto ret = state.builder.AllocateId();
+	state.builder.AddFunction({OpBitcast, TypeU32(state), ret, value});
+	return ret;
+}
+
+uint32_t EmitBitcastU32ToF32(EmitterState& state, uint32_t value) {
+	const auto ret = state.builder.AllocateId();
+	state.builder.AddFunction({OpBitcast, TypeF32(state), ret, value});
+	return ret;
+}
+
+uint32_t EmitAndU32(EmitterState& state, uint32_t lhs, uint32_t rhs) {
+	const auto ret = state.builder.AllocateId();
+	state.builder.AddFunction({OpBitwiseAnd, TypeU32(state), ret, lhs, rhs});
+	return ret;
+}
+
+uint32_t EmitLogicalAndBool(EmitterState& state, uint32_t lhs, uint32_t rhs) {
+	const auto ret = state.builder.AllocateId();
+	state.builder.AddFunction({OpLogicalAnd, TypeBool(state), ret, lhs, rhs});
+	return ret;
+}
+
+uint32_t EmitLogicalOrBool(EmitterState& state, uint32_t lhs, uint32_t rhs) {
+	const auto ret = state.builder.AllocateId();
+	state.builder.AddFunction({OpLogicalOr, TypeBool(state), ret, lhs, rhs});
+	return ret;
+}
+
+uint32_t EmitLogicalNotBool(EmitterState& state, uint32_t value) {
+	const auto ret = state.builder.AllocateId();
+	state.builder.AddFunction({OpLogicalNot, TypeBool(state), ret, value});
+	return ret;
+}
+
+F32Class EmitClassifyF32(EmitterState& state, uint32_t value) {
+	F32Class cls;
+	cls.bits                 = EmitBitcastF32ToU32(state, value);
+	const auto abs_bits      = EmitAndConstant(state, cls.bits, 0x7fffffffu);
+	const auto exponent_bits = EmitAndConstant(state, abs_bits, 0x7f800000u);
+	const auto mantissa_bits = EmitAndConstant(state, abs_bits, 0x007fffffu);
+	const auto exponent_max  = EmitCompareU32Constant(state, OpIEqual, exponent_bits, 0x7f800000u);
+	const auto mantissa_nonzero = EmitCompareU32Constant(state, OpINotEqual, mantissa_bits, 0);
+	cls.nan  = EmitLogicalAndBool(state, exponent_max, mantissa_nonzero);
+	cls.zero = EmitCompareU32Constant(state, OpIEqual, abs_bits, 0);
+	return cls;
+}
+
+uint32_t EmitClassMaskBitMatch(EmitterState& state, uint32_t mask, uint32_t bit,
+                               uint32_t class_match) {
+	const auto selected =
+	    EmitCompareU32Constant(state, OpINotEqual, EmitAndConstant(state, mask, 1u << bit), 0);
+	return EmitLogicalAndBool(state, selected, class_match);
+}
+
+uint32_t EmitClassMaskF32(EmitterState& state, uint32_t value, uint32_t mask) {
+	const auto bits          = EmitBitcastF32ToU32(state, value);
+	const auto sign_bits     = EmitAndConstant(state, bits, 0x80000000u);
+	const auto abs_bits      = EmitAndConstant(state, bits, 0x7fffffffu);
+	const auto exponent_bits = EmitAndConstant(state, abs_bits, 0x7f800000u);
+	const auto mantissa_bits = EmitAndConstant(state, abs_bits, 0x007fffffu);
+	const auto quiet_bits    = EmitAndConstant(state, mantissa_bits, 0x00400000u);
+
+	const auto sign             = EmitCompareU32Constant(state, OpINotEqual, sign_bits, 0);
+	const auto positive         = EmitLogicalNotBool(state, sign);
+	const auto exponent_zero    = EmitCompareU32Constant(state, OpIEqual, exponent_bits, 0);
+	const auto exponent_nonzero = EmitLogicalNotBool(state, exponent_zero);
+	const auto exponent_inf = EmitCompareU32Constant(state, OpIEqual, exponent_bits, 0x7f800000u);
+	const auto finite_exponent  = EmitLogicalNotBool(state, exponent_inf);
+	const auto mantissa_zero    = EmitCompareU32Constant(state, OpIEqual, mantissa_bits, 0);
+	const auto mantissa_nonzero = EmitLogicalNotBool(state, mantissa_zero);
+	const auto quiet            = EmitCompareU32Constant(state, OpINotEqual, quiet_bits, 0);
+
+	const auto nan_common = EmitLogicalAndBool(state, exponent_inf, mantissa_nonzero);
+	const auto snan       = EmitLogicalAndBool(state, nan_common, EmitLogicalNotBool(state, quiet));
+	const auto qnan       = EmitLogicalAndBool(state, nan_common, quiet);
+	const auto inf        = EmitLogicalAndBool(state, exponent_inf, mantissa_zero);
+	const auto normal     = EmitLogicalAndBool(state, exponent_nonzero, finite_exponent);
+	const auto denorm     = EmitLogicalAndBool(state, exponent_zero, mantissa_nonzero);
+	const auto zero       = EmitCompareU32Constant(state, OpIEqual, abs_bits, 0);
+
+	uint32_t match = EmitClassMaskBitMatch(state, mask, 0, snan);
+	match          = EmitLogicalOrBool(state, match, EmitClassMaskBitMatch(state, mask, 1, qnan));
+	match          = EmitLogicalOrBool(
+	    state, match, EmitClassMaskBitMatch(state, mask, 2, EmitLogicalAndBool(state, inf, sign)));
+	match = EmitLogicalOrBool(
+	    state, match,
+	    EmitClassMaskBitMatch(state, mask, 3, EmitLogicalAndBool(state, normal, sign)));
+	match = EmitLogicalOrBool(
+	    state, match,
+	    EmitClassMaskBitMatch(state, mask, 4, EmitLogicalAndBool(state, denorm, sign)));
+	match = EmitLogicalOrBool(
+	    state, match, EmitClassMaskBitMatch(state, mask, 5, EmitLogicalAndBool(state, zero, sign)));
+	match = EmitLogicalOrBool(
+	    state, match,
+	    EmitClassMaskBitMatch(state, mask, 6, EmitLogicalAndBool(state, zero, positive)));
+	match = EmitLogicalOrBool(
+	    state, match,
+	    EmitClassMaskBitMatch(state, mask, 7, EmitLogicalAndBool(state, denorm, positive)));
+	match = EmitLogicalOrBool(
+	    state, match,
+	    EmitClassMaskBitMatch(state, mask, 8, EmitLogicalAndBool(state, normal, positive)));
+	return EmitLogicalOrBool(
+	    state, match,
+	    EmitClassMaskBitMatch(state, mask, 9, EmitLogicalAndBool(state, inf, positive)));
+}
+
+uint32_t EmitMinMaxF32Value(EmitterState& state, uint32_t lhs, uint32_t rhs, bool max_value) {
+	const auto lhs_class = EmitClassifyF32(state, lhs);
+	const auto rhs_class = EmitClassifyF32(state, rhs);
+
+	const auto numeric_cond = state.builder.AllocateId();
+	state.builder.AddFunction({max_value ? OpFOrdGreaterThanEqual : OpFOrdLessThan, TypeBool(state),
+	                           numeric_cond, lhs, rhs});
+	const auto ordered_bits =
+	    EmitSelectValueU32(state, numeric_cond, lhs_class.bits, rhs_class.bits);
+
+	const auto both_zero    = EmitLogicalAndBool(state, lhs_class.zero, rhs_class.zero);
+	const auto zero_bits    = max_value ? EmitAndU32(state, lhs_class.bits, rhs_class.bits)
+	                                    : EmitOrU32(state, lhs_class.bits, rhs_class.bits);
+	const auto numeric_bits = EmitSelectValueU32(state, both_zero, zero_bits, ordered_bits);
+
+	const auto rhs_nan_bits =
+	    EmitSelectValueU32(state, rhs_class.nan, lhs_class.bits, numeric_bits);
+	const auto result_bits = EmitSelectValueU32(state, lhs_class.nan, rhs_class.bits, rhs_nan_bits);
+	return EmitBitcastU32ToF32(state, result_bits);
+}
+
+uint32_t EmitTruncF32Value(EmitterState& state, uint32_t value) {
+	const auto ret = state.builder.AllocateId();
+	state.builder.AddFunction(
+	    {OpExtInst, TypeF32(state), ret, GlslStd450(state), GlslTrunc, value});
+	return ret;
+}
+
+uint32_t EmitFlushF32DenormToSignedZero(EmitterState& state, uint32_t value) {
+	const auto bits      = state.builder.AllocateId();
+	const auto abs_bits  = state.builder.AllocateId();
+	const auto sign_bits = state.builder.AllocateId();
+	const auto non_zero  = state.builder.AllocateId();
+	const auto subnormal = state.builder.AllocateId();
+	const auto flush     = state.builder.AllocateId();
+	const auto selected  = state.builder.AllocateId();
+	const auto ret       = state.builder.AllocateId();
+	state.builder.AddFunction({OpBitcast, TypeU32(state), bits, value});
+	state.builder.AddFunction(
+	    {OpBitwiseAnd, TypeU32(state), abs_bits, bits, ConstantU32(state, 0x7fffffffu)});
+	state.builder.AddFunction(
+	    {OpBitwiseAnd, TypeU32(state), sign_bits, bits, ConstantU32(state, 0x80000000u)});
+	state.builder.AddFunction(
+	    {OpINotEqual, TypeBool(state), non_zero, abs_bits, ConstantU32(state, 0)});
+	state.builder.AddFunction(
+	    {OpULessThan, TypeBool(state), subnormal, abs_bits, ConstantU32(state, 0x00800000u)});
+	state.builder.AddFunction({OpLogicalAnd, TypeBool(state), flush, non_zero, subnormal});
+	state.builder.AddFunction({OpSelect, TypeU32(state), selected, flush, sign_bits, bits});
+	state.builder.AddFunction({OpBitcast, TypeF32(state), ret, selected});
+	return ret;
+}
+
+uint32_t EmitTrigCycleF32(EmitterState& state, uint32_t src, bool preserve_signed_zero) {
+	const auto fract        = state.builder.AllocateId();
+	const auto bits         = state.builder.AllocateId();
+	const auto abs_bits     = state.builder.AllocateId();
+	const auto large        = state.builder.AllocateId();
+	const auto finite       = state.builder.AllocateId();
+	const auto large_finite = state.builder.AllocateId();
+	const auto reduced      = state.builder.AllocateId();
+	state.builder.AddFunction(
+	    {OpExtInst, TypeF32(state), fract, GlslStd450(state), GlslFract, src});
+	state.builder.AddFunction({OpBitcast, TypeU32(state), bits, src});
+	state.builder.AddFunction(
+	    {OpBitwiseAnd, TypeU32(state), abs_bits, bits, ConstantU32(state, 0x7fffffffu)});
+	state.builder.AddFunction(
+	    {OpUGreaterThanEqual, TypeBool(state), large, abs_bits, ConstantU32(state, 0x4b000000u)});
+	state.builder.AddFunction(
+	    {OpULessThan, TypeBool(state), finite, abs_bits, ConstantU32(state, 0x7f800000u)});
+	state.builder.AddFunction({OpLogicalAnd, TypeBool(state), large_finite, large, finite});
+	state.builder.AddFunction(
+	    {OpSelect, TypeF32(state), reduced, large_finite, ConstantF32(state, 0), fract});
+	if (!preserve_signed_zero) {
+		return reduced;
+	}
+	const auto zero = state.builder.AllocateId();
+	const auto ret  = state.builder.AllocateId();
+	state.builder.AddFunction({OpIEqual, TypeBool(state), zero, abs_bits, ConstantU32(state, 0)});
+	state.builder.AddFunction({OpSelect, TypeF32(state), ret, zero, src, reduced});
+	return ret;
+}
+
+uint32_t EmitFNegateValue(EmitterState& state, uint32_t value) {
+	const auto ret = state.builder.AllocateId();
+	state.builder.AddFunction({OpFNegate, TypeF32(state), ret, value});
+	return ret;
+}
+
+uint32_t EmitFAbsValue(EmitterState& state, uint32_t value) {
+	const auto ret = state.builder.AllocateId();
+	state.builder.AddFunction({OpExtInst, TypeF32(state), ret, GlslStd450(state), GlslFAbs, value});
+	return ret;
+}
+
+uint32_t EmitF16BitsToF32(EmitterState& state, uint32_t bits) {
+	const auto unpacked = state.builder.AllocateId();
+	const auto ret      = state.builder.AllocateId();
+	state.builder.AddFunction({OpExtInst, TypeF32Vector(state, 2), unpacked, GlslStd450(state),
+	                           GlslUnpackHalf2x16, bits});
+	state.builder.AddFunction({OpCompositeExtract, TypeF32(state), ret, unpacked, 0});
+	return ret;
+}
+
+} // namespace Libs::Graphics::ShaderRecompiler::Spirv::Emitter

src/graphics/shader/recompiler/backend/spirv/spirvEmitterAnalysis.cpp

@@ -0,0 +1,343 @@
+#include "common/assert.h"
+#include "graphics/guest_gpu/gpu_defs.h"
+#include "graphics/shader/recompiler/backend/spirv/SpirvEmitter.h"
+#include "graphics/shader/recompiler/backend/spirv/spirvEmitterInternal.h"
+#include "graphics/shader/recompiler/ir/ValueProgram.h"
+
+namespace Libs::Graphics::ShaderRecompiler::Spirv::Emitter {
+
+uint32_t PixelParameterMappedLocation(const EmitterState& state, uint32_t attr) {
+	if (state.stage != ShaderType::Pixel) {
+		return attr;
+	}
+	return ShaderPixelParameterMappedLocation(*state.input_info.pixel, attr);
+}
+
+uint32_t PixelParameterLocation(const EmitterState& state, uint32_t attr) {
+	std::array<uint32_t, 32> active_inputs {};
+	uint32_t                 active_count = 0;
+	for (const auto& input: state.inputs) {
+		if (input.kind == IR::StageInputKind::Parameter) {
+			active_inputs[active_count++] = input.location;
+		}
+	}
+	return state.stage == ShaderType::Pixel
+	           ? ShaderPixelParameterLocation(*state.input_info.pixel,
+	                                          {active_inputs.data(), active_count}, attr)
+	           : attr;
+}
+
+bool PixelParameterIsFlat(const EmitterState& state, uint32_t attr) {
+	return state.stage == ShaderType::Pixel &&
+	       ShaderPixelParameterIsFlat(*state.input_info.pixel, attr);
+}
+
+bool PixelParameterIsCustom(const EmitterState& state, uint32_t attr) {
+	return state.stage == ShaderType::Pixel &&
+	       ShaderPixelParameterIsCustom(*state.input_info.pixel, attr);
+}
+
+void SetError(std::string* error, const char* message) {
+	if (error != nullptr) {
+		*error = message;
+	}
+}
+
+bool HasOutput(const std::vector<OutputBinding>& outputs, IR::StageOutputKind kind,
+               uint32_t index) {
+	return std::any_of(outputs.begin(), outputs.end(), [kind, index](const OutputBinding& binding) {
+		return binding.kind == kind && binding.index == index;
+	});
+}
+
+void CopyProgramInputsAndOutputs(EmitterState& state, const IR::Program& program) {
+	for (const auto& input: program.info.inputs) {
+		state.inputs.push_back({input.kind, input.location, input.component_count, 0,
+		                        input.debug_name, input.per_vertex});
+	}
+	for (const auto& output: program.info.outputs) {
+		if (HasOutput(state.outputs, output.kind, output.index)) {
+			continue;
+		}
+		state.outputs.push_back({output.kind, output.index, output.location, 0, output.debug_name});
+	}
+}
+
+uint32_t OutputVariableForExport(const EmitterState& state, const IR::ExportInfo& exp) {
+	if (exp.kind == IR::ExportTargetKind::Position && exp.index == 0) {
+		return state.per_vertex_variable;
+	}
+	if (exp.kind == IR::ExportTargetKind::MrtZ) {
+		return state.depth_variable;
+	}
+	for (const auto& binding: state.outputs) {
+		const auto expected_kind = exp.kind == IR::ExportTargetKind::Mrt
+		                               ? IR::StageOutputKind::Mrt
+		                               : IR::StageOutputKind::Parameter;
+		if (binding.kind == expected_kind && binding.index == exp.index) {
+			return binding.variable_id;
+		}
+	}
+	return 0;
+}
+
+uint32_t          ConstantU32(EmitterState& state, uint32_t value);
+[[noreturn]] void ExitDescriptorBindingFailure(const EmitterState&       state,
+                                               IR::DescriptorBindingKind kind, uint32_t resource,
+                                               const char* reason) {
+	EXIT("shader binding resolution failed during SPIR-V emit: hash=0x%016" PRIx64
+	     " stage=%u resource=%" PRIu32 " binding_kind=%u reason=%s\n",
+	     state.program.shader_hash, static_cast<unsigned>(state.stage), resource,
+	     static_cast<unsigned>(kind), reason);
+	std::abort();
+}
+
+DescriptorResourceBinding ResourceForDescriptor(const EmitterState&       state,
+                                                IR::DescriptorBindingKind kind, uint32_t resource) {
+	const auto* descriptor = IR::FindBinding(state.program.bindings, kind);
+	if (descriptor == nullptr) {
+		ExitDescriptorBindingFailure(state, kind, resource, "descriptor group was not allocated");
+	}
+	const auto found =
+	    std::find(descriptor->resources.begin(), descriptor->resources.end(), resource);
+	if (found == descriptor->resources.end()) {
+		ExitDescriptorBindingFailure(state, kind, resource,
+		                             "resource is absent from descriptor group");
+	}
+	return {descriptor, static_cast<uint32_t>(found - descriptor->resources.begin())};
+}
+
+uint32_t DescriptorElementPointer(EmitterState& state, uint32_t result_ptr_type,
+                                  uint32_t variable_id, uint32_t array_index,
+                                  IR::DescriptorBindingKind kind, uint32_t resource,
+                                  const char* variable_name) {
+	if (variable_id == 0) {
+		ExitDescriptorBindingFailure(state, kind, resource, variable_name);
+	}
+	const auto pointer = state.builder.AllocateId();
+	state.builder.AddFunction(
+	    {OpAccessChain, result_ptr_type, pointer, variable_id, ConstantU32(state, array_index)});
+	return pointer;
+}
+
+ImageViewKind ImageViewKindFromDimension(Decoder::ImageDimension dimension) {
+	switch (dimension) {
+		case Decoder::ImageDimension::Dim1D: return ImageViewKind::Dim1D;
+		case Decoder::ImageDimension::Dim1DArray: return ImageViewKind::Dim1DArray;
+		case Decoder::ImageDimension::Dim2DArray: return ImageViewKind::Dim2DArray;
+		case Decoder::ImageDimension::Dim3D: return ImageViewKind::Dim3D;
+		case Decoder::ImageDimension::Dim2DMsaa: return ImageViewKind::Dim2DMsaa;
+		case Decoder::ImageDimension::Dim2DMsaaArray: return ImageViewKind::Dim2DMsaaArray;
+		default: return ImageViewKind::Dim2D;
+	}
+}
+
+ImageViewKind SampledImageViewKind(const EmitterState& state, const IR::MemoryInfo& mem,
+                                   uint32_t use_pc) {
+	(void)state;
+	(void)use_pc;
+	return ImageViewKindFromDimension(mem.image_dimension);
+}
+
+ImageViewKind StorageImageViewKind(const IR::MemoryInfo& mem) {
+	return ImageViewKindFromDimension(mem.image_dimension);
+}
+
+uint32_t ImageViewCoordinateComponents(ImageViewKind view) {
+	switch (view) {
+		case ImageViewKind::Dim1D: return 1u;
+		case ImageViewKind::Dim1DArray:
+		case ImageViewKind::Dim2D: return 2u;
+		case ImageViewKind::Dim2DArray:
+		case ImageViewKind::Dim2DMsaaArray:
+		case ImageViewKind::Dim3D: return 3u;
+		case ImageViewKind::Dim2DMsaa: return 2u;
+		default: return 0u;
+	}
+}
+
+uint32_t ImageViewSpatialComponents(ImageViewKind view) {
+	switch (view) {
+		case ImageViewKind::Dim1D:
+		case ImageViewKind::Dim1DArray: return 1u;
+		case ImageViewKind::Dim2D:
+		case ImageViewKind::Dim2DArray:
+		case ImageViewKind::Dim2DMsaa:
+		case ImageViewKind::Dim2DMsaaArray: return 2u;
+		case ImageViewKind::Dim3D: return 3u;
+		default: return 0u;
+	}
+}
+
+uint32_t ImageViewImageType(EmitterState& state, ImageViewKind view, bool integer) {
+	const auto component = integer ? TypeU32(state) : TypeF32(state);
+	return state.builder.Type(OpTypeImage,
+	                          {component, ImageSpirvDimension(view), 0, ImageSpirvArrayed(view),
+	                           ImageSpirvMultisampled(view), 1, ImageFormatUnknown});
+}
+
+uint32_t ImageViewSampledImageType(EmitterState& state, ImageViewKind view, bool integer) {
+	return state.builder.Type(OpTypeSampledImage, {ImageViewImageType(state, view, integer)});
+}
+
+uint32_t ImageViewSizeType(EmitterState& state, ImageViewKind view) {
+	switch (ImageViewCoordinateComponents(view)) {
+		case 1u: return TypeU32(state);
+		case 2u: return TypeU32Vector(state, 2);
+		case 3u: return TypeU32Vector(state, 3);
+		default: return 0;
+	}
+}
+
+uint32_t StorageImageType(EmitterState& state, StorageImageClass image_class, ImageViewKind view) {
+	uint32_t component = TypeF32(state);
+	uint32_t format    = ImageFormatUnknown;
+	if (image_class == StorageImageClass::FormatlessUint) {
+		component = TypeU32(state);
+	} else if (image_class == StorageImageClass::AtomicUint) {
+		component = TypeU32(state);
+		format    = ImageFormatR32ui;
+	}
+	return state.builder.Type(OpTypeImage, {component, ImageSpirvDimension(view), 0,
+	                                        ImageSpirvArrayed(view), 0, 2, format});
+}
+
+uint32_t StorageImagePointerType(EmitterState& state, StorageImageClass image_class,
+                                 ImageViewKind view) {
+	return state.builder.Type(
+	    OpTypePointer, {StorageClassUniformConstant, StorageImageType(state, image_class, view)});
+}
+
+StorageImageClass StorageImageClassForResource(const EmitterState& state, uint32_t resource) {
+	const auto& image      = state.program.info.images.at(resource);
+	const bool  uint_image = image.kind == IR::ResourceKind::StorageImageUint;
+	EXIT_IF(image.kind != IR::ResourceKind::StorageImage && !uint_image);
+	EXIT_IF(image.atomic && !uint_image);
+	return StorageImageClassFor(uint_image, image.atomic);
+}
+
+uint32_t StorageImageVariable(const EmitterState& state, StorageImageClass image_class,
+                              ImageViewKind view) {
+	return state.storage_image_variables[StorageImageIndex(image_class, view)];
+}
+
+uint32_t LoadSampledImageDescriptor(EmitterState& state, const IR::MemoryInfo& mem, uint32_t use_pc,
+                                    ImageViewKind view) {
+	(void)use_pc;
+	const bool integer      = mem.kind == IR::ResourceKind::ImageUint;
+	const auto kind         = SampledBindingKind(integer, view);
+	const auto binding      = ResourceForDescriptor(state, kind, mem.resource);
+	const auto variable     = state.sampled_image_variables[SampledImageIndex(integer, view)];
+	const auto pointer_type = state.builder.Type(
+	    OpTypePointer, {StorageClassUniformConstant, ImageViewImageType(state, view, integer)});
+	const auto pointer =
+	    DescriptorElementPointer(state, pointer_type, variable, binding.array_index, kind,
+	                             mem.resource, "sampled image descriptor array was not emitted");
+	const auto image = state.builder.AllocateId();
+	state.builder.AddFunction({OpLoad, ImageViewImageType(state, view, integer), image, pointer});
+	return image;
+}
+
+uint32_t LoadSamplerDescriptor(EmitterState& state, uint32_t sampler, uint32_t use_pc) {
+	(void)use_pc;
+	const auto binding = ResourceForDescriptor(state, IR::DescriptorBindingKind::Samplers, sampler);
+	const auto sampler_type = state.builder.Type(OpTypeSampler);
+	const auto pointer_type =
+	    state.builder.Type(OpTypePointer, {StorageClassUniformConstant, sampler_type});
+	const auto pointer = DescriptorElementPointer(
+	    state, pointer_type, state.sampler_variable, binding.array_index,
+	    IR::DescriptorBindingKind::Samplers, sampler, "sampler descriptor array was not emitted");
+	const auto sampler_id = state.builder.AllocateId();
+	state.builder.AddFunction({OpLoad, sampler_type, sampler_id, pointer});
+	return sampler_id;
+}
+
+uint32_t MakeSampledImage(EmitterState& state, const IR::MemoryInfo& mem, uint32_t use_pc,
+                          ImageViewKind view) {
+	const auto image   = LoadSampledImageDescriptor(state, mem, use_pc, view);
+	const auto sampler = LoadSamplerDescriptor(state, mem.sampler, use_pc);
+	if (image == 0 || sampler == 0) {
+		ExitDescriptorBindingFailure(
+		    state, SampledBindingKind(mem.kind == IR::ResourceKind::ImageUint, view), mem.resource,
+		    "sampled image or sampler descriptor load failed");
+	}
+	const auto sampled_image = state.builder.AllocateId();
+	state.builder.AddFunction(
+	    {OpSampledImage,
+	     ImageViewSampledImageType(state, view, mem.kind == IR::ResourceKind::ImageUint),
+	     sampled_image, image, sampler});
+	return sampled_image;
+}
+
+uint32_t MakeSampledImage(EmitterState& state, const IR::MemoryInfo& mem, uint32_t use_pc,
+                          ImageViewKind view, uint32_t image_resource) {
+	auto selected     = mem;
+	selected.resource = image_resource;
+	return MakeSampledImage(state, selected, use_pc, view);
+}
+
+uint32_t StorageImageDescriptorPointer(EmitterState& state, uint32_t resource, ImageViewKind view) {
+	const auto image_class = StorageImageClassForResource(state, resource);
+	const auto kind        = StorageBindingKind(image_class, view);
+	const auto binding  = ResourceForDescriptor(state, kind, resource);
+	const auto ptr_type    = StorageImagePointerType(state, image_class, view);
+	const auto variable    = StorageImageVariable(state, image_class, view);
+	return DescriptorElementPointer(state, ptr_type, variable, binding.array_index, kind, resource,
+	                                "storage image descriptor array was not emitted");
+}
+
+void EmitStorageImageWrite(EmitterState& state, uint32_t resource, ImageViewKind view,
+                           uint32_t mip_lod, uint32_t coord, uint32_t texel) {
+	const auto image_class = StorageImageClassForResource(state, resource);
+	if (image_class != StorageImageClass::AtomicUint) {
+		state.builder.RequireCapability(CapabilityStorageImageWriteWithoutFormat);
+	}
+	const auto  kind    = StorageBindingKind(image_class, view);
+	const auto  binding = ResourceForDescriptor(state, kind, resource);
+	const auto& image   = state.program.info.images.at(resource);
+	const auto  LoadAt  = [&](uint32_t array_index) {
+		const auto pointer = DescriptorElementPointer(
+		    state, StorageImagePointerType(state, image_class, view),
+		    StorageImageVariable(state, image_class, view), array_index, kind, resource,
+		    "storage image descriptor array was not emitted");
+		const auto descriptor = state.builder.AllocateId();
+		state.builder.AddFunction(
+		    {OpLoad, StorageImageType(state, image_class, view), descriptor, pointer});
+		return descriptor;
+	};
+	if (image.mip_mode != IR::ImageMipMode::DynamicStorage) {
+		state.builder.AddFunction({OpImageWrite, LoadAt(binding.array_index), coord, texel});
+		return;
+	}
+	if (image.mip_count == 0u) {
+		ExitDescriptorBindingFailure(state, kind, resource,
+		                             "dynamic storage image has no mip descriptors");
+	}
+
+	const auto            merge_label = state.builder.AllocateId();
+	std::vector<uint32_t> labels(image.mip_count);
+	std::vector<uint32_t> words {OpSwitch, mip_lod, merge_label};
+	for (uint32_t mip = 0; mip < image.mip_count; mip++) {
+		labels[mip] = state.builder.AllocateId();
+		words.push_back(mip);
+		words.push_back(labels[mip]);
+	}
+	state.builder.AddFunction({OpSelectionMerge, merge_label, SelectionControlNone});
+	state.builder.AddFunction(words);
+	for (uint32_t mip = 0; mip < image.mip_count; mip++) {
+		EmitLabel(state, labels[mip]);
+		state.builder.AddFunction({OpImageWrite, LoadAt(binding.array_index + mip), coord, texel});
+		state.builder.AddFunction({OpBranch, merge_label});
+	}
+	EmitLabel(state, merge_label);
+}
+
+uint32_t ExecutionModelForStage(ShaderType stage) {
+	switch (stage) {
+		case ShaderType::Vertex: return ExecutionModelVertex;
+		case ShaderType::Pixel: return ExecutionModelFragment;
+		default: return ExecutionModelGLCompute;
+	}
+}
+
+} // namespace Libs::Graphics::ShaderRecompiler::Spirv::Emitter