gremling/KytyPS5
1
0 Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-27 16:02:37 +00:00
Code Issues Packages Projects Releases Wiki Activity

shader: link multiple compute helpers without synthetic return loops

Browse Source
This commit is contained in:
nmzik committed 2026-10-19 02:34:25 +02:00
1 parent b2d3ce6dfc
commit ab8566591f
8 files changed
+363 -109
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

tests/shaderCfgTests.cpp
+7 -7
View File
@@ -10805,8 +10805,8 @@ void TestMemoryFedScalarLeafCall() {
IR::SrtRuntime runtime{.read_memory = read, .userdata = &memory, .read_specialization_memory = read};
const auto options = MakeCompileOptions(ShaderType::Compute);
auto source = PrepareShaderSource(shader, options);
Check(source.call && source.decoded.has_swap_pc &&
source.decoded.instructions[source.call->instruction].branch_target == UINT32_MAX,
Check(source.calls.size() == 1 && source.decoded.has_swap_pc &&
source.decoded.instructions[source.calls.front().instruction].branch_target == UINT32_MAX,
"memory-fed SWAPPC did not retain an unresolved target");
#if KYTY_PLATFORM != KYTY_PLATFORM_WINDOWS
ExpectFatal([&] { (void)TranslateProgram(source.decoded, options); },
@@ -10814,7 +10814,7 @@ void TestMemoryFedScalarLeafCall() {
#endif
const auto& linked = RefreshShaderSource(source, runtime);
Check(linked.has_swap_pc && source.revision == 1 && linked.instructions.back().branch_target == 0x40 &&
linked.instructions[source.call->instruction].branch_target == 0x44 &&
linked.instructions[source.calls.front().instruction].branch_target == 0x44 &&
source.reads.back() == std::pair<uint64_t, uint64_t>{base + 0x100, 12},
"aliased call target, exact leaf range, or continuation was lost");
const auto* storage = linked.instructions.data();
@@ -10899,14 +10899,14 @@ void TestMemoryFedScalarLeafCall() {
memory.words[96] = EncodeSMovB32(14, 129);
ExpectFatal([&] { (void)RefreshShaderSource(source, runtime); },
"leaf clobbering its return pair was linked");
// AND_B64 writes the high half despite the decoder's generic scalar width.
// AND_B64 must replace the preceding high-half definition.
const uint32_t high_writer[] = {
EncodeSMovB32(15, 129), EncodeSop2(0x0f, 14, 128, 128),
EncodeSop1(0x21, 14, 14), EncodeSopp(0x01)};
auto pair_source = PrepareShaderSource(high_writer, options);
IR::SrtWalker pair_walker(pair_source.call_targets, runtime);
uint32_t high = 1;
Check(pair_walker.Evaluate(pair_source.call->target[1], high) && high == 0,
Check(pair_walker.Evaluate(pair_source.calls.front().target[1], high) && high == 0,
"scalar pair producer used a stale preceding high-half definition");
const uint32_t joined[] = {
EncodeSopp(0x04, 2), EncodeSMovB32(14, 129), EncodeSopp(0x02, 1),
@@ -10915,7 +10915,7 @@ void TestMemoryFedScalarLeafCall() {
"ambiguous reaching definitions were accepted for a call target");
const uint32_t multiple[] = {EncodeSop1(0x21, 14, 14), EncodeSop1(0x21, 14, 14), EncodeSopp(0x01)};
ExpectFatal([&] { (void)PrepareShaderSource(multiple, options); },
"call metadata bypassed the multiple-call rejection");
"a native return pair was accepted as a later call target");
#endif
}
@@ -10962,7 +10962,7 @@ void TestFusedShaderHandoffPreservesRegisters() {
for (const auto handoff: {EncodeSop1(0x20, 0, 6), 0xbefd2106u}) {
front[2] = handoff; // SETPC or captured SWAPPC with NULL destination
auto source = PrepareShaderSource(front, options);
Check(!source.decoded.has_swap_pc && !source.call,
Check(!source.decoded.has_swap_pc && source.calls.empty(),
"fused SETPC handoff entered scalar call preparation");
auto translated = TranslateProgram(source.decoded, options);
uint32_t allocations = 0;