gremling/KytyPS5
1
0 Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-28 16:23:05 +00:00
Code Issues Packages Projects Releases Wiki Activity

shader: separate legacy FP32 MAD rounding from fused FMA

Browse Source 
Co-authored-by: shadowbeat070 <70843316+shadowbeat070@users.noreply.github.com>
This commit is contained in:
nmzikandshadowbeat070 committed 2026-10-05 04:11:48 +04:00
1 parent bbb98bd490
commit d8be245c22
7 files changed
+134 -8
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

src/graphics/shader/recompiler/frontend/decode/ShaderDecoder.h
+3
View File
@@ -224,6 +224,9 @@ enum class Opcode {
V_MAC_F32,
V_MADMK_F32,
V_MADAK_F32,
V_FMAC_F32,
V_FMAMK_F32,
V_FMAAK_F32,
V_MIN_F32,
V_MAX_F32,
V_ADD_F16,
src/graphics/shader/recompiler/frontend/decode/VectorAluOps.cpp
+10 -3
View File
@@ -65,9 +65,9 @@ constexpr Vop2OpcodeInfo VOP2_OPCODE_LIST[] = {
{0x2au, Opcode::V_SUBREV_CO_CI_U32},
{0x26u, Opcode::V_SUB_NC_U32, Vop2SdwaProfile::IntegerPartialDestination},
{0x27u, Opcode::V_SUBREV_NC_U32, Vop2SdwaProfile::IntegerFullDestination},
{0x2bu, Opcode::V_MAC_F32},
{0x2cu, Opcode::V_MADMK_F32},
{0x2du, Opcode::V_MADAK_F32},
{0x2bu, Opcode::V_FMAC_F32},
{0x2cu, Opcode::V_FMAMK_F32},
{0x2du, Opcode::V_FMAAK_F32},
{0x2fu, Opcode::V_CVT_PKRTZ_F16_F32, Vop2SdwaProfile::PackedFloat16},
{0x32u, Opcode::V_ADD_F16, Vop2SdwaProfile::Float16},
{0x33u, Opcode::V_SUB_F16, Vop2SdwaProfile::Float16},
@@ -800,6 +800,9 @@ bool IsVop2FloatOpcode(Opcode opcode) {
case Opcode::V_MAC_F32:
case Opcode::V_MADMK_F32:
case Opcode::V_MADAK_F32:
case Opcode::V_FMAC_F32:
case Opcode::V_FMAMK_F32:
case Opcode::V_FMAAK_F32:
case Opcode::V_CVT_PKRTZ_F16_F32:
case Opcode::V_ADD_F16:
case Opcode::V_SUB_F16:
@@ -1057,6 +1060,7 @@ void FinalizeVop2Instruction(std::span<const uint32_t> code, uint32_t word_index
Instruction& inst) {
switch (inst.opcode) {
case Opcode::V_MADMK_F32:
case Opcode::V_FMAMK_F32:
case Opcode::V_FMAMK_F16:
inst.src2 = inst.src1;
inst.src1 = {};
@@ -1064,6 +1068,7 @@ void FinalizeVop2Instruction(std::span<const uint32_t> code, uint32_t word_index
inst.src_count = 3;
break;
case Opcode::V_MADAK_F32:
case Opcode::V_FMAAK_F32:
case Opcode::V_FMAAK_F16:
inst.src2 = {};
inst.src2.kind = OperandKind::LiteralConstant;
@@ -1401,6 +1406,7 @@ bool SupportsNativeVop3SourceModifiers(Opcode opcode) {
case Opcode::V_MIN_F32:
case Opcode::V_MAX_F32:
case Opcode::V_MAC_F32:
case Opcode::V_FMAC_F32:
case Opcode::V_MAD_F32:
case Opcode::V_ADD_F64:
case Opcode::V_MUL_F64:
@@ -1436,6 +1442,7 @@ bool SupportsNativeVop3ResultModifiers(Opcode opcode) {
case Opcode::V_MIN_F32:
case Opcode::V_MAX_F32:
case Opcode::V_MAC_F32:
case Opcode::V_FMAC_F32:
case Opcode::V_MAD_F32:
case Opcode::V_FMA_F32:
case Opcode::V_FMA_F16: