gremling/KytyPS5
1
0 Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-30 09:08:19 +00:00
Code Issues Packages Projects Releases Wiki Activity

shader: support high products of 24-bit integers

Browse Source
This commit is contained in:
nmzik committed 2026-10-28 22:49:21 +01:00
1 parent 22610cb93e
commit 54b937c85f
7 files changed
+128 -4
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
+2
View File
@@ -249,7 +249,9 @@ enum class Opcode {
V_MIN_I32,
V_MAX_I32,
V_MUL_I32_I24,
V_MUL_HI_I32_I24,
V_MUL_U32_U24,
V_MUL_HI_U32_U24,
V_MUL_LO_U32,
V_MUL_HI_U32,
V_MUL_LO_I32,
src/graphics/shader/recompiler/frontend/decode/VectorAluOps.cpp
+8 -1
View File
@@ -35,7 +35,9 @@ constexpr Vop2OpcodeInfo VOP2_OPCODE_LIST[] = {
{0x05u, Opcode::V_SUBREV_F32},
{0x08u, Opcode::V_MUL_F32, Vop2SdwaProfile::Float32},
{0x09u, Opcode::V_MUL_I32_I24, Vop2SdwaProfile::IntegerPartialDestination},
{0x0au, Opcode::V_MUL_HI_I32_I24, Vop2SdwaProfile::IntegerPartialDestination},
{0x0bu, Opcode::V_MUL_U32_U24, Vop2SdwaProfile::IntegerPartialDestination},
{0x0cu, Opcode::V_MUL_HI_U32_U24, Vop2SdwaProfile::IntegerPartialDestination},
{0x0fu, Opcode::V_MIN_F32},
{0x10u, Opcode::V_MAX_F32},
{0x11u, Opcode::V_MIN_I32, Vop2SdwaProfile::IntegerFullDestination},
@@ -1539,7 +1541,8 @@ bool SupportsNativeVop3Clamp(Opcode opcode) {
return SupportsNativeVop3ResultModifiers(opcode) || UsesInexactClampControl(opcode) ||
opcode == Opcode::V_MAD_I16 || opcode == Opcode::V_MAD_U16 ||
opcode == Opcode::V_MIN3_I16 || opcode == Opcode::V_MAX3_I16 ||
opcode == Opcode::V_MAX3_U16 || opcode == Opcode::V_MED3_U16;
opcode == Opcode::V_MAX3_U16 || opcode == Opcode::V_MED3_U16 ||
opcode == Opcode::V_MUL_HI_I32_I24 || opcode == Opcode::V_MUL_HI_U32_U24;
}
bool HasUnsupportedNativeVop3Modifiers(Opcode opcode, bool permlane, bool mad_mix,
@@ -1570,6 +1573,10 @@ bool HasUnsupportedNativeVop3Modifiers(Opcode opcode, bool permlane, bool mad_mi
return clamp != 0u || omod != 0u || neg != 0u;
}
switch (opcode) {
case Opcode::V_MUL_HI_I32_I24:
case Opcode::V_MUL_HI_U32_U24:
// The high product fits in 32 bits, so clamp has no effect.
return abs != 0u || op_sel != 0u || omod != 0u || neg != 0u;
case Opcode::V_LDEXP_F32: return (abs & ~1u) != 0u || op_sel != 0u || (neg & ~1u) != 0u;
case Opcode::V_CVT_F32_UBYTE0:
case Opcode::V_CVT_F32_UBYTE1:
src/graphics/shader/recompiler/frontend/translate/Integer.cpp
+4 -2
View File
@@ -269,7 +269,7 @@ void Translator::V_FFBH_32(const Decoder::Instruction& inst, bool sign) {
WriteOperand(DestinationOperand(inst), result);
}
void Translator::Integer24(const Decoder::Instruction& inst, bool sign, bool addend) {
void Translator::Integer24(const Decoder::Instruction& inst, bool sign, bool addend, bool high) {
const auto extract24 = [&](IR::U32 value) {
return IR::U32(
ir.Emit(sign ? IR::ValueOpcode::BitFieldSExtract : IR::ValueOpcode::BitFieldUExtract,
@@ -277,7 +277,9 @@ void Translator::Integer24(const Decoder::Instruction& inst, bool sign, bool add
};
const auto lhs = extract24(ReadU32(inst.src0));
const auto rhs = extract24(ReadU32(inst.src1));
auto result = ir.IMul(lhs, rhs);
const auto multiply = high ? (sign ? IR::ValueOpcode::SMulHi : IR::ValueOpcode::UMulHi)
: IR::ValueOpcode::IMul32;
auto result = IR::U32(ir.Emit(multiply, {lhs, rhs}));
if (addend) {
result = ir.IAdd(result, ReadU32(inst.src2));
}
src/graphics/shader/recompiler/frontend/translate/Translator.h
+1 -1
View File
@@ -187,7 +187,7 @@ private:
void V_XOR3_B32(const Decoder::Instruction& inst);
void S_FF_I32_B64(const Decoder::Instruction& inst, bool find_zero, bool from_msb, bool sign);
void V_FFBH_32(const Decoder::Instruction& inst, bool sign);
void Integer24(const Decoder::Instruction& inst, bool sign, bool addend);
void Integer24(const Decoder::Instruction& inst, bool sign, bool addend, bool high = false);
void V_MAD_U64_U32(const Decoder::Instruction& inst);
void V_SAD_U32(const Decoder::Instruction& inst);
void V_ADD3_U32(const Decoder::Instruction& inst);
src/graphics/shader/recompiler/frontend/translate/Vector.cpp
+2
View File
@@ -586,6 +586,8 @@ void Translator::EmitVector(const Decoder::Instruction& inst) {
case O::V_MAD_U32_U24: return Integer24(inst, false, true);
case O::V_MUL_I32_I24: return Integer24(inst, true, false);
case O::V_MUL_U32_U24: return Integer24(inst, false, false);
case O::V_MUL_HI_I32_I24: return Integer24(inst, true, false, true);
case O::V_MUL_HI_U32_U24: return Integer24(inst, false, false, true);
case O::V_MAD_U64_U32: return V_MAD_U64_U32(inst);
case O::V_SAD_U32: return V_SAD_U32(inst);
case O::V_ADD3_U32: return V_ADD3_U32(inst);