gremling/KytyPS5
1
0 Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-29 16:51:20 +00:00
Code Issues Packages Projects Releases Wiki Activity

shader: support unsigned 16-bit multiply-add

Browse Source
This commit is contained in:
nmzik committed 2026-10-28 22:41:29 +01:00
1 parent 0386f33bf3
commit 22610cb93e
7 files changed
+79 -11
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
+1
View File
@@ -258,6 +258,7 @@ enum class Opcode {
V_CVT_PK_U8_F32,
V_MAD_F32,
V_MAD_I16,
V_MAD_U16,
V_MAD_I32_I24,
V_MAD_U32_U24,
V_MAD_U64_U32,
src/graphics/shader/recompiler/frontend/decode/VectorAluOps.cpp
+5 -4
View File
@@ -380,6 +380,7 @@ constexpr OpcodeMap VOP3_OPCODE_LIST[] = {
{0x311u, Opcode::V_PACK_B32_F16},
{0x314u, Opcode::V_LSHLREV_B16},
{0x319u, Opcode::V_SUBREV_I32},
{0x340u, Opcode::V_MAD_U16},
{0x345u, Opcode::V_XAD_U32},
{0x346u, Opcode::V_LSHL_ADD_U32},
{0x347u, Opcode::V_ADD_LSHL_U32},
@@ -501,7 +502,7 @@ bool IsNativeVop3B16TernaryOpcode(Opcode opcode) {
return opcode == Opcode::V_MIN3_I16 || opcode == Opcode::V_MIN3_U16 ||
opcode == Opcode::V_MAX3_I16 || opcode == Opcode::V_MAX3_U16 ||
opcode == Opcode::V_MED3_I16 || opcode == Opcode::V_MED3_U16 ||
opcode == Opcode::V_MAD_I16;
opcode == Opcode::V_MAD_I16 || opcode == Opcode::V_MAD_U16;
}
bool IsNativeVop3B16BinaryOpcode(Opcode opcode) {
@@ -1536,9 +1537,9 @@ bool SupportsNativeVop3ResultModifiers(Opcode opcode) {
bool SupportsNativeVop3Clamp(Opcode opcode) {
return SupportsNativeVop3ResultModifiers(opcode) || UsesInexactClampControl(opcode) ||
opcode == Opcode::V_MAD_I16 || opcode == Opcode::V_MIN3_I16 ||
opcode == Opcode::V_MAX3_I16 || opcode == Opcode::V_MAX3_U16 ||
opcode == Opcode::V_MED3_U16;
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;
}
bool HasUnsupportedNativeVop3Modifiers(Opcode opcode, bool permlane, bool mad_mix,
src/graphics/shader/recompiler/frontend/translate/Integer.cpp
+5 -5
View File
@@ -20,12 +20,12 @@ void Translator::Integer16Binary(const Decoder::Instruction& inst, IR::ValueOpco
Write16Bits(DestinationOperand(inst), result);
}
void Translator::V_MAD_I16(const Decoder::Instruction& inst) {
const auto lhs = ReadU16AsU32(inst.src0, true);
const auto rhs = ReadU16AsU32(inst.src1, true);
auto result = ir.IAdd(ir.IMul(lhs, rhs), ReadU16AsU32(inst.src2, true));
void Translator::Integer16Mad(const Decoder::Instruction& inst, bool sign) {
const auto lhs = ReadU16AsU32(inst.src0, sign);
const auto rhs = ReadU16AsU32(inst.src1, sign);
auto result = ir.IAdd(ir.IMul(lhs, rhs), ReadU16AsU32(inst.src2, sign));
if (inst.dst.clamp) {
result = Saturate16(result, true);
result = Saturate16(result, sign);
}
Write16Bits(DestinationOperand(inst), result);
}
src/graphics/shader/recompiler/frontend/translate/Translator.h
+1 -1
View File
@@ -167,7 +167,7 @@ private:
void Integer16Shift(const Decoder::Instruction& inst, IR::ValueOpcode opcode, bool arithmetic);
void Integer16Binary(const Decoder::Instruction& inst, IR::ValueOpcode opcode, bool sign);
void Integer16Ternary(const Decoder::Instruction& inst, IR::ValueOpcode opcode, bool sign);
void V_MAD_I16(const Decoder::Instruction& inst);
void Integer16Mad(const Decoder::Instruction& inst, bool sign);
void PackedInteger16Shift(const Decoder::Instruction& inst, IR::ValueOpcode opcode,
bool arithmetic);
void PackedInteger16Binary(const Decoder::Instruction& inst, IR::ValueOpcode opcode);
src/graphics/shader/recompiler/frontend/translate/Vector.cpp
+2 -1
View File
@@ -375,7 +375,8 @@ void Translator::EmitVector(const Decoder::Instruction& inst) {
case O::V_SUB_NC_U16:
case O::V_SUB_NC_I16: return Integer16Binary(inst, IR::ValueOpcode::ISub32, false);
case O::V_MUL_LO_U16: return Integer16Binary(inst, IR::ValueOpcode::IMul32, false);
case O::V_MAD_I16: return V_MAD_I16(inst);
case O::V_MAD_I16: return Integer16Mad(inst, true);
case O::V_MAD_U16: return Integer16Mad(inst, false);
case O::V_MED3_I16: return Integer16Ternary(inst, IR::ValueOpcode::SMedTri32, true);
case O::V_MIN3_U16: return Integer16Ternary(inst, IR::ValueOpcode::UMinTri32, false);
case O::V_MIN3_I16: return Integer16Ternary(inst, IR::ValueOpcode::SMinTri32, true);