raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-21 11:26:56 +02:00
Code Issues Packages Projects Releases Wiki Activity

shader: preserve untouched half in VOP1 B16 writes (#370)

Browse Source 
shader: preserve native 16-bit destinations

Co-authored-by: nmzik <Nmzik@mail.ru>
This commit is contained in:
GodsWill Dev
2026-08-30 11:33:26 -04:00
committed by GitHub
parent 18af08460f
commit 9992ab18fc
8 changed files with 244 additions and 90 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/shader/recompiler/frontend/translate/Convert.cpp
+4 -2
View File
@@ -141,8 +141,10 @@ void Translator::V_CVT_OFF_F32_I4(const Decoder::Instruction& inst) {
}
void Translator::V_CVT_PKRTZ_F16_F32(const Decoder::Instruction& inst) {
const auto lhs = ReadOperand(SourceAt(inst, 0), IR::Type::F32);
const auto rhs = ReadOperand(SourceAt(inst, 1), IR::Type::F32);
const auto lhs = ApplyF32ResultModifiers(
inst.dst, IR::F32(ReadOperand(SourceAt(inst, 0), IR::Type::F32)));
const auto rhs = ApplyF32ResultModifiers(
inst.dst, IR::F32(ReadOperand(SourceAt(inst, 1), IR::Type::F32)));
WriteOperand(DestinationOperand(inst), ir.Emit(IR::ValueOpcode::PackFloat2x16Rtz, {lhs, rhs}));
}
src/graphics/shader/recompiler/frontend/translate/Memory.cpp
+1
View File
@@ -37,6 +37,7 @@ Decoder::Operand OffsetDecodedRegister(const Decoder::Operand& operand, uint32_t
result.dpp_ctrl = 0;
result.dpp_row_mask = 0xf;
result.dpp_bank_mask = 0xf;
result.explicit_sdwa_dst = false;
result.dpp_fetch_inactive = false;
result.dpp_bound_ctrl = false;
result.dpp = false;
src/graphics/shader/recompiler/frontend/translate/Translate.cpp
+22 -26
View File
@@ -104,6 +104,7 @@ Decoder::Operand Translator::PlainOperand(const Decoder::Operand& operand) {
result.dpp_ctrl = 0;
result.dpp_row_mask = 0xf;
result.dpp_bank_mask = 0xf;
result.explicit_sdwa_dst = false;
result.dpp_fetch_inactive = false;
result.dpp_bound_ctrl = false;
result.dpp = false;
@@ -392,12 +393,13 @@ void Translator::WriteOperand(const Decoder::Operand& operand, IR::Value value)
}
}
if (type == IR::Type::U16) {
WriteRawU32(operand, IR::U32(ir.Emit(IR::ValueOpcode::ConvertU32U16, {IR::U16(value)})));
Write16Bits(operand,
IR::U32(ir.Emit(IR::ValueOpcode::ConvertU32U16, {IR::U16(value)})));
return;
}
if (type == IR::Type::F16) {
const auto bits = IR::U16(ir.Emit(IR::ValueOpcode::BitCastU16F16, {value}));
WriteRawU32(operand, IR::U32(ir.Emit(IR::ValueOpcode::ConvertU32U16, {bits})));
Write16Bits(operand, IR::U32(ir.Emit(IR::ValueOpcode::ConvertU32U16, {bits})));
return;
}
if (type == IR::Type::U64) {
@@ -417,38 +419,32 @@ IR::U32 Translator::PackHalf2x16(IR::F32 low, IR::F32 high) {
return IR::U32(ir.Emit(IR::ValueOpcode::PackHalf2x16, {pair}));
}
void Translator::Write16Bits(const Decoder::Operand& operand, IR::U32 value) {
auto destination = operand;
// Native GFX10 16-bit results preserve the unselected half. Plain/DPP destinations use the
// low half; native VOP3 may already select either half. Explicit SDWA retains encoded DST_U.
if (!destination.explicit_sdwa_dst) {
if (destination.sdwa_sel == 6u) {
destination.sdwa_sel = 4u;
}
destination.sdwa_dst_unused = 2u;
}
destination.omod = 0u;
destination.op_sel = false;
destination.clamp = false;
WriteRawU32(destination, ir.BitwiseAnd(value, IR::U32(IR::Value(0xffffu))));
}
void Translator::WriteF16(const Decoder::Operand& operand, IR::F32 value) {
value = ApplyF32ResultModifiers(operand, value);
const auto half = IR::F16(ir.Emit(IR::ValueOpcode::ConvertF16F32, {value}));
const auto bits = IR::U32(ir.Emit(IR::ValueOpcode::ConvertU32U16,
{IR::U16(ir.Emit(IR::ValueOpcode::BitCastU16F16, {half}))}));
auto merged = bits;
if (operand.sdwa_sel == 4u || operand.sdwa_sel == 5u) {
merged = IR::U32(ir.Emit(IR::ValueOpcode::BitFieldInsert,
{ReadRawU32(PlainOperand(operand)), bits,
IR::Value(operand.sdwa_sel == 5u ? 16u : 0u), IR::Value(16u)}));
}
auto raw = operand;
raw.sdwa_sel = 6u;
raw.omod = 0u;
raw.op_sel = false;
raw.clamp = false;
WriteRawU32(raw, merged);
Write16Bits(operand, bits);
}
void Translator::WriteU16(const Decoder::Operand& operand, IR::U32 value) {
auto merged = ir.BitwiseAnd(value, IR::U32(IR::Value(0xffffu)));
if (operand.sdwa_sel == 4u || operand.sdwa_sel == 5u) {
merged = IR::U32(ir.Emit(IR::ValueOpcode::BitFieldInsert,
{ReadRawU32(PlainOperand(operand)), merged,
IR::Value(operand.sdwa_sel == 5u ? 16u : 0u), IR::Value(16u)}));
}
auto raw = operand;
raw.sdwa_sel = 6u;
raw.omod = 0u;
raw.op_sel = false;
raw.clamp = false;
WriteRawU32(raw, merged);
Write16Bits(operand, value);
}
IR::U32 Translator::ReadU32(const Decoder::Operand& operand) {
src/graphics/shader/recompiler/frontend/translate/Translator.h
+1
View File
@@ -34,6 +34,7 @@ private:
IR::F32 ApplyF32ResultModifiers(const Decoder::Operand& operand, IR::F32 value);
void WriteOperand(const Decoder::Operand& operand, IR::Value value);
IR::U32 PackHalf2x16(IR::F32 low, IR::F32 high);
void Write16Bits(const Decoder::Operand& operand, IR::U32 value);
void WriteF16(const Decoder::Operand& operand, IR::F32 value);
void WriteU16(const Decoder::Operand& operand, IR::U32 value);
IR::U32 ReadU32(const Decoder::Operand& operand);