raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-20 10:56:58 +02:00
Code Issues Packages Projects Releases Wiki Activity

shader: add scalar zero-bit count and search operations

Browse Source
This commit is contained in:
fxpw
2026-10-10 04:24:33 +02:00
committed by nmzik
parent 3b9e176b70
commit fa56c5eae9
7 changed files with 106 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/graphics/shader/recompiler/frontend/decode/ScalarAluOps.cpp
+4
View File
@@ -47,8 +47,12 @@ constexpr OpcodeMap SOP1_OPCODE_LIST[] = {
{0x0au, Opcode::S_WQM_B64},
{0x0bu, Opcode::S_BREV_B32},
{0x0cu, Opcode::S_BREV_B64},
{0x0du, Opcode::S_BCNT0_I32_B32},
{0x0eu, Opcode::S_BCNT0_I32_B64},
{0x0fu, Opcode::S_BCNT1_I32_B32},
{0x10u, Opcode::S_BCNT1_I32_B64},
{0x11u, Opcode::S_FF0_I32_B32},
{0x12u, Opcode::S_FF0_I32_B64},
{0x13u, Opcode::S_FF1_I32_B32},
{0x14u, Opcode::S_FF1_I32_B64},
{0x15u, Opcode::S_FLBIT_I32_B32},
src/graphics/shader/recompiler/frontend/decode/ShaderDecoder.cpp
+4
View File
@@ -510,6 +510,10 @@ std::string InstructionToString(const Instruction& inst) {
case Opcode::S_SEXT_I32_I16:
case Opcode::S_BREV_B32:
case Opcode::S_BREV_B64:
case Opcode::S_BCNT0_I32_B32:
case Opcode::S_BCNT0_I32_B64:
case Opcode::S_FF0_I32_B32:
case Opcode::S_FF0_I32_B64:
case Opcode::S_BCNT1_I32_B32:
case Opcode::S_FLBIT_I32_B32:
case Opcode::S_FF1_I32_B32:
src/graphics/shader/recompiler/frontend/decode/ShaderDecoder.h
+4
View File
@@ -50,8 +50,12 @@ enum class Opcode {
S_ABSDIFF_I32,
S_BREV_B32,
S_BREV_B64,
S_BCNT0_I32_B32,
S_BCNT0_I32_B64,
S_BCNT1_I32_B32,
S_BCNT1_I32_B64,
S_FF0_I32_B32,
S_FF0_I32_B64,
S_FF1_I32_B32,
S_FF1_I32_B64,
S_FLBIT_I32_B32,
src/graphics/shader/recompiler/frontend/translate/Integer.cpp
+6 -2
View File
@@ -221,8 +221,12 @@ void Translator::V_XOR3_B32(const Decoder::Instruction& inst) {
WriteOperand(DestinationOperand(inst), result);
}
void Translator::S_FF1_I32_B64(const Decoder::Instruction& inst) {
const auto source = ExtractU64(ReadU64(inst.src0));
void Translator::S_FF_I32_B64(const Decoder::Instruction& inst, bool find_zero) {
auto source = ReadU32Pair(inst.src0);
if (find_zero) {
source[0] = ir.BitwiseNot(source[0]);
source[1] = ir.BitwiseNot(source[1]);
}
const auto low_lsb = IR::U32(ir.Emit(IR::ValueOpcode::FindILsb32, {source[0]}));
const auto high_lsb = IR::U32(ir.Emit(IR::ValueOpcode::FindILsb32, {source[1]}));
const auto high_position = ir.IAdd(high_lsb, IR::U32(IR::Value(32u)));
src/graphics/shader/recompiler/frontend/translate/Scalar.cpp
+20 -2
View File
@@ -178,12 +178,29 @@ void Translator::EmitScalar(const Decoder::Instruction& inst) {
IR::U32(ir.Emit(IR::ValueOpcode::BitReverse32, {value[0]}))});
return;
}
case O::S_BCNT0_I32_B32:
case O::S_FF0_I32_B32: {
const auto value = ir.BitwiseNot(ReadU32(inst.src0));
const bool count = inst.opcode == O::S_BCNT0_I32_B32;
const auto result = IR::U32(ir.Emit(count ? IR::ValueOpcode::BitCount32
: IR::ValueOpcode::FindILsb32, {value}));
WriteOperand(inst.dst, result);
if (count) {
ir.SetScc(ir.INotEqual(result, IR::U32(IR::Value(0u))));
}
return;
}
case O::S_BCNT1_I32_B32:
return SimpleInteger(inst, IR::ValueOpcode::BitCount32, IR::Type::U32, false, false,
true);
case O::S_BCNT0_I32_B64:
case O::S_BCNT1_I32_B64: {
// Vulkan bit counts operate on 32-bit words; avoid packing only to split again.
const auto value = ReadU32Pair(inst.src0);
auto value = ReadU32Pair(inst.src0);
if (inst.opcode == O::S_BCNT0_I32_B64) {
value[0] = ir.BitwiseNot(value[0]);
value[1] = ir.BitwiseNot(value[1]);
}
const auto low = IR::U32(ir.Emit(IR::ValueOpcode::BitCount32, {value[0]}));
const auto high = IR::U32(ir.Emit(IR::ValueOpcode::BitCount32, {value[1]}));
const auto result = ir.IAdd(low, high);
@@ -221,7 +238,8 @@ void Translator::EmitScalar(const Decoder::Instruction& inst) {
return ComposedIntegerBinary(inst, IR::ValueOpcode::BitwiseOr32, false, true, true);
case O::S_XNOR_B32:
return ComposedIntegerBinary(inst, IR::ValueOpcode::BitwiseXor32, false, true, true);
case O::S_FF1_I32_B64: return S_FF1_I32_B64(inst);
case O::S_FF0_I32_B64: return S_FF_I32_B64(inst, true);
case O::S_FF1_I32_B64: return S_FF_I32_B64(inst, false);
case O::S_FLBIT_I32_B32: return V_FFBH_32(inst, false);
case O::S_FLBIT_I32_B64: return S_FLBIT_I32_B64(inst);
src/graphics/shader/recompiler/frontend/translate/Translator.h
+1 -1
View File
@@ -187,7 +187,7 @@ private:
void V_AND_OR_B32(const Decoder::Instruction& inst);
void V_OR3_B32(const Decoder::Instruction& inst);
void V_XOR3_B32(const Decoder::Instruction& inst);
void S_FF1_I32_B64(const Decoder::Instruction& inst);
void S_FF_I32_B64(const Decoder::Instruction& inst, bool find_zero);
void V_FFBH_32(const Decoder::Instruction& inst, bool sign);
void S_FLBIT_I32_B64(const Decoder::Instruction& inst);
void Integer24(const Decoder::Instruction& inst, bool sign, bool addend);