raphiaura/KytyPS5
1
0
Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-26 21:06:53 +01:00
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/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);