gremling/KytyPS5
1
0 Fork 0
mirror of https://github.com/KytyPS5/KytyPS5.git synced 2026-10-28 08:13:05 +00:00
Code Issues Packages Projects Releases Wiki Activity

shader: saturate V_CVT_PK_U16_U32 and V_CVT_PK_I16_I32

Browse Source
This commit is contained in:
TheCruZ authored and nmzik committed 2026-10-25 02:20:44 +02:00
1 parent 66e21a6245
commit 9210264d47
5 files changed
+58 -17
Download Patch File Download Diff File Expand all files Collapse all files

No files matched your search

src/graphics/shader/recompiler/frontend/translate/Convert.cpp
+18 -11
View File
@@ -49,18 +49,11 @@ void Translator::V_CVT_F16_16(const Decoder::Instruction& inst, bool signed_valu
}
void Translator::V_CVT_16_F16(const Decoder::Instruction& inst, bool signed_value) {
const auto value = ReadF16AsF32(inst.src0);
// The 32-bit conversion already saturates and maps NaN to zero.
if (signed_value) {
const auto converted = ir.Emit(IR::ValueOpcode::ConvertS32F32, {value});
const auto low = ir.Emit(IR::ValueOpcode::SMax32, {converted, IR::Value(0xffff8000u)});
Write16Bits(DestinationOperand(inst),
IR::U32(ir.Emit(IR::ValueOpcode::SMin32, {low, IR::Value(0x7fffu)})));
return;
}
const auto converted = ir.Emit(IR::ValueOpcode::ConvertU32F32, {value});
Write16Bits(DestinationOperand(inst),
IR::U32(ir.Emit(IR::ValueOpcode::UMin32, {converted, IR::Value(0xffffu)})));
const auto converted = IR::U32(
ir.Emit(signed_value ? IR::ValueOpcode::ConvertS32F32 : IR::ValueOpcode::ConvertU32F32,
{ReadF16AsF32(inst.src0)}));
Write16Bits(DestinationOperand(inst), Saturate16(converted, signed_value));
}
void Translator::V_CVT_RPI_I32_F32(const Decoder::Instruction& inst) {
@@ -130,6 +123,12 @@ void Translator::V_CVT_PK_U8_F32(const Decoder::Instruction& inst) {
{ReadU32(SourceAt(inst, 2)), byte, shift, IR::Value(8u)}));
}
void Translator::V_CVT_PK_16_32(const Decoder::Instruction& inst, bool signed_value) {
WriteOperand(DestinationOperand(inst),
PackU16Lanes(Saturate16(ReadU32(inst.src0), signed_value),
Saturate16(ReadU32(inst.src1), signed_value)));
}
void Translator::V_PACK_B32_F16(const Decoder::Instruction& inst) {
WriteOperand(DestinationOperand(inst),
PackU16Lanes(Read16LaneBits(inst.src0, false), Read16LaneBits(inst.src1, false)));
@@ -140,4 +139,12 @@ IR::U32 Translator::PackU16Lanes(IR::U32 low, IR::U32 high) {
ir.Emit(IR::ValueOpcode::BitFieldInsert, {low, high, IR::Value(16u), IR::Value(16u)}));
}
IR::U32 Translator::Saturate16(IR::U32 value, bool signed_value) {
if (!signed_value) {
return IR::U32(ir.Emit(IR::ValueOpcode::UMin32, {value, IR::Value(0xffffu)}));
}
const auto low = ir.Emit(IR::ValueOpcode::SMax32, {value, IR::Value(0xffff8000u)});
return IR::U32(ir.Emit(IR::ValueOpcode::SMin32, {low, IR::Value(0x7fffu)}));
}
} // namespace Libs::Graphics::ShaderRecompiler::Frontend