shadps4-emu/shadPS4
1
0
Fork 0
mirror of https://github.com/shadps4-emu/shadPS4.git synced 2024-11-23 03:09:55 +00:00
Code Issues Actions 4 Packages Projects Releases Wiki Activity

Implement V_MED3_U32 vector ALU Opcode (#1553)

Browse Source
This commit is contained in:
Ruah Devlin 2024-11-20 16:23:59 +00:00 committed by GitHub
parent da00235eaf
commit 96cd79f272
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 11 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
src/shader_recompiler/frontend/translate/translate.h
View File

@ -225,6 +225,7 @@ public:
void V_MAX3_U32(bool is_signed, const GcnInst& inst); void V_MAX3_U32(bool is_signed, const GcnInst& inst);
void V_MED3_F32(const GcnInst& inst); void V_MED3_F32(const GcnInst& inst);
void V_MED3_I32(const GcnInst& inst); void V_MED3_I32(const GcnInst& inst);
void V_MED3_U32(const GcnInst& inst);
void V_SAD(const GcnInst& inst); void V_SAD(const GcnInst& inst);
void V_SAD_U32(const GcnInst& inst); void V_SAD_U32(const GcnInst& inst);
void V_CVT_PK_U16_U32(const GcnInst& inst); void V_CVT_PK_U16_U32(const GcnInst& inst);

10
src/shader_recompiler/frontend/translate/vector_alu.cpp
View File

@ -357,6 +357,8 @@ void Translator::EmitVectorAlu(const GcnInst& inst) {
return V_MED3_F32(inst); return V_MED3_F32(inst);
case Opcode::V_MED3_I32: case Opcode::V_MED3_I32:
return V_MED3_I32(inst); return V_MED3_I32(inst);
case Opcode::V_MED3_U32:
return V_MED3_U32(inst);
case Opcode::V_SAD_U32: case Opcode::V_SAD_U32:
return V_SAD_U32(inst); return V_SAD_U32(inst);
case Opcode::V_CVT_PK_U16_U32: case Opcode::V_CVT_PK_U16_U32:
@ -1092,6 +1094,14 @@ void Translator::V_MED3_I32(const GcnInst& inst) {
SetDst(inst.dst[0], ir.SMax(ir.SMin(src0, src1), mmx)); SetDst(inst.dst[0], ir.SMax(ir.SMin(src0, src1), mmx));
} }
void Translator::V_MED3_U32(const GcnInst& inst) {
const IR::U32 src0{GetSrc(inst.src[0])};
const IR::U32 src1{GetSrc(inst.src[1])};
const IR::U32 src2{GetSrc(inst.src[2])};
const IR::U32 mmx = ir.UMin(ir.UMax(src0, src1), src2);
SetDst(inst.dst[0], ir.UMax(ir.UMin(src0, src1), mmx));
}
void Translator::V_SAD(const GcnInst& inst) { void Translator::V_SAD(const GcnInst& inst) {
const IR::U32 abs_diff = ir.IAbs(ir.ISub(GetSrc(inst.src[0]), GetSrc(inst.src[1]))); const IR::U32 abs_diff = ir.IAbs(ir.ISub(GetSrc(inst.src[0]), GetSrc(inst.src[1])));
SetDst(inst.dst[0], ir.IAdd(abs_diff, GetSrc(inst.src[2]))); SetDst(inst.dst[0], ir.IAdd(abs_diff, GetSrc(inst.src[2])));