sneedmca/Lime3DS
1
0
Fork 0
mirror of https://github.com/Lime3DS/Lime3DS synced 2024-12-28 01:52:29 -06:00
Code Issues Projects Releases Packages Wiki Activity Actions

Merge pull request #1055 from aroulin/shader-sge-sgei-slt

Browse source 
Shader: Implement SGE, SGEI and SLT in interpreter/JIT
This commit is contained in:
bunnei 2015-08-19 19:58:49 -04:00
commit a575399fbc
3 changed files with 50 additions and 15 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

14
src/video_core/shader/shader_interpreter.cpp
View file

@ -278,6 +278,20 @@ void RunInterpreter(UnitState<Debug>& state) {
break; break;
} }
case OpCode::Id::SGE:
case OpCode::Id::SGEI:
Record<DebugDataRecord::SRC1>(state.debug, iteration, src1);
Record<DebugDataRecord::SRC2>(state.debug, iteration, src2);
Record<DebugDataRecord::DEST_IN>(state.debug, iteration, dest);
for (int i = 0; i < 4; ++i) {
if (!swizzle.DestComponentEnabled(i))
continue;
dest[i] = (src1[i] >= src2[i]) ? float24::FromFloat32(1.0f) : float24::FromFloat32(0.0f);
}
Record<DebugDataRecord::DEST_OUT>(state.debug, iteration, dest);
break;
case OpCode::Id::SLT: case OpCode::Id::SLT:
case OpCode::Id::SLTI: case OpCode::Id::SLTI:
Record<DebugDataRecord::SRC1>(state.debug, iteration, src1); Record<DebugDataRecord::SRC1>(state.debug, iteration, src1);

48
src/video_core/shader/shader_jit_x64.cpp
View file

@ -29,8 +29,8 @@ const JitFunction instr_table[64] = {
&JitCompiler::Compile_LG2, // lg2 &JitCompiler::Compile_LG2, // lg2
nullptr, // unknown nullptr, // unknown
&JitCompiler::Compile_MUL, // mul &JitCompiler::Compile_MUL, // mul
nullptr, // lge &JitCompiler::Compile_SGE, // sge
nullptr, // slt &JitCompiler::Compile_SLT, // slt
&JitCompiler::Compile_FLR, // flr &JitCompiler::Compile_FLR, // flr
&JitCompiler::Compile_MAX, // max &JitCompiler::Compile_MAX, // max
&JitCompiler::Compile_MIN, // min &JitCompiler::Compile_MIN, // min
@ -46,8 +46,8 @@ const JitFunction instr_table[64] = {
nullptr, // unknown nullptr, // unknown
nullptr, // dphi nullptr, // dphi
nullptr, // unknown nullptr, // unknown
nullptr, // sgei &JitCompiler::Compile_SGE, // sgei
&JitCompiler::Compile_SLTI, // slti &JitCompiler::Compile_SLT, // slti
nullptr, // unknown nullptr, // unknown
nullptr, // unknown nullptr, // unknown
nullptr, // unknown nullptr, // unknown
@ -386,6 +386,36 @@ void JitCompiler::Compile_MUL(Instruction instr) {
Compile_DestEnable(instr, SRC1); Compile_DestEnable(instr, SRC1);
} }
void JitCompiler::Compile_SGE(Instruction instr) {
if (instr.opcode.Value().EffectiveOpCode() == OpCode::Id::SGEI) {
Compile_SwizzleSrc(instr, 1, instr.common.src1i, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2i, SRC2);
} else {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2, SRC2);
}
CMPPS(SRC1, R(SRC2), CMP_NLT);
ANDPS(SRC1, R(ONE));
Compile_DestEnable(instr, SRC1);
}
void JitCompiler::Compile_SLT(Instruction instr) {
if (instr.opcode.Value().EffectiveOpCode() == OpCode::Id::SLTI) {
Compile_SwizzleSrc(instr, 1, instr.common.src1i, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2i, SRC2);
} else {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
Compile_SwizzleSrc(instr, 2, instr.common.src2, SRC2);
}
CMPPS(SRC1, R(SRC2), CMP_LT);
ANDPS(SRC1, R(ONE));
Compile_DestEnable(instr, SRC1);
}
void JitCompiler::Compile_FLR(Instruction instr) { void JitCompiler::Compile_FLR(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1); Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);
@ -463,16 +493,6 @@ void JitCompiler::Compile_MOV(Instruction instr) {
Compile_DestEnable(instr, SRC1); Compile_DestEnable(instr, SRC1);
} }
void JitCompiler::Compile_SLTI(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1i, SRC1);
Compile_SwizzleSrc(instr, 1, instr.common.src2i, SRC2);
CMPSS(SRC1, R(SRC2), CMP_LT);
ANDPS(SRC1, R(ONE));
Compile_DestEnable(instr, SRC1);
}
void JitCompiler::Compile_RCP(Instruction instr) { void JitCompiler::Compile_RCP(Instruction instr) {
Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1); Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1);

3
src/video_core/shader/shader_jit_x64.h
View file

@ -40,6 +40,8 @@ public:
void Compile_EX2(Instruction instr); void Compile_EX2(Instruction instr);
void Compile_LG2(Instruction instr); void Compile_LG2(Instruction instr);
void Compile_MUL(Instruction instr); void Compile_MUL(Instruction instr);
void Compile_SGE(Instruction instr);
void Compile_SLT(Instruction instr);
void Compile_FLR(Instruction instr); void Compile_FLR(Instruction instr);
void Compile_MAX(Instruction instr); void Compile_MAX(Instruction instr);
void Compile_MIN(Instruction instr); void Compile_MIN(Instruction instr);
@ -47,7 +49,6 @@ public:
void Compile_RSQ(Instruction instr); void Compile_RSQ(Instruction instr);
void Compile_MOVA(Instruction instr); void Compile_MOVA(Instruction instr);
void Compile_MOV(Instruction instr); void Compile_MOV(Instruction instr);
void Compile_SLTI(Instruction instr);
void Compile_NOP(Instruction instr); void Compile_NOP(Instruction instr);
void Compile_END(Instruction instr); void Compile_END(Instruction instr);
void Compile_CALL(Instruction instr); void Compile_CALL(Instruction instr);