N-archive/citra
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

shader_jit_x64_compiler: Use haddps for horizontal summation

Browse source
This commit is contained in:
MerryMage 2017-12-10 22:00:04 +00:00
parent af45f2b2de
commit efec8fe513
1 changed files with 22 additions and 12 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

34
src/video_core/shader/shader_jit_x64_compiler.cpp
View file

@ -387,13 +387,18 @@ void JitShader::Compile_DP4(Instruction instr) {
Compile_SanitizedMul(SRC1, SRC2, SCRATCH);
movaps(SRC2, SRC1);
shufps(SRC1, SRC1, _MM_SHUFFLE(2, 3, 0, 1)); // XYZW -> ZWXY
addps(SRC1, SRC2);
if (Common::GetCPUCaps().sse3) {
haddps(SRC1, SRC1);
haddps(SRC1, SRC1);
} else {
movaps(SRC2, SRC1);
shufps(SRC1, SRC1, _MM_SHUFFLE(2, 3, 0, 1)); // XYZW -> ZWXY
addps(SRC1, SRC2);
movaps(SRC2, SRC1);
shufps(SRC1, SRC1, _MM_SHUFFLE(0, 1, 2, 3)); // XYZW -> WZYX
addps(SRC1, SRC2);
movaps(SRC2, SRC1);
shufps(SRC1, SRC1, _MM_SHUFFLE(0, 1, 2, 3)); // XYZW -> WZYX
addps(SRC1, SRC2);
}
Compile_DestEnable(instr, SRC1);
}
@ -419,13 +424,18 @@ void JitShader::Compile_DPH(Instruction instr) {
Compile_SanitizedMul(SRC1, SRC2, SCRATCH);
movaps(SRC2, SRC1);
shufps(SRC1, SRC1, _MM_SHUFFLE(2, 3, 0, 1)); // XYZW -> ZWXY
addps(SRC1, SRC2);
if (Common::GetCPUCaps().sse3) {
haddps(SRC1, SRC1);
haddps(SRC1, SRC1);
} else {
movaps(SRC2, SRC1);
shufps(SRC1, SRC1, _MM_SHUFFLE(2, 3, 0, 1)); // XYZW -> ZWXY
addps(SRC1, SRC2);
movaps(SRC2, SRC1);
shufps(SRC1, SRC1, _MM_SHUFFLE(0, 1, 2, 3)); // XYZW -> WZYX
addps(SRC1, SRC2);
movaps(SRC2, SRC1);
shufps(SRC1, SRC1, _MM_SHUFFLE(0, 1, 2, 3)); // XYZW -> WZYX
addps(SRC1, SRC2);
}
Compile_DestEnable(instr, SRC1);
}