diff options
| author | bunnei <bunneidev@gmail.com> | 2015-08-19 04:02:25 +0200 |
|---|---|---|
| committer | bunnei <bunneidev@gmail.com> | 2015-08-19 04:02:25 +0200 |
| commit | 3c5ff418ca9a0cf3af0a8eeae76bb111a7af73bc (patch) | |
| tree | 079c8e9671bf16e9467248598d7d471dcac57943 /src | |
| parent | Merge pull request #1037 from aroulin/shader-ex2-lg2 (diff) | |
| parent | Shader: Save caller-saved registers in JIT before a CALL (diff) | |
| download | yuzu-3c5ff418ca9a0cf3af0a8eeae76bb111a7af73bc.tar yuzu-3c5ff418ca9a0cf3af0a8eeae76bb111a7af73bc.tar.gz yuzu-3c5ff418ca9a0cf3af0a8eeae76bb111a7af73bc.tar.bz2 yuzu-3c5ff418ca9a0cf3af0a8eeae76bb111a7af73bc.tar.lz yuzu-3c5ff418ca9a0cf3af0a8eeae76bb111a7af73bc.tar.xz yuzu-3c5ff418ca9a0cf3af0a8eeae76bb111a7af73bc.tar.zst yuzu-3c5ff418ca9a0cf3af0a8eeae76bb111a7af73bc.zip | |
Diffstat (limited to 'src')
| -rw-r--r-- | src/video_core/shader/shader_jit_x64.cpp | 30 | ||||
| -rw-r--r-- | src/video_core/shader/shader_jit_x64.h | 3 |
2 files changed, 33 insertions, 0 deletions
diff --git a/src/video_core/shader/shader_jit_x64.cpp b/src/video_core/shader/shader_jit_x64.cpp index 93f608584..e4b8295b3 100644 --- a/src/video_core/shader/shader_jit_x64.cpp +++ b/src/video_core/shader/shader_jit_x64.cpp @@ -280,6 +280,22 @@ void JitCompiler::Compile_UniformCondition(Instruction instr) { CMP(sizeof(bool) * 8, MDisp(UNIFORMS, offset), Imm8(0)); } +void JitCompiler::Compile_PushCallerSavedXMM() { +#ifndef _WIN32 + SUB(64, R(RSP), Imm8(2 * 16)); + MOVUPS(MDisp(RSP, 16), ONE); + MOVUPS(MDisp(RSP, 0), NEGBIT); +#endif +} + +void JitCompiler::Compile_PopCallerSavedXMM() { +#ifndef _WIN32 + MOVUPS(NEGBIT, MDisp(RSP, 0)); + MOVUPS(ONE, MDisp(RSP, 16)); + ADD(64, R(RSP), Imm8(2 * 16)); +#endif +} + void JitCompiler::Compile_ADD(Instruction instr) { Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1); Compile_SwizzleSrc(instr, 2, instr.common.src2, SRC2); @@ -334,7 +350,14 @@ void JitCompiler::Compile_DP4(Instruction instr) { void JitCompiler::Compile_EX2(Instruction instr) { Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1); MOVSS(XMM0, R(SRC1)); + + // The following will actually break the stack alignment + ABI_PushAllCallerSavedRegsAndAdjustStack(); + Compile_PushCallerSavedXMM(); ABI_CallFunction(reinterpret_cast<const void*>(exp2f)); + Compile_PopCallerSavedXMM(); + ABI_PopAllCallerSavedRegsAndAdjustStack(); + SHUFPS(XMM0, R(XMM0), _MM_SHUFFLE(0, 0, 0, 0)); MOVAPS(SRC1, R(XMM0)); Compile_DestEnable(instr, SRC1); @@ -343,7 +366,14 @@ void JitCompiler::Compile_EX2(Instruction instr) { void JitCompiler::Compile_LG2(Instruction instr) { Compile_SwizzleSrc(instr, 1, instr.common.src1, SRC1); MOVSS(XMM0, R(SRC1)); + + // The following will actually break the stack alignment + ABI_PushAllCallerSavedRegsAndAdjustStack(); + Compile_PushCallerSavedXMM(); ABI_CallFunction(reinterpret_cast<const void*>(log2f)); + Compile_PopCallerSavedXMM(); + ABI_PopAllCallerSavedRegsAndAdjustStack(); + SHUFPS(XMM0, R(XMM0), _MM_SHUFFLE(0, 0, 0, 0)); MOVAPS(SRC1, R(XMM0)); Compile_DestEnable(instr, SRC1); diff --git a/src/video_core/shader/shader_jit_x64.h b/src/video_core/shader/shader_jit_x64.h index 104f9f466..a6ae7fbf1 100644 --- a/src/video_core/shader/shader_jit_x64.h +++ b/src/video_core/shader/shader_jit_x64.h @@ -69,6 +69,9 @@ private: void Compile_EvaluateCondition(Instruction instr); void Compile_UniformCondition(Instruction instr); + void Compile_PushCallerSavedXMM(); + void Compile_PopCallerSavedXMM(); + /// Pointer to the variable that stores the current Pica code offset. Used to handle nested code blocks. unsigned* offset_ptr = nullptr; |