diff options
6 files changed, 3 insertions, 14 deletions
diff --git a/src/gallium/drivers/swr/rasterizer/codegen/gen_llvm_ir_macros.py b/src/gallium/drivers/swr/rasterizer/codegen/gen_llvm_ir_macros.py index 3578a7534ae..485403ae1ec 100644 --- a/src/gallium/drivers/swr/rasterizer/codegen/gen_llvm_ir_macros.py +++ b/src/gallium/drivers/swr/rasterizer/codegen/gen_llvm_ir_macros.py @@ -57,7 +57,6 @@ intrinsics = [ ['VHSUBPS', ['a', 'b'], 'a'], ['VPTESTC', ['a', 'b'], 'mInt32Ty'], ['VPTESTZ', ['a', 'b'], 'mInt32Ty'], - ['VFMADDPS', ['a', 'b', 'c'], 'a'], ['VPHADDD', ['a', 'b'], 'a'], ['PDEP32', ['a', 'b'], 'a'], ['RDTSC', [], 'mInt64Ty'], @@ -71,6 +70,7 @@ llvm_intrinsics = [ ['STACKRESTORE', 'stackrestore', ['a'], []], ['VMINPS', 'minnum', ['a', 'b'], ['a']], ['VMAXPS', 'maxnum', ['a', 'b'], ['a']], + ['VFMADDPS', 'fmuladd', ['a', 'b', 'c'], ['a']], ['DEBUGTRAP', 'debugtrap', [], []], ['POPCNT', 'ctpop', ['a'], ['a']], ['LOG2', 'log2', ['a'], ['a']], diff --git a/src/gallium/drivers/swr/rasterizer/jitter/blend_jit.cpp b/src/gallium/drivers/swr/rasterizer/jitter/blend_jit.cpp index f89c502db7d..d5328c8e4e6 100644 --- a/src/gallium/drivers/swr/rasterizer/jitter/blend_jit.cpp +++ b/src/gallium/drivers/swr/rasterizer/jitter/blend_jit.cpp @@ -870,7 +870,6 @@ struct BlendJit : public Builder passes.add(createCFGSimplificationPass()); passes.add(createEarlyCSEPass()); passes.add(createInstructionCombiningPass()); - passes.add(createInstructionSimplifierPass()); passes.add(createConstantPropagationPass()); passes.add(createSCCPPass()); passes.add(createAggressiveDCEPass()); diff --git a/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.cpp b/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.cpp index 4116dad4430..26d8688f5e9 100644 --- a/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.cpp +++ b/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.cpp @@ -755,15 +755,8 @@ namespace SwrJit Value* Builder::FMADDPS(Value* a, Value* b, Value* c) { Value* vOut; - // use FMADs if available - if (JM()->mArch.AVX2()) - { - vOut = VFMADDPS(a, b, c); - } - else - { - vOut = FADD(FMUL(a, b), c); - } + // This maps to LLVM fmuladd intrinsic + vOut = VFMADDPS(a, b, c); return vOut; } diff --git a/src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.cpp b/src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.cpp index b4d326ebdcc..3ad0fabe81f 100644 --- a/src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.cpp +++ b/src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.cpp @@ -294,7 +294,6 @@ Function* FetchJit::Create(const FETCH_COMPILE_STATE& fetchState) optPasses.add(createCFGSimplificationPass()); optPasses.add(createEarlyCSEPass()); optPasses.add(createInstructionCombiningPass()); - optPasses.add(createInstructionSimplifierPass()); optPasses.add(createConstantPropagationPass()); optPasses.add(createSCCPPass()); optPasses.add(createAggressiveDCEPass()); diff --git a/src/gallium/drivers/swr/rasterizer/jitter/functionpasses/lower_x86.cpp b/src/gallium/drivers/swr/rasterizer/jitter/functionpasses/lower_x86.cpp index 7605823c04d..c34959d35ee 100644 --- a/src/gallium/drivers/swr/rasterizer/jitter/functionpasses/lower_x86.cpp +++ b/src/gallium/drivers/swr/rasterizer/jitter/functionpasses/lower_x86.cpp @@ -76,7 +76,6 @@ namespace SwrJit {"meta.intrinsic.VCVTPS2PH", Intrinsic::x86_vcvtps2ph_256}, {"meta.intrinsic.VPTESTC", Intrinsic::x86_avx_ptestc_256}, {"meta.intrinsic.VPTESTZ", Intrinsic::x86_avx_ptestz_256}, - {"meta.intrinsic.VFMADDPS", Intrinsic::x86_fma_vfmadd_ps_256}, {"meta.intrinsic.VPHADDD", Intrinsic::x86_avx2_phadd_d}, {"meta.intrinsic.PDEP32", Intrinsic::x86_bmi_pdep_32}, {"meta.intrinsic.RDTSC", Intrinsic::x86_rdtsc}, diff --git a/src/gallium/drivers/swr/rasterizer/jitter/streamout_jit.cpp b/src/gallium/drivers/swr/rasterizer/jitter/streamout_jit.cpp index 8f86af2a4b4..11ad36521b3 100644 --- a/src/gallium/drivers/swr/rasterizer/jitter/streamout_jit.cpp +++ b/src/gallium/drivers/swr/rasterizer/jitter/streamout_jit.cpp @@ -306,7 +306,6 @@ struct StreamOutJit : public Builder passes.add(createCFGSimplificationPass()); passes.add(createEarlyCSEPass()); passes.add(createInstructionCombiningPass()); - passes.add(createInstructionSimplifierPass()); passes.add(createConstantPropagationPass()); passes.add(createSCCPPass()); passes.add(createAggressiveDCEPass()); |