diff options
| author | Craig Topper <craig.topper@gmail.com> | 2017-03-17 06:10:37 +0000 |
|---|---|---|
| committer | Craig Topper <craig.topper@gmail.com> | 2017-03-17 06:10:37 +0000 |
| commit | 6a1290a0fd35776f87c81dbf631f1b73c6a9b59a (patch) | |
| tree | d6a9b34b067391d6996d94782b76db4bacf4bf26 /llvm | |
| parent | c1338f21ed8f3f06067b058914ad05ee2abeb7e5 (diff) | |
| download | bcm5719-llvm-6a1290a0fd35776f87c81dbf631f1b73c6a9b59a.tar.gz bcm5719-llvm-6a1290a0fd35776f87c81dbf631f1b73c6a9b59a.zip | |
[AVX-512] Give priority to EVEX encoded scalar FMA instructions when we have FMA, AVX512 and no VLX.
We were giving priority if VLX was enabled.
llvm-svn: 298046
Diffstat (limited to 'llvm')
| -rw-r--r-- | llvm/lib/Target/X86/X86InstrFMA.td | 16 | ||||
| -rw-r--r-- | llvm/test/CodeGen/X86/fma.ll | 4 |
2 files changed, 11 insertions, 9 deletions
diff --git a/llvm/lib/Target/X86/X86InstrFMA.td b/llvm/lib/Target/X86/X86InstrFMA.td index 4b19f801dae..1941ae57f0f 100644 --- a/llvm/lib/Target/X86/X86InstrFMA.td +++ b/llvm/lib/Target/X86/X86InstrFMA.td @@ -191,13 +191,15 @@ multiclass fma3s_rm_int<bits<8> opc, string OpcodeStr, multiclass fma3s_forms<bits<8> opc132, bits<8> opc213, bits<8> opc231, string OpStr, string PackTy, string Suff, SDNode OpNode, RegisterClass RC, - X86MemOperand x86memop> { - defm NAME#132#Suff : fma3s_rm<opc132, !strconcat(OpStr, "132", PackTy), - x86memop, RC>; - defm NAME#213#Suff : fma3s_rm<opc213, !strconcat(OpStr, "213", PackTy), - x86memop, RC, OpNode>; - defm NAME#231#Suff : fma3s_rm<opc231, !strconcat(OpStr, "231", PackTy), - x86memop, RC>; + X86MemOperand x86memop> { + let Predicates = [HasFMA, NoAVX512] in { + defm NAME#132#Suff : fma3s_rm<opc132, !strconcat(OpStr, "132", PackTy), + x86memop, RC>; + defm NAME#213#Suff : fma3s_rm<opc213, !strconcat(OpStr, "213", PackTy), + x86memop, RC, OpNode>; + defm NAME#231#Suff : fma3s_rm<opc231, !strconcat(OpStr, "231", PackTy), + x86memop, RC>; + } } // The FMA 213 form is created for lowering of scalar FMA intrinscis diff --git a/llvm/test/CodeGen/X86/fma.ll b/llvm/test/CodeGen/X86/fma.ll index 8c50677b935..f5002c01846 100644 --- a/llvm/test/CodeGen/X86/fma.ll +++ b/llvm/test/CodeGen/X86/fma.ll @@ -46,7 +46,7 @@ define float @test_f32(float %a, float %b, float %c) #0 { ; ; AVX51264-LABEL: test_f32: ; AVX51264: ## BB#0: ## %entry -; AVX51264-NEXT: vfmadd213ss %xmm2, %xmm1, %xmm0 ## encoding: [0xc4,0xe2,0x71,0xa9,0xc2] +; AVX51264-NEXT: vfmadd213ss %xmm2, %xmm1, %xmm0 ## EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xa9,0xc2] ; AVX51264-NEXT: retq ## encoding: [0xc3] entry: %call = call float @llvm.fma.f32(float %a, float %b, float %c) @@ -91,7 +91,7 @@ define double @test_f64(double %a, double %b, double %c) #0 { ; ; AVX51264-LABEL: test_f64: ; AVX51264: ## BB#0: ## %entry -; AVX51264-NEXT: vfmadd213sd %xmm2, %xmm1, %xmm0 ## encoding: [0xc4,0xe2,0xf1,0xa9,0xc2] +; AVX51264-NEXT: vfmadd213sd %xmm2, %xmm1, %xmm0 ## EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xa9,0xc2] ; AVX51264-NEXT: retq ## encoding: [0xc3] entry: %call = call double @llvm.fma.f64(double %a, double %b, double %c) |
