maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity

[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
This commit is contained in:
Craig Topper 2017-03-17 06:10:37 +00:00
parent c1338f21ed
commit 6a1290a0fd
2 changed files with 11 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
llvm/lib/Target/X86/X86InstrFMA.td
View File

@ -192,12 +192,14 @@ multiclass fma3s_forms<bits<8> opc132, bits<8> opc213, bits<8> opc231,
string OpStr, string PackTy, string Suff, string OpStr, string PackTy, string Suff,
SDNode OpNode, RegisterClass RC, SDNode OpNode, RegisterClass RC,
X86MemOperand x86memop> { X86MemOperand x86memop> {
let Predicates = [HasFMA, NoAVX512] in {
defm NAME#132#Suff : fma3s_rm<opc132, !strconcat(OpStr, "132", PackTy), defm NAME#132#Suff : fma3s_rm<opc132, !strconcat(OpStr, "132", PackTy),
x86memop, RC>; x86memop, RC>;
defm NAME#213#Suff : fma3s_rm<opc213, !strconcat(OpStr, "213", PackTy), defm NAME#213#Suff : fma3s_rm<opc213, !strconcat(OpStr, "213", PackTy),
x86memop, RC, OpNode>; x86memop, RC, OpNode>;
defm NAME#231#Suff : fma3s_rm<opc231, !strconcat(OpStr, "231", PackTy), defm NAME#231#Suff : fma3s_rm<opc231, !strconcat(OpStr, "231", PackTy),
x86memop, RC>; x86memop, RC>;
}
} }
// The FMA 213 form is created for lowering of scalar FMA intrinscis // The FMA 213 form is created for lowering of scalar FMA intrinscis

4
llvm/test/CodeGen/X86/fma.ll
View File

@ -46,7 +46,7 @@ define float @test_f32(float %a, float %b, float %c) #0 {
; ;
; AVX51264-LABEL: test_f32: ; AVX51264-LABEL: test_f32:
; AVX51264: ## BB#0: ## %entry ; 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] ; AVX51264-NEXT: retq ## encoding: [0xc3]
entry: entry:
%call = call float @llvm.fma.f32(float %a, float %b, float %c) %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-LABEL: test_f64:
; AVX51264: ## BB#0: ## %entry ; 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] ; AVX51264-NEXT: retq ## encoding: [0xc3]
entry: entry:
%call = call double @llvm.fma.f64(double %a, double %b, double %c) %call = call double @llvm.fma.f64(double %a, double %b, double %c)