R600: Use shared op optimization when checking cycle compatibility

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@189981 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Vincent Lejeune 2013-09-04 19:53:54 +00:00
parent bb25a01d23
commit f94eea9e11
2 changed files with 40 additions and 0 deletions

View File

@ -336,6 +336,8 @@ R600InstrInfo::ExtractSrcs(MachineInstr *MI,
static std::vector<std::pair<int, unsigned> >
Swizzle(std::vector<std::pair<int, unsigned> > Src,
R600InstrInfo::BankSwizzle Swz) {
if (Src[0] == Src[1])
Src[1].first = -1;
switch (Swz) {
case R600InstrInfo::ALU_VEC_012_SCL_210:
break;

View File

@ -0,0 +1,38 @@
; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck %s
; CHECK: @main
; CHECK: MULADD_IEEE *
; CHECK-NOT: MULADD_IEEE *
define void @main() {
%w0 = call float @llvm.R600.load.input(i32 3)
%w1 = call float @llvm.R600.load.input(i32 7)
%w2 = call float @llvm.R600.load.input(i32 11)
%sq0 = fmul float %w0, %w0
%r0 = fadd float %sq0, 2.0
%sq1 = fmul float %w1, %w1
%r1 = fadd float %sq1, 2.0
%sq2 = fmul float %w2, %w2
%r2 = fadd float %sq2, 2.0
%v0 = insertelement <4 x float> undef, float %r0, i32 0
%v1 = insertelement <4 x float> %v0, float %r1, i32 1
%v2 = insertelement <4 x float> %v1, float %r2, i32 2
%res = call float @llvm.AMDGPU.dp4(<4 x float> %v2, <4 x float> %v2)
%vecres = insertelement <4 x float> undef, float %res, i32 0
call void @llvm.R600.store.swizzle(<4 x float> %vecres, i32 0, i32 2)
ret void
}
; Function Attrs: readnone
declare float @llvm.R600.load.input(i32) #1
; Function Attrs: readnone
declare float @llvm.AMDGPU.dp4(<4 x float>, <4 x float>) #1
declare void @llvm.R600.store.swizzle(<4 x float>, i32, i32)
attributes #0 = { "ShaderType"="1" }
attributes #1 = { readnone }
attributes #2 = { readonly }
attributes #3 = { nounwind readonly }