[x86] Teach the vector combiner that picks a canonical shuffle from to

[oota-llvm.git] / test / CodeGen / X86 / vec_extract.ll

diff --git a/test/CodeGen/X86/vec_extract.ll b/test/CodeGen/X86/vec_extract.ll
index b45f9398e4fbb443436a524580c214886569597a..6391ef61682a2e9a447d70732c6c244204aa836b 100644 (file)
--- a/test/CodeGen/X86/vec_extract.ll
+++ b/test/CodeGen/X86/vec_extract.ll
@@ -1,25 +1,37 @@
-; RUN: llvm-as < %s | llc -march=x86 -mattr=+sse2 -o %t -f
-; RUN: grep movss    %t | count 3
-; RUN: grep movhlps  %t | count 1
-; RUN: grep pshufd   %t | count 1
-; RUN: grep unpckhpd %t | count 1
+; RUN: llc < %s -mcpu=corei7 -march=x86 -mattr=+sse2,-sse4.1 | FileCheck %s
 
 define void @test1(<4 x float>* %F, float* %f) nounwind {
+; CHECK-LABEL: test1:
+; CHECK:         addps %[[X:xmm[0-9]+]], %[[X]]
+; CHECK-NEXT:    movss %[[X]], {{.*}}(%{{.*}})
+; CHECK-NEXT:    retl
+entry:
        %tmp = load <4 x float>* %F             ; <<4 x float>> [#uses=2]
-       %tmp7 = add <4 x float> %tmp, %tmp              ; <<4 x float>> [#uses=1]
+       %tmp7 = fadd <4 x float> %tmp, %tmp             ; <<4 x float>> [#uses=1]
        %tmp2 = extractelement <4 x float> %tmp7, i32 0         ; <float> [#uses=1]
        store float %tmp2, float* %f
        ret void
 }
 
 define float @test2(<4 x float>* %F, float* %f) nounwind {
+; CHECK-LABEL: test2:
+; CHECK:         addps %[[X:xmm[0-9]+]], %[[X]]
+; CHECK-NEXT:    movhlps %[[X]], %[[X2:xmm[0-9]+]]
+; CHECK-NEXT:    movss %[[X2]], [[mem:.*\(%.*\)]]
+; CHECK-NEXT:    flds [[mem]]
+entry:
        %tmp = load <4 x float>* %F             ; <<4 x float>> [#uses=2]
-       %tmp7 = add <4 x float> %tmp, %tmp              ; <<4 x float>> [#uses=1]
+       %tmp7 = fadd <4 x float> %tmp, %tmp             ; <<4 x float>> [#uses=1]
        %tmp2 = extractelement <4 x float> %tmp7, i32 2         ; <float> [#uses=1]
        ret float %tmp2
 }
 
 define void @test3(float* %R, <4 x float>* %P1) nounwind {
+; CHECK-LABEL: test3:
+; CHECK:         movss {{.*}}(%{{.*}}), %[[X:xmm[0-9]+]]
+; CHECK-NEXT:    movss %[[X]], {{.*}}(%{{.*}})
+; CHECK-NEXT:    retl
+entry:
        %X = load <4 x float>* %P1              ; <<4 x float>> [#uses=1]
        %tmp = extractelement <4 x float> %X, i32 3             ; <float> [#uses=1]
        store float %tmp, float* %R
@@ -27,9 +39,16 @@ define void @test3(float* %R, <4 x float>* %P1) nounwind {
 }
 
 define double @test4(double %A) nounwind {
+; CHECK-LABEL: test4:
+; CHECK:         calll {{.*}}foo
+; CHECK-NEXT:    movhlps %[[X:xmm[0-9]+]], %[[X]]
+; CHECK-NEXT:    addsd {{.*}}(%{{.*}}), %[[X2]]
+; CHECK-NEXT:    movsd %[[X2]], [[mem:.*\(%.*\)]]
+; CHECK-NEXT:    fldl [[mem]]
+entry:
        %tmp1 = call <2 x double> @foo( )               ; <<2 x double>> [#uses=1]
        %tmp2 = extractelement <2 x double> %tmp1, i32 1                ; <double> [#uses=1]
-       %tmp3 = add double %tmp2, %A            ; <double> [#uses=1]
+       %tmp3 = fadd double %tmp2, %A           ; <double> [#uses=1]
        ret double %tmp3
 }