[x86] Implement a faster vector population count based on the PSHUFB

[oota-llvm.git] / test / CodeGen / X86 / vec_set-3.ll

diff --git a/test/CodeGen/X86/vec_set-3.ll b/test/CodeGen/X86/vec_set-3.ll
index ada17e0092a80a6a0b78bb17bb6e0d48afe907e0..a13c813ea7b0f48a833bee5e1c8893473f365e58 100644 (file)
--- a/test/CodeGen/X86/vec_set-3.ll
+++ b/test/CodeGen/X86/vec_set-3.ll
@@ -1,17 +1,37 @@
-; RUN: llc < %s -march=x86 -mattr=+sse2 -o %t
-; RUN: grep pshufd %t | count 2
+; RUN: llc < %s -march=x86 -mattr=+sse2 -mcpu=penryn | FileCheck %s
 
-define <4 x float> @test(float %a) nounwind {
-        %tmp = insertelement <4 x float> zeroinitializer, float %a, i32 1               ; <<4 x float>> [#uses=1]
-        %tmp5 = insertelement <4 x float> %tmp, float 0.000000e+00, i32 2               ; <<4 x float>> [#uses=1]
-        %tmp6 = insertelement <4 x float> %tmp5, float 0.000000e+00, i32 3              ; <<4 x float>> [#uses=1]
-        ret <4 x float> %tmp6
+define <4 x float> @test(float %a) {
+; CHECK-LABEL: test:
+; CHECK:         insertps $29, {{.*}}, %xmm0
+; CHECK-NEXT:    retl
+
+entry:
+  %tmp = insertelement <4 x float> zeroinitializer, float %a, i32 1
+  %tmp5 = insertelement <4 x float> %tmp, float 0.000000e+00, i32 2
+  %tmp6 = insertelement <4 x float> %tmp5, float 0.000000e+00, i32 3
+  ret <4 x float> %tmp6
 }
 
-define <2 x i64> @test2(i32 %a) nounwind {
-        %tmp7 = insertelement <4 x i32> zeroinitializer, i32 %a, i32 2          ; <<4 x i32>> [#uses=1]
-        %tmp9 = insertelement <4 x i32> %tmp7, i32 0, i32 3             ; <<4 x i32>> [#uses=1]
-        %tmp10 = bitcast <4 x i32> %tmp9 to <2 x i64>           ; <<2 x i64>> [#uses=1]
-        ret <2 x i64> %tmp10
+define <2 x i64> @test2(i32 %a) {
+; CHECK-LABEL: test2:
+; CHECK:         movd {{.*}}, %xmm0
+; CHECK-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,0,1]
+; CHECK-NEXT:    retl
+
+entry:
+  %tmp7 = insertelement <4 x i32> zeroinitializer, i32 %a, i32 2
+  %tmp9 = insertelement <4 x i32> %tmp7, i32 0, i32 3
+  %tmp10 = bitcast <4 x i32> %tmp9 to <2 x i64>
+  ret <2 x i64> %tmp10
 }
 
+define <4 x float> @test3(<4 x float> %A) {
+; CHECK-LABEL: test3:
+; CHECK:         insertps {{.*#+}} xmm0 = zero,xmm0[0],zero,zero
+; CHECK-NEXT:    retl
+
+  %tmp0 = extractelement <4 x float> %A, i32 0
+  %tmp1 = insertelement <4 x float> <float 0.000000e+00, float undef, float undef, float undef >, float %tmp0, i32 1
+  %tmp2 = insertelement <4 x float> %tmp1, float 0.000000e+00, i32 2
+  ret <4 x float> %tmp2
+}