1 ; RUN: opt < %s -scalarrepl -S | FileCheck %s
2 target datalayout = "E-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64"
3 target triple = "x86_64-apple-darwin10.0.0"
5 define void @test1(<4 x float>* %F, float %f) {
7 %G = alloca <4 x float>, align 16 ; <<4 x float>*> [#uses=3]
8 %tmp = load <4 x float>* %F ; <<4 x float>> [#uses=2]
9 %tmp3 = fadd <4 x float> %tmp, %tmp ; <<4 x float>> [#uses=1]
10 store <4 x float> %tmp3, <4 x float>* %G
11 %G.upgrd.1 = getelementptr <4 x float>* %G, i32 0, i32 0 ; <float*> [#uses=1]
12 store float %f, float* %G.upgrd.1
13 %tmp4 = load <4 x float>* %G ; <<4 x float>> [#uses=2]
14 %tmp6 = fadd <4 x float> %tmp4, %tmp4 ; <<4 x float>> [#uses=1]
15 store <4 x float> %tmp6, <4 x float>* %F
17 ; CHECK-LABEL: @test1(
19 ; CHECK: %tmp = load <4 x float>* %F
20 ; CHECK: fadd <4 x float> %tmp, %tmp
21 ; CHECK-NEXT: insertelement <4 x float> %tmp3, float %f, i32 0
24 define void @test2(<4 x float>* %F, float %f) {
26 %G = alloca <4 x float>, align 16 ; <<4 x float>*> [#uses=3]
27 %tmp = load <4 x float>* %F ; <<4 x float>> [#uses=2]
28 %tmp3 = fadd <4 x float> %tmp, %tmp ; <<4 x float>> [#uses=1]
29 store <4 x float> %tmp3, <4 x float>* %G
30 %tmp.upgrd.2 = getelementptr <4 x float>* %G, i32 0, i32 2 ; <float*> [#uses=1]
31 store float %f, float* %tmp.upgrd.2
32 %tmp4 = load <4 x float>* %G ; <<4 x float>> [#uses=2]
33 %tmp6 = fadd <4 x float> %tmp4, %tmp4 ; <<4 x float>> [#uses=1]
34 store <4 x float> %tmp6, <4 x float>* %F
36 ; CHECK-LABEL: @test2(
38 ; CHECK: %tmp = load <4 x float>* %F
39 ; CHECK: fadd <4 x float> %tmp, %tmp
40 ; CHECK-NEXT: insertelement <4 x float> %tmp3, float %f, i32 2
43 define void @test3(<4 x float>* %F, float* %f) {
45 %G = alloca <4 x float>, align 16 ; <<4 x float>*> [#uses=2]
46 %tmp = load <4 x float>* %F ; <<4 x float>> [#uses=2]
47 %tmp3 = fadd <4 x float> %tmp, %tmp ; <<4 x float>> [#uses=1]
48 store <4 x float> %tmp3, <4 x float>* %G
49 %tmp.upgrd.3 = getelementptr <4 x float>* %G, i32 0, i32 2 ; <float*> [#uses=1]
50 %tmp.upgrd.4 = load float* %tmp.upgrd.3 ; <float> [#uses=1]
51 store float %tmp.upgrd.4, float* %f
53 ; CHECK-LABEL: @test3(
55 ; CHECK: %tmp = load <4 x float>* %F
56 ; CHECK: fadd <4 x float> %tmp, %tmp
57 ; CHECK-NEXT: extractelement <4 x float> %tmp3, i32 2
60 define void @test4(<4 x float>* %F, float* %f) {
62 %G = alloca <4 x float>, align 16 ; <<4 x float>*> [#uses=2]
63 %tmp = load <4 x float>* %F ; <<4 x float>> [#uses=2]
64 %tmp3 = fadd <4 x float> %tmp, %tmp ; <<4 x float>> [#uses=1]
65 store <4 x float> %tmp3, <4 x float>* %G
66 %G.upgrd.5 = getelementptr <4 x float>* %G, i32 0, i32 0 ; <float*> [#uses=1]
67 %tmp.upgrd.6 = load float* %G.upgrd.5 ; <float> [#uses=1]
68 store float %tmp.upgrd.6, float* %f
70 ; CHECK-LABEL: @test4(
72 ; CHECK: %tmp = load <4 x float>* %F
73 ; CHECK: fadd <4 x float> %tmp, %tmp
74 ; CHECK-NEXT: extractelement <4 x float> %tmp3, i32 0
77 define i32 @test5(float %X) { ;; should turn into bitcast.
78 %X_addr = alloca [4 x float]
79 %X1 = getelementptr [4 x float]* %X_addr, i32 0, i32 2
80 store float %X, float* %X1
81 %a = bitcast float* %X1 to i32*
84 ; CHECK-LABEL: @test5(
85 ; CHECK-NEXT: bitcast float %X to i32
89 define i64 @test6(<2 x float> %X) {
90 %X_addr = alloca <2 x float>
91 store <2 x float> %X, <2 x float>* %X_addr
92 %P = bitcast <2 x float>* %X_addr to i64*
95 ; CHECK-LABEL: @test6(
96 ; CHECK: bitcast <2 x float> %X to i64
100 %struct.test7 = type { [6 x i32] }
102 define void @test7() {
104 %memtmp = alloca %struct.test7, align 16
105 %0 = bitcast %struct.test7* %memtmp to <4 x i32>*
106 store <4 x i32> zeroinitializer, <4 x i32>* %0, align 16
107 %1 = getelementptr inbounds %struct.test7* %memtmp, i64 0, i32 0, i64 5
108 store i32 0, i32* %1, align 4
110 ; CHECK-LABEL: @test7(
115 ; When promoting an alloca to a 1-element vector type, instructions that
116 ; produce that same vector type should not be changed to insert one element
117 ; into a new vector. <rdar://problem/14249078>
118 define <1 x i64> @test8(<1 x i64> %a) {
120 %a.addr = alloca <1 x i64>, align 8
121 %__a = alloca <1 x i64>, align 8
122 %tmp = alloca <1 x i64>, align 8
123 store <1 x i64> %a, <1 x i64>* %a.addr, align 8
124 %0 = load <1 x i64>* %a.addr, align 8
125 store <1 x i64> %0, <1 x i64>* %__a, align 8
126 %1 = load <1 x i64>* %__a, align 8
127 %2 = bitcast <1 x i64> %1 to <8 x i8>
128 %3 = bitcast <8 x i8> %2 to <1 x i64>
129 %vshl_n = shl <1 x i64> %3, <i64 4>
130 store <1 x i64> %vshl_n, <1 x i64>* %tmp
131 %4 = load <1 x i64>* %tmp
133 ; CHECK-LABEL: @test8(
135 ; CHECK-NOT: insertelement
136 ; CHECK: ret <1 x i64>