1 ; RUN: opt < %s -instcombine -S | FileCheck %s
3 target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64"
4 target triple = "x86_64-apple-darwin10.0.0"
6 ; Bitcasts between vectors and scalars are valid.
8 define i32 @test1(i64 %a) {
9 %t1 = bitcast i64 %a to <2 x i32>
10 %t2 = bitcast i64 %a to <2 x i32>
11 %t3 = xor <2 x i32> %t1, %t2
12 %t4 = extractelement <2 x i32> %t3, i32 0
15 ; CHECK-LABEL: @test1(
19 ; Optimize bitcasts that are extracting low element of vector. This happens
22 define float @test2(<2 x float> %A, <2 x i32> %B) {
23 %tmp28 = bitcast <2 x float> %A to i64 ; <i64> [#uses=2]
24 %tmp23 = trunc i64 %tmp28 to i32 ; <i32> [#uses=1]
25 %tmp24 = bitcast i32 %tmp23 to float ; <float> [#uses=1]
27 %tmp = bitcast <2 x i32> %B to i64
28 %tmp2 = trunc i64 %tmp to i32 ; <i32> [#uses=1]
29 %tmp4 = bitcast i32 %tmp2 to float ; <float> [#uses=1]
31 %add = fadd float %tmp24, %tmp4
34 ; CHECK-LABEL: @test2(
35 ; CHECK-NEXT: %tmp24 = extractelement <2 x float> %A, i32 0
36 ; CHECK-NEXT: bitcast <2 x i32> %B to <2 x float>
37 ; CHECK-NEXT: %tmp4 = extractelement <2 x float> {{.*}}, i32 0
38 ; CHECK-NEXT: %add = fadd float %tmp24, %tmp4
39 ; CHECK-NEXT: ret float %add
42 ; Optimize bitcasts that are extracting other elements of a vector. This
43 ; happens because of SRoA.
45 define float @test3(<2 x float> %A, <2 x i64> %B) {
46 %tmp28 = bitcast <2 x float> %A to i64
47 %tmp29 = lshr i64 %tmp28, 32
48 %tmp23 = trunc i64 %tmp29 to i32
49 %tmp24 = bitcast i32 %tmp23 to float
51 %tmp = bitcast <2 x i64> %B to i128
52 %tmp1 = lshr i128 %tmp, 64
53 %tmp2 = trunc i128 %tmp1 to i32
54 %tmp4 = bitcast i32 %tmp2 to float
56 %add = fadd float %tmp24, %tmp4
59 ; CHECK-LABEL: @test3(
60 ; CHECK-NEXT: %tmp24 = extractelement <2 x float> %A, i32 1
61 ; CHECK-NEXT: bitcast <2 x i64> %B to <4 x float>
62 ; CHECK-NEXT: %tmp4 = extractelement <4 x float> {{.*}}, i32 2
63 ; CHECK-NEXT: %add = fadd float %tmp24, %tmp4
64 ; CHECK-NEXT: ret float %add
67 ; Both bitcasts are unnecessary; change the extractelement.
69 define float @bitcast_extelt1(<2 x float> %A) {
70 %bc1 = bitcast <2 x float> %A to <2 x i32>
71 %ext = extractelement <2 x i32> %bc1, i32 0
72 %bc2 = bitcast i32 %ext to float
75 ; CHECK-LABEL: @bitcast_extelt1(
76 ; CHECK-NEXT: %bc2 = extractelement <2 x float> %A, i32 0
77 ; CHECK-NEXT: ret float %bc2
80 ; Second bitcast can be folded into the first.
82 define i64 @bitcast_extelt2(<4 x float> %A) {
83 %bc1 = bitcast <4 x float> %A to <2 x double>
84 %ext = extractelement <2 x double> %bc1, i32 1
85 %bc2 = bitcast double %ext to i64
88 ; CHECK-LABEL: @bitcast_extelt2(
89 ; CHECK-NEXT: %bc = bitcast <4 x float> %A to <2 x i64>
90 ; CHECK-NEXT: %bc2 = extractelement <2 x i64> %bc, i32 1
91 ; CHECK-NEXT: ret i64 %bc2
94 define <2 x i32> @test4(i32 %A, i32 %B){
95 %tmp38 = zext i32 %A to i64
96 %tmp32 = zext i32 %B to i64
97 %tmp33 = shl i64 %tmp32, 32
98 %ins35 = or i64 %tmp33, %tmp38
99 %tmp43 = bitcast i64 %ins35 to <2 x i32>
101 ; CHECK-LABEL: @test4(
102 ; CHECK-NEXT: insertelement <2 x i32> undef, i32 %A, i32 0
103 ; CHECK-NEXT: insertelement <2 x i32> {{.*}}, i32 %B, i32 1
104 ; CHECK-NEXT: ret <2 x i32>
109 define <2 x float> @test5(float %A, float %B) {
110 %tmp37 = bitcast float %A to i32
111 %tmp38 = zext i32 %tmp37 to i64
112 %tmp31 = bitcast float %B to i32
113 %tmp32 = zext i32 %tmp31 to i64
114 %tmp33 = shl i64 %tmp32, 32
115 %ins35 = or i64 %tmp33, %tmp38
116 %tmp43 = bitcast i64 %ins35 to <2 x float>
117 ret <2 x float> %tmp43
118 ; CHECK-LABEL: @test5(
119 ; CHECK-NEXT: insertelement <2 x float> undef, float %A, i32 0
120 ; CHECK-NEXT: insertelement <2 x float> {{.*}}, float %B, i32 1
121 ; CHECK-NEXT: ret <2 x float>
124 define <2 x float> @test6(float %A){
125 %tmp23 = bitcast float %A to i32 ; <i32> [#uses=1]
126 %tmp24 = zext i32 %tmp23 to i64 ; <i64> [#uses=1]
127 %tmp25 = shl i64 %tmp24, 32 ; <i64> [#uses=1]
128 %mask20 = or i64 %tmp25, 1109917696 ; <i64> [#uses=1]
129 %tmp35 = bitcast i64 %mask20 to <2 x float> ; <<2 x float>> [#uses=1]
130 ret <2 x float> %tmp35
131 ; CHECK-LABEL: @test6(
132 ; CHECK-NEXT: insertelement <2 x float> <float 4.200000e+01, float undef>, float %A, i32 1
136 define i64 @ISPC0(i64 %in) {
137 %out = and i64 %in, xor (i64 bitcast (<4 x i16> <i16 -1, i16 -1, i16 -1, i16 -1> to i64), i64 -1)
139 ; CHECK-LABEL: @ISPC0(
144 define i64 @Vec2(i64 %in) {
145 %out = and i64 %in, xor (i64 bitcast (<4 x i16> <i16 0, i16 0, i16 0, i16 0> to i64), i64 0)
147 ; CHECK-LABEL: @Vec2(
151 define i64 @All11(i64 %in) {
152 %out = and i64 %in, xor (i64 bitcast (<2 x float> bitcast (i64 -1 to <2 x float>) to i64), i64 -1)
154 ; CHECK-LABEL: @All11(
159 define i32 @All111(i32 %in) {
160 %out = and i32 %in, xor (i32 bitcast (<1 x float> bitcast (i32 -1 to <1 x float>) to i32), i32 -1)
162 ; CHECK-LABEL: @All111(
166 define <2 x i16> @BitcastInsert(i32 %a) {
167 %v = insertelement <1 x i32> undef, i32 %a, i32 0
168 %r = bitcast <1 x i32> %v to <2 x i16>
170 ; CHECK-LABEL: @BitcastInsert(
171 ; CHECK: bitcast i32 %a to <2 x i16>
175 define <2 x i64> @test7(<2 x i8*>* %arg) nounwind {
176 %cast = bitcast <2 x i8*>* %arg to <2 x i64>*
177 %load = load <2 x i64>, <2 x i64>* %cast, align 16