| blob | b9b8f413c97a34370a45f01be43aa46fcc5156e5 |
1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=CHECK,AVX1
3 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=CHECK,AVX2
5 define void @PR32957(ptr %in, ptr %out) {
6 ; CHECK-LABEL: PR32957:
7 ; CHECK: # %bb.0:
8 ; CHECK-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
9 ; CHECK-NEXT: vmovaps %ymm0, (%rsi)
10 ; CHECK-NEXT: vzeroupper
11 ; CHECK-NEXT: retq
12 %ld = load <2 x float>, ptr %in, align 8
13 %ext = extractelement <2 x float> %ld, i64 0
14 %ext2 = extractelement <2 x float> %ld, i64 1
15 %ins = insertelement <8 x float> <float undef, float undef, float 0.0, float 0.0, float 0.0, float 0.0, float 0.0, float 0.0>, float %ext, i64 0
16 %ins2 = insertelement <8 x float> %ins, float %ext2, i64 1
17 store <8 x float> %ins2, ptr %out, align 32
18 ret void
19 }
21 declare { i8, double } @fun()
23 ; Check that this does not fail to combine concat_vectors of a value from
24 ; merge_values through a bitcast.
25 define void @d(i1 %cmp) {
26 ; CHECK-LABEL: d:
27 ; CHECK: # %bb.0: # %bar
28 ; CHECK-NEXT: pushq %rax
29 ; CHECK-NEXT: .cfi_def_cfa_offset 16
30 ; CHECK-NEXT: callq fun@PLT
31 bar:
32 %val = call { i8, double } @fun()
33 %extr = extractvalue { i8, double } %val, 1
34 %bc = bitcast double %extr to <2 x float>
35 br label %baz
37 baz:
38 %extr1 = extractelement <2 x float> %bc, i64 0
39 unreachable
40 }
42 @qa_ = external unnamed_addr global [49216 x i8], align 32
44 define void @concat_of_broadcast_v2f64_v4f64() {
45 ; AVX1-LABEL: concat_of_broadcast_v2f64_v4f64:
46 ; AVX1: # %bb.0: # %alloca_0
47 ; AVX1-NEXT: movq qa_@GOTPCREL(%rip), %rax
48 ; AVX1-NEXT: movl $1091567616, 30256(%rax) # imm = 0x41100000
49 ; AVX1-NEXT: movabsq $4294967297, %rcx # imm = 0x100000001
50 ; AVX1-NEXT: movq %rcx, 46348(%rax)
51 ; AVX1-NEXT: vbroadcastf128 {{.*#+}} ymm0 = [7.812501848093234E-3,7.812501848093234E-3,7.812501848093234E-3,7.812501848093234E-3]
52 ; AVX1-NEXT: # ymm0 = mem[0,1,0,1]
53 ; AVX1-NEXT: vmovups %ymm0, 48296(%rax)
54 ; AVX1-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
55 ; AVX1-NEXT: vmovsd %xmm0, 47372(%rax)
56 ; AVX1-NEXT: vzeroupper
57 ; AVX1-NEXT: retq
58 ;
59 ; AVX2-LABEL: concat_of_broadcast_v2f64_v4f64:
60 ; AVX2: # %bb.0: # %alloca_0
61 ; AVX2-NEXT: movq qa_@GOTPCREL(%rip), %rax
62 ; AVX2-NEXT: movl $1091567616, 30256(%rax) # imm = 0x41100000
63 ; AVX2-NEXT: movabsq $4294967297, %rcx # imm = 0x100000001
64 ; AVX2-NEXT: movq %rcx, 46348(%rax)
65 ; AVX2-NEXT: vbroadcastss {{.*#+}} xmm0 = [1.0E+0,1.0E+0,1.0E+0,1.0E+0]
66 ; AVX2-NEXT: vbroadcastsd %xmm0, %ymm1
67 ; AVX2-NEXT: vmovups %ymm1, 48296(%rax)
68 ; AVX2-NEXT: vmovlps %xmm0, 47372(%rax)
69 ; AVX2-NEXT: vzeroupper
70 ; AVX2-NEXT: retq
71 alloca_0:
72 store float 9.000000e+00, ptr getelementptr inbounds ([49216 x i8], ptr @qa_, i64 0, i64 30256), align 16
73 store <2 x i32> <i32 1, i32 1>, ptr getelementptr inbounds ([49216 x i8], ptr @qa_, i64 0, i64 46348), align 4
74 br label %loop.4942
76 loop.4942: ; preds = %loop.4942, %alloca_0
77 br i1 undef, label %loop.4942, label %ifmerge.1298
79 ifmerge.1298: ; preds = %loop.4942
80 %gepload4638 = load float, ptr getelementptr inbounds ([49216 x i8], ptr @qa_, i64 0, i64 28324), align 4
81 store <2 x float> <float 1.000000e+00, float 1.000000e+00>, ptr getelementptr inbounds ([49216 x i8], ptr @qa_, i64 0, i64 48296), align 8
82 store <2 x float> <float 1.000000e+00, float 1.000000e+00>, ptr getelementptr inbounds ([49216 x i8], ptr @qa_, i64 0, i64 48304), align 16
83 store <2 x float> <float 1.000000e+00, float 1.000000e+00>, ptr getelementptr inbounds ([49216 x i8], ptr @qa_, i64 0, i64 48312), align 8
84 store <2 x float> <float 1.000000e+00, float 1.000000e+00>, ptr getelementptr inbounds ([49216 x i8], ptr @qa_, i64 0, i64 48320), align 32
85 store <2 x float> <float 1.000000e+00, float 1.000000e+00>, ptr getelementptr inbounds ([49216 x i8], ptr @qa_, i64 0, i64 47372), align 4
86 ret void
87 }
89 define <4 x float> @concat_of_broadcast_v4f32_v8f32(ptr %a0, ptr %a1, ptr %a2) {
90 ; AVX1-LABEL: concat_of_broadcast_v4f32_v8f32:
91 ; AVX1: # %bb.0:
92 ; AVX1-NEXT: vmovaps (%rdi), %ymm0
93 ; AVX1-NEXT: vmovaps (%rsi), %ymm1
94 ; AVX1-NEXT: vmovaps (%rdx), %ymm2
95 ; AVX1-NEXT: vpermilps {{.*#+}} xmm1 = xmm1[0,0,0,0]
96 ; AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm1, %ymm1
97 ; AVX1-NEXT: vblendps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,5],ymm1[6,7]
98 ; AVX1-NEXT: vblendps {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4],ymm2[5,6],ymm0[7]
99 ; AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0
100 ; AVX1-NEXT: vshufps {{.*#+}} xmm1 = xmm2[3,0],xmm0[0,0]
101 ; AVX1-NEXT: vshufps {{.*#+}} xmm0 = xmm0[2,3],xmm1[2,0]
102 ; AVX1-NEXT: vzeroupper
103 ; AVX1-NEXT: retq
104 ;
105 ; AVX2-LABEL: concat_of_broadcast_v4f32_v8f32:
106 ; AVX2: # %bb.0:
107 ; AVX2-NEXT: vmovaps (%rdi), %ymm0
108 ; AVX2-NEXT: vunpcklps {{.*#+}} ymm0 = ymm0[0],mem[0],ymm0[1],mem[1],ymm0[4],mem[4],ymm0[5],mem[5]
109 ; AVX2-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,1,2,0]
110 ; AVX2-NEXT: vmovaps {{.*#+}} xmm1 = [6,7,4,3]
111 ; AVX2-NEXT: vblendps {{.*#+}} ymm0 = mem[0,1,2,3],ymm0[4],mem[5,6],ymm0[7]
112 ; AVX2-NEXT: vpermps %ymm0, %ymm1, %ymm0
113 ; AVX2-NEXT: # kill: def $xmm0 killed $xmm0 killed $ymm0
114 ; AVX2-NEXT: vzeroupper
115 ; AVX2-NEXT: retq
116 %ld0 = load volatile <8 x float>, ptr %a0
117 %ld1 = load volatile <8 x float>, ptr %a1
118 %ld2 = load volatile <8 x float>, ptr %a2
119 %shuffle = shufflevector <8 x float> %ld0, <8 x float> %ld1, <8 x i32> <i32 undef, i32 undef, i32 undef, i32 undef, i32 4, i32 undef, i32 undef, i32 8>
120 %shuffle1 = shufflevector <8 x float> %ld2, <8 x float> %shuffle, <4 x i32> <i32 6, i32 15, i32 12, i32 3>
121 ret <4 x float> %shuffle1
122 }