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=ALL,AVX
3 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=ALL,AVX2
5 ; https://bugs.llvm.org/show_bug.cgi?id=51615
6 ; We can not replace a wide volatile load with a broadcast-from-memory,
7 ; because that would narrow the load, which isn't legal for volatiles.
9 @g0 = external dso_local global <2 x double>, align 16
10 define void @volatile_load_2_elts() {
11 ; AVX-LABEL: volatile_load_2_elts:
13 ; AVX-NEXT: vmovaps g0(%rip), %xmm0
14 ; AVX-NEXT: vmovddup {{.*#+}} xmm1 = xmm0[0,0]
15 ; AVX-NEXT: vxorps %xmm2, %xmm2, %xmm2
16 ; AVX-NEXT: vblendps {{.*#+}} ymm1 = ymm2[0,1],ymm1[2,3],ymm2[4,5,6,7]
17 ; AVX-NEXT: vinsertf128 $1, %xmm1, %ymm1, %ymm1
18 ; AVX-NEXT: vperm2f128 {{.*#+}} ymm0 = zero,zero,ymm0[0,1]
19 ; AVX-NEXT: vblendps {{.*#+}} ymm0 = ymm2[0,1,2,3],ymm0[4,5],ymm2[6,7]
20 ; AVX-NEXT: vmovaps %ymm0, (%rax)
21 ; AVX-NEXT: vmovaps %ymm1, (%rax)
22 ; AVX-NEXT: vzeroupper
25 ; AVX2-LABEL: volatile_load_2_elts:
27 ; AVX2-NEXT: vmovaps g0(%rip), %xmm0
28 ; AVX2-NEXT: vbroadcastsd %xmm0, %ymm0
29 ; AVX2-NEXT: vxorps %xmm1, %xmm1, %xmm1
30 ; AVX2-NEXT: vblendps {{.*#+}} ymm2 = ymm1[0,1,2,3],ymm0[4,5],ymm1[6,7]
31 ; AVX2-NEXT: vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5],ymm0[6,7]
32 ; AVX2-NEXT: vmovaps %ymm0, (%rax)
33 ; AVX2-NEXT: vmovaps %ymm2, (%rax)
34 ; AVX2-NEXT: vzeroupper
36 %i = load volatile <2 x double>, ptr @g0, align 16
37 %i1 = shufflevector <2 x double> %i, <2 x double> poison, <4 x i32> <i32 undef, i32 0, i32 undef, i32 0>
38 %shuffle1 = shufflevector <4 x double> %i1, <4 x double> zeroinitializer, <8 x i32> <i32 6, i32 7, i32 3, i32 6, i32 7, i32 1, i32 7, i32 1>
39 store volatile <8 x double> %shuffle1, ptr undef, align 64
43 @g1 = external dso_local global <1 x double>, align 16
44 define void @volatile_load_1_elt() {
45 ; ALL-LABEL: volatile_load_1_elt:
47 ; ALL-NEXT: vbroadcastsd g1(%rip), %ymm0
48 ; ALL-NEXT: vxorps %xmm1, %xmm1, %xmm1
49 ; ALL-NEXT: vblendps {{.*#+}} ymm2 = ymm1[0,1,2,3],ymm0[4,5],ymm1[6,7]
50 ; ALL-NEXT: vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5],ymm0[6,7]
51 ; ALL-NEXT: vmovaps %ymm0, (%rax)
52 ; ALL-NEXT: vmovaps %ymm2, (%rax)
53 ; ALL-NEXT: vzeroupper
55 %i = load volatile <1 x double>, ptr @g1, align 16
56 %i1 = shufflevector <1 x double> %i, <1 x double> poison, <4 x i32> <i32 undef, i32 0, i32 undef, i32 0>
57 %shuffle1 = shufflevector <4 x double> %i1, <4 x double> zeroinitializer, <8 x i32> <i32 6, i32 7, i32 3, i32 6, i32 7, i32 1, i32 7, i32 1>
58 store volatile <8 x double> %shuffle1, ptr undef, align 64
62 @g2 = external dso_local global <2 x float>, align 16
63 define void @volatile_load_2_elts_bitcast() {
64 ; ALL-LABEL: volatile_load_2_elts_bitcast:
66 ; ALL-NEXT: vbroadcastsd g2(%rip), %ymm0
67 ; ALL-NEXT: vxorps %xmm1, %xmm1, %xmm1
68 ; ALL-NEXT: vblendps {{.*#+}} ymm2 = ymm1[0,1,2,3],ymm0[4,5],ymm1[6,7]
69 ; ALL-NEXT: vblendps {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5],ymm0[6,7]
70 ; ALL-NEXT: vmovaps %ymm0, (%rax)
71 ; ALL-NEXT: vmovaps %ymm2, (%rax)
72 ; ALL-NEXT: vzeroupper
74 %i0 = load volatile <2 x float>, ptr @g2, align 16
75 %i = bitcast <2 x float> %i0 to <1 x double>
76 %i1 = shufflevector <1 x double> %i, <1 x double> poison, <4 x i32> <i32 undef, i32 0, i32 undef, i32 0>
77 %shuffle1 = shufflevector <4 x double> %i1, <4 x double> zeroinitializer, <8 x i32> <i32 6, i32 7, i32 3, i32 6, i32 7, i32 1, i32 7, i32 1>
78 store volatile <8 x double> %shuffle1, ptr undef, align 64
82 define void @elts_from_consecutive_loads(ptr %arg, ptr %arg12, ptr %arg13, float %arg14, i1 %arg15) {
83 ; ALL-LABEL: elts_from_consecutive_loads:
85 ; ALL-NEXT: vxorps %xmm1, %xmm1, %xmm1
86 ; ALL-NEXT: .p2align 4
87 ; ALL-NEXT: .LBB3_1: # %bb16
88 ; ALL-NEXT: # =>This Loop Header: Depth=1
89 ; ALL-NEXT: # Child Loop BB3_2 Depth 2
90 ; ALL-NEXT: testb $1, %cl
91 ; ALL-NEXT: je .LBB3_1
92 ; ALL-NEXT: .p2align 4
93 ; ALL-NEXT: .LBB3_2: # %bb17
94 ; ALL-NEXT: # Parent Loop BB3_1 Depth=1
95 ; ALL-NEXT: # => This Inner Loop Header: Depth=2
96 ; ALL-NEXT: movl (%rdi), %eax
97 ; ALL-NEXT: vbroadcastss (%rdi), %ymm2
98 ; ALL-NEXT: movl %eax, (%rsi)
99 ; ALL-NEXT: vmovaps %ymm2, (%rdx)
100 ; ALL-NEXT: vucomiss %xmm1, %xmm0
101 ; ALL-NEXT: jne .LBB3_2
102 ; ALL-NEXT: jp .LBB3_2
103 ; ALL-NEXT: jmp .LBB3_1
107 bb16: ; preds = %bb17, %bb16, %bb
108 br i1 %arg15, label %bb17, label %bb16
110 bb17: ; preds = %bb17, %bb16
111 %tmp = load <2 x i64>, ptr %arg, align 16
112 %tmp18 = extractelement <2 x i64> %tmp, i32 0
113 %tmp19 = trunc i64 %tmp18 to i32
114 store i32 %tmp19, ptr %arg12, align 4
115 %tmp20 = insertelement <8 x i32> <i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>, i32 %tmp19, i32 0
116 %tmp21 = shufflevector <8 x i32> %tmp20, <8 x i32> <i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1, i32 1>, <8 x i32> <i32 0, i32 undef, i32 undef, i32 undef, i32 0, i32 undef, i32 undef, i32 undef>
117 store <8 x i32> %tmp21, ptr %arg13, align 32
118 %tmp22 = fcmp une float %arg14, 0.000000e+00
119 br i1 %tmp22, label %bb17, label %bb16