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
3
4; In the following 4 tests, the existing call to VZU/VZA ensures clean state before
5; the call to the unknown, so we don't need to insert a second VZU at that point.
6
7define <4 x float> @zeroupper_v4f32(<8 x float> *%x, <8 x float> %y) nounwind {
8; CHECK-LABEL: zeroupper_v4f32:
9; CHECK:       # %bb.0:
10; CHECK-NEXT:    pushq %rbx
11; CHECK-NEXT:    subq $32, %rsp
12; CHECK-NEXT:    vmovups %ymm0, (%rsp) # 32-byte Spill
13; CHECK-NEXT:    movq %rdi, %rbx
14; CHECK-NEXT:    vzeroupper
15; CHECK-NEXT:    callq the_unknown@PLT
16; CHECK-NEXT:    vmovups (%rsp), %ymm0 # 32-byte Reload
17; CHECK-NEXT:    vaddps (%rbx), %ymm0, %ymm0
18; CHECK-NEXT:    vextractf128 $1, %ymm0, %xmm1
19; CHECK-NEXT:    vaddps %xmm1, %xmm0, %xmm0
20; CHECK-NEXT:    addq $32, %rsp
21; CHECK-NEXT:    popq %rbx
22; CHECK-NEXT:    vzeroupper
23; CHECK-NEXT:    retq
24  call void @llvm.x86.avx.vzeroupper()
25  call void @the_unknown()
26  %loadx = load <8 x float>, <8 x float> *%x, align 32
27  %sum = fadd <8 x float> %loadx, %y
28  %lo = shufflevector <8 x float> %sum, <8 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
29  %hi = shufflevector <8 x float> %sum, <8 x float> undef, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
30  %res = fadd <4 x float> %lo, %hi
31  ret <4 x float> %res
32}
33
34define <8 x float> @zeroupper_v8f32(<8 x float> %x) nounwind {
35; CHECK-LABEL: zeroupper_v8f32:
36; CHECK:       # %bb.0:
37; CHECK-NEXT:    subq $40, %rsp
38; CHECK-NEXT:    vmovups %ymm0, (%rsp) # 32-byte Spill
39; CHECK-NEXT:    vzeroupper
40; CHECK-NEXT:    callq the_unknown@PLT
41; CHECK-NEXT:    vmovups (%rsp), %ymm0 # 32-byte Reload
42; CHECK-NEXT:    addq $40, %rsp
43; CHECK-NEXT:    retq
44  call void @llvm.x86.avx.vzeroupper()
45  call void @the_unknown()
46  ret <8 x float> %x
47}
48
49define <4 x float> @zeroall_v4f32(<8 x float> *%x, <8 x float> %y) nounwind {
50; CHECK-LABEL: zeroall_v4f32:
51; CHECK:       # %bb.0:
52; CHECK-NEXT:    pushq %rbx
53; CHECK-NEXT:    subq $32, %rsp
54; CHECK-NEXT:    vmovups %ymm0, (%rsp) # 32-byte Spill
55; CHECK-NEXT:    movq %rdi, %rbx
56; CHECK-NEXT:    vzeroall
57; CHECK-NEXT:    callq the_unknown@PLT
58; CHECK-NEXT:    vmovups (%rsp), %ymm0 # 32-byte Reload
59; CHECK-NEXT:    vaddps (%rbx), %ymm0, %ymm0
60; CHECK-NEXT:    vextractf128 $1, %ymm0, %xmm1
61; CHECK-NEXT:    vaddps %xmm1, %xmm0, %xmm0
62; CHECK-NEXT:    addq $32, %rsp
63; CHECK-NEXT:    popq %rbx
64; CHECK-NEXT:    vzeroupper
65; CHECK-NEXT:    retq
66  call void @llvm.x86.avx.vzeroall()
67  call void @the_unknown()
68  %loadx = load <8 x float>, <8 x float> *%x, align 32
69  %sum = fadd <8 x float> %loadx, %y
70  %lo = shufflevector <8 x float> %sum, <8 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
71  %hi = shufflevector <8 x float> %sum, <8 x float> undef, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
72  %res = fadd <4 x float> %lo, %hi
73  ret <4 x float> %res
74}
75
76define <8 x float> @zeroall_v8f32(<8 x float> %x) nounwind {
77; CHECK-LABEL: zeroall_v8f32:
78; CHECK:       # %bb.0:
79; CHECK-NEXT:    subq $40, %rsp
80; CHECK-NEXT:    vmovups %ymm0, (%rsp) # 32-byte Spill
81; CHECK-NEXT:    vzeroall
82; CHECK-NEXT:    callq the_unknown@PLT
83; CHECK-NEXT:    vmovups (%rsp), %ymm0 # 32-byte Reload
84; CHECK-NEXT:    addq $40, %rsp
85; CHECK-NEXT:    retq
86  call void @llvm.x86.avx.vzeroall()
87  call void @the_unknown()
88  ret <8 x float> %x
89}
90
91declare void @llvm.x86.avx.vzeroupper() nounwind readnone
92declare void @llvm.x86.avx.vzeroall() nounwind readnone
93declare void @the_unknown() nounwind
94
95