| blob | 889ab6a0818e2f98d6dae0b1a5f038ab79f4469a |
1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefix=X86-SSE
3 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+sse2 | FileCheck %s --check-prefix=X64-SSE
4 ; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=X86-AVX
5 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=X64-AVX
7 define i64 @extract_any_extend_vector_inreg_v16i64(<16 x i64> %a0, i32 %a1) nounwind {
8 ; X86-SSE-LABEL: extract_any_extend_vector_inreg_v16i64:
9 ; X86-SSE: # %bb.0:
10 ; X86-SSE-NEXT: pushl %ebp
11 ; X86-SSE-NEXT: movl %esp, %ebp
12 ; X86-SSE-NEXT: andl $-16, %esp
13 ; X86-SSE-NEXT: subl $272, %esp # imm = 0x110
14 ; X86-SSE-NEXT: movl 88(%ebp), %ecx
15 ; X86-SSE-NEXT: movdqa 72(%ebp), %xmm0
16 ; X86-SSE-NEXT: psrldq {{.*#+}} xmm0 = xmm0[8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero
17 ; X86-SSE-NEXT: xorps %xmm1, %xmm1
18 ; X86-SSE-NEXT: movaps %xmm1, {{[0-9]+}}(%esp)
19 ; X86-SSE-NEXT: movaps %xmm1, {{[0-9]+}}(%esp)
20 ; X86-SSE-NEXT: movaps %xmm1, {{[0-9]+}}(%esp)
21 ; X86-SSE-NEXT: movaps %xmm1, {{[0-9]+}}(%esp)
22 ; X86-SSE-NEXT: movaps %xmm1, {{[0-9]+}}(%esp)
23 ; X86-SSE-NEXT: movaps %xmm1, {{[0-9]+}}(%esp)
24 ; X86-SSE-NEXT: movaps %xmm1, {{[0-9]+}}(%esp)
25 ; X86-SSE-NEXT: movdqa %xmm0, {{[0-9]+}}(%esp)
26 ; X86-SSE-NEXT: movaps %xmm1, {{[0-9]+}}(%esp)
27 ; X86-SSE-NEXT: movaps %xmm1, {{[0-9]+}}(%esp)
28 ; X86-SSE-NEXT: movaps %xmm1, {{[0-9]+}}(%esp)
29 ; X86-SSE-NEXT: movaps %xmm1, {{[0-9]+}}(%esp)
30 ; X86-SSE-NEXT: movaps %xmm1, {{[0-9]+}}(%esp)
31 ; X86-SSE-NEXT: movaps %xmm1, {{[0-9]+}}(%esp)
32 ; X86-SSE-NEXT: movaps %xmm1, (%esp)
33 ; X86-SSE-NEXT: movdqa %xmm0, {{[0-9]+}}(%esp)
34 ; X86-SSE-NEXT: leal (%ecx,%ecx), %eax
35 ; X86-SSE-NEXT: andl $31, %eax
36 ; X86-SSE-NEXT: movl 128(%esp,%eax,4), %eax
37 ; X86-SSE-NEXT: leal 1(%ecx,%ecx), %ecx
38 ; X86-SSE-NEXT: andl $31, %ecx
39 ; X86-SSE-NEXT: movl (%esp,%ecx,4), %edx
40 ; X86-SSE-NEXT: movl %ebp, %esp
41 ; X86-SSE-NEXT: popl %ebp
42 ; X86-SSE-NEXT: retl
43 ;
44 ; X64-SSE-LABEL: extract_any_extend_vector_inreg_v16i64:
45 ; X64-SSE: # %bb.0:
46 ; X64-SSE-NEXT: pushq %rax
47 ; X64-SSE-NEXT: # kill: def $edi killed $edi def $rdi
48 ; X64-SSE-NEXT: psrldq {{.*#+}} xmm7 = xmm7[8,9,10,11,12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero
49 ; X64-SSE-NEXT: xorps %xmm0, %xmm0
50 ; X64-SSE-NEXT: movaps %xmm0, -{{[0-9]+}}(%rsp)
51 ; X64-SSE-NEXT: movaps %xmm0, -{{[0-9]+}}(%rsp)
52 ; X64-SSE-NEXT: movaps %xmm0, -{{[0-9]+}}(%rsp)
53 ; X64-SSE-NEXT: movaps %xmm0, -{{[0-9]+}}(%rsp)
54 ; X64-SSE-NEXT: movaps %xmm0, -{{[0-9]+}}(%rsp)
55 ; X64-SSE-NEXT: movaps %xmm0, -{{[0-9]+}}(%rsp)
56 ; X64-SSE-NEXT: movaps %xmm0, -{{[0-9]+}}(%rsp)
57 ; X64-SSE-NEXT: movdqa %xmm7, -{{[0-9]+}}(%rsp)
58 ; X64-SSE-NEXT: andl $15, %edi
59 ; X64-SSE-NEXT: movq -128(%rsp,%rdi,8), %rax
60 ; X64-SSE-NEXT: popq %rcx
61 ; X64-SSE-NEXT: retq
62 ;
63 ; X86-AVX-LABEL: extract_any_extend_vector_inreg_v16i64:
64 ; X86-AVX: # %bb.0:
65 ; X86-AVX-NEXT: pushl %ebp
66 ; X86-AVX-NEXT: movl %esp, %ebp
67 ; X86-AVX-NEXT: andl $-32, %esp
68 ; X86-AVX-NEXT: subl $288, %esp # imm = 0x120
69 ; X86-AVX-NEXT: movl 40(%ebp), %ecx
70 ; X86-AVX-NEXT: vmovsd {{.*#+}} xmm0 = mem[0],zero
71 ; X86-AVX-NEXT: vxorps %xmm1, %xmm1, %xmm1
72 ; X86-AVX-NEXT: vmovaps %ymm1, {{[0-9]+}}(%esp)
73 ; X86-AVX-NEXT: vmovaps %ymm1, {{[0-9]+}}(%esp)
74 ; X86-AVX-NEXT: vmovaps %ymm1, {{[0-9]+}}(%esp)
75 ; X86-AVX-NEXT: vmovaps %ymm0, {{[0-9]+}}(%esp)
76 ; X86-AVX-NEXT: vmovaps %ymm1, {{[0-9]+}}(%esp)
77 ; X86-AVX-NEXT: vmovaps %ymm1, {{[0-9]+}}(%esp)
78 ; X86-AVX-NEXT: vmovaps %ymm1, (%esp)
79 ; X86-AVX-NEXT: vmovaps %ymm0, {{[0-9]+}}(%esp)
80 ; X86-AVX-NEXT: leal (%ecx,%ecx), %eax
81 ; X86-AVX-NEXT: andl $31, %eax
82 ; X86-AVX-NEXT: movl 128(%esp,%eax,4), %eax
83 ; X86-AVX-NEXT: leal 1(%ecx,%ecx), %ecx
84 ; X86-AVX-NEXT: andl $31, %ecx
85 ; X86-AVX-NEXT: movl (%esp,%ecx,4), %edx
86 ; X86-AVX-NEXT: movl %ebp, %esp
87 ; X86-AVX-NEXT: popl %ebp
88 ; X86-AVX-NEXT: vzeroupper
89 ; X86-AVX-NEXT: retl
90 ;
91 ; X64-AVX-LABEL: extract_any_extend_vector_inreg_v16i64:
92 ; X64-AVX: # %bb.0:
93 ; X64-AVX-NEXT: pushq %rbp
94 ; X64-AVX-NEXT: movq %rsp, %rbp
95 ; X64-AVX-NEXT: andq $-32, %rsp
96 ; X64-AVX-NEXT: subq $160, %rsp
97 ; X64-AVX-NEXT: # kill: def $edi killed $edi def $rdi
98 ; X64-AVX-NEXT: vpermq {{.*#+}} ymm0 = ymm3[3,3,3,3]
99 ; X64-AVX-NEXT: vmovq {{.*#+}} xmm0 = xmm0[0],zero
100 ; X64-AVX-NEXT: vxorps %xmm1, %xmm1, %xmm1
101 ; X64-AVX-NEXT: vmovaps %ymm1, {{[0-9]+}}(%rsp)
102 ; X64-AVX-NEXT: vmovaps %ymm1, {{[0-9]+}}(%rsp)
103 ; X64-AVX-NEXT: vmovaps %ymm1, (%rsp)
104 ; X64-AVX-NEXT: vmovdqa %ymm0, {{[0-9]+}}(%rsp)
105 ; X64-AVX-NEXT: andl $15, %edi
106 ; X64-AVX-NEXT: movq (%rsp,%rdi,8), %rax
107 ; X64-AVX-NEXT: movq %rbp, %rsp
108 ; X64-AVX-NEXT: popq %rbp
109 ; X64-AVX-NEXT: vzeroupper
110 ; X64-AVX-NEXT: retq
111 %1 = extractelement <16 x i64> %a0, i32 15
112 %2 = insertelement <16 x i64> zeroinitializer, i64 %1, i32 4
113 %3 = extractelement <16 x i64> %2, i32 %a1
114 ret i64 %3
115 }