|
1 | 1 | ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
|
2 |
| -; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=avx2 -O0 | FileCheck %s |
| 2 | +; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=avx2 -O0 | FileCheck %s --check-prefixes=CHECK-O0 |
| 3 | +; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=avx2 -O3 | FileCheck %s --check-prefixes=CHECK-O3 |
3 | 4 |
|
4 | 5 | define <16 x i64> @pluto(<16 x i64> %arg, <16 x i64> %arg1, <16 x i64> %arg2, <16 x i64> %arg3, <16 x i64> %arg4) nounwind {
|
5 |
| -; CHECK-LABEL: pluto: |
6 |
| -; CHECK: # %bb.0: # %bb |
7 |
| -; CHECK-NEXT: pushq %rbp |
8 |
| -; CHECK-NEXT: movq %rsp, %rbp |
9 |
| -; CHECK-NEXT: andq $-32, %rsp |
10 |
| -; CHECK-NEXT: subq $32, %rsp |
11 |
| -; CHECK-NEXT: vmovaps %ymm4, %ymm10 |
12 |
| -; CHECK-NEXT: vmovaps %ymm3, %ymm9 |
13 |
| -; CHECK-NEXT: vmovaps %ymm1, %ymm8 |
14 |
| -; CHECK-NEXT: vmovaps 240(%rbp), %ymm4 |
15 |
| -; CHECK-NEXT: vmovaps 208(%rbp), %ymm3 |
16 |
| -; CHECK-NEXT: vmovaps 176(%rbp), %ymm1 |
17 |
| -; CHECK-NEXT: vmovaps 144(%rbp), %ymm1 |
18 |
| -; CHECK-NEXT: vmovaps 112(%rbp), %ymm11 |
19 |
| -; CHECK-NEXT: vmovaps 80(%rbp), %ymm11 |
20 |
| -; CHECK-NEXT: vmovaps 48(%rbp), %ymm11 |
21 |
| -; CHECK-NEXT: vmovaps 16(%rbp), %ymm11 |
22 |
| -; CHECK-NEXT: vpblendd {{.*#+}} ymm0 = ymm6[0,1,2,3,4,5],ymm2[6,7] |
23 |
| -; CHECK-NEXT: vpunpcklqdq {{.*#+}} ymm1 = ymm1[0],ymm3[0],ymm1[2],ymm3[2] |
24 |
| -; CHECK-NEXT: vpermq {{.*#+}} ymm1 = ymm1[0,2,1,3] |
25 |
| -; CHECK-NEXT: vpermq {{.*#+}} ymm0 = ymm0[3,1,2,1] |
26 |
| -; CHECK-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3,4,5],ymm0[6,7] |
27 |
| -; CHECK-NEXT: vperm2i128 {{.*#+}} ymm1 = ymm7[2,3],ymm6[0,1] |
28 |
| -; CHECK-NEXT: vxorps %xmm2, %xmm2, %xmm2 |
29 |
| -; CHECK-NEXT: vpblendd {{.*#+}} ymm2 = ymm1[0,1],ymm2[2,3],ymm1[4,5,6,7] |
30 |
| -; CHECK-NEXT: vpunpcklqdq {{.*#+}} ymm1 = ymm7[0],ymm5[0],ymm7[2],ymm5[2] |
31 |
| -; CHECK-NEXT: vpermq {{.*#+}} ymm1 = ymm1[2,1,2,3] |
32 |
| -; CHECK-NEXT: vpermq {{.*#+}} ymm4 = ymm4[1,1,1,1] |
33 |
| -; CHECK-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1],ymm4[2,3,4,5],ymm1[6,7] |
34 |
| -; CHECK-NEXT: vmovaps %xmm3, %xmm4 |
35 |
| -; CHECK-NEXT: vmovaps %xmm7, %xmm3 |
36 |
| -; CHECK-NEXT: vpblendd {{.*#+}} xmm4 = xmm3[0,1],xmm4[2,3] |
37 |
| -; CHECK-NEXT: # implicit-def: $ymm3 |
38 |
| -; CHECK-NEXT: vmovaps %xmm4, %xmm3 |
39 |
| -; CHECK-NEXT: vpermq {{.*#+}} ymm4 = ymm3[0,0,1,3] |
40 |
| -; CHECK-NEXT: vpslldq {{.*#+}} ymm3 = zero,zero,zero,zero,zero,zero,zero,zero,ymm5[0,1,2,3,4,5,6,7],zero,zero,zero,zero,zero,zero,zero,zero,ymm5[16,17,18,19,20,21,22,23] |
41 |
| -; CHECK-NEXT: vpblendd {{.*#+}} ymm3 = ymm3[0,1],ymm4[2,3,4,5],ymm3[6,7] |
42 |
| -; CHECK-NEXT: movq %rbp, %rsp |
43 |
| -; CHECK-NEXT: popq %rbp |
44 |
| -; CHECK-NEXT: retq |
| 6 | +; CHECK-O0-LABEL: pluto: |
| 7 | +; CHECK-O0: # %bb.0: # %bb |
| 8 | +; CHECK-O0-NEXT: pushq %rbp |
| 9 | +; CHECK-O0-NEXT: movq %rsp, %rbp |
| 10 | +; CHECK-O0-NEXT: andq $-32, %rsp |
| 11 | +; CHECK-O0-NEXT: subq $32, %rsp |
| 12 | +; CHECK-O0-NEXT: vmovaps %ymm4, %ymm10 |
| 13 | +; CHECK-O0-NEXT: vmovaps %ymm3, %ymm9 |
| 14 | +; CHECK-O0-NEXT: vmovaps %ymm1, %ymm8 |
| 15 | +; CHECK-O0-NEXT: vmovaps 240(%rbp), %ymm4 |
| 16 | +; CHECK-O0-NEXT: vmovaps 208(%rbp), %ymm3 |
| 17 | +; CHECK-O0-NEXT: vmovaps 176(%rbp), %ymm1 |
| 18 | +; CHECK-O0-NEXT: vmovaps 144(%rbp), %ymm1 |
| 19 | +; CHECK-O0-NEXT: vmovaps 112(%rbp), %ymm11 |
| 20 | +; CHECK-O0-NEXT: vmovaps 80(%rbp), %ymm11 |
| 21 | +; CHECK-O0-NEXT: vmovaps 48(%rbp), %ymm11 |
| 22 | +; CHECK-O0-NEXT: vmovaps 16(%rbp), %ymm11 |
| 23 | +; CHECK-O0-NEXT: vpblendd {{.*#+}} ymm0 = ymm6[0,1,2,3,4,5],ymm2[6,7] |
| 24 | +; CHECK-O0-NEXT: vpunpcklqdq {{.*#+}} ymm1 = ymm1[0],ymm3[0],ymm1[2],ymm3[2] |
| 25 | +; CHECK-O0-NEXT: vpermq {{.*#+}} ymm1 = ymm1[0,2,1,3] |
| 26 | +; CHECK-O0-NEXT: vpermq {{.*#+}} ymm0 = ymm0[3,1,2,1] |
| 27 | +; CHECK-O0-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3,4,5],ymm0[6,7] |
| 28 | +; CHECK-O0-NEXT: vperm2i128 {{.*#+}} ymm1 = ymm7[2,3],ymm6[0,1] |
| 29 | +; CHECK-O0-NEXT: vxorps %xmm2, %xmm2, %xmm2 |
| 30 | +; CHECK-O0-NEXT: vpblendd {{.*#+}} ymm2 = ymm1[0,1],ymm2[2,3],ymm1[4,5,6,7] |
| 31 | +; CHECK-O0-NEXT: vpunpcklqdq {{.*#+}} ymm1 = ymm7[0],ymm5[0],ymm7[2],ymm5[2] |
| 32 | +; CHECK-O0-NEXT: vpermq {{.*#+}} ymm1 = ymm1[2,1,2,3] |
| 33 | +; CHECK-O0-NEXT: vpermq {{.*#+}} ymm4 = ymm4[1,1,1,1] |
| 34 | +; CHECK-O0-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1],ymm4[2,3,4,5],ymm1[6,7] |
| 35 | +; CHECK-O0-NEXT: vmovaps %xmm3, %xmm4 |
| 36 | +; CHECK-O0-NEXT: vmovaps %xmm7, %xmm3 |
| 37 | +; CHECK-O0-NEXT: vpblendd {{.*#+}} xmm4 = xmm3[0,1],xmm4[2,3] |
| 38 | +; CHECK-O0-NEXT: # implicit-def: $ymm3 |
| 39 | +; CHECK-O0-NEXT: vmovaps %xmm4, %xmm3 |
| 40 | +; CHECK-O0-NEXT: vpermq {{.*#+}} ymm4 = ymm3[0,0,1,3] |
| 41 | +; CHECK-O0-NEXT: vpslldq {{.*#+}} ymm3 = zero,zero,zero,zero,zero,zero,zero,zero,ymm5[0,1,2,3,4,5,6,7],zero,zero,zero,zero,zero,zero,zero,zero,ymm5[16,17,18,19,20,21,22,23] |
| 42 | +; CHECK-O0-NEXT: vpblendd {{.*#+}} ymm3 = ymm3[0,1],ymm4[2,3,4,5],ymm3[6,7] |
| 43 | +; CHECK-O0-NEXT: movq %rbp, %rsp |
| 44 | +; CHECK-O0-NEXT: popq %rbp |
| 45 | +; CHECK-O0-NEXT: retq |
| 46 | +; |
| 47 | +; CHECK-O3-LABEL: pluto: |
| 48 | +; CHECK-O3: # %bb.0: # %bb |
| 49 | +; CHECK-O3-NEXT: pushq %rbp |
| 50 | +; CHECK-O3-NEXT: movq %rsp, %rbp |
| 51 | +; CHECK-O3-NEXT: andq $-32, %rsp |
| 52 | +; CHECK-O3-NEXT: subq $32, %rsp |
| 53 | +; CHECK-O3-NEXT: vmovdqa 208(%rbp), %ymm3 |
| 54 | +; CHECK-O3-NEXT: vmovdqa 144(%rbp), %ymm0 |
| 55 | +; CHECK-O3-NEXT: vpblendd {{.*#+}} ymm1 = ymm6[0,1,2,3,4,5],ymm2[6,7] |
| 56 | +; CHECK-O3-NEXT: vpunpcklqdq {{.*#+}} ymm0 = ymm0[0],ymm3[0],ymm0[2],ymm3[2] |
| 57 | +; CHECK-O3-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3] |
| 58 | +; CHECK-O3-NEXT: vpermq {{.*#+}} ymm1 = ymm1[3,1,2,1] |
| 59 | +; CHECK-O3-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3,4,5],ymm1[6,7] |
| 60 | +; CHECK-O3-NEXT: vperm2i128 {{.*#+}} ymm1 = ymm7[2,3],ymm6[0,1] |
| 61 | +; CHECK-O3-NEXT: vpxor %xmm2, %xmm2, %xmm2 |
| 62 | +; CHECK-O3-NEXT: vpblendd {{.*#+}} ymm2 = ymm1[0,1],ymm2[2,3],ymm1[4,5,6,7] |
| 63 | +; CHECK-O3-NEXT: vpunpcklqdq {{.*#+}} ymm1 = ymm7[0],ymm5[0],ymm7[2],ymm5[2] |
| 64 | +; CHECK-O3-NEXT: vpermq {{.*#+}} ymm1 = ymm1[2,1,2,3] |
| 65 | +; CHECK-O3-NEXT: vpbroadcastq 248(%rbp), %ymm4 |
| 66 | +; CHECK-O3-NEXT: vpblendd {{.*#+}} ymm1 = ymm1[0,1],ymm4[2,3,4,5],ymm1[6,7] |
| 67 | +; CHECK-O3-NEXT: vpblendd {{.*#+}} xmm3 = xmm7[0,1],xmm3[2,3] |
| 68 | +; CHECK-O3-NEXT: vpermq {{.*#+}} ymm3 = ymm3[0,0,1,3] |
| 69 | +; CHECK-O3-NEXT: vpslldq {{.*#+}} ymm4 = zero,zero,zero,zero,zero,zero,zero,zero,ymm5[0,1,2,3,4,5,6,7],zero,zero,zero,zero,zero,zero,zero,zero,ymm5[16,17,18,19,20,21,22,23] |
| 70 | +; CHECK-O3-NEXT: vpblendd {{.*#+}} ymm3 = ymm4[0,1],ymm3[2,3,4,5],ymm4[6,7] |
| 71 | +; CHECK-O3-NEXT: movq %rbp, %rsp |
| 72 | +; CHECK-O3-NEXT: popq %rbp |
| 73 | +; CHECK-O3-NEXT: retq |
45 | 74 | bb:
|
46 | 75 | %tmp = select <16 x i1> <i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false, i1 true, i1 true, i1 false, i1 false, i1 false, i1 false>, <16 x i64> %arg, <16 x i64> %arg1
|
47 | 76 | %tmp5 = select <16 x i1> <i1 true, i1 false, i1 false, i1 true, i1 true, i1 false, i1 false, i1 true, i1 false, i1 true, i1 false, i1 false, i1 false, i1 false, i1 false, i1 false>, <16 x i64> %arg2, <16 x i64> zeroinitializer
|
|
0 commit comments