• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1; RUN: llc -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -mattr=sse2 < %s | FileCheck %s --check-prefix=SSE
2; RUN: llc -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -mattr=avx2 < %s | FileCheck %s --check-prefix=AVX
3
4; Verify that 128-bit vector logical ops are reassociated.
5
6define <4 x i32> @reassociate_and_v4i32(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2, <4 x i32> %x3) {
7; SSE-LABEL: reassociate_and_v4i32:
8; SSE:       # BB#0:
9; SSE-NEXT:    paddd %xmm1, %xmm0
10; SSE-NEXT:    pand %xmm3, %xmm2
11; SSE-NEXT:    pand %xmm2, %xmm0
12; SSE-NEXT:    retq
13;
14; AVX-LABEL: reassociate_and_v4i32:
15; AVX:       # BB#0:
16; AVX-NEXT:    vpaddd %xmm1, %xmm0, %xmm0
17; AVX-NEXT:    vpand %xmm3, %xmm2, %xmm1
18; AVX-NEXT:    vpand %xmm1, %xmm0, %xmm0
19; AVX-NEXT:    retq
20
21  %t0 = add <4 x i32> %x0, %x1
22  %t1 = and <4 x i32> %x2, %t0
23  %t2 = and <4 x i32> %x3, %t1
24  ret <4 x i32> %t2
25}
26
27define <4 x i32> @reassociate_or_v4i32(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2, <4 x i32> %x3) {
28; SSE-LABEL: reassociate_or_v4i32:
29; SSE:       # BB#0:
30; SSE-NEXT:    paddd %xmm1, %xmm0
31; SSE-NEXT:    por %xmm3, %xmm2
32; SSE-NEXT:    por %xmm2, %xmm0
33; SSE-NEXT:    retq
34;
35; AVX-LABEL: reassociate_or_v4i32:
36; AVX:       # BB#0:
37; AVX-NEXT:    vpaddd %xmm1, %xmm0, %xmm0
38; AVX-NEXT:    vpor %xmm3, %xmm2, %xmm1
39; AVX-NEXT:    vpor %xmm1, %xmm0, %xmm0
40; AVX-NEXT:    retq
41
42  %t0 = add <4 x i32> %x0, %x1
43  %t1 = or <4 x i32> %x2, %t0
44  %t2 = or <4 x i32> %x3, %t1
45  ret <4 x i32> %t2
46}
47
48define <4 x i32> @reassociate_xor_v4i32(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2, <4 x i32> %x3) {
49; SSE-LABEL: reassociate_xor_v4i32:
50; SSE:       # BB#0:
51; SSE-NEXT:    paddd %xmm1, %xmm0
52; SSE-NEXT:    pxor %xmm3, %xmm2
53; SSE-NEXT:    pxor %xmm2, %xmm0
54; SSE-NEXT:    retq
55;
56; AVX-LABEL: reassociate_xor_v4i32:
57; AVX:       # BB#0:
58; AVX-NEXT:    vpaddd %xmm1, %xmm0, %xmm0
59; AVX-NEXT:    vpxor %xmm3, %xmm2, %xmm1
60; AVX-NEXT:    vpxor %xmm1, %xmm0, %xmm0
61; AVX-NEXT:    retq
62
63  %t0 = add <4 x i32> %x0, %x1
64  %t1 = xor <4 x i32> %x2, %t0
65  %t2 = xor <4 x i32> %x3, %t1
66  ret <4 x i32> %t2
67}
68
69; Verify that 256-bit vector logical ops are reassociated.
70
71define <8 x i32> @reassociate_and_v8i32(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2, <8 x i32> %x3) {
72; AVX-LABEL: reassociate_and_v8i32:
73; AVX:       # BB#0:
74; AVX-NEXT:    vpaddd %ymm1, %ymm0, %ymm0
75; AVX-NEXT:    vpand %ymm3, %ymm2, %ymm1
76; AVX-NEXT:    vpand %ymm1, %ymm0, %ymm0
77; AVX-NEXT:    retq
78
79  %t0 = add <8 x i32> %x0, %x1
80  %t1 = and <8 x i32> %x2, %t0
81  %t2 = and <8 x i32> %x3, %t1
82  ret <8 x i32> %t2
83}
84
85define <8 x i32> @reassociate_or_v8i32(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2, <8 x i32> %x3) {
86; AVX-LABEL: reassociate_or_v8i32:
87; AVX:       # BB#0:
88; AVX-NEXT:    vpaddd %ymm1, %ymm0, %ymm0
89; AVX-NEXT:    vpor %ymm3, %ymm2, %ymm1
90; AVX-NEXT:    vpor %ymm1, %ymm0, %ymm0
91; AVX-NEXT:    retq
92
93  %t0 = add <8 x i32> %x0, %x1
94  %t1 = or <8 x i32> %x2, %t0
95  %t2 = or <8 x i32> %x3, %t1
96  ret <8 x i32> %t2
97}
98
99define <8 x i32> @reassociate_xor_v8i32(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2, <8 x i32> %x3) {
100; AVX-LABEL: reassociate_xor_v8i32:
101; AVX:       # BB#0:
102; AVX-NEXT:    vpaddd %ymm1, %ymm0, %ymm0
103; AVX-NEXT:    vpxor %ymm3, %ymm2, %ymm1
104; AVX-NEXT:    vpxor %ymm1, %ymm0, %ymm0
105; AVX-NEXT:    retq
106
107  %t0 = add <8 x i32> %x0, %x1
108  %t1 = xor <8 x i32> %x2, %t0
109  %t2 = xor <8 x i32> %x3, %t1
110  ret <8 x i32> %t2
111}
112
113