| blob | b0fe43a647716573fd7bc5321cf11a1eed14de31 |
1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc < %s -mtriple=x86_64-- -mcpu=x86-64 | FileCheck %s --check-prefixes=SSE,SSE2
3 ; RUN: llc < %s -mtriple=x86_64-- -mcpu=x86-64-v2 | FileCheck %s --check-prefixes=SSE,SSE41
4 ; RUN: llc < %s -mtriple=x86_64-- -mcpu=x86-64-v3 | FileCheck %s --check-prefixes=AVX2
5 ; RUN: llc < %s -mtriple=x86_64-- -mcpu=x86-64-v4 | FileCheck %s --check-prefixes=AVX512
7 define <8 x i16> @pow2_mask_v16i8(i8 zeroext %0) {
8 ; SSE2-LABEL: pow2_mask_v16i8:
9 ; SSE2: # %bb.0:
10 ; SSE2-NEXT: movd %edi, %xmm0
11 ; SSE2-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
12 ; SSE2-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
13 ; SSE2-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
14 ; SSE2-NEXT: movdqa {{.*#+}} xmm1 = [128,128,64,64,32,32,16,16,8,8,4,4,2,2,1,1]
15 ; SSE2-NEXT: pand %xmm1, %xmm0
16 ; SSE2-NEXT: pcmpeqb %xmm1, %xmm0
17 ; SSE2-NEXT: retq
18 ;
19 ; SSE41-LABEL: pow2_mask_v16i8:
20 ; SSE41: # %bb.0:
21 ; SSE41-NEXT: movd %edi, %xmm0
22 ; SSE41-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
23 ; SSE41-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
24 ; SSE41-NEXT: movq {{.*#+}} xmm1 = [128,64,32,16,8,4,2,1,0,0,0,0,0,0,0,0]
25 ; SSE41-NEXT: pand %xmm1, %xmm0
26 ; SSE41-NEXT: pcmpeqb %xmm1, %xmm0
27 ; SSE41-NEXT: pmovsxbw %xmm0, %xmm0
28 ; SSE41-NEXT: retq
29 ;
30 ; AVX2-LABEL: pow2_mask_v16i8:
31 ; AVX2: # %bb.0:
32 ; AVX2-NEXT: vmovd %edi, %xmm0
33 ; AVX2-NEXT: vpbroadcastb %xmm0, %xmm0
34 ; AVX2-NEXT: vpbroadcastq {{.*#+}} xmm1 = [128,64,32,16,8,4,2,1,128,64,32,16,8,4,2,1]
35 ; AVX2-NEXT: vpand %xmm1, %xmm0, %xmm0
36 ; AVX2-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm0
37 ; AVX2-NEXT: vpmovsxbw %xmm0, %xmm0
38 ; AVX2-NEXT: retq
39 ;
40 ; AVX512-LABEL: pow2_mask_v16i8:
41 ; AVX512: # %bb.0:
42 ; AVX512-NEXT: vpbroadcastb %edi, %xmm0
43 ; AVX512-NEXT: vptestmb {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %k0
44 ; AVX512-NEXT: vpmovm2w %k0, %xmm0
45 ; AVX512-NEXT: retq
46 %vec = insertelement <1 x i8> poison, i8 %0, i64 0
47 %splat = shufflevector <1 x i8> %vec, <1 x i8> poison, <8 x i32> zeroinitializer
48 %mask = and <8 x i8> %splat, <i8 -128, i8 64, i8 32, i8 16, i8 8, i8 4, i8 2, i8 1>
49 %not = icmp ne <8 x i8> %mask, zeroinitializer
50 %ext = sext <8 x i1> %not to <8 x i16>
51 ret <8 x i16> %ext
52 }
54 define <16 x i16> @pow2_mask_v16i16(i16 zeroext %0) {
55 ; SSE2-LABEL: pow2_mask_v16i16:
56 ; SSE2: # %bb.0:
57 ; SSE2-NEXT: movd %edi, %xmm0
58 ; SSE2-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
59 ; SSE2-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
60 ; SSE2-NEXT: movdqa {{.*#+}} xmm2 = [128,64,32,16,8,4,2,1]
61 ; SSE2-NEXT: movdqa %xmm0, %xmm1
62 ; SSE2-NEXT: pand %xmm2, %xmm1
63 ; SSE2-NEXT: movdqa {{.*#+}} xmm3 = [32768,16384,8192,4096,2048,1024,512,256]
64 ; SSE2-NEXT: pand %xmm3, %xmm0
65 ; SSE2-NEXT: pcmpeqw %xmm3, %xmm0
66 ; SSE2-NEXT: pcmpeqw %xmm2, %xmm1
67 ; SSE2-NEXT: retq
68 ;
69 ; SSE41-LABEL: pow2_mask_v16i16:
70 ; SSE41: # %bb.0:
71 ; SSE41-NEXT: movd %edi, %xmm0
72 ; SSE41-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
73 ; SSE41-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,0,0,0]
74 ; SSE41-NEXT: pmovzxbw {{.*#+}} xmm2 = [128,64,32,16,8,4,2,1]
75 ; SSE41-NEXT: movdqa %xmm0, %xmm1
76 ; SSE41-NEXT: pand %xmm2, %xmm1
77 ; SSE41-NEXT: movdqa {{.*#+}} xmm3 = [32768,16384,8192,4096,2048,1024,512,256]
78 ; SSE41-NEXT: pand %xmm3, %xmm0
79 ; SSE41-NEXT: pcmpeqw %xmm3, %xmm0
80 ; SSE41-NEXT: pcmpeqw %xmm2, %xmm1
81 ; SSE41-NEXT: retq
82 ;
83 ; AVX2-LABEL: pow2_mask_v16i16:
84 ; AVX2: # %bb.0:
85 ; AVX2-NEXT: vmovd %edi, %xmm0
86 ; AVX2-NEXT: vpbroadcastw %xmm0, %ymm0
87 ; AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = [32768,16384,8192,4096,2048,1024,512,256,128,64,32,16,8,4,2,1]
88 ; AVX2-NEXT: vpand %ymm1, %ymm0, %ymm0
89 ; AVX2-NEXT: vpcmpeqw %ymm1, %ymm0, %ymm0
90 ; AVX2-NEXT: retq
91 ;
92 ; AVX512-LABEL: pow2_mask_v16i16:
93 ; AVX512: # %bb.0:
94 ; AVX512-NEXT: vpbroadcastw %edi, %ymm0
95 ; AVX512-NEXT: vmovdqa {{.*#+}} ymm1 = [32768,16384,8192,4096,2048,1024,512,256,128,64,32,16,8,4,2,1]
96 ; AVX512-NEXT: vpand %ymm1, %ymm0, %ymm0
97 ; AVX512-NEXT: vpcmpeqw %ymm1, %ymm0, %ymm0
98 ; AVX512-NEXT: retq
99 %vec = insertelement <1 x i16> poison, i16 %0, i64 0
100 %splat = shufflevector <1 x i16> %vec, <1 x i16> poison, <16 x i32> zeroinitializer
101 %mask = and <16 x i16> %splat, <i16 -32768, i16 16384, i16 8192, i16 4096, i16 2048, i16 1024, i16 512, i16 256, i16 128, i16 64, i16 32, i16 16, i16 8, i16 4, i16 2, i16 1>
102 %not = icmp ne <16 x i16> %mask, zeroinitializer
103 %ext = sext <16 x i1> %not to <16 x i16>
104 ret <16 x i16> %ext
105 }
107 ; PR78888
108 define i64 @pow2_mask_v8i8(i8 zeroext %0) {
109 ; SSE-LABEL: pow2_mask_v8i8:
110 ; SSE: # %bb.0:
111 ; SSE-NEXT: movd %edi, %xmm0
112 ; SSE-NEXT: punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
113 ; SSE-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,0,0,0,4,5,6,7]
114 ; SSE-NEXT: movq {{.*#+}} xmm1 = [128,64,32,16,8,4,2,1,0,0,0,0,0,0,0,0]
115 ; SSE-NEXT: pand %xmm1, %xmm0
116 ; SSE-NEXT: pcmpeqb %xmm1, %xmm0
117 ; SSE-NEXT: movq %xmm0, %rax
118 ; SSE-NEXT: retq
119 ;
120 ; AVX2-LABEL: pow2_mask_v8i8:
121 ; AVX2: # %bb.0:
122 ; AVX2-NEXT: vmovd %edi, %xmm0
123 ; AVX2-NEXT: vpbroadcastb %xmm0, %xmm0
124 ; AVX2-NEXT: vpbroadcastq {{.*#+}} xmm1 = [128,64,32,16,8,4,2,1,128,64,32,16,8,4,2,1]
125 ; AVX2-NEXT: vpand %xmm1, %xmm0, %xmm0
126 ; AVX2-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm0
127 ; AVX2-NEXT: vmovq %xmm0, %rax
128 ; AVX2-NEXT: retq
129 ;
130 ; AVX512-LABEL: pow2_mask_v8i8:
131 ; AVX512: # %bb.0:
132 ; AVX512-NEXT: vpbroadcastb %edi, %xmm0
133 ; AVX512-NEXT: vpbroadcastq {{.*#+}} xmm1 = [128,64,32,16,8,4,2,1,128,64,32,16,8,4,2,1]
134 ; AVX512-NEXT: vpand %xmm1, %xmm0, %xmm0
135 ; AVX512-NEXT: vpcmpeqb %xmm1, %xmm0, %xmm0
136 ; AVX512-NEXT: vmovq %xmm0, %rax
137 ; AVX512-NEXT: retq
138 %vec = insertelement <1 x i8> poison, i8 %0, i64 0
139 %splat = shufflevector <1 x i8> %vec, <1 x i8> poison, <8 x i32> zeroinitializer
140 %mask = and <8 x i8> %splat, <i8 -128, i8 64, i8 32, i8 16, i8 8, i8 4, i8 2, i8 1>
141 %not = icmp ne <8 x i8> %mask, zeroinitializer
142 %ext = sext <8 x i1> %not to <8 x i8>
143 %res = bitcast <8 x i8> %ext to i64
144 ret i64 %res
145 }