xref: /llvm-project/llvm/test/Transforms/PhaseOrdering/X86/blendv-select.ll (revision da286c8bf69684d1612d1fc440bd9c6f1a4326df) 
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
2; RUN: opt < %s -O3 -S -mtriple=x86_64-- -mcpu=x86-64-v2 | FileCheck %s
3; RUN: opt < %s -O3 -S -mtriple=x86_64-- -mcpu=x86-64-v3 | FileCheck %s
4; RUN: opt < %s -O3 -S -mtriple=x86_64-- -mcpu=x86-64-v4 | FileCheck %s
5
6;
7; PR58895 - replace shuffled _mm_blendv_epi8+icmp with select+icmp
8;
9
10;
11; 256-bits types split to use 128-bit PBLENDVB selects
12;
13
14define <4 x double> @x86_pblendvb_v4f64_v2f64(<4 x double> %a, <4 x double> %b, <4 x double> %c, <4 x double> %d) {
15; CHECK-LABEL: @x86_pblendvb_v4f64_v2f64(
16; CHECK-NEXT:    [[CMP:%.*]] = fcmp olt <4 x double> [[C:%.*]], [[D:%.*]]
17; CHECK-NEXT:    [[DOTV:%.*]] = select <4 x i1> [[CMP]], <4 x double> [[B:%.*]], <4 x double> [[A:%.*]]
18; CHECK-NEXT:    ret <4 x double> [[DOTV]]
19;
20  %a.bc = bitcast <4 x double> %a to <32 x i8>
21  %b.bc = bitcast <4 x double> %b to <32 x i8>
22  %a.lo = shufflevector <32 x i8> %a.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
23  %b.lo = shufflevector <32 x i8> %b.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
24  %a.hi = shufflevector <32 x i8> %a.bc, <32 x i8> poison, <16 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
25  %b.hi = shufflevector <32 x i8> %b.bc, <32 x i8> poison, <16 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
26  %cmp = fcmp olt <4 x double> %c, %d
27  %sext = sext <4 x i1> %cmp to <4 x i64>
28  %sext.bc = bitcast <4 x i64> %sext to <32 x i8>
29  %sext.lo = shufflevector <32 x i8> %sext.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
30  %sext.hi = shufflevector <32 x i8> %sext.bc, <32 x i8> poison, <16 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
31  %sel.lo = tail call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %a.lo, <16 x i8> %b.lo, <16 x i8> %sext.lo)
32  %sel.hi = tail call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %a.hi, <16 x i8> %b.hi, <16 x i8> %sext.hi)
33  %concat = shufflevector <16 x i8> %sel.lo, <16 x i8> %sel.hi, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
34  %res = bitcast <32 x i8> %concat to <4 x double>
35  ret <4 x double> %res
36}
37
38define <8 x float> @x86_pblendvb_v8f32_v4f32(<8 x float> %a, <8 x float> %b, <8 x float> %c, <8 x float> %d) {
39; CHECK-LABEL: @x86_pblendvb_v8f32_v4f32(
40; CHECK-NEXT:    [[CMP:%.*]] = fcmp olt <8 x float> [[C:%.*]], [[D:%.*]]
41; CHECK-NEXT:    [[DOTV:%.*]] = select <8 x i1> [[CMP]], <8 x float> [[B:%.*]], <8 x float> [[A:%.*]]
42; CHECK-NEXT:    ret <8 x float> [[DOTV]]
43;
44  %a.bc = bitcast <8 x float> %a to <32 x i8>
45  %b.bc = bitcast <8 x float> %b to <32 x i8>
46  %a.lo = shufflevector <32 x i8> %a.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
47  %b.lo = shufflevector <32 x i8> %b.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
48  %a.hi = shufflevector <32 x i8> %a.bc, <32 x i8> poison, <16 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
49  %b.hi = shufflevector <32 x i8> %b.bc, <32 x i8> poison, <16 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
50  %cmp = fcmp olt <8 x float> %c, %d
51  %sext = sext <8 x i1> %cmp to <8 x i32>
52  %sext.bc = bitcast <8 x i32> %sext to <32 x i8>
53  %sext.lo = shufflevector <32 x i8> %sext.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
54  %sext.hi = shufflevector <32 x i8> %sext.bc, <32 x i8> poison, <16 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
55  %sel.lo = tail call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %a.lo, <16 x i8> %b.lo, <16 x i8> %sext.lo)
56  %sel.hi = tail call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %a.hi, <16 x i8> %b.hi, <16 x i8> %sext.hi)
57  %concat = shufflevector <16 x i8> %sel.lo, <16 x i8> %sel.hi, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
58  %res = bitcast <32 x i8> %concat to <8 x float>
59  ret <8 x float> %res
60}
61
62define <4 x i64> @x86_pblendvb_v4i64_v2i64(<4 x i64> %a, <4 x i64> %b, <4 x i64> %c, <4 x i64> %d) {
63; CHECK-LABEL: @x86_pblendvb_v4i64_v2i64(
64; CHECK-NEXT:    [[CMP:%.*]] = icmp slt <4 x i64> [[C:%.*]], [[D:%.*]]
65; CHECK-NEXT:    [[TMP1:%.*]] = select <4 x i1> [[CMP]], <4 x i64> [[B:%.*]], <4 x i64> [[A:%.*]]
66; CHECK-NEXT:    ret <4 x i64> [[TMP1]]
67;
68  %a.bc = bitcast <4 x i64> %a to <32 x i8>
69  %b.bc = bitcast <4 x i64> %b to <32 x i8>
70  %a.lo = shufflevector <32 x i8> %a.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
71  %b.lo = shufflevector <32 x i8> %b.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
72  %a.hi = shufflevector <32 x i8> %a.bc, <32 x i8> poison, <16 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
73  %b.hi = shufflevector <32 x i8> %b.bc, <32 x i8> poison, <16 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
74  %cmp = icmp slt <4 x i64> %c, %d
75  %sext = sext <4 x i1> %cmp to <4 x i64>
76  %sext.bc = bitcast <4 x i64> %sext to <32 x i8>
77  %sext.lo = shufflevector <32 x i8> %sext.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
78  %sext.hi = shufflevector <32 x i8> %sext.bc, <32 x i8> poison, <16 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
79  %sel.lo = tail call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %a.lo, <16 x i8> %b.lo, <16 x i8> %sext.lo)
80  %sel.hi = tail call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %a.hi, <16 x i8> %b.hi, <16 x i8> %sext.hi)
81  %concat = shufflevector <16 x i8> %sel.lo, <16 x i8> %sel.hi, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
82  %res = bitcast <32 x i8> %concat to <4 x i64>
83  ret <4 x i64> %res
84}
85
86define <4 x i64> @x86_pblendvb_v8i32_v4i32(<4 x i64> %a, <4 x i64> %b, <4 x i64> %c, <4 x i64> %d) {
87; CHECK-LABEL: @x86_pblendvb_v8i32_v4i32(
88; CHECK-NEXT:    [[C_BC:%.*]] = bitcast <4 x i64> [[C:%.*]] to <8 x i32>
89; CHECK-NEXT:    [[D_BC:%.*]] = bitcast <4 x i64> [[D:%.*]] to <8 x i32>
90; CHECK-NEXT:    [[CMP:%.*]] = icmp slt <8 x i32> [[C_BC]], [[D_BC]]
91; CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i64> [[A:%.*]] to <8 x i32>
92; CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i64> [[B:%.*]] to <8 x i32>
93; CHECK-NEXT:    [[TMP3:%.*]] = select <8 x i1> [[CMP]], <8 x i32> [[TMP2]], <8 x i32> [[TMP1]]
94; CHECK-NEXT:    [[RES:%.*]] = bitcast <8 x i32> [[TMP3]] to <4 x i64>
95; CHECK-NEXT:    ret <4 x i64> [[RES]]
96;
97  %a.bc = bitcast <4 x i64> %a to <32 x i8>
98  %b.bc = bitcast <4 x i64> %b to <32 x i8>
99  %c.bc = bitcast <4 x i64> %c to <8 x i32>
100  %d.bc = bitcast <4 x i64> %d to <8 x i32>
101  %a.lo = shufflevector <32 x i8> %a.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
102  %b.lo = shufflevector <32 x i8> %b.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
103  %a.hi = shufflevector <32 x i8> %a.bc, <32 x i8> poison, <16 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
104  %b.hi = shufflevector <32 x i8> %b.bc, <32 x i8> poison, <16 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
105  %cmp = icmp slt <8 x i32> %c.bc, %d.bc
106  %sext = sext <8 x i1> %cmp to <8 x i32>
107  %sext.bc = bitcast <8 x i32> %sext to <32 x i8>
108  %sext.lo = shufflevector <32 x i8> %sext.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
109  %sext.hi = shufflevector <32 x i8> %sext.bc, <32 x i8> poison, <16 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
110  %sel.lo = tail call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %a.lo, <16 x i8> %b.lo, <16 x i8> %sext.lo)
111  %sel.hi = tail call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %a.hi, <16 x i8> %b.hi, <16 x i8> %sext.hi)
112  %concat = shufflevector <16 x i8> %sel.lo, <16 x i8> %sel.hi, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
113  %res = bitcast <32 x i8> %concat to <4 x i64>
114  ret <4 x i64> %res
115}
116
117define <4 x i64> @x86_pblendvb_v16i16_v8i16(<4 x i64> %a, <4 x i64> %b, <4 x i64> %c, <4 x i64> %d) {
118; CHECK-LABEL: @x86_pblendvb_v16i16_v8i16(
119; CHECK-NEXT:    [[C_BC:%.*]] = bitcast <4 x i64> [[C:%.*]] to <16 x i16>
120; CHECK-NEXT:    [[D_BC:%.*]] = bitcast <4 x i64> [[D:%.*]] to <16 x i16>
121; CHECK-NEXT:    [[CMP:%.*]] = icmp slt <16 x i16> [[C_BC]], [[D_BC]]
122; CHECK-NEXT:    [[TMP1:%.*]] = bitcast <4 x i64> [[A:%.*]] to <16 x i16>
123; CHECK-NEXT:    [[TMP2:%.*]] = bitcast <4 x i64> [[B:%.*]] to <16 x i16>
124; CHECK-NEXT:    [[TMP3:%.*]] = select <16 x i1> [[CMP]], <16 x i16> [[TMP2]], <16 x i16> [[TMP1]]
125; CHECK-NEXT:    [[RES:%.*]] = bitcast <16 x i16> [[TMP3]] to <4 x i64>
126; CHECK-NEXT:    ret <4 x i64> [[RES]]
127;
128  %a.bc = bitcast <4 x i64> %a to <32 x i8>
129  %b.bc = bitcast <4 x i64> %b to <32 x i8>
130  %c.bc = bitcast <4 x i64> %c to <16 x i16>
131  %d.bc = bitcast <4 x i64> %d to <16 x i16>
132  %a.lo = shufflevector <32 x i8> %a.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
133  %b.lo = shufflevector <32 x i8> %b.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
134  %a.hi = shufflevector <32 x i8> %a.bc, <32 x i8> poison, <16 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
135  %b.hi = shufflevector <32 x i8> %b.bc, <32 x i8> poison, <16 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
136  %cmp = icmp slt <16 x i16> %c.bc, %d.bc
137  %sext = sext <16 x i1> %cmp to <16 x i16>
138  %sext.bc = bitcast <16 x i16> %sext to <32 x i8>
139  %sext.lo = shufflevector <32 x i8> %sext.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
140  %sext.hi = shufflevector <32 x i8> %sext.bc, <32 x i8> poison, <16 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
141  %sel.lo = tail call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %a.lo, <16 x i8> %b.lo, <16 x i8> %sext.lo)
142  %sel.hi = tail call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %a.hi, <16 x i8> %b.hi, <16 x i8> %sext.hi)
143  %concat = shufflevector <16 x i8> %sel.lo, <16 x i8> %sel.hi, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
144  %res = bitcast <32 x i8> %concat to <4 x i64>
145  ret <4 x i64> %res
146}
147
148define <4 x i64> @x86_pblendvb_v32i8_v16i8(<4 x i64> %a, <4 x i64> %b, <4 x i64> %c, <4 x i64> %d) {
149; CHECK-LABEL: @x86_pblendvb_v32i8_v16i8(
150; CHECK-NEXT:    [[A_BC:%.*]] = bitcast <4 x i64> [[A:%.*]] to <32 x i8>
151; CHECK-NEXT:    [[B_BC:%.*]] = bitcast <4 x i64> [[B:%.*]] to <32 x i8>
152; CHECK-NEXT:    [[C_BC:%.*]] = bitcast <4 x i64> [[C:%.*]] to <32 x i8>
153; CHECK-NEXT:    [[D_BC:%.*]] = bitcast <4 x i64> [[D:%.*]] to <32 x i8>
154; CHECK-NEXT:    [[CMP:%.*]] = icmp slt <32 x i8> [[C_BC]], [[D_BC]]
155; CHECK-NEXT:    [[CONCAT:%.*]] = select <32 x i1> [[CMP]], <32 x i8> [[B_BC]], <32 x i8> [[A_BC]]
156; CHECK-NEXT:    [[RES:%.*]] = bitcast <32 x i8> [[CONCAT]] to <4 x i64>
157; CHECK-NEXT:    ret <4 x i64> [[RES]]
158;
159  %a.bc = bitcast <4 x i64> %a to <32 x i8>
160  %b.bc = bitcast <4 x i64> %b to <32 x i8>
161  %c.bc = bitcast <4 x i64> %c to <32 x i8>
162  %d.bc = bitcast <4 x i64> %d to <32 x i8>
163  %a.lo = shufflevector <32 x i8> %a.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
164  %b.lo = shufflevector <32 x i8> %b.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
165  %a.hi = shufflevector <32 x i8> %a.bc, <32 x i8> poison, <16 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
166  %b.hi = shufflevector <32 x i8> %b.bc, <32 x i8> poison, <16 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
167  %cmp = icmp slt <32 x i8> %c.bc, %d.bc
168  %sext = sext <32 x i1> %cmp to <32 x i8>
169  %sext.lo = shufflevector <32 x i8> %sext, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
170  %sext.hi = shufflevector <32 x i8> %sext, <32 x i8> poison, <16 x i32> <i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
171  %sel.lo = tail call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %a.lo, <16 x i8> %b.lo, <16 x i8> %sext.lo)
172  %sel.hi = tail call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %a.hi, <16 x i8> %b.hi, <16 x i8> %sext.hi)
173  %concat = shufflevector <16 x i8> %sel.lo, <16 x i8> %sel.hi, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
174  %res = bitcast <32 x i8> %concat to <4 x i64>
175  ret <4 x i64> %res
176}
177
178;
179; 512-bits types split to use 256-bit PBLENDVB selects
180;
181
182define <8 x double> @x86_pblendvb_v8f64_v4f64(<8 x double> %a, <8 x double> %b, <8 x double> %c, <8 x double> %d) {
183; CHECK-LABEL: @x86_pblendvb_v8f64_v4f64(
184; CHECK-NEXT:    [[CMP:%.*]] = fcmp olt <8 x double> [[C:%.*]], [[D:%.*]]
185; CHECK-NEXT:    [[DOTV:%.*]] = select <8 x i1> [[CMP]], <8 x double> [[B:%.*]], <8 x double> [[A:%.*]]
186; CHECK-NEXT:    ret <8 x double> [[DOTV]]
187;
188  %a.bc = bitcast <8 x double> %a to <64 x i8>
189  %b.bc = bitcast <8 x double> %b to <64 x i8>
190  %a.lo = shufflevector <64 x i8> %a.bc, <64 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
191  %b.lo = shufflevector <64 x i8> %b.bc, <64 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
192  %a.hi = shufflevector <64 x i8> %a.bc, <64 x i8> poison, <32 x i32> <i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
193  %b.hi = shufflevector <64 x i8> %b.bc, <64 x i8> poison, <32 x i32> <i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
194  %cmp = fcmp olt <8 x double> %c, %d
195  %sext = sext <8 x i1> %cmp to <8 x i64>
196  %sext.bc = bitcast <8 x i64> %sext to <64 x i8>
197  %sext.lo = shufflevector <64 x i8> %sext.bc, <64 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
198  %sext.hi = shufflevector <64 x i8> %sext.bc, <64 x i8> poison, <32 x i32> <i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
199  %sel.lo = tail call <32 x i8> @llvm.x86.avx2.pblendvb(<32 x i8> %a.lo, <32 x i8> %b.lo, <32 x i8> %sext.lo)
200  %sel.hi = tail call <32 x i8> @llvm.x86.avx2.pblendvb(<32 x i8> %a.hi, <32 x i8> %b.hi, <32 x i8> %sext.hi)
201  %concat = shufflevector <32 x i8> %sel.lo, <32 x i8> %sel.hi, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
202  %res = bitcast <64 x i8> %concat to <8 x double>
203  ret <8 x double> %res
204}
205
206define <16 x float> @x86_pblendvb_v16f32_v8f32(<16 x float> %a, <16 x float> %b, <16 x float> %c, <16 x float> %d) {
207; CHECK-LABEL: @x86_pblendvb_v16f32_v8f32(
208; CHECK-NEXT:    [[CMP:%.*]] = fcmp olt <16 x float> [[C:%.*]], [[D:%.*]]
209; CHECK-NEXT:    [[DOTV:%.*]] = select <16 x i1> [[CMP]], <16 x float> [[B:%.*]], <16 x float> [[A:%.*]]
210; CHECK-NEXT:    ret <16 x float> [[DOTV]]
211;
212  %a.bc = bitcast <16 x float> %a to <64 x i8>
213  %b.bc = bitcast <16 x float> %b to <64 x i8>
214  %a.lo = shufflevector <64 x i8> %a.bc, <64 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
215  %b.lo = shufflevector <64 x i8> %b.bc, <64 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
216  %a.hi = shufflevector <64 x i8> %a.bc, <64 x i8> poison, <32 x i32> <i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
217  %b.hi = shufflevector <64 x i8> %b.bc, <64 x i8> poison, <32 x i32> <i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
218  %cmp = fcmp olt <16 x float> %c, %d
219  %sext = sext <16 x i1> %cmp to <16 x i32>
220  %sext.bc = bitcast <16 x i32> %sext to <64 x i8>
221  %sext.lo = shufflevector <64 x i8> %sext.bc, <64 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
222  %sext.hi = shufflevector <64 x i8> %sext.bc, <64 x i8> poison, <32 x i32> <i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
223  %sel.lo = tail call <32 x i8> @llvm.x86.avx2.pblendvb(<32 x i8> %a.lo, <32 x i8> %b.lo, <32 x i8> %sext.lo)
224  %sel.hi = tail call <32 x i8> @llvm.x86.avx2.pblendvb(<32 x i8> %a.hi, <32 x i8> %b.hi, <32 x i8> %sext.hi)
225  %concat = shufflevector <32 x i8> %sel.lo, <32 x i8> %sel.hi, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
226  %res = bitcast <64 x i8> %concat to <16 x float>
227  ret <16 x float> %res
228}
229
230define <8 x i64> @x86_pblendvb_v8i64_v4i64(<8 x i64> %a, <8 x i64> %b, <8 x i64> %c, <8 x i64> %d) {
231; CHECK-LABEL: @x86_pblendvb_v8i64_v4i64(
232; CHECK-NEXT:    [[CMP:%.*]] = icmp slt <8 x i64> [[C:%.*]], [[D:%.*]]
233; CHECK-NEXT:    [[TMP1:%.*]] = select <8 x i1> [[CMP]], <8 x i64> [[B:%.*]], <8 x i64> [[A:%.*]]
234; CHECK-NEXT:    ret <8 x i64> [[TMP1]]
235;
236  %a.bc = bitcast <8 x i64> %a to <64 x i8>
237  %b.bc = bitcast <8 x i64> %b to <64 x i8>
238  %a.lo = shufflevector <64 x i8> %a.bc, <64 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
239  %b.lo = shufflevector <64 x i8> %b.bc, <64 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
240  %a.hi = shufflevector <64 x i8> %a.bc, <64 x i8> poison, <32 x i32> <i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
241  %b.hi = shufflevector <64 x i8> %b.bc, <64 x i8> poison, <32 x i32> <i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
242  %cmp = icmp slt <8 x i64> %c, %d
243  %sext = sext <8 x i1> %cmp to <8 x i64>
244  %sext.bc = bitcast <8 x i64> %sext to <64 x i8>
245  %sext.lo = shufflevector <64 x i8> %sext.bc, <64 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
246  %sext.hi = shufflevector <64 x i8> %sext.bc, <64 x i8> poison, <32 x i32> <i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
247  %sel.lo = tail call <32 x i8> @llvm.x86.avx2.pblendvb(<32 x i8> %a.lo, <32 x i8> %b.lo, <32 x i8> %sext.lo)
248  %sel.hi = tail call <32 x i8> @llvm.x86.avx2.pblendvb(<32 x i8> %a.hi, <32 x i8> %b.hi, <32 x i8> %sext.hi)
249  %concat = shufflevector <32 x i8> %sel.lo, <32 x i8> %sel.hi, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
250  %res = bitcast <64 x i8> %concat to <8 x i64>
251  ret <8 x i64> %res
252}
253
254define <8 x i64> @x86_pblendvb_v16i32_v8i32(<8 x i64> %a, <8 x i64> %b, <8 x i64> %c, <8 x i64> %d) {
255; CHECK-LABEL: @x86_pblendvb_v16i32_v8i32(
256; CHECK-NEXT:    [[C_BC:%.*]] = bitcast <8 x i64> [[C:%.*]] to <16 x i32>
257; CHECK-NEXT:    [[D_BC:%.*]] = bitcast <8 x i64> [[D:%.*]] to <16 x i32>
258; CHECK-NEXT:    [[CMP:%.*]] = icmp slt <16 x i32> [[C_BC]], [[D_BC]]
259; CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i64> [[A:%.*]] to <16 x i32>
260; CHECK-NEXT:    [[TMP2:%.*]] = bitcast <8 x i64> [[B:%.*]] to <16 x i32>
261; CHECK-NEXT:    [[TMP3:%.*]] = select <16 x i1> [[CMP]], <16 x i32> [[TMP2]], <16 x i32> [[TMP1]]
262; CHECK-NEXT:    [[RES:%.*]] = bitcast <16 x i32> [[TMP3]] to <8 x i64>
263; CHECK-NEXT:    ret <8 x i64> [[RES]]
264;
265  %a.bc = bitcast <8 x i64> %a to <64 x i8>
266  %b.bc = bitcast <8 x i64> %b to <64 x i8>
267  %c.bc = bitcast <8 x i64> %c to <16 x i32>
268  %d.bc = bitcast <8 x i64> %d to <16 x i32>
269  %a.lo = shufflevector <64 x i8> %a.bc, <64 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
270  %b.lo = shufflevector <64 x i8> %b.bc, <64 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
271  %a.hi = shufflevector <64 x i8> %a.bc, <64 x i8> poison, <32 x i32> <i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
272  %b.hi = shufflevector <64 x i8> %b.bc, <64 x i8> poison, <32 x i32> <i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
273  %cmp = icmp slt <16 x i32> %c.bc, %d.bc
274  %sext = sext <16 x i1> %cmp to <16 x i32>
275  %sext.bc = bitcast <16 x i32> %sext to <64 x i8>
276  %sext.lo = shufflevector <64 x i8> %sext.bc, <64 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
277  %sext.hi = shufflevector <64 x i8> %sext.bc, <64 x i8> poison, <32 x i32> <i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
278  %sel.lo = tail call <32 x i8> @llvm.x86.avx2.pblendvb(<32 x i8> %a.lo, <32 x i8> %b.lo, <32 x i8> %sext.lo)
279  %sel.hi = tail call <32 x i8> @llvm.x86.avx2.pblendvb(<32 x i8> %a.hi, <32 x i8> %b.hi, <32 x i8> %sext.hi)
280  %concat = shufflevector <32 x i8> %sel.lo, <32 x i8> %sel.hi, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
281  %res = bitcast <64 x i8> %concat to <8 x i64>
282  ret <8 x i64> %res
283}
284
285define <8 x i64> @x86_pblendvb_v32i16_v16i16(<8 x i64> %a, <8 x i64> %b, <8 x i64> %c, <8 x i64> %d) {
286; CHECK-LABEL: @x86_pblendvb_v32i16_v16i16(
287; CHECK-NEXT:    [[C_BC:%.*]] = bitcast <8 x i64> [[C:%.*]] to <32 x i16>
288; CHECK-NEXT:    [[D_BC:%.*]] = bitcast <8 x i64> [[D:%.*]] to <32 x i16>
289; CHECK-NEXT:    [[CMP:%.*]] = icmp slt <32 x i16> [[C_BC]], [[D_BC]]
290; CHECK-NEXT:    [[TMP1:%.*]] = bitcast <8 x i64> [[A:%.*]] to <32 x i16>
291; CHECK-NEXT:    [[TMP2:%.*]] = bitcast <8 x i64> [[B:%.*]] to <32 x i16>
292; CHECK-NEXT:    [[TMP3:%.*]] = select <32 x i1> [[CMP]], <32 x i16> [[TMP2]], <32 x i16> [[TMP1]]
293; CHECK-NEXT:    [[RES:%.*]] = bitcast <32 x i16> [[TMP3]] to <8 x i64>
294; CHECK-NEXT:    ret <8 x i64> [[RES]]
295;
296  %a.bc = bitcast <8 x i64> %a to <64 x i8>
297  %b.bc = bitcast <8 x i64> %b to <64 x i8>
298  %c.bc = bitcast <8 x i64> %c to <32 x i16>
299  %d.bc = bitcast <8 x i64> %d to <32 x i16>
300  %a.lo = shufflevector <64 x i8> %a.bc, <64 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
301  %b.lo = shufflevector <64 x i8> %b.bc, <64 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
302  %a.hi = shufflevector <64 x i8> %a.bc, <64 x i8> poison, <32 x i32> <i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
303  %b.hi = shufflevector <64 x i8> %b.bc, <64 x i8> poison, <32 x i32> <i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
304  %cmp = icmp slt <32 x i16> %c.bc, %d.bc
305  %sext = sext <32 x i1> %cmp to <32 x i16>
306  %sext.bc = bitcast <32 x i16> %sext to <64 x i8>
307  %sext.lo = shufflevector <64 x i8> %sext.bc, <64 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
308  %sext.hi = shufflevector <64 x i8> %sext.bc, <64 x i8> poison, <32 x i32> <i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
309  %sel.lo = tail call <32 x i8> @llvm.x86.avx2.pblendvb(<32 x i8> %a.lo, <32 x i8> %b.lo, <32 x i8> %sext.lo)
310  %sel.hi = tail call <32 x i8> @llvm.x86.avx2.pblendvb(<32 x i8> %a.hi, <32 x i8> %b.hi, <32 x i8> %sext.hi)
311  %concat = shufflevector <32 x i8> %sel.lo, <32 x i8> %sel.hi, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
312  %res = bitcast <64 x i8> %concat to <8 x i64>
313  ret <8 x i64> %res
314}
315
316define <8 x i64> @x86_pblendvb_v64i8_v32i8(<8 x i64> %a, <8 x i64> %b, <8 x i64> %c, <8 x i64> %d) {
317; CHECK-LABEL: @x86_pblendvb_v64i8_v32i8(
318; CHECK-NEXT:    [[A_BC:%.*]] = bitcast <8 x i64> [[A:%.*]] to <64 x i8>
319; CHECK-NEXT:    [[B_BC:%.*]] = bitcast <8 x i64> [[B:%.*]] to <64 x i8>
320; CHECK-NEXT:    [[C_BC:%.*]] = bitcast <8 x i64> [[C:%.*]] to <64 x i8>
321; CHECK-NEXT:    [[D_BC:%.*]] = bitcast <8 x i64> [[D:%.*]] to <64 x i8>
322; CHECK-NEXT:    [[CMP:%.*]] = icmp slt <64 x i8> [[C_BC]], [[D_BC]]
323; CHECK-NEXT:    [[CONCAT:%.*]] = select <64 x i1> [[CMP]], <64 x i8> [[B_BC]], <64 x i8> [[A_BC]]
324; CHECK-NEXT:    [[RES:%.*]] = bitcast <64 x i8> [[CONCAT]] to <8 x i64>
325; CHECK-NEXT:    ret <8 x i64> [[RES]]
326;
327  %a.bc = bitcast <8 x i64> %a to <64 x i8>
328  %b.bc = bitcast <8 x i64> %b to <64 x i8>
329  %c.bc = bitcast <8 x i64> %c to <64 x i8>
330  %d.bc = bitcast <8 x i64> %d to <64 x i8>
331  %a.lo = shufflevector <64 x i8> %a.bc, <64 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
332  %b.lo = shufflevector <64 x i8> %b.bc, <64 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
333  %a.hi = shufflevector <64 x i8> %a.bc, <64 x i8> poison, <32 x i32> <i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
334  %b.hi = shufflevector <64 x i8> %b.bc, <64 x i8> poison, <32 x i32> <i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
335  %cmp = icmp slt <64 x i8> %c.bc, %d.bc
336  %sext = sext <64 x i1> %cmp to <64 x i8>
337  %sext.lo = shufflevector <64 x i8> %sext, <64 x i8> poison, <32 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
338  %sext.hi = shufflevector <64 x i8> %sext, <64 x i8> poison, <32 x i32> <i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
339  %sel.lo = tail call <32 x i8> @llvm.x86.avx2.pblendvb(<32 x i8> %a.lo, <32 x i8> %b.lo, <32 x i8> %sext.lo)
340  %sel.hi = tail call <32 x i8> @llvm.x86.avx2.pblendvb(<32 x i8> %a.hi, <32 x i8> %b.hi, <32 x i8> %sext.hi)
341  %concat = shufflevector <32 x i8> %sel.lo, <32 x i8> %sel.hi, <64 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31, i32 32, i32 33, i32 34, i32 35, i32 36, i32 37, i32 38, i32 39, i32 40, i32 41, i32 42, i32 43, i32 44, i32 45, i32 46, i32 47, i32 48, i32 49, i32 50, i32 51, i32 52, i32 53, i32 54, i32 55, i32 56, i32 57, i32 58, i32 59, i32 60, i32 61, i32 62, i32 63>
342  %res = bitcast <64 x i8> %concat to <8 x i64>
343  ret <8 x i64> %res
344}
345
346;
347; Negative Tests
348;
349
350; non-pow2 source
351define <2 x double> @x86_pblendvb_v3f64_v2f64(<3 x double> %a, <3 x double> %b, <3 x double> %c, <3 x double> %d) {
352; CHECK-LABEL: @x86_pblendvb_v3f64_v2f64(
353; CHECK-NEXT:    [[A_BC:%.*]] = bitcast <3 x double> [[A:%.*]] to <24 x i8>
354; CHECK-NEXT:    [[B_BC:%.*]] = bitcast <3 x double> [[B:%.*]] to <24 x i8>
355; CHECK-NEXT:    [[A_LO:%.*]] = shufflevector <24 x i8> [[A_BC]], <24 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
356; CHECK-NEXT:    [[B_LO:%.*]] = shufflevector <24 x i8> [[B_BC]], <24 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
357; CHECK-NEXT:    [[CMP:%.*]] = fcmp olt <3 x double> [[C:%.*]], [[D:%.*]]
358; CHECK-NEXT:    [[SEXT:%.*]] = sext <3 x i1> [[CMP]] to <3 x i64>
359; CHECK-NEXT:    [[SEXT_BC:%.*]] = bitcast <3 x i64> [[SEXT]] to <24 x i8>
360; CHECK-NEXT:    [[SEXT_LO:%.*]] = shufflevector <24 x i8> [[SEXT_BC]], <24 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
361; CHECK-NEXT:    [[SEL_LO:%.*]] = tail call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> [[A_LO]], <16 x i8> [[B_LO]], <16 x i8> [[SEXT_LO]])
362; CHECK-NEXT:    [[RES:%.*]] = bitcast <16 x i8> [[SEL_LO]] to <2 x double>
363; CHECK-NEXT:    ret <2 x double> [[RES]]
364;
365  %a.bc = bitcast <3 x double> %a to <24 x i8>
366  %b.bc = bitcast <3 x double> %b to <24 x i8>
367  %a.lo = shufflevector <24 x i8> %a.bc, <24 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
368  %b.lo = shufflevector <24 x i8> %b.bc, <24 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
369  %cmp = fcmp olt <3 x double> %c, %d
370  %sext = sext <3 x i1> %cmp to <3 x i64>
371  %sext.bc = bitcast <3 x i64> %sext to <24 x i8>
372  %sext.lo = shufflevector <24 x i8> %sext.bc, <24 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
373  %sel.lo = tail call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %a.lo, <16 x i8> %b.lo, <16 x i8> %sext.lo)
374  %res = bitcast <16 x i8> %sel.lo to <2 x double>
375  ret <2 x double> %res
376}
377
378; shuffle masks contain undef
379define <2 x i64> @x86_pblendvb_v32i8_v16i8_undefs(<4 x i64> %a, <4 x i64> %b, <4 x i64> %c, <4 x i64> %d) {
380; CHECK-LABEL: @x86_pblendvb_v32i8_v16i8_undefs(
381; CHECK-NEXT:    [[A_BC:%.*]] = bitcast <4 x i64> [[A:%.*]] to <32 x i8>
382; CHECK-NEXT:    [[B_BC:%.*]] = bitcast <4 x i64> [[B:%.*]] to <32 x i8>
383; CHECK-NEXT:    [[C_BC:%.*]] = bitcast <4 x i64> [[C:%.*]] to <32 x i8>
384; CHECK-NEXT:    [[D_BC:%.*]] = bitcast <4 x i64> [[D:%.*]] to <32 x i8>
385; CHECK-NEXT:    [[A_LO:%.*]] = shufflevector <32 x i8> [[A_BC]], <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
386; CHECK-NEXT:    [[B_LO:%.*]] = shufflevector <32 x i8> [[B_BC]], <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
387; CHECK-NEXT:    [[CMP:%.*]] = icmp slt <32 x i8> [[C_BC]], [[D_BC]]
388; CHECK-NEXT:    [[SEXT:%.*]] = sext <32 x i1> [[CMP]] to <32 x i8>
389; CHECK-NEXT:    [[SEXT_LO:%.*]] = shufflevector <32 x i8> [[SEXT]], <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 poison, i32 poison, i32 poison, i32 poison>
390; CHECK-NEXT:    [[SEL_LO:%.*]] = tail call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> [[A_LO]], <16 x i8> [[B_LO]], <16 x i8> [[SEXT_LO]])
391; CHECK-NEXT:    [[RES:%.*]] = bitcast <16 x i8> [[SEL_LO]] to <2 x i64>
392; CHECK-NEXT:    ret <2 x i64> [[RES]]
393;
394  %a.bc = bitcast <4 x i64> %a to <32 x i8>
395  %b.bc = bitcast <4 x i64> %b to <32 x i8>
396  %c.bc = bitcast <4 x i64> %c to <32 x i8>
397  %d.bc = bitcast <4 x i64> %d to <32 x i8>
398  %a.lo = shufflevector <32 x i8> %a.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
399  %b.lo = shufflevector <32 x i8> %b.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
400  %cmp = icmp slt <32 x i8> %c.bc, %d.bc
401  %sext = sext <32 x i1> %cmp to <32 x i8>
402  %sext.lo = shufflevector <32 x i8> %sext, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 undef, i32 undef, i32 undef, i32 undef>
403  %sel.lo = tail call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %a.lo, <16 x i8> %b.lo, <16 x i8> %sext.lo)
404  %res = bitcast <16 x i8> %sel.lo to <2 x i64>
405  ret <2 x i64> %res
406}
407
408; masks elements too small
409define <2 x i64> @x86_pblendvb_v32i8_v32i4(<4 x i64> %a, <4 x i64> %b, <4 x i64> %c, <4 x i64> %d) {
410; CHECK-LABEL: @x86_pblendvb_v32i8_v32i4(
411; CHECK-NEXT:    [[A_BC:%.*]] = bitcast <4 x i64> [[A:%.*]] to <32 x i8>
412; CHECK-NEXT:    [[B_BC:%.*]] = bitcast <4 x i64> [[B:%.*]] to <32 x i8>
413; CHECK-NEXT:    [[C_BC:%.*]] = bitcast <4 x i64> [[C:%.*]] to <64 x i4>
414; CHECK-NEXT:    [[D_BC:%.*]] = bitcast <4 x i64> [[D:%.*]] to <64 x i4>
415; CHECK-NEXT:    [[A_LO:%.*]] = shufflevector <32 x i8> [[A_BC]], <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
416; CHECK-NEXT:    [[B_LO:%.*]] = shufflevector <32 x i8> [[B_BC]], <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
417; CHECK-NEXT:    [[CMP:%.*]] = icmp slt <64 x i4> [[C_BC]], [[D_BC]]
418; CHECK-NEXT:    [[SEXT:%.*]] = sext <64 x i1> [[CMP]] to <64 x i4>
419; CHECK-NEXT:    [[SEXT_BC:%.*]] = bitcast <64 x i4> [[SEXT]] to <32 x i8>
420; CHECK-NEXT:    [[SEXT_LO:%.*]] = shufflevector <32 x i8> [[SEXT_BC]], <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
421; CHECK-NEXT:    [[SEL_LO:%.*]] = tail call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> [[A_LO]], <16 x i8> [[B_LO]], <16 x i8> [[SEXT_LO]])
422; CHECK-NEXT:    [[RES:%.*]] = bitcast <16 x i8> [[SEL_LO]] to <2 x i64>
423; CHECK-NEXT:    ret <2 x i64> [[RES]]
424;
425  %a.bc = bitcast <4 x i64> %a to <32 x i8>
426  %b.bc = bitcast <4 x i64> %b to <32 x i8>
427  %c.bc = bitcast <4 x i64> %c to <64 x i4>
428  %d.bc = bitcast <4 x i64> %d to <64 x i4>
429  %a.lo = shufflevector <32 x i8> %a.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
430  %b.lo = shufflevector <32 x i8> %b.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
431  %cmp = icmp slt <64 x i4> %c.bc, %d.bc
432  %sext = sext <64 x i1> %cmp to <64 x i4>
433  %sext.bc = bitcast <64 x i4> %sext to <32 x i8>
434  %sext.lo = shufflevector <32 x i8> %sext.bc, <32 x i8> poison, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>
435  %sel.lo = tail call <16 x i8> @llvm.x86.sse41.pblendvb(<16 x i8> %a.lo, <16 x i8> %b.lo, <16 x i8> %sext.lo)
436  %res = bitcast <16 x i8> %sel.lo to <2 x i64>
437  ret <2 x i64> %res
438}
439