1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
|
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=CHECK,AVX1
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=CHECK,AVX2
declare i32 @llvm.x86.avx.movmsk.pd.256(<4 x double>)
declare i32 @llvm.x86.avx.movmsk.ps.256(<8 x float>)
; Use widest possible vector for movmsk comparisons (PR37087)
define i1 @movmskps_noneof_bitcast_v4f64(<4 x double> %a0) {
; CHECK-LABEL: movmskps_noneof_bitcast_v4f64:
; CHECK: # %bb.0:
; CHECK-NEXT: vxorpd %xmm1, %xmm1, %xmm1
; CHECK-NEXT: vcmpeqpd %ymm1, %ymm0, %ymm0
; CHECK-NEXT: vmovmskpd %ymm0, %eax
; CHECK-NEXT: testl %eax, %eax
; CHECK-NEXT: sete %al
; CHECK-NEXT: vzeroupper
; CHECK-NEXT: retq
%1 = fcmp oeq <4 x double> %a0, zeroinitializer
%2 = sext <4 x i1> %1 to <4 x i64>
%3 = bitcast <4 x i64> %2 to <8 x float>
%4 = tail call i32 @llvm.x86.avx.movmsk.ps.256(<8 x float> %3)
%5 = icmp eq i32 %4, 0
ret i1 %5
}
define i1 @movmskps_allof_bitcast_v4f64(<4 x double> %a0) {
; CHECK-LABEL: movmskps_allof_bitcast_v4f64:
; CHECK: # %bb.0:
; CHECK-NEXT: vxorpd %xmm1, %xmm1, %xmm1
; CHECK-NEXT: vcmpeqpd %ymm1, %ymm0, %ymm0
; CHECK-NEXT: vmovmskpd %ymm0, %eax
; CHECK-NEXT: cmpl $15, %eax
; CHECK-NEXT: sete %al
; CHECK-NEXT: vzeroupper
; CHECK-NEXT: retq
%1 = fcmp oeq <4 x double> %a0, zeroinitializer
%2 = sext <4 x i1> %1 to <4 x i64>
%3 = bitcast <4 x i64> %2 to <8 x float>
%4 = tail call i32 @llvm.x86.avx.movmsk.ps.256(<8 x float> %3)
%5 = icmp eq i32 %4, 255
ret i1 %5
}
;
; TODO - Avoid sign extension ops when just extracting the sign bits.
;
define i32 @movmskpd_cmpgt_v4i64(<4 x i64> %a0) {
; AVX1-LABEL: movmskpd_cmpgt_v4i64:
; AVX1: # %bb.0:
; AVX1-NEXT: vpxor %xmm1, %xmm1, %xmm1
; AVX1-NEXT: vpcmpgtq %xmm0, %xmm1, %xmm1
; AVX1-NEXT: vblendpd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3]
; AVX1-NEXT: vmovmskpd %ymm0, %eax
; AVX1-NEXT: vzeroupper
; AVX1-NEXT: retq
;
; AVX2-LABEL: movmskpd_cmpgt_v4i64:
; AVX2: # %bb.0:
; AVX2-NEXT: vmovmskpd %ymm0, %eax
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
%1 = icmp sgt <4 x i64> zeroinitializer, %a0
%2 = sext <4 x i1> %1 to <4 x i64>
%3 = bitcast <4 x i64> %2 to <4 x double>
%4 = tail call i32 @llvm.x86.avx.movmsk.pd.256(<4 x double> %3)
ret i32 %4
}
define i32 @movmskps_ashr_v8i32(<8 x i32> %a0) {
; AVX1-LABEL: movmskps_ashr_v8i32:
; AVX1: # %bb.0:
; AVX1-NEXT: vpsrad $31, %xmm0, %xmm1
; AVX1-NEXT: vblendps {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]
; AVX1-NEXT: vmovmskps %ymm0, %eax
; AVX1-NEXT: vzeroupper
; AVX1-NEXT: retq
;
; AVX2-LABEL: movmskps_ashr_v8i32:
; AVX2: # %bb.0:
; AVX2-NEXT: vmovmskps %ymm0, %eax
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
%1 = ashr <8 x i32> %a0, <i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31, i32 31>
%2 = bitcast <8 x i32> %1 to <8 x float>
%3 = tail call i32 @llvm.x86.avx.movmsk.ps.256(<8 x float> %2)
ret i32 %3
}
define i32 @movmskps_sext_v4i64(<4 x i32> %a0) {
; AVX1-LABEL: movmskps_sext_v4i64:
; AVX1: # %bb.0:
; AVX1-NEXT: vpmovsxdq %xmm0, %xmm1
; AVX1-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[2,3,2,3]
; AVX1-NEXT: vpmovsxdq %xmm0, %xmm0
; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0
; AVX1-NEXT: vmovmskpd %ymm0, %eax
; AVX1-NEXT: vzeroupper
; AVX1-NEXT: retq
;
; AVX2-LABEL: movmskps_sext_v4i64:
; AVX2: # %bb.0:
; AVX2-NEXT: vpmovsxdq %xmm0, %ymm0
; AVX2-NEXT: vmovmskpd %ymm0, %eax
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
%1 = sext <4 x i32> %a0 to <4 x i64>
%2 = bitcast <4 x i64> %1 to <4 x double>
%3 = tail call i32 @llvm.x86.avx.movmsk.pd.256(<4 x double> %2)
ret i32 %3
}
define i32 @movmskps_sext_v8i32(<8 x i16> %a0) {
; AVX1-LABEL: movmskps_sext_v8i32:
; AVX1: # %bb.0:
; AVX1-NEXT: vpmovsxwd %xmm0, %xmm1
; AVX1-NEXT: vpshufd {{.*#+}} xmm0 = xmm0[2,3,2,3]
; AVX1-NEXT: vpmovsxwd %xmm0, %xmm0
; AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0
; AVX1-NEXT: vmovmskps %ymm0, %eax
; AVX1-NEXT: vzeroupper
; AVX1-NEXT: retq
;
; AVX2-LABEL: movmskps_sext_v8i32:
; AVX2: # %bb.0:
; AVX2-NEXT: vpmovsxwd %xmm0, %ymm0
; AVX2-NEXT: vmovmskps %ymm0, %eax
; AVX2-NEXT: vzeroupper
; AVX2-NEXT: retq
%1 = sext <8 x i16> %a0 to <8 x i32>
%2 = bitcast <8 x i32> %1 to <8 x float>
%3 = tail call i32 @llvm.x86.avx.movmsk.ps.256(<8 x float> %2)
ret i32 %3
}
define i32 @movmskps_concat_v4f32(<4 x float> %a0, <4 x float> %a1) {
; CHECK-LABEL: movmskps_concat_v4f32:
; CHECK: # %bb.0:
; CHECK-NEXT: vorps %xmm1, %xmm0, %xmm0
; CHECK-NEXT: vmovmskps %xmm0, %ecx
; CHECK-NEXT: xorl %eax, %eax
; CHECK-NEXT: negl %ecx
; CHECK-NEXT: sbbl %eax, %eax
; CHECK-NEXT: retq
%1 = shufflevector <4 x float> %a0, <4 x float> %a1, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
%2 = tail call i32 @llvm.x86.avx.movmsk.ps.256(<8 x float> %1)
%3 = icmp ne i32 %2, 0
%4 = sext i1 %3 to i32
ret i32 %4
}
define i32 @movmskps_demanded_concat_v4f32(<4 x float> %a0, <4 x float> %a1) {
; CHECK-LABEL: movmskps_demanded_concat_v4f32:
; CHECK: # %bb.0:
; CHECK-NEXT: vmovmskps %xmm0, %ecx
; CHECK-NEXT: andl $3, %ecx
; CHECK-NEXT: xorl %eax, %eax
; CHECK-NEXT: negl %ecx
; CHECK-NEXT: sbbl %eax, %eax
; CHECK-NEXT: retq
%1 = shufflevector <4 x float> %a0, <4 x float> %a1, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
%2 = tail call i32 @llvm.x86.avx.movmsk.ps.256(<8 x float> %1)
%3 = and i32 %2, 3
%4 = icmp ne i32 %3, 0
%5 = sext i1 %4 to i32
ret i32 %5
}
|
