83
|
1 ; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=skx --show-mc-encoding| FileCheck %s
|
|
2
|
|
3 define i64 @test_pcmpeq_b(<64 x i8> %a, <64 x i8> %b) {
|
|
4 ; CHECK-LABEL: test_pcmpeq_b
|
|
5 ; CHECK: vpcmpeqb %zmm1, %zmm0, %k0 ##
|
|
6 %res = call i64 @llvm.x86.avx512.mask.pcmpeq.b.512(<64 x i8> %a, <64 x i8> %b, i64 -1)
|
|
7 ret i64 %res
|
|
8 }
|
|
9
|
|
10 define i64 @test_mask_pcmpeq_b(<64 x i8> %a, <64 x i8> %b, i64 %mask) {
|
|
11 ; CHECK-LABEL: test_mask_pcmpeq_b
|
|
12 ; CHECK: vpcmpeqb %zmm1, %zmm0, %k0 {%k1} ##
|
|
13 %res = call i64 @llvm.x86.avx512.mask.pcmpeq.b.512(<64 x i8> %a, <64 x i8> %b, i64 %mask)
|
|
14 ret i64 %res
|
|
15 }
|
|
16
|
|
17 declare i64 @llvm.x86.avx512.mask.pcmpeq.b.512(<64 x i8>, <64 x i8>, i64)
|
|
18
|
|
19 define i32 @test_pcmpeq_w(<32 x i16> %a, <32 x i16> %b) {
|
|
20 ; CHECK-LABEL: test_pcmpeq_w
|
|
21 ; CHECK: vpcmpeqw %zmm1, %zmm0, %k0 ##
|
|
22 %res = call i32 @llvm.x86.avx512.mask.pcmpeq.w.512(<32 x i16> %a, <32 x i16> %b, i32 -1)
|
|
23 ret i32 %res
|
|
24 }
|
|
25
|
|
26 define i32 @test_mask_pcmpeq_w(<32 x i16> %a, <32 x i16> %b, i32 %mask) {
|
|
27 ; CHECK-LABEL: test_mask_pcmpeq_w
|
|
28 ; CHECK: vpcmpeqw %zmm1, %zmm0, %k0 {%k1} ##
|
|
29 %res = call i32 @llvm.x86.avx512.mask.pcmpeq.w.512(<32 x i16> %a, <32 x i16> %b, i32 %mask)
|
|
30 ret i32 %res
|
|
31 }
|
|
32
|
|
33 declare i32 @llvm.x86.avx512.mask.pcmpeq.w.512(<32 x i16>, <32 x i16>, i32)
|
|
34
|
|
35 define i64 @test_pcmpgt_b(<64 x i8> %a, <64 x i8> %b) {
|
|
36 ; CHECK-LABEL: test_pcmpgt_b
|
|
37 ; CHECK: vpcmpgtb %zmm1, %zmm0, %k0 ##
|
|
38 %res = call i64 @llvm.x86.avx512.mask.pcmpgt.b.512(<64 x i8> %a, <64 x i8> %b, i64 -1)
|
|
39 ret i64 %res
|
|
40 }
|
|
41
|
|
42 define i64 @test_mask_pcmpgt_b(<64 x i8> %a, <64 x i8> %b, i64 %mask) {
|
|
43 ; CHECK-LABEL: test_mask_pcmpgt_b
|
|
44 ; CHECK: vpcmpgtb %zmm1, %zmm0, %k0 {%k1} ##
|
|
45 %res = call i64 @llvm.x86.avx512.mask.pcmpgt.b.512(<64 x i8> %a, <64 x i8> %b, i64 %mask)
|
|
46 ret i64 %res
|
|
47 }
|
|
48
|
|
49 declare i64 @llvm.x86.avx512.mask.pcmpgt.b.512(<64 x i8>, <64 x i8>, i64)
|
|
50
|
|
51 define i32 @test_pcmpgt_w(<32 x i16> %a, <32 x i16> %b) {
|
|
52 ; CHECK-LABEL: test_pcmpgt_w
|
|
53 ; CHECK: vpcmpgtw %zmm1, %zmm0, %k0 ##
|
|
54 %res = call i32 @llvm.x86.avx512.mask.pcmpgt.w.512(<32 x i16> %a, <32 x i16> %b, i32 -1)
|
|
55 ret i32 %res
|
|
56 }
|
|
57
|
|
58 define i32 @test_mask_pcmpgt_w(<32 x i16> %a, <32 x i16> %b, i32 %mask) {
|
|
59 ; CHECK-LABEL: test_mask_pcmpgt_w
|
|
60 ; CHECK: vpcmpgtw %zmm1, %zmm0, %k0 {%k1} ##
|
|
61 %res = call i32 @llvm.x86.avx512.mask.pcmpgt.w.512(<32 x i16> %a, <32 x i16> %b, i32 %mask)
|
|
62 ret i32 %res
|
|
63 }
|
|
64
|
|
65 declare i32 @llvm.x86.avx512.mask.pcmpgt.w.512(<32 x i16>, <32 x i16>, i32)
|
|
66
|
|
67 define <8 x i64> @test_cmp_b_512(<64 x i8> %a0, <64 x i8> %a1) {
|
|
68 ; CHECK_LABEL: test_cmp_b_512
|
|
69 ; CHECK: vpcmpeqb %zmm1, %zmm0, %k0 ##
|
|
70 %res0 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i8 0, i64 -1)
|
|
71 %vec0 = insertelement <8 x i64> undef, i64 %res0, i32 0
|
|
72 ; CHECK: vpcmpltb %zmm1, %zmm0, %k0 ##
|
|
73 %res1 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i8 1, i64 -1)
|
|
74 %vec1 = insertelement <8 x i64> %vec0, i64 %res1, i32 1
|
|
75 ; CHECK: vpcmpleb %zmm1, %zmm0, %k0 ##
|
|
76 %res2 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i8 2, i64 -1)
|
|
77 %vec2 = insertelement <8 x i64> %vec1, i64 %res2, i32 2
|
|
78 ; CHECK: vpcmpunordb %zmm1, %zmm0, %k0 ##
|
|
79 %res3 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i8 3, i64 -1)
|
|
80 %vec3 = insertelement <8 x i64> %vec2, i64 %res3, i32 3
|
|
81 ; CHECK: vpcmpneqb %zmm1, %zmm0, %k0 ##
|
|
82 %res4 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i8 4, i64 -1)
|
|
83 %vec4 = insertelement <8 x i64> %vec3, i64 %res4, i32 4
|
|
84 ; CHECK: vpcmpnltb %zmm1, %zmm0, %k0 ##
|
|
85 %res5 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i8 5, i64 -1)
|
|
86 %vec5 = insertelement <8 x i64> %vec4, i64 %res5, i32 5
|
|
87 ; CHECK: vpcmpnleb %zmm1, %zmm0, %k0 ##
|
|
88 %res6 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i8 6, i64 -1)
|
|
89 %vec6 = insertelement <8 x i64> %vec5, i64 %res6, i32 6
|
|
90 ; CHECK: vpcmpordb %zmm1, %zmm0, %k0 ##
|
|
91 %res7 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i8 7, i64 -1)
|
|
92 %vec7 = insertelement <8 x i64> %vec6, i64 %res7, i32 7
|
|
93 ret <8 x i64> %vec7
|
|
94 }
|
|
95
|
|
96 define <8 x i64> @test_mask_cmp_b_512(<64 x i8> %a0, <64 x i8> %a1, i64 %mask) {
|
|
97 ; CHECK_LABEL: test_mask_cmp_b_512
|
|
98 ; CHECK: vpcmpeqb %zmm1, %zmm0, %k0 {%k1} ##
|
|
99 %res0 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i8 0, i64 %mask)
|
|
100 %vec0 = insertelement <8 x i64> undef, i64 %res0, i32 0
|
|
101 ; CHECK: vpcmpltb %zmm1, %zmm0, %k0 {%k1} ##
|
|
102 %res1 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i8 1, i64 %mask)
|
|
103 %vec1 = insertelement <8 x i64> %vec0, i64 %res1, i32 1
|
|
104 ; CHECK: vpcmpleb %zmm1, %zmm0, %k0 {%k1} ##
|
|
105 %res2 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i8 2, i64 %mask)
|
|
106 %vec2 = insertelement <8 x i64> %vec1, i64 %res2, i32 2
|
|
107 ; CHECK: vpcmpunordb %zmm1, %zmm0, %k0 {%k1} ##
|
|
108 %res3 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i8 3, i64 %mask)
|
|
109 %vec3 = insertelement <8 x i64> %vec2, i64 %res3, i32 3
|
|
110 ; CHECK: vpcmpneqb %zmm1, %zmm0, %k0 {%k1} ##
|
|
111 %res4 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i8 4, i64 %mask)
|
|
112 %vec4 = insertelement <8 x i64> %vec3, i64 %res4, i32 4
|
|
113 ; CHECK: vpcmpnltb %zmm1, %zmm0, %k0 {%k1} ##
|
|
114 %res5 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i8 5, i64 %mask)
|
|
115 %vec5 = insertelement <8 x i64> %vec4, i64 %res5, i32 5
|
|
116 ; CHECK: vpcmpnleb %zmm1, %zmm0, %k0 {%k1} ##
|
|
117 %res6 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i8 6, i64 %mask)
|
|
118 %vec6 = insertelement <8 x i64> %vec5, i64 %res6, i32 6
|
|
119 ; CHECK: vpcmpordb %zmm1, %zmm0, %k0 {%k1} ##
|
|
120 %res7 = call i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i8 7, i64 %mask)
|
|
121 %vec7 = insertelement <8 x i64> %vec6, i64 %res7, i32 7
|
|
122 ret <8 x i64> %vec7
|
|
123 }
|
|
124
|
|
125 declare i64 @llvm.x86.avx512.mask.cmp.b.512(<64 x i8>, <64 x i8>, i8, i64) nounwind readnone
|
|
126
|
|
127 define <8 x i64> @test_ucmp_b_512(<64 x i8> %a0, <64 x i8> %a1) {
|
|
128 ; CHECK_LABEL: test_ucmp_b_512
|
|
129 ; CHECK: vpcmpequb %zmm1, %zmm0, %k0 ##
|
|
130 %res0 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i8 0, i64 -1)
|
|
131 %vec0 = insertelement <8 x i64> undef, i64 %res0, i32 0
|
|
132 ; CHECK: vpcmpltub %zmm1, %zmm0, %k0 ##
|
|
133 %res1 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i8 1, i64 -1)
|
|
134 %vec1 = insertelement <8 x i64> %vec0, i64 %res1, i32 1
|
|
135 ; CHECK: vpcmpleub %zmm1, %zmm0, %k0 ##
|
|
136 %res2 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i8 2, i64 -1)
|
|
137 %vec2 = insertelement <8 x i64> %vec1, i64 %res2, i32 2
|
|
138 ; CHECK: vpcmpunordub %zmm1, %zmm0, %k0 ##
|
|
139 %res3 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i8 3, i64 -1)
|
|
140 %vec3 = insertelement <8 x i64> %vec2, i64 %res3, i32 3
|
|
141 ; CHECK: vpcmpnequb %zmm1, %zmm0, %k0 ##
|
|
142 %res4 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i8 4, i64 -1)
|
|
143 %vec4 = insertelement <8 x i64> %vec3, i64 %res4, i32 4
|
|
144 ; CHECK: vpcmpnltub %zmm1, %zmm0, %k0 ##
|
|
145 %res5 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i8 5, i64 -1)
|
|
146 %vec5 = insertelement <8 x i64> %vec4, i64 %res5, i32 5
|
|
147 ; CHECK: vpcmpnleub %zmm1, %zmm0, %k0 ##
|
|
148 %res6 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i8 6, i64 -1)
|
|
149 %vec6 = insertelement <8 x i64> %vec5, i64 %res6, i32 6
|
|
150 ; CHECK: vpcmpordub %zmm1, %zmm0, %k0 ##
|
|
151 %res7 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i8 7, i64 -1)
|
|
152 %vec7 = insertelement <8 x i64> %vec6, i64 %res7, i32 7
|
|
153 ret <8 x i64> %vec7
|
|
154 }
|
|
155
|
|
156 define <8 x i64> @test_mask_x86_avx512_ucmp_b_512(<64 x i8> %a0, <64 x i8> %a1, i64 %mask) {
|
|
157 ; CHECK_LABEL: test_mask_ucmp_b_512
|
|
158 ; CHECK: vpcmpequb %zmm1, %zmm0, %k0 {%k1} ##
|
|
159 %res0 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i8 0, i64 %mask)
|
|
160 %vec0 = insertelement <8 x i64> undef, i64 %res0, i32 0
|
|
161 ; CHECK: vpcmpltub %zmm1, %zmm0, %k0 {%k1} ##
|
|
162 %res1 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i8 1, i64 %mask)
|
|
163 %vec1 = insertelement <8 x i64> %vec0, i64 %res1, i32 1
|
|
164 ; CHECK: vpcmpleub %zmm1, %zmm0, %k0 {%k1} ##
|
|
165 %res2 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i8 2, i64 %mask)
|
|
166 %vec2 = insertelement <8 x i64> %vec1, i64 %res2, i32 2
|
|
167 ; CHECK: vpcmpunordub %zmm1, %zmm0, %k0 {%k1} ##
|
|
168 %res3 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i8 3, i64 %mask)
|
|
169 %vec3 = insertelement <8 x i64> %vec2, i64 %res3, i32 3
|
|
170 ; CHECK: vpcmpnequb %zmm1, %zmm0, %k0 {%k1} ##
|
|
171 %res4 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i8 4, i64 %mask)
|
|
172 %vec4 = insertelement <8 x i64> %vec3, i64 %res4, i32 4
|
|
173 ; CHECK: vpcmpnltub %zmm1, %zmm0, %k0 {%k1} ##
|
|
174 %res5 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i8 5, i64 %mask)
|
|
175 %vec5 = insertelement <8 x i64> %vec4, i64 %res5, i32 5
|
|
176 ; CHECK: vpcmpnleub %zmm1, %zmm0, %k0 {%k1} ##
|
|
177 %res6 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i8 6, i64 %mask)
|
|
178 %vec6 = insertelement <8 x i64> %vec5, i64 %res6, i32 6
|
|
179 ; CHECK: vpcmpordub %zmm1, %zmm0, %k0 {%k1} ##
|
|
180 %res7 = call i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8> %a0, <64 x i8> %a1, i8 7, i64 %mask)
|
|
181 %vec7 = insertelement <8 x i64> %vec6, i64 %res7, i32 7
|
|
182 ret <8 x i64> %vec7
|
|
183 }
|
|
184
|
|
185 declare i64 @llvm.x86.avx512.mask.ucmp.b.512(<64 x i8>, <64 x i8>, i8, i64) nounwind readnone
|
|
186
|
|
187 define <8 x i32> @test_cmp_w_512(<32 x i16> %a0, <32 x i16> %a1) {
|
|
188 ; CHECK_LABEL: test_cmp_w_512
|
|
189 ; CHECK: vpcmpeqw %zmm1, %zmm0, %k0 ##
|
|
190 %res0 = call i32 @llvm.x86.avx512.mask.cmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i8 0, i32 -1)
|
|
191 %vec0 = insertelement <8 x i32> undef, i32 %res0, i32 0
|
|
192 ; CHECK: vpcmpltw %zmm1, %zmm0, %k0 ##
|
|
193 %res1 = call i32 @llvm.x86.avx512.mask.cmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i8 1, i32 -1)
|
|
194 %vec1 = insertelement <8 x i32> %vec0, i32 %res1, i32 1
|
|
195 ; CHECK: vpcmplew %zmm1, %zmm0, %k0 ##
|
|
196 %res2 = call i32 @llvm.x86.avx512.mask.cmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i8 2, i32 -1)
|
|
197 %vec2 = insertelement <8 x i32> %vec1, i32 %res2, i32 2
|
|
198 ; CHECK: vpcmpunordw %zmm1, %zmm0, %k0 ##
|
|
199 %res3 = call i32 @llvm.x86.avx512.mask.cmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i8 3, i32 -1)
|
|
200 %vec3 = insertelement <8 x i32> %vec2, i32 %res3, i32 3
|
|
201 ; CHECK: vpcmpneqw %zmm1, %zmm0, %k0 ##
|
|
202 %res4 = call i32 @llvm.x86.avx512.mask.cmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i8 4, i32 -1)
|
|
203 %vec4 = insertelement <8 x i32> %vec3, i32 %res4, i32 4
|
|
204 ; CHECK: vpcmpnltw %zmm1, %zmm0, %k0 ##
|
|
205 %res5 = call i32 @llvm.x86.avx512.mask.cmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i8 5, i32 -1)
|
|
206 %vec5 = insertelement <8 x i32> %vec4, i32 %res5, i32 5
|
|
207 ; CHECK: vpcmpnlew %zmm1, %zmm0, %k0 ##
|
|
208 %res6 = call i32 @llvm.x86.avx512.mask.cmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i8 6, i32 -1)
|
|
209 %vec6 = insertelement <8 x i32> %vec5, i32 %res6, i32 6
|
|
210 ; CHECK: vpcmpordw %zmm1, %zmm0, %k0 ##
|
|
211 %res7 = call i32 @llvm.x86.avx512.mask.cmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i8 7, i32 -1)
|
|
212 %vec7 = insertelement <8 x i32> %vec6, i32 %res7, i32 7
|
|
213 ret <8 x i32> %vec7
|
|
214 }
|
|
215
|
|
216 define <8 x i32> @test_mask_cmp_w_512(<32 x i16> %a0, <32 x i16> %a1, i32 %mask) {
|
|
217 ; CHECK_LABEL: test_mask_cmp_w_512
|
|
218 ; CHECK: vpcmpeqw %zmm1, %zmm0, %k0 {%k1} ##
|
|
219 %res0 = call i32 @llvm.x86.avx512.mask.cmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i8 0, i32 %mask)
|
|
220 %vec0 = insertelement <8 x i32> undef, i32 %res0, i32 0
|
|
221 ; CHECK: vpcmpltw %zmm1, %zmm0, %k0 {%k1} ##
|
|
222 %res1 = call i32 @llvm.x86.avx512.mask.cmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i8 1, i32 %mask)
|
|
223 %vec1 = insertelement <8 x i32> %vec0, i32 %res1, i32 1
|
|
224 ; CHECK: vpcmplew %zmm1, %zmm0, %k0 {%k1} ##
|
|
225 %res2 = call i32 @llvm.x86.avx512.mask.cmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i8 2, i32 %mask)
|
|
226 %vec2 = insertelement <8 x i32> %vec1, i32 %res2, i32 2
|
|
227 ; CHECK: vpcmpunordw %zmm1, %zmm0, %k0 {%k1} ##
|
|
228 %res3 = call i32 @llvm.x86.avx512.mask.cmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i8 3, i32 %mask)
|
|
229 %vec3 = insertelement <8 x i32> %vec2, i32 %res3, i32 3
|
|
230 ; CHECK: vpcmpneqw %zmm1, %zmm0, %k0 {%k1} ##
|
|
231 %res4 = call i32 @llvm.x86.avx512.mask.cmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i8 4, i32 %mask)
|
|
232 %vec4 = insertelement <8 x i32> %vec3, i32 %res4, i32 4
|
|
233 ; CHECK: vpcmpnltw %zmm1, %zmm0, %k0 {%k1} ##
|
|
234 %res5 = call i32 @llvm.x86.avx512.mask.cmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i8 5, i32 %mask)
|
|
235 %vec5 = insertelement <8 x i32> %vec4, i32 %res5, i32 5
|
|
236 ; CHECK: vpcmpnlew %zmm1, %zmm0, %k0 {%k1} ##
|
|
237 %res6 = call i32 @llvm.x86.avx512.mask.cmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i8 6, i32 %mask)
|
|
238 %vec6 = insertelement <8 x i32> %vec5, i32 %res6, i32 6
|
|
239 ; CHECK: vpcmpordw %zmm1, %zmm0, %k0 {%k1} ##
|
|
240 %res7 = call i32 @llvm.x86.avx512.mask.cmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i8 7, i32 %mask)
|
|
241 %vec7 = insertelement <8 x i32> %vec6, i32 %res7, i32 7
|
|
242 ret <8 x i32> %vec7
|
|
243 }
|
|
244
|
|
245 declare i32 @llvm.x86.avx512.mask.cmp.w.512(<32 x i16>, <32 x i16>, i8, i32) nounwind readnone
|
|
246
|
|
247 define <8 x i32> @test_ucmp_w_512(<32 x i16> %a0, <32 x i16> %a1) {
|
|
248 ; CHECK_LABEL: test_ucmp_w_512
|
|
249 ; CHECK: vpcmpequw %zmm1, %zmm0, %k0 ##
|
|
250 %res0 = call i32 @llvm.x86.avx512.mask.ucmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i8 0, i32 -1)
|
|
251 %vec0 = insertelement <8 x i32> undef, i32 %res0, i32 0
|
|
252 ; CHECK: vpcmpltuw %zmm1, %zmm0, %k0 ##
|
|
253 %res1 = call i32 @llvm.x86.avx512.mask.ucmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i8 1, i32 -1)
|
|
254 %vec1 = insertelement <8 x i32> %vec0, i32 %res1, i32 1
|
|
255 ; CHECK: vpcmpleuw %zmm1, %zmm0, %k0 ##
|
|
256 %res2 = call i32 @llvm.x86.avx512.mask.ucmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i8 2, i32 -1)
|
|
257 %vec2 = insertelement <8 x i32> %vec1, i32 %res2, i32 2
|
|
258 ; CHECK: vpcmpunorduw %zmm1, %zmm0, %k0 ##
|
|
259 %res3 = call i32 @llvm.x86.avx512.mask.ucmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i8 3, i32 -1)
|
|
260 %vec3 = insertelement <8 x i32> %vec2, i32 %res3, i32 3
|
|
261 ; CHECK: vpcmpnequw %zmm1, %zmm0, %k0 ##
|
|
262 %res4 = call i32 @llvm.x86.avx512.mask.ucmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i8 4, i32 -1)
|
|
263 %vec4 = insertelement <8 x i32> %vec3, i32 %res4, i32 4
|
|
264 ; CHECK: vpcmpnltuw %zmm1, %zmm0, %k0 ##
|
|
265 %res5 = call i32 @llvm.x86.avx512.mask.ucmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i8 5, i32 -1)
|
|
266 %vec5 = insertelement <8 x i32> %vec4, i32 %res5, i32 5
|
|
267 ; CHECK: vpcmpnleuw %zmm1, %zmm0, %k0 ##
|
|
268 %res6 = call i32 @llvm.x86.avx512.mask.ucmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i8 6, i32 -1)
|
|
269 %vec6 = insertelement <8 x i32> %vec5, i32 %res6, i32 6
|
|
270 ; CHECK: vpcmporduw %zmm1, %zmm0, %k0 ##
|
|
271 %res7 = call i32 @llvm.x86.avx512.mask.ucmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i8 7, i32 -1)
|
|
272 %vec7 = insertelement <8 x i32> %vec6, i32 %res7, i32 7
|
|
273 ret <8 x i32> %vec7
|
|
274 }
|
|
275
|
|
276 define <8 x i32> @test_mask_ucmp_w_512(<32 x i16> %a0, <32 x i16> %a1, i32 %mask) {
|
|
277 ; CHECK_LABEL: test_mask_ucmp_w_512
|
|
278 ; CHECK: vpcmpequw %zmm1, %zmm0, %k0 {%k1} ##
|
|
279 %res0 = call i32 @llvm.x86.avx512.mask.ucmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i8 0, i32 %mask)
|
|
280 %vec0 = insertelement <8 x i32> undef, i32 %res0, i32 0
|
|
281 ; CHECK: vpcmpltuw %zmm1, %zmm0, %k0 {%k1} ##
|
|
282 %res1 = call i32 @llvm.x86.avx512.mask.ucmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i8 1, i32 %mask)
|
|
283 %vec1 = insertelement <8 x i32> %vec0, i32 %res1, i32 1
|
|
284 ; CHECK: vpcmpleuw %zmm1, %zmm0, %k0 {%k1} ##
|
|
285 %res2 = call i32 @llvm.x86.avx512.mask.ucmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i8 2, i32 %mask)
|
|
286 %vec2 = insertelement <8 x i32> %vec1, i32 %res2, i32 2
|
|
287 ; CHECK: vpcmpunorduw %zmm1, %zmm0, %k0 {%k1} ##
|
|
288 %res3 = call i32 @llvm.x86.avx512.mask.ucmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i8 3, i32 %mask)
|
|
289 %vec3 = insertelement <8 x i32> %vec2, i32 %res3, i32 3
|
|
290 ; CHECK: vpcmpnequw %zmm1, %zmm0, %k0 {%k1} ##
|
|
291 %res4 = call i32 @llvm.x86.avx512.mask.ucmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i8 4, i32 %mask)
|
|
292 %vec4 = insertelement <8 x i32> %vec3, i32 %res4, i32 4
|
|
293 ; CHECK: vpcmpnltuw %zmm1, %zmm0, %k0 {%k1} ##
|
|
294 %res5 = call i32 @llvm.x86.avx512.mask.ucmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i8 5, i32 %mask)
|
|
295 %vec5 = insertelement <8 x i32> %vec4, i32 %res5, i32 5
|
|
296 ; CHECK: vpcmpnleuw %zmm1, %zmm0, %k0 {%k1} ##
|
|
297 %res6 = call i32 @llvm.x86.avx512.mask.ucmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i8 6, i32 %mask)
|
|
298 %vec6 = insertelement <8 x i32> %vec5, i32 %res6, i32 6
|
|
299 ; CHECK: vpcmporduw %zmm1, %zmm0, %k0 {%k1} ##
|
|
300 %res7 = call i32 @llvm.x86.avx512.mask.ucmp.w.512(<32 x i16> %a0, <32 x i16> %a1, i8 7, i32 %mask)
|
|
301 %vec7 = insertelement <8 x i32> %vec6, i32 %res7, i32 7
|
|
302 ret <8 x i32> %vec7
|
|
303 }
|
|
304
|
|
305 declare i32 @llvm.x86.avx512.mask.ucmp.w.512(<32 x i16>, <32 x i16>, i8, i32) nounwind readnone
|
|
306
|
|
307 ; CHECK-LABEL: test_x86_mask_blend_b_256
|
|
308 ; CHECK: vpblendmb
|
|
309 define <32 x i8> @test_x86_mask_blend_b_256(i32 %a0, <32 x i8> %a1, <32 x i8> %a2) {
|
|
310 %res = call <32 x i8> @llvm.x86.avx512.mask.blend.b.256(<32 x i8> %a1, <32 x i8> %a2, i32 %a0) ; <<32 x i8>> [#uses=1]
|
|
311 ret <32 x i8> %res
|
|
312 }
|
|
313 declare <32 x i8> @llvm.x86.avx512.mask.blend.b.256(<32 x i8>, <32 x i8>, i32) nounwind readonly
|
|
314
|
|
315 ; CHECK-LABEL: test_x86_mask_blend_w_256
|
|
316 define <16 x i16> @test_x86_mask_blend_w_256(i16 %mask, <16 x i16> %a1, <16 x i16> %a2) {
|
|
317 ; CHECK: vpblendmw
|
|
318 %res = call <16 x i16> @llvm.x86.avx512.mask.blend.w.256(<16 x i16> %a1, <16 x i16> %a2, i16 %mask) ; <<16 x i16>> [#uses=1]
|
|
319 ret <16 x i16> %res
|
|
320 }
|
|
321 declare <16 x i16> @llvm.x86.avx512.mask.blend.w.256(<16 x i16>, <16 x i16>, i16) nounwind readonly
|
|
322
|
|
323 ; CHECK-LABEL: test_x86_mask_blend_b_512
|
|
324 ; CHECK: vpblendmb
|
|
325 define <64 x i8> @test_x86_mask_blend_b_512(i64 %a0, <64 x i8> %a1, <64 x i8> %a2) {
|
|
326 %res = call <64 x i8> @llvm.x86.avx512.mask.blend.b.512(<64 x i8> %a1, <64 x i8> %a2, i64 %a0) ; <<64 x i8>> [#uses=1]
|
|
327 ret <64 x i8> %res
|
|
328 }
|
|
329 declare <64 x i8> @llvm.x86.avx512.mask.blend.b.512(<64 x i8>, <64 x i8>, i64) nounwind readonly
|
|
330
|
|
331 ; CHECK-LABEL: test_x86_mask_blend_w_512
|
|
332 define <32 x i16> @test_x86_mask_blend_w_512(i32 %mask, <32 x i16> %a1, <32 x i16> %a2) {
|
|
333 ; CHECK: vpblendmw
|
|
334 %res = call <32 x i16> @llvm.x86.avx512.mask.blend.w.512(<32 x i16> %a1, <32 x i16> %a2, i32 %mask) ; <<32 x i16>> [#uses=1]
|
|
335 ret <32 x i16> %res
|
|
336 }
|
|
337 declare <32 x i16> @llvm.x86.avx512.mask.blend.w.512(<32 x i16>, <32 x i16>, i32) nounwind readonly
|
|
338
|
|
339 ; CHECK-LABEL: test_x86_mask_blend_b_128
|
|
340 ; CHECK: vpblendmb
|
|
341 define <16 x i8> @test_x86_mask_blend_b_128(i16 %a0, <16 x i8> %a1, <16 x i8> %a2) {
|
|
342 %res = call <16 x i8> @llvm.x86.avx512.mask.blend.b.128(<16 x i8> %a1, <16 x i8> %a2, i16 %a0) ; <<16 x i8>> [#uses=1]
|
|
343 ret <16 x i8> %res
|
|
344 }
|
|
345 declare <16 x i8> @llvm.x86.avx512.mask.blend.b.128(<16 x i8>, <16 x i8>, i16) nounwind readonly
|
|
346
|
|
347 ; CHECK-LABEL: test_x86_mask_blend_w_128
|
|
348 define <8 x i16> @test_x86_mask_blend_w_128(i8 %mask, <8 x i16> %a1, <8 x i16> %a2) {
|
|
349 ; CHECK: vpblendmw
|
|
350 %res = call <8 x i16> @llvm.x86.avx512.mask.blend.w.128(<8 x i16> %a1, <8 x i16> %a2, i8 %mask) ; <<8 x i16>> [#uses=1]
|
|
351 ret <8 x i16> %res
|
|
352 }
|
|
353 declare <8 x i16> @llvm.x86.avx512.mask.blend.w.128(<8 x i16>, <8 x i16>, i8) nounwind readonly
|