xref: /llvm-project/llvm/test/CodeGen/X86/avx512bf16-vl-intrinsics-upgrade.ll (revision bc1819389fb4701cdeba5e093278e32dd668d6d5) 
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx512bf16 -mattr=+avx512vl --show-mc-encoding | FileCheck %s --check-prefixes=CHECK,X86
3; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bf16 -mattr=+avx512vl --show-mc-encoding | FileCheck %s --check-prefixes=CHECK,X64
4
5declare <8 x i16> @llvm.x86.avx512bf16.cvtne2ps2bf16.128(<4 x float>, <4 x float>) #1
6
7define <2 x i64> @test_mm_cvtne2ps2bf16_128(<4 x float> %A, <4 x float> %B) local_unnamed_addr #0 {
8; CHECK-LABEL: test_mm_cvtne2ps2bf16_128:
9; CHECK:       # %bb.0: # %entry
10; CHECK-NEXT:    vcvtne2ps2bf16 %xmm1, %xmm0, %xmm0 # encoding: [0x62,0xf2,0x7f,0x08,0x72,0xc1]
11; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
12entry:
13  %0 = tail call <8 x i16> @llvm.x86.avx512bf16.cvtne2ps2bf16.128(<4 x float> %A, <4 x float> %B) #2
14  %1 = bitcast <8 x i16> %0 to <2 x i64>
15  ret <2 x i64> %1
16}
17
18define <2 x i64> @test_mm_maskz_cvtne2ps2bf16_128(<4 x float> %A, <4 x float> %B, i8 zeroext %U) local_unnamed_addr #0 {
19; X86-LABEL: test_mm_maskz_cvtne2ps2bf16_128:
20; X86:       # %bb.0: # %entry
21; X86-NEXT:    vcvtne2ps2bf16 %xmm1, %xmm0, %xmm0 # encoding: [0x62,0xf2,0x7f,0x08,0x72,0xc1]
22; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
23; X86-NEXT:    kmovd %eax, %k1 # encoding: [0xc5,0xfb,0x92,0xc8]
24; X86-NEXT:    vmovdqu16 %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0xff,0x89,0x6f,0xc0]
25; X86-NEXT:    retl # encoding: [0xc3]
26;
27; X64-LABEL: test_mm_maskz_cvtne2ps2bf16_128:
28; X64:       # %bb.0: # %entry
29; X64-NEXT:    vcvtne2ps2bf16 %xmm1, %xmm0, %xmm0 # encoding: [0x62,0xf2,0x7f,0x08,0x72,0xc1]
30; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
31; X64-NEXT:    vmovdqu16 %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0xff,0x89,0x6f,0xc0]
32; X64-NEXT:    retq # encoding: [0xc3]
33entry:
34  %0 = tail call <8 x i16> @llvm.x86.avx512bf16.cvtne2ps2bf16.128(<4 x float> %A, <4 x float> %B) #2
35  %1 = bitcast i8 %U to <8 x i1>
36  %2 = select <8 x i1> %1, <8 x i16> %0, <8 x i16> zeroinitializer
37  %3 = bitcast <8 x i16> %2 to <2 x i64>
38  ret <2 x i64> %3
39}
40
41define <2 x i64> @test_mm_mask_cvtne2ps2bf16_128(<2 x i64> %C, i8 zeroext %U, <4 x float> %A, <4 x float> %B) local_unnamed_addr #0 {
42; X86-LABEL: test_mm_mask_cvtne2ps2bf16_128:
43; X86:       # %bb.0: # %entry
44; X86-NEXT:    vcvtne2ps2bf16 %xmm2, %xmm1, %xmm1 # encoding: [0x62,0xf2,0x77,0x08,0x72,0xca]
45; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
46; X86-NEXT:    kmovd %eax, %k1 # encoding: [0xc5,0xfb,0x92,0xc8]
47; X86-NEXT:    vmovdqu16 %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf1,0xff,0x09,0x6f,0xc1]
48; X86-NEXT:    retl # encoding: [0xc3]
49;
50; X64-LABEL: test_mm_mask_cvtne2ps2bf16_128:
51; X64:       # %bb.0: # %entry
52; X64-NEXT:    vcvtne2ps2bf16 %xmm2, %xmm1, %xmm1 # encoding: [0x62,0xf2,0x77,0x08,0x72,0xca]
53; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
54; X64-NEXT:    vmovdqu16 %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf1,0xff,0x09,0x6f,0xc1]
55; X64-NEXT:    retq # encoding: [0xc3]
56entry:
57  %0 = tail call <8 x i16> @llvm.x86.avx512bf16.cvtne2ps2bf16.128(<4 x float> %A, <4 x float> %B) #2
58  %1 = bitcast <2 x i64> %C to <8 x i16>
59  %2 = bitcast i8 %U to <8 x i1>
60  %3 = select <8 x i1> %2, <8 x i16> %0, <8 x i16> %1
61  %4 = bitcast <8 x i16> %3 to <2 x i64>
62  ret <2 x i64> %4
63}
64
65declare <16 x i16> @llvm.x86.avx512bf16.cvtne2ps2bf16.256(<8 x float>, <8 x float>) #3
66
67define <4 x i64> @test_mm256_cvtne2ps2bf16_256(<8 x float> %A, <8 x float> %B) local_unnamed_addr #1 {
68; CHECK-LABEL: test_mm256_cvtne2ps2bf16_256:
69; CHECK:       # %bb.0: # %entry
70; CHECK-NEXT:    vcvtne2ps2bf16 %ymm1, %ymm0, %ymm0 # encoding: [0x62,0xf2,0x7f,0x28,0x72,0xc1]
71; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
72entry:
73  %0 = tail call <16 x i16> @llvm.x86.avx512bf16.cvtne2ps2bf16.256(<8 x float> %A, <8 x float> %B) #4
74  %1 = bitcast <16 x i16> %0 to <4 x i64>
75  ret <4 x i64> %1
76}
77
78define <4 x i64> @test_mm256_maskz_cvtne2ps2bf16_256(<8 x float> %A, <8 x float> %B, i16 zeroext %U) local_unnamed_addr #1 {
79; X86-LABEL: test_mm256_maskz_cvtne2ps2bf16_256:
80; X86:       # %bb.0: # %entry
81; X86-NEXT:    vcvtne2ps2bf16 %ymm1, %ymm0, %ymm0 # encoding: [0x62,0xf2,0x7f,0x28,0x72,0xc1]
82; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
83; X86-NEXT:    vmovdqu16 %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf1,0xff,0xa9,0x6f,0xc0]
84; X86-NEXT:    retl # encoding: [0xc3]
85;
86; X64-LABEL: test_mm256_maskz_cvtne2ps2bf16_256:
87; X64:       # %bb.0: # %entry
88; X64-NEXT:    vcvtne2ps2bf16 %ymm1, %ymm0, %ymm0 # encoding: [0x62,0xf2,0x7f,0x28,0x72,0xc1]
89; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
90; X64-NEXT:    vmovdqu16 %ymm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf1,0xff,0xa9,0x6f,0xc0]
91; X64-NEXT:    retq # encoding: [0xc3]
92entry:
93  %0 = tail call <16 x i16> @llvm.x86.avx512bf16.cvtne2ps2bf16.256(<8 x float> %A, <8 x float> %B) #4
94  %1 = bitcast i16 %U to <16 x i1>
95  %2 = select <16 x i1> %1, <16 x i16> %0, <16 x i16> zeroinitializer
96  %3 = bitcast <16 x i16> %2 to <4 x i64>
97  ret <4 x i64> %3
98}
99
100define <4 x i64> @test_mm256_mask_cvtne2ps2bf16_256(<4 x i64> %C, i16 zeroext %U, <8 x float> %A, <8 x float> %B) local_unnamed_addr #1 {
101; X86-LABEL: test_mm256_mask_cvtne2ps2bf16_256:
102; X86:       # %bb.0: # %entry
103; X86-NEXT:    vcvtne2ps2bf16 %ymm2, %ymm1, %ymm1 # encoding: [0x62,0xf2,0x77,0x28,0x72,0xca]
104; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
105; X86-NEXT:    vmovdqu16 %ymm1, %ymm0 {%k1} # encoding: [0x62,0xf1,0xff,0x29,0x6f,0xc1]
106; X86-NEXT:    retl # encoding: [0xc3]
107;
108; X64-LABEL: test_mm256_mask_cvtne2ps2bf16_256:
109; X64:       # %bb.0: # %entry
110; X64-NEXT:    vcvtne2ps2bf16 %ymm2, %ymm1, %ymm1 # encoding: [0x62,0xf2,0x77,0x28,0x72,0xca]
111; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
112; X64-NEXT:    vmovdqu16 %ymm1, %ymm0 {%k1} # encoding: [0x62,0xf1,0xff,0x29,0x6f,0xc1]
113; X64-NEXT:    retq # encoding: [0xc3]
114entry:
115  %0 = tail call <16 x i16> @llvm.x86.avx512bf16.cvtne2ps2bf16.256(<8 x float> %A, <8 x float> %B) #4
116  %1 = bitcast <4 x i64> %C to <16 x i16>
117  %2 = bitcast i16 %U to <16 x i1>
118  %3 = select <16 x i1> %2, <16 x i16> %0, <16 x i16> %1
119  %4 = bitcast <16 x i16> %3 to <4 x i64>
120  ret <4 x i64> %4
121}
122
123declare <8 x i16> @llvm.x86.avx512bf16.cvtneps2bf16.256(<8 x float>) #3
124
125define <2 x i64> @test_mm256_cvtneps2bf16_256(<8 x float> %A) local_unnamed_addr #2 {
126; CHECK-LABEL: test_mm256_cvtneps2bf16_256:
127; CHECK:       # %bb.0: # %entry
128; CHECK-NEXT:    vcvtneps2bf16 %ymm0, %xmm0 # encoding: [0x62,0xf2,0x7e,0x28,0x72,0xc0]
129; CHECK-NEXT:    vzeroupper # encoding: [0xc5,0xf8,0x77]
130; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
131entry:
132  %0 = tail call <8 x i16> @llvm.x86.avx512bf16.cvtneps2bf16.256(<8 x float> %A) #4
133  %1 = bitcast <8 x i16> %0 to <2 x i64>
134  ret <2 x i64> %1
135}
136
137define <2 x i64> @test_mm256_maskz_cvtneps2bf16_256(<8 x float> %A, i8 zeroext %U) local_unnamed_addr #2 {
138; X86-LABEL: test_mm256_maskz_cvtneps2bf16_256:
139; X86:       # %bb.0: # %entry
140; X86-NEXT:    vcvtneps2bf16 %ymm0, %xmm0 # encoding: [0x62,0xf2,0x7e,0x28,0x72,0xc0]
141; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
142; X86-NEXT:    kmovd %eax, %k1 # encoding: [0xc5,0xfb,0x92,0xc8]
143; X86-NEXT:    vmovdqu16 %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0xff,0x89,0x6f,0xc0]
144; X86-NEXT:    vzeroupper # encoding: [0xc5,0xf8,0x77]
145; X86-NEXT:    retl # encoding: [0xc3]
146;
147; X64-LABEL: test_mm256_maskz_cvtneps2bf16_256:
148; X64:       # %bb.0: # %entry
149; X64-NEXT:    vcvtneps2bf16 %ymm0, %xmm0 # encoding: [0x62,0xf2,0x7e,0x28,0x72,0xc0]
150; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
151; X64-NEXT:    vmovdqu16 %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf1,0xff,0x89,0x6f,0xc0]
152; X64-NEXT:    vzeroupper # encoding: [0xc5,0xf8,0x77]
153; X64-NEXT:    retq # encoding: [0xc3]
154entry:
155  %0 = tail call <8 x i16> @llvm.x86.avx512bf16.cvtneps2bf16.256(<8 x float> %A) #4
156  %1 = bitcast i8 %U to <8 x i1>
157  %2 = select <8 x i1> %1, <8 x i16> %0, <8 x i16> zeroinitializer
158  %3 = bitcast <8 x i16> %2 to <2 x i64>
159  ret <2 x i64> %3
160}
161
162define <2 x i64> @test_mm256_mask_cvtneps2bf16_256(<2 x i64> %C, i8 zeroext %U, <8 x float> %A) local_unnamed_addr #2 {
163; X86-LABEL: test_mm256_mask_cvtneps2bf16_256:
164; X86:       # %bb.0: # %entry
165; X86-NEXT:    vcvtneps2bf16 %ymm1, %xmm1 # encoding: [0x62,0xf2,0x7e,0x28,0x72,0xc9]
166; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
167; X86-NEXT:    kmovd %eax, %k1 # encoding: [0xc5,0xfb,0x92,0xc8]
168; X86-NEXT:    vmovdqu16 %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf1,0xff,0x09,0x6f,0xc1]
169; X86-NEXT:    vzeroupper # encoding: [0xc5,0xf8,0x77]
170; X86-NEXT:    retl # encoding: [0xc3]
171;
172; X64-LABEL: test_mm256_mask_cvtneps2bf16_256:
173; X64:       # %bb.0: # %entry
174; X64-NEXT:    vcvtneps2bf16 %ymm1, %xmm1 # encoding: [0x62,0xf2,0x7e,0x28,0x72,0xc9]
175; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
176; X64-NEXT:    vmovdqu16 %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf1,0xff,0x09,0x6f,0xc1]
177; X64-NEXT:    vzeroupper # encoding: [0xc5,0xf8,0x77]
178; X64-NEXT:    retq # encoding: [0xc3]
179entry:
180  %0 = tail call <8 x i16> @llvm.x86.avx512bf16.cvtneps2bf16.256(<8 x float> %A) #4
181  %1 = bitcast <2 x i64> %C to <8 x i16>
182  %2 = bitcast i8 %U to <8 x i1>
183  %3 = select <8 x i1> %2, <8 x i16> %0, <8 x i16> %1
184  %4 = bitcast <8 x i16> %3 to <2 x i64>
185  ret <2 x i64> %4
186}
187
188declare <8 x i16> @llvm.x86.avx512bf16.mask.cvtneps2bf16.128(<4 x float>, <8 x i16>, <4 x i1>) #3
189
190define <2 x i64> @test_mm128_cvtneps2bf16_128(<4 x float> %A) local_unnamed_addr #2 {
191; CHECK-LABEL: test_mm128_cvtneps2bf16_128:
192; CHECK:       # %bb.0: # %entry
193; CHECK-NEXT:    vcvtneps2bf16 %xmm0, %xmm0 # encoding: [0x62,0xf2,0x7e,0x08,0x72,0xc0]
194; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
195entry:
196  %0 = tail call <8 x i16> @llvm.x86.avx512bf16.mask.cvtneps2bf16.128(<4 x float> %A, <8 x i16> undef, <4 x i1> <i1 true, i1 true, i1 true, i1 true>) #4
197  %1 = bitcast <8 x i16> %0 to <2 x i64>
198  ret <2 x i64> %1
199}
200
201define <2 x i64> @test_mm128_maskz_cvtneps2bf16_128(<4 x float> %A, i8 zeroext %U) local_unnamed_addr #2 {
202; X86-LABEL: test_mm128_maskz_cvtneps2bf16_128:
203; X86:       # %bb.0: # %entry
204; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
205; X86-NEXT:    kmovd %eax, %k1 # encoding: [0xc5,0xfb,0x92,0xc8]
206; X86-NEXT:    vcvtneps2bf16 %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0x7e,0x89,0x72,0xc0]
207; X86-NEXT:    retl # encoding: [0xc3]
208;
209; X64-LABEL: test_mm128_maskz_cvtneps2bf16_128:
210; X64:       # %bb.0: # %entry
211; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
212; X64-NEXT:    vcvtneps2bf16 %xmm0, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0x7e,0x89,0x72,0xc0]
213; X64-NEXT:    retq # encoding: [0xc3]
214entry:
215  %0 = bitcast i8 %U to <8 x i1>
216  %1 = shufflevector <8 x i1> %0, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
217  %2 = tail call <8 x i16> @llvm.x86.avx512bf16.mask.cvtneps2bf16.128(<4 x float> %A, <8 x i16> zeroinitializer, <4 x i1> %1) #4
218  %3 = bitcast <8 x i16> %2 to <2 x i64>
219  ret <2 x i64> %3
220}
221
222define <2 x i64> @test_mm128_mask_cvtneps2bf16_128(<2 x i64> %C, i8 zeroext %U, <4 x float> %A) local_unnamed_addr #2 {
223; X86-LABEL: test_mm128_mask_cvtneps2bf16_128:
224; X86:       # %bb.0: # %entry
225; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
226; X86-NEXT:    kmovd %eax, %k1 # encoding: [0xc5,0xfb,0x92,0xc8]
227; X86-NEXT:    vcvtneps2bf16 %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf2,0x7e,0x09,0x72,0xc1]
228; X86-NEXT:    retl # encoding: [0xc3]
229;
230; X64-LABEL: test_mm128_mask_cvtneps2bf16_128:
231; X64:       # %bb.0: # %entry
232; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
233; X64-NEXT:    vcvtneps2bf16 %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf2,0x7e,0x09,0x72,0xc1]
234; X64-NEXT:    retq # encoding: [0xc3]
235entry:
236  %0 = bitcast i8 %U to <8 x i1>
237  %1 = shufflevector <8 x i1> %0, <8 x i1> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
238  %2 = bitcast <2 x i64> %C to <8 x i16>
239  %3 = tail call <8 x i16> @llvm.x86.avx512bf16.mask.cvtneps2bf16.128(<4 x float> %A, <8 x i16> %2, <4 x i1> %1) #4
240  %4 = bitcast <8 x i16> %3 to <2 x i64>
241  ret <2 x i64> %4
242}
243
244; Make sure we don't fold a select into the 128 bit form of cvtneps2bf16. It
245; always writes zeros to bits 127:64 regardless of mask.
246define <2 x i64> @test_mm128_cvtneps2bf16_128_select(<2 x i64> %C, i8 zeroext %U, <4 x float> %A) local_unnamed_addr #2 {
247; X86-LABEL: test_mm128_cvtneps2bf16_128_select:
248; X86:       # %bb.0: # %entry
249; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
250; X86-NEXT:    kmovd %eax, %k1 # encoding: [0xc5,0xfb,0x92,0xc8]
251; X86-NEXT:    vcvtneps2bf16 %xmm1, %xmm1 # encoding: [0x62,0xf2,0x7e,0x08,0x72,0xc9]
252; X86-NEXT:    vmovdqu16 %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf1,0xff,0x09,0x6f,0xc1]
253; X86-NEXT:    retl # encoding: [0xc3]
254;
255; X64-LABEL: test_mm128_cvtneps2bf16_128_select:
256; X64:       # %bb.0: # %entry
257; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
258; X64-NEXT:    vcvtneps2bf16 %xmm1, %xmm1 # encoding: [0x62,0xf2,0x7e,0x08,0x72,0xc9]
259; X64-NEXT:    vmovdqu16 %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf1,0xff,0x09,0x6f,0xc1]
260; X64-NEXT:    retq # encoding: [0xc3]
261entry:
262  %0 = bitcast i8 %U to <8 x i1>
263  %1 = bitcast <2 x i64> %C to <8 x i16>
264  %2 = tail call <8 x i16> @llvm.x86.avx512bf16.mask.cvtneps2bf16.128(<4 x float> %A, <8 x i16> undef, <4 x i1> <i1 true, i1 true, i1 true, i1 true>) #4
265  %3 = select <8 x i1> %0, <8 x i16> %2, <8 x i16> %1
266  %4 = bitcast <8 x i16> %3 to <2 x i64>
267  ret <2 x i64> %4
268}
269
270declare <8 x float> @llvm.x86.avx512bf16.dpbf16ps.256(<8 x float>, <8 x i32>, <8 x i32>) #3
271
272define <8 x float> @test_mm256_dpbf16ps_256(<8 x float> %E, <8 x i32> %A, <8 x i32> %B) local_unnamed_addr #2 {
273; CHECK-LABEL: test_mm256_dpbf16ps_256:
274; CHECK:       # %bb.0: # %entry
275; CHECK-NEXT:    vdpbf16ps %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf2,0x76,0x28,0x52,0xc2]
276; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
277entry:
278  %0 = tail call <8 x float> @llvm.x86.avx512bf16.dpbf16ps.256(<8 x float> %E, <8 x i32> %A, <8 x i32> %B) #4
279  ret <8 x float> %0
280}
281
282define <8 x float> @test_mm256_maskz_dpbf16ps_256(<8 x float> %E, <8 x i32> %A, <8 x i32> %B, i8 zeroext %U) local_unnamed_addr #2 {
283; X86-LABEL: test_mm256_maskz_dpbf16ps_256:
284; X86:       # %bb.0: # %entry
285; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
286; X86-NEXT:    kmovd %eax, %k1 # encoding: [0xc5,0xfb,0x92,0xc8]
287; X86-NEXT:    vdpbf16ps %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0x76,0xa9,0x52,0xc2]
288; X86-NEXT:    retl # encoding: [0xc3]
289;
290; X64-LABEL: test_mm256_maskz_dpbf16ps_256:
291; X64:       # %bb.0: # %entry
292; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
293; X64-NEXT:    vdpbf16ps %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0x76,0xa9,0x52,0xc2]
294; X64-NEXT:    retq # encoding: [0xc3]
295entry:
296  %0 = tail call <8 x float> @llvm.x86.avx512bf16.dpbf16ps.256(<8 x float> %E, <8 x i32> %A, <8 x i32> %B) #4
297  %1 = bitcast i8 %U to <8 x i1>
298  %2 = select <8 x i1> %1, <8 x float> %0, <8 x float> zeroinitializer
299  ret <8 x float> %2
300}
301define <8 x float> @test_mm256_mask_dpbf16ps_256(i8 zeroext %U, <8 x float> %E, <8 x i32> %A, <8 x i32> %B) local_unnamed_addr #2 {
302; X86-LABEL: test_mm256_mask_dpbf16ps_256:
303; X86:       # %bb.0: # %entry
304; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
305; X86-NEXT:    kmovd %eax, %k1 # encoding: [0xc5,0xfb,0x92,0xc8]
306; X86-NEXT:    vdpbf16ps %ymm2, %ymm1, %ymm0 {%k1} # encoding: [0x62,0xf2,0x76,0x29,0x52,0xc2]
307; X86-NEXT:    retl # encoding: [0xc3]
308;
309; X64-LABEL: test_mm256_mask_dpbf16ps_256:
310; X64:       # %bb.0: # %entry
311; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
312; X64-NEXT:    vdpbf16ps %ymm2, %ymm1, %ymm0 {%k1} # encoding: [0x62,0xf2,0x76,0x29,0x52,0xc2]
313; X64-NEXT:    retq # encoding: [0xc3]
314entry:
315  %0 = tail call <8 x float> @llvm.x86.avx512bf16.dpbf16ps.256(<8 x float> %E, <8 x i32> %A, <8 x i32> %B) #4
316  %1 = bitcast i8 %U to <8 x i1>
317  %2 = select <8 x i1> %1, <8 x float> %0, <8 x float> %E
318  ret <8 x float> %2
319}
320
321declare <4 x float> @llvm.x86.avx512bf16.dpbf16ps.128(<4 x float>, <4 x i32>, <4 x i32>) #3
322
323define <4 x float> @test_mm128_dpbf16ps_128(<4 x float> %E, <4 x i32> %A, <4 x i32> %B) local_unnamed_addr #2 {
324; CHECK-LABEL: test_mm128_dpbf16ps_128:
325; CHECK:       # %bb.0: # %entry
326; CHECK-NEXT:    vdpbf16ps %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf2,0x76,0x08,0x52,0xc2]
327; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
328entry:
329  %0 = tail call <4 x float> @llvm.x86.avx512bf16.dpbf16ps.128(<4 x float> %E, <4 x i32> %A, <4x i32> %B) #4
330  ret <4 x float> %0
331}
332
333define <4 x float> @test_mm128_maskz_dpbf16ps_128(<4 x float> %E, <4 x i32> %A, <4 x i32> %B, i4 zeroext %U) local_unnamed_addr #2 {
334; X86-LABEL: test_mm128_maskz_dpbf16ps_128:
335; X86:       # %bb.0: # %entry
336; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
337; X86-NEXT:    kmovd %eax, %k1 # encoding: [0xc5,0xfb,0x92,0xc8]
338; X86-NEXT:    vdpbf16ps %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0x76,0x89,0x52,0xc2]
339; X86-NEXT:    retl # encoding: [0xc3]
340;
341; X64-LABEL: test_mm128_maskz_dpbf16ps_128:
342; X64:       # %bb.0: # %entry
343; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
344; X64-NEXT:    vdpbf16ps %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0x76,0x89,0x52,0xc2]
345; X64-NEXT:    retq # encoding: [0xc3]
346entry:
347  %0 = tail call <4 x float> @llvm.x86.avx512bf16.dpbf16ps.128(<4 x float> %E, <4 x i32> %A, <4 x i32> %B) #4
348  %1 = bitcast i4 %U to <4 x i1>
349  %2 = select <4 x i1> %1, <4 x float> %0, <4 x float> zeroinitializer
350  ret <4 x float> %2
351}
352define <4 x float> @test_mm128_mask_dpbf16ps_128(i4 zeroext %U, <4 x float> %E, <4 x i32> %A, <4 x i32> %B) local_unnamed_addr #2 {
353; X86-LABEL: test_mm128_mask_dpbf16ps_128:
354; X86:       # %bb.0: # %entry
355; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
356; X86-NEXT:    kmovd %eax, %k1 # encoding: [0xc5,0xfb,0x92,0xc8]
357; X86-NEXT:    vdpbf16ps %xmm2, %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf2,0x76,0x09,0x52,0xc2]
358; X86-NEXT:    retl # encoding: [0xc3]
359;
360; X64-LABEL: test_mm128_mask_dpbf16ps_128:
361; X64:       # %bb.0: # %entry
362; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
363; X64-NEXT:    vdpbf16ps %xmm2, %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf2,0x76,0x09,0x52,0xc2]
364; X64-NEXT:    retq # encoding: [0xc3]
365entry:
366  %0 = tail call <4 x float> @llvm.x86.avx512bf16.dpbf16ps.128(<4 x float> %E, <4 x i32> %A, <4 x i32> %B) #4
367  %1 = bitcast i4 %U to <4 x i1>
368  %2 = select <4 x i1> %1, <4 x float> %0, <4 x float> %E
369  ret <4 x float> %2
370}
371