llvm.org GIT mirror llvm / 60b616c
[X86] Add AVX512VL command line to f16c intrinsic test to show missed EVEX opportunities for the legacy intrinsics. git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@317543 91177308-0d34-0410-b5e6-96231b3b80d8 Craig Topper 3 years ago
1 changed file(s) with 186 addition(s) and 57 deletion(s). Raw diff Collapse all Expand all
0 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
1 ; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx,+f16c | FileCheck %s --check-prefix=X32
2 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx,+f16c | FileCheck %s --check-prefix=X64
1 ; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx,+f16c -show-mc-encoding | FileCheck %s --check-prefix=X32
2 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx,+f16c -show-mc-encoding | FileCheck %s --check-prefix=X64
3 ; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx512vl -show-mc-encoding | FileCheck %s --check-prefix=X32-AVX512VL
4 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl -show-mc-encoding | FileCheck %s --check-prefix=X64-AVX512VL
35
46 define <4 x float> @test_x86_vcvtph2ps_128(<8 x i16> %a0) {
57 ; X32-LABEL: test_x86_vcvtph2ps_128:
68 ; X32: # BB#0:
7 ; X32-NEXT: vcvtph2ps %xmm0, %xmm0
8 ; X32-NEXT: retl
9 ; X32-NEXT: vcvtph2ps %xmm0, %xmm0 # encoding: [0xc4,0xe2,0x79,0x13,0xc0]
10 ; X32-NEXT: retl # encoding: [0xc3]
911 ;
1012 ; X64-LABEL: test_x86_vcvtph2ps_128:
1113 ; X64: # BB#0:
12 ; X64-NEXT: vcvtph2ps %xmm0, %xmm0
13 ; X64-NEXT: retq
14 ; X64-NEXT: vcvtph2ps %xmm0, %xmm0 # encoding: [0xc4,0xe2,0x79,0x13,0xc0]
15 ; X64-NEXT: retq # encoding: [0xc3]
16 ;
17 ; X32-AVX512VL-LABEL: test_x86_vcvtph2ps_128:
18 ; X32-AVX512VL: # BB#0:
19 ; X32-AVX512VL-NEXT: vcvtph2ps %xmm0, %xmm0 # encoding: [0xc4,0xe2,0x79,0x13,0xc0]
20 ; X32-AVX512VL-NEXT: retl # encoding: [0xc3]
21 ;
22 ; X64-AVX512VL-LABEL: test_x86_vcvtph2ps_128:
23 ; X64-AVX512VL: # BB#0:
24 ; X64-AVX512VL-NEXT: vcvtph2ps %xmm0, %xmm0 # encoding: [0xc4,0xe2,0x79,0x13,0xc0]
25 ; X64-AVX512VL-NEXT: retq # encoding: [0xc3]
1426 %res = call <4 x float> @llvm.x86.vcvtph2ps.128(<8 x i16> %a0) ; <<4 x float>> [#uses=1]
1527 ret <4 x float> %res
1628 }
2032 define <8 x float> @test_x86_vcvtph2ps_256(<8 x i16> %a0) {
2133 ; X32-LABEL: test_x86_vcvtph2ps_256:
2234 ; X32: # BB#0:
23 ; X32-NEXT: vcvtph2ps %xmm0, %ymm0
24 ; X32-NEXT: retl
35 ; X32-NEXT: vcvtph2ps %xmm0, %ymm0 # encoding: [0xc4,0xe2,0x7d,0x13,0xc0]
36 ; X32-NEXT: retl # encoding: [0xc3]
2537 ;
2638 ; X64-LABEL: test_x86_vcvtph2ps_256:
2739 ; X64: # BB#0:
28 ; X64-NEXT: vcvtph2ps %xmm0, %ymm0
29 ; X64-NEXT: retq
40 ; X64-NEXT: vcvtph2ps %xmm0, %ymm0 # encoding: [0xc4,0xe2,0x7d,0x13,0xc0]
41 ; X64-NEXT: retq # encoding: [0xc3]
42 ;
43 ; X32-AVX512VL-LABEL: test_x86_vcvtph2ps_256:
44 ; X32-AVX512VL: # BB#0:
45 ; X32-AVX512VL-NEXT: vcvtph2ps %xmm0, %ymm0 # encoding: [0xc4,0xe2,0x7d,0x13,0xc0]
46 ; X32-AVX512VL-NEXT: retl # encoding: [0xc3]
47 ;
48 ; X64-AVX512VL-LABEL: test_x86_vcvtph2ps_256:
49 ; X64-AVX512VL: # BB#0:
50 ; X64-AVX512VL-NEXT: vcvtph2ps %xmm0, %ymm0 # encoding: [0xc4,0xe2,0x7d,0x13,0xc0]
51 ; X64-AVX512VL-NEXT: retq # encoding: [0xc3]
3052 %res = call <8 x float> @llvm.x86.vcvtph2ps.256(<8 x i16> %a0) ; <<8 x float>> [#uses=1]
3153 ret <8 x float> %res
3254 }
3557 define <8 x float> @test_x86_vcvtph2ps_256_m(<8 x i16>* nocapture %a) nounwind {
3658 ; X32-LABEL: test_x86_vcvtph2ps_256_m:
3759 ; X32: # BB#0:
38 ; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
39 ; X32-NEXT: vcvtph2ps (%eax), %ymm0
40 ; X32-NEXT: retl
60 ; X32-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
61 ; X32-NEXT: vcvtph2ps (%eax), %ymm0 # encoding: [0xc4,0xe2,0x7d,0x13,0x00]
62 ; X32-NEXT: retl # encoding: [0xc3]
4163 ;
4264 ; X64-LABEL: test_x86_vcvtph2ps_256_m:
4365 ; X64: # BB#0:
44 ; X64-NEXT: vcvtph2ps (%rdi), %ymm0
45 ; X64-NEXT: retq
66 ; X64-NEXT: vcvtph2ps (%rdi), %ymm0 # encoding: [0xc4,0xe2,0x7d,0x13,0x07]
67 ; X64-NEXT: retq # encoding: [0xc3]
68 ;
69 ; X32-AVX512VL-LABEL: test_x86_vcvtph2ps_256_m:
70 ; X32-AVX512VL: # BB#0:
71 ; X32-AVX512VL-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
72 ; X32-AVX512VL-NEXT: vcvtph2ps (%eax), %ymm0 # encoding: [0xc4,0xe2,0x7d,0x13,0x00]
73 ; X32-AVX512VL-NEXT: retl # encoding: [0xc3]
74 ;
75 ; X64-AVX512VL-LABEL: test_x86_vcvtph2ps_256_m:
76 ; X64-AVX512VL: # BB#0:
77 ; X64-AVX512VL-NEXT: vcvtph2ps (%rdi), %ymm0 # encoding: [0xc4,0xe2,0x7d,0x13,0x07]
78 ; X64-AVX512VL-NEXT: retq # encoding: [0xc3]
4679 %load = load <8 x i16>, <8 x i16>* %a, align 16
4780 %res = tail call <8 x float> @llvm.x86.vcvtph2ps.256(<8 x i16> %load)
4881 ret <8 x float> %res
5184 define <8 x i16> @test_x86_vcvtps2ph_128(<4 x float> %a0) {
5285 ; X32-LABEL: test_x86_vcvtps2ph_128:
5386 ; X32: # BB#0:
54 ; X32-NEXT: vcvtps2ph $0, %xmm0, %xmm0
55 ; X32-NEXT: retl
87 ; X32-NEXT: vcvtps2ph $0, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x1d,0xc0,0x00]
88 ; X32-NEXT: retl # encoding: [0xc3]
5689 ;
5790 ; X64-LABEL: test_x86_vcvtps2ph_128:
5891 ; X64: # BB#0:
59 ; X64-NEXT: vcvtps2ph $0, %xmm0, %xmm0
60 ; X64-NEXT: retq
92 ; X64-NEXT: vcvtps2ph $0, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x1d,0xc0,0x00]
93 ; X64-NEXT: retq # encoding: [0xc3]
94 ;
95 ; X32-AVX512VL-LABEL: test_x86_vcvtps2ph_128:
96 ; X32-AVX512VL: # BB#0:
97 ; X32-AVX512VL-NEXT: vcvtps2ph $0, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x1d,0xc0,0x00]
98 ; X32-AVX512VL-NEXT: retl # encoding: [0xc3]
99 ;
100 ; X64-AVX512VL-LABEL: test_x86_vcvtps2ph_128:
101 ; X64-AVX512VL: # BB#0:
102 ; X64-AVX512VL-NEXT: vcvtps2ph $0, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x1d,0xc0,0x00]
103 ; X64-AVX512VL-NEXT: retq # encoding: [0xc3]
61104 %res = call <8 x i16> @llvm.x86.vcvtps2ph.128(<4 x float> %a0, i32 0) ; <<8 x i16>> [#uses=1]
62105 ret <8 x i16> %res
63106 }
66109 define <8 x i16> @test_x86_vcvtps2ph_256(<8 x float> %a0) {
67110 ; X32-LABEL: test_x86_vcvtps2ph_256:
68111 ; X32: # BB#0:
69 ; X32-NEXT: vcvtps2ph $0, %ymm0, %xmm0
70 ; X32-NEXT: vzeroupper
71 ; X32-NEXT: retl
112 ; X32-NEXT: vcvtps2ph $0, %ymm0, %xmm0 # encoding: [0xc4,0xe3,0x7d,0x1d,0xc0,0x00]
113 ; X32-NEXT: vzeroupper # encoding: [0xc5,0xf8,0x77]
114 ; X32-NEXT: retl # encoding: [0xc3]
72115 ;
73116 ; X64-LABEL: test_x86_vcvtps2ph_256:
74117 ; X64: # BB#0:
75 ; X64-NEXT: vcvtps2ph $0, %ymm0, %xmm0
76 ; X64-NEXT: vzeroupper
77 ; X64-NEXT: retq
118 ; X64-NEXT: vcvtps2ph $0, %ymm0, %xmm0 # encoding: [0xc4,0xe3,0x7d,0x1d,0xc0,0x00]
119 ; X64-NEXT: vzeroupper # encoding: [0xc5,0xf8,0x77]
120 ; X64-NEXT: retq # encoding: [0xc3]
121 ;
122 ; X32-AVX512VL-LABEL: test_x86_vcvtps2ph_256:
123 ; X32-AVX512VL: # BB#0:
124 ; X32-AVX512VL-NEXT: vcvtps2ph $0, %ymm0, %xmm0 # encoding: [0xc4,0xe3,0x7d,0x1d,0xc0,0x00]
125 ; X32-AVX512VL-NEXT: vzeroupper # encoding: [0xc5,0xf8,0x77]
126 ; X32-AVX512VL-NEXT: retl # encoding: [0xc3]
127 ;
128 ; X64-AVX512VL-LABEL: test_x86_vcvtps2ph_256:
129 ; X64-AVX512VL: # BB#0:
130 ; X64-AVX512VL-NEXT: vcvtps2ph $0, %ymm0, %xmm0 # encoding: [0xc4,0xe3,0x7d,0x1d,0xc0,0x00]
131 ; X64-AVX512VL-NEXT: vzeroupper # encoding: [0xc5,0xf8,0x77]
132 ; X64-AVX512VL-NEXT: retq # encoding: [0xc3]
78133 %res = call <8 x i16> @llvm.x86.vcvtps2ph.256(<8 x float> %a0, i32 0) ; <<8 x i16>> [#uses=1]
79134 ret <8 x i16> %res
80135 }
83138 define <4 x float> @test_x86_vcvtps2ph_128_scalar(i64* %ptr) {
84139 ; X32-LABEL: test_x86_vcvtps2ph_128_scalar:
85140 ; X32: # BB#0:
86 ; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
87 ; X32-NEXT: vcvtph2ps (%eax), %xmm0
88 ; X32-NEXT: retl
141 ; X32-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
142 ; X32-NEXT: vcvtph2ps (%eax), %xmm0 # encoding: [0xc4,0xe2,0x79,0x13,0x00]
143 ; X32-NEXT: retl # encoding: [0xc3]
89144 ;
90145 ; X64-LABEL: test_x86_vcvtps2ph_128_scalar:
91146 ; X64: # BB#0:
92 ; X64-NEXT: vcvtph2ps (%rdi), %xmm0
93 ; X64-NEXT: retq
147 ; X64-NEXT: vcvtph2ps (%rdi), %xmm0 # encoding: [0xc4,0xe2,0x79,0x13,0x07]
148 ; X64-NEXT: retq # encoding: [0xc3]
149 ;
150 ; X32-AVX512VL-LABEL: test_x86_vcvtps2ph_128_scalar:
151 ; X32-AVX512VL: # BB#0:
152 ; X32-AVX512VL-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
153 ; X32-AVX512VL-NEXT: vcvtph2ps (%eax), %xmm0 # encoding: [0xc4,0xe2,0x79,0x13,0x00]
154 ; X32-AVX512VL-NEXT: retl # encoding: [0xc3]
155 ;
156 ; X64-AVX512VL-LABEL: test_x86_vcvtps2ph_128_scalar:
157 ; X64-AVX512VL: # BB#0:
158 ; X64-AVX512VL-NEXT: vcvtph2ps (%rdi), %xmm0 # encoding: [0xc4,0xe2,0x79,0x13,0x07]
159 ; X64-AVX512VL-NEXT: retq # encoding: [0xc3]
94160 %load = load i64, i64* %ptr
95161 %ins1 = insertelement <2 x i64> undef, i64 %load, i32 0
96162 %ins2 = insertelement <2 x i64> %ins1, i64 0, i32 1
102168 define <4 x float> @test_x86_vcvtps2ph_128_scalar2(i64* %ptr) {
103169 ; X32-LABEL: test_x86_vcvtps2ph_128_scalar2:
104170 ; X32: # BB#0:
105 ; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
106 ; X32-NEXT: vcvtph2ps (%eax), %xmm0
107 ; X32-NEXT: retl
171 ; X32-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
172 ; X32-NEXT: vcvtph2ps (%eax), %xmm0 # encoding: [0xc4,0xe2,0x79,0x13,0x00]
173 ; X32-NEXT: retl # encoding: [0xc3]
108174 ;
109175 ; X64-LABEL: test_x86_vcvtps2ph_128_scalar2:
110176 ; X64: # BB#0:
111 ; X64-NEXT: vcvtph2ps (%rdi), %xmm0
112 ; X64-NEXT: retq
177 ; X64-NEXT: vcvtph2ps (%rdi), %xmm0 # encoding: [0xc4,0xe2,0x79,0x13,0x07]
178 ; X64-NEXT: retq # encoding: [0xc3]
179 ;
180 ; X32-AVX512VL-LABEL: test_x86_vcvtps2ph_128_scalar2:
181 ; X32-AVX512VL: # BB#0:
182 ; X32-AVX512VL-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
183 ; X32-AVX512VL-NEXT: vcvtph2ps (%eax), %xmm0 # encoding: [0xc4,0xe2,0x79,0x13,0x00]
184 ; X32-AVX512VL-NEXT: retl # encoding: [0xc3]
185 ;
186 ; X64-AVX512VL-LABEL: test_x86_vcvtps2ph_128_scalar2:
187 ; X64-AVX512VL: # BB#0:
188 ; X64-AVX512VL-NEXT: vcvtph2ps (%rdi), %xmm0 # encoding: [0xc4,0xe2,0x79,0x13,0x07]
189 ; X64-AVX512VL-NEXT: retq # encoding: [0xc3]
113190 %load = load i64, i64* %ptr
114191 %ins = insertelement <2 x i64> undef, i64 %load, i32 0
115192 %bc = bitcast <2 x i64> %ins to <8 x i16>
120197 define void @test_x86_vcvtps2ph_256_m(<8 x i16>* nocapture %d, <8 x float> %a) nounwind {
121198 ; X32-LABEL: test_x86_vcvtps2ph_256_m:
122199 ; X32: # BB#0: # %entry
123 ; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
124 ; X32-NEXT: vcvtps2ph $3, %ymm0, (%eax)
125 ; X32-NEXT: vzeroupper
126 ; X32-NEXT: retl
200 ; X32-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
201 ; X32-NEXT: vcvtps2ph $3, %ymm0, (%eax) # encoding: [0xc4,0xe3,0x7d,0x1d,0x00,0x03]
202 ; X32-NEXT: vzeroupper # encoding: [0xc5,0xf8,0x77]
203 ; X32-NEXT: retl # encoding: [0xc3]
127204 ;
128205 ; X64-LABEL: test_x86_vcvtps2ph_256_m:
129206 ; X64: # BB#0: # %entry
130 ; X64-NEXT: vcvtps2ph $3, %ymm0, (%rdi)
131 ; X64-NEXT: vzeroupper
132 ; X64-NEXT: retq
207 ; X64-NEXT: vcvtps2ph $3, %ymm0, (%rdi) # encoding: [0xc4,0xe3,0x7d,0x1d,0x07,0x03]
208 ; X64-NEXT: vzeroupper # encoding: [0xc5,0xf8,0x77]
209 ; X64-NEXT: retq # encoding: [0xc3]
210 ;
211 ; X32-AVX512VL-LABEL: test_x86_vcvtps2ph_256_m:
212 ; X32-AVX512VL: # BB#0: # %entry
213 ; X32-AVX512VL-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
214 ; X32-AVX512VL-NEXT: vcvtps2ph $3, %ymm0, (%eax) # encoding: [0xc4,0xe3,0x7d,0x1d,0x00,0x03]
215 ; X32-AVX512VL-NEXT: vzeroupper # encoding: [0xc5,0xf8,0x77]
216 ; X32-AVX512VL-NEXT: retl # encoding: [0xc3]
217 ;
218 ; X64-AVX512VL-LABEL: test_x86_vcvtps2ph_256_m:
219 ; X64-AVX512VL: # BB#0: # %entry
220 ; X64-AVX512VL-NEXT: vcvtps2ph $3, %ymm0, (%rdi) # encoding: [0xc4,0xe3,0x7d,0x1d,0x07,0x03]
221 ; X64-AVX512VL-NEXT: vzeroupper # encoding: [0xc5,0xf8,0x77]
222 ; X64-AVX512VL-NEXT: retq # encoding: [0xc3]
133223 entry:
134224 %0 = tail call <8 x i16> @llvm.x86.vcvtps2ph.256(<8 x float> %a, i32 3)
135225 store <8 x i16> %0, <8 x i16>* %d, align 16
139229 define void @test_x86_vcvtps2ph_128_m(<4 x i16>* nocapture %d, <4 x float> %a) nounwind {
140230 ; X32-LABEL: test_x86_vcvtps2ph_128_m:
141231 ; X32: # BB#0: # %entry
142 ; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
143 ; X32-NEXT: vcvtps2ph $3, %xmm0, (%eax)
144 ; X32-NEXT: retl
232 ; X32-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
233 ; X32-NEXT: vcvtps2ph $3, %xmm0, (%eax) # encoding: [0xc4,0xe3,0x79,0x1d,0x00,0x03]
234 ; X32-NEXT: retl # encoding: [0xc3]
145235 ;
146236 ; X64-LABEL: test_x86_vcvtps2ph_128_m:
147237 ; X64: # BB#0: # %entry
148 ; X64-NEXT: vcvtps2ph $3, %xmm0, (%rdi)
149 ; X64-NEXT: retq
238 ; X64-NEXT: vcvtps2ph $3, %xmm0, (%rdi) # encoding: [0xc4,0xe3,0x79,0x1d,0x07,0x03]
239 ; X64-NEXT: retq # encoding: [0xc3]
240 ;
241 ; X32-AVX512VL-LABEL: test_x86_vcvtps2ph_128_m:
242 ; X32-AVX512VL: # BB#0: # %entry
243 ; X32-AVX512VL-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
244 ; X32-AVX512VL-NEXT: vcvtps2ph $3, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x1d,0xc0,0x03]
245 ; X32-AVX512VL-NEXT: vpmovzxwd %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x79,0x33,0xc0]
246 ; X32-AVX512VL-NEXT: # xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
247 ; X32-AVX512VL-NEXT: vpmovdw %xmm0, (%eax) # encoding: [0x62,0xf2,0x7e,0x08,0x33,0x00]
248 ; X32-AVX512VL-NEXT: retl # encoding: [0xc3]
249 ;
250 ; X64-AVX512VL-LABEL: test_x86_vcvtps2ph_128_m:
251 ; X64-AVX512VL: # BB#0: # %entry
252 ; X64-AVX512VL-NEXT: vcvtps2ph $3, %xmm0, %xmm0 # encoding: [0xc4,0xe3,0x79,0x1d,0xc0,0x03]
253 ; X64-AVX512VL-NEXT: vpmovzxwd %xmm0, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x79,0x33,0xc0]
254 ; X64-AVX512VL-NEXT: # xmm0 = xmm0[0],zero,xmm0[1],zero,xmm0[2],zero,xmm0[3],zero
255 ; X64-AVX512VL-NEXT: vpmovdw %xmm0, (%rdi) # encoding: [0x62,0xf2,0x7e,0x08,0x33,0x07]
256 ; X64-AVX512VL-NEXT: retq # encoding: [0xc3]
150257 entry:
151258 %0 = tail call <8 x i16> @llvm.x86.vcvtps2ph.128(<4 x float> %a, i32 3)
152259 %1 = shufflevector <8 x i16> %0, <8 x i16> undef, <4 x i32>
157264 define void @test_x86_vcvtps2ph_128_m2(double* nocapture %hf4x16, <4 x float> %f4x32) #0 {
158265 ; X32-LABEL: test_x86_vcvtps2ph_128_m2:
159266 ; X32: # BB#0: # %entry
160 ; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
161 ; X32-NEXT: vcvtps2ph $3, %xmm0, (%eax)
162 ; X32-NEXT: retl
267 ; X32-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
268 ; X32-NEXT: vcvtps2ph $3, %xmm0, (%eax) # encoding: [0xc4,0xe3,0x79,0x1d,0x00,0x03]
269 ; X32-NEXT: retl # encoding: [0xc3]
163270 ;
164271 ; X64-LABEL: test_x86_vcvtps2ph_128_m2:
165272 ; X64: # BB#0: # %entry
166 ; X64-NEXT: vcvtps2ph $3, %xmm0, (%rdi)
167 ; X64-NEXT: retq
273 ; X64-NEXT: vcvtps2ph $3, %xmm0, (%rdi) # encoding: [0xc4,0xe3,0x79,0x1d,0x07,0x03]
274 ; X64-NEXT: retq # encoding: [0xc3]
275 ;
276 ; X32-AVX512VL-LABEL: test_x86_vcvtps2ph_128_m2:
277 ; X32-AVX512VL: # BB#0: # %entry
278 ; X32-AVX512VL-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
279 ; X32-AVX512VL-NEXT: vcvtps2ph $3, %xmm0, (%eax) # encoding: [0xc4,0xe3,0x79,0x1d,0x00,0x03]
280 ; X32-AVX512VL-NEXT: retl # encoding: [0xc3]
281 ;
282 ; X64-AVX512VL-LABEL: test_x86_vcvtps2ph_128_m2:
283 ; X64-AVX512VL: # BB#0: # %entry
284 ; X64-AVX512VL-NEXT: vcvtps2ph $3, %xmm0, (%rdi) # encoding: [0xc4,0xe3,0x79,0x1d,0x07,0x03]
285 ; X64-AVX512VL-NEXT: retq # encoding: [0xc3]
168286 entry:
169287 %0 = tail call <8 x i16> @llvm.x86.vcvtps2ph.128(<4 x float> %f4x32, i32 3)
170288 %1 = bitcast <8 x i16> %0 to <2 x double>
176294 define void @test_x86_vcvtps2ph_128_m3(i64* nocapture %hf4x16, <4 x float> %f4x32) #0 {
177295 ; X32-LABEL: test_x86_vcvtps2ph_128_m3:
178296 ; X32: # BB#0: # %entry
179 ; X32-NEXT: movl {{[0-9]+}}(%esp), %eax
180 ; X32-NEXT: vcvtps2ph $3, %xmm0, (%eax)
181 ; X32-NEXT: retl
297 ; X32-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
298 ; X32-NEXT: vcvtps2ph $3, %xmm0, (%eax) # encoding: [0xc4,0xe3,0x79,0x1d,0x00,0x03]
299 ; X32-NEXT: retl # encoding: [0xc3]
182300 ;
183301 ; X64-LABEL: test_x86_vcvtps2ph_128_m3:
184302 ; X64: # BB#0: # %entry
185 ; X64-NEXT: vcvtps2ph $3, %xmm0, (%rdi)
186 ; X64-NEXT: retq
303 ; X64-NEXT: vcvtps2ph $3, %xmm0, (%rdi) # encoding: [0xc4,0xe3,0x79,0x1d,0x07,0x03]
304 ; X64-NEXT: retq # encoding: [0xc3]
305 ;
306 ; X32-AVX512VL-LABEL: test_x86_vcvtps2ph_128_m3:
307 ; X32-AVX512VL: # BB#0: # %entry
308 ; X32-AVX512VL-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
309 ; X32-AVX512VL-NEXT: vcvtps2ph $3, %xmm0, (%eax) # encoding: [0xc4,0xe3,0x79,0x1d,0x00,0x03]
310 ; X32-AVX512VL-NEXT: retl # encoding: [0xc3]
311 ;
312 ; X64-AVX512VL-LABEL: test_x86_vcvtps2ph_128_m3:
313 ; X64-AVX512VL: # BB#0: # %entry
314 ; X64-AVX512VL-NEXT: vcvtps2ph $3, %xmm0, (%rdi) # encoding: [0xc4,0xe3,0x79,0x1d,0x07,0x03]
315 ; X64-AVX512VL-NEXT: retq # encoding: [0xc3]
187316 entry:
188317 %0 = tail call <8 x i16> @llvm.x86.vcvtps2ph.128(<4 x float> %f4x32, i32 3)
189318 %1 = bitcast <8 x i16> %0 to <2 x i64>