| blob | 2a3dae1b64ea781730b987ceaa836bacbc95e78f |
1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=x86-64 -mattr=+sse3 | FileCheck %s --check-prefix=CHECK --check-prefix=GENERIC
3 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=atom | FileCheck %s --check-prefix=CHECK --check-prefix=ATOM
4 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=slm | FileCheck %s --check-prefix=CHECK --check-prefix=SLM
5 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=sandybridge | FileCheck %s --check-prefix=CHECK --check-prefix=SANDY
6 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=ivybridge | FileCheck %s --check-prefix=CHECK --check-prefix=SANDY
7 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=haswell | FileCheck %s --check-prefix=CHECK --check-prefix=HASWELL
8 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=broadwell | FileCheck %s --check-prefix=CHECK --check-prefix=BROADWELL
9 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skylake | FileCheck %s --check-prefix=CHECK --check-prefix=SKYLAKE
10 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=skx | FileCheck %s --check-prefix=CHECK --check-prefix=SKX
11 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=btver2 | FileCheck %s --check-prefix=CHECK --check-prefix=BTVER2
12 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -print-schedule -mcpu=znver1 | FileCheck %s --check-prefix=CHECK --check-prefix=ZNVER1
14 define <2 x double> @test_addsubpd(<2 x double> %a0, <2 x double> %a1, <2 x double> *%a2) {
15 ; GENERIC-LABEL: test_addsubpd:
16 ; GENERIC: # BB#0:
17 ; GENERIC-NEXT: addsubpd %xmm1, %xmm0 # sched: [3:1.00]
18 ; GENERIC-NEXT: addsubpd (%rdi), %xmm0 # sched: [9:1.00]
19 ; GENERIC-NEXT: retq # sched: [1:1.00]
20 ;
21 ; ATOM-LABEL: test_addsubpd:
22 ; ATOM: # BB#0:
23 ; ATOM-NEXT: addsubpd %xmm1, %xmm0 # sched: [6:3.00]
24 ; ATOM-NEXT: addsubpd (%rdi), %xmm0 # sched: [6:3.00]
25 ; ATOM-NEXT: retq # sched: [79:39.50]
26 ;
27 ; SLM-LABEL: test_addsubpd:
28 ; SLM: # BB#0:
29 ; SLM-NEXT: addsubpd %xmm1, %xmm0 # sched: [3:1.00]
30 ; SLM-NEXT: addsubpd (%rdi), %xmm0 # sched: [6:1.00]
31 ; SLM-NEXT: retq # sched: [4:1.00]
32 ;
33 ; SANDY-LABEL: test_addsubpd:
34 ; SANDY: # BB#0:
35 ; SANDY-NEXT: vaddsubpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
36 ; SANDY-NEXT: vaddsubpd (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
37 ; SANDY-NEXT: retq # sched: [1:1.00]
38 ;
39 ; HASWELL-LABEL: test_addsubpd:
40 ; HASWELL: # BB#0:
41 ; HASWELL-NEXT: vaddsubpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
42 ; HASWELL-NEXT: vaddsubpd (%rdi), %xmm0, %xmm0 # sched: [3:1.00]
43 ; HASWELL-NEXT: retq # sched: [2:1.00]
44 ;
45 ; BROADWELL-LABEL: test_addsubpd:
46 ; BROADWELL: # BB#0:
47 ; BROADWELL-NEXT: vaddsubpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
48 ; BROADWELL-NEXT: vaddsubpd (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
49 ; BROADWELL-NEXT: retq # sched: [7:1.00]
50 ;
51 ; SKYLAKE-LABEL: test_addsubpd:
52 ; SKYLAKE: # BB#0:
53 ; SKYLAKE-NEXT: vaddsubpd %xmm1, %xmm0, %xmm0 # sched: [4:0.50]
54 ; SKYLAKE-NEXT: vaddsubpd (%rdi), %xmm0, %xmm0 # sched: [10:0.50]
55 ; SKYLAKE-NEXT: retq # sched: [7:1.00]
56 ;
57 ; SKX-LABEL: test_addsubpd:
58 ; SKX: # BB#0:
59 ; SKX-NEXT: vaddsubpd %xmm1, %xmm0, %xmm0 # sched: [4:0.33]
60 ; SKX-NEXT: vaddsubpd (%rdi), %xmm0, %xmm0 # sched: [10:0.50]
61 ; SKX-NEXT: retq # sched: [7:1.00]
62 ;
63 ; BTVER2-LABEL: test_addsubpd:
64 ; BTVER2: # BB#0:
65 ; BTVER2-NEXT: vaddsubpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
66 ; BTVER2-NEXT: vaddsubpd (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
67 ; BTVER2-NEXT: retq # sched: [4:1.00]
68 ;
69 ; ZNVER1-LABEL: test_addsubpd:
70 ; ZNVER1: # BB#0:
71 ; ZNVER1-NEXT: vaddsubpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
72 ; ZNVER1-NEXT: vaddsubpd (%rdi), %xmm0, %xmm0 # sched: [10:1.00]
73 ; ZNVER1-NEXT: retq # sched: [1:0.50]
74 %1 = call <2 x double> @llvm.x86.sse3.addsub.pd(<2 x double> %a0, <2 x double> %a1)
75 %2 = load <2 x double>, <2 x double> *%a2, align 16
76 %3 = call <2 x double> @llvm.x86.sse3.addsub.pd(<2 x double> %1, <2 x double> %2)
77 ret <2 x double> %3
78 }
79 declare <2 x double> @llvm.x86.sse3.addsub.pd(<2 x double>, <2 x double>) nounwind readnone
81 define <4 x float> @test_addsubps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a2) {
82 ; GENERIC-LABEL: test_addsubps:
83 ; GENERIC: # BB#0:
84 ; GENERIC-NEXT: addsubps %xmm1, %xmm0 # sched: [3:1.00]
85 ; GENERIC-NEXT: addsubps (%rdi), %xmm0 # sched: [9:1.00]
86 ; GENERIC-NEXT: retq # sched: [1:1.00]
87 ;
88 ; ATOM-LABEL: test_addsubps:
89 ; ATOM: # BB#0:
90 ; ATOM-NEXT: addsubps %xmm1, %xmm0 # sched: [5:5.00]
91 ; ATOM-NEXT: addsubps (%rdi), %xmm0 # sched: [5:5.00]
92 ; ATOM-NEXT: retq # sched: [79:39.50]
93 ;
94 ; SLM-LABEL: test_addsubps:
95 ; SLM: # BB#0:
96 ; SLM-NEXT: addsubps %xmm1, %xmm0 # sched: [3:1.00]
97 ; SLM-NEXT: addsubps (%rdi), %xmm0 # sched: [6:1.00]
98 ; SLM-NEXT: retq # sched: [4:1.00]
99 ;
100 ; SANDY-LABEL: test_addsubps:
101 ; SANDY: # BB#0:
102 ; SANDY-NEXT: vaddsubps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
103 ; SANDY-NEXT: vaddsubps (%rdi), %xmm0, %xmm0 # sched: [9:1.00]
104 ; SANDY-NEXT: retq # sched: [1:1.00]
105 ;
106 ; HASWELL-LABEL: test_addsubps:
107 ; HASWELL: # BB#0:
108 ; HASWELL-NEXT: vaddsubps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
109 ; HASWELL-NEXT: vaddsubps (%rdi), %xmm0, %xmm0 # sched: [3:1.00]
110 ; HASWELL-NEXT: retq # sched: [2:1.00]
111 ;
112 ; BROADWELL-LABEL: test_addsubps:
113 ; BROADWELL: # BB#0:
114 ; BROADWELL-NEXT: vaddsubps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
115 ; BROADWELL-NEXT: vaddsubps (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
116 ; BROADWELL-NEXT: retq # sched: [7:1.00]
117 ;
118 ; SKYLAKE-LABEL: test_addsubps:
119 ; SKYLAKE: # BB#0:
120 ; SKYLAKE-NEXT: vaddsubps %xmm1, %xmm0, %xmm0 # sched: [4:0.50]
121 ; SKYLAKE-NEXT: vaddsubps (%rdi), %xmm0, %xmm0 # sched: [10:0.50]
122 ; SKYLAKE-NEXT: retq # sched: [7:1.00]
123 ;
124 ; SKX-LABEL: test_addsubps:
125 ; SKX: # BB#0:
126 ; SKX-NEXT: vaddsubps %xmm1, %xmm0, %xmm0 # sched: [4:0.33]
127 ; SKX-NEXT: vaddsubps (%rdi), %xmm0, %xmm0 # sched: [10:0.50]
128 ; SKX-NEXT: retq # sched: [7:1.00]
129 ;
130 ; BTVER2-LABEL: test_addsubps:
131 ; BTVER2: # BB#0:
132 ; BTVER2-NEXT: vaddsubps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
133 ; BTVER2-NEXT: vaddsubps (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
134 ; BTVER2-NEXT: retq # sched: [4:1.00]
135 ;
136 ; ZNVER1-LABEL: test_addsubps:
137 ; ZNVER1: # BB#0:
138 ; ZNVER1-NEXT: vaddsubps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
139 ; ZNVER1-NEXT: vaddsubps (%rdi), %xmm0, %xmm0 # sched: [10:1.00]
140 ; ZNVER1-NEXT: retq # sched: [1:0.50]
141 %1 = call <4 x float> @llvm.x86.sse3.addsub.ps(<4 x float> %a0, <4 x float> %a1)
142 %2 = load <4 x float>, <4 x float> *%a2, align 16
143 %3 = call <4 x float> @llvm.x86.sse3.addsub.ps(<4 x float> %1, <4 x float> %2)
144 ret <4 x float> %3
145 }
146 declare <4 x float> @llvm.x86.sse3.addsub.ps(<4 x float>, <4 x float>) nounwind readnone
148 define <2 x double> @test_haddpd(<2 x double> %a0, <2 x double> %a1, <2 x double> *%a2) {
149 ; GENERIC-LABEL: test_haddpd:
150 ; GENERIC: # BB#0:
151 ; GENERIC-NEXT: haddpd %xmm1, %xmm0 # sched: [5:2.00]
152 ; GENERIC-NEXT: haddpd (%rdi), %xmm0 # sched: [11:2.00]
153 ; GENERIC-NEXT: retq # sched: [1:1.00]
154 ;
155 ; ATOM-LABEL: test_haddpd:
156 ; ATOM: # BB#0:
157 ; ATOM-NEXT: haddpd %xmm1, %xmm0 # sched: [8:4.00]
158 ; ATOM-NEXT: haddpd (%rdi), %xmm0 # sched: [9:4.50]
159 ; ATOM-NEXT: retq # sched: [79:39.50]
160 ;
161 ; SLM-LABEL: test_haddpd:
162 ; SLM: # BB#0:
163 ; SLM-NEXT: haddpd %xmm1, %xmm0 # sched: [3:1.00]
164 ; SLM-NEXT: haddpd (%rdi), %xmm0 # sched: [6:1.00]
165 ; SLM-NEXT: retq # sched: [4:1.00]
166 ;
167 ; SANDY-LABEL: test_haddpd:
168 ; SANDY: # BB#0:
169 ; SANDY-NEXT: vhaddpd %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
170 ; SANDY-NEXT: vhaddpd (%rdi), %xmm0, %xmm0 # sched: [11:2.00]
171 ; SANDY-NEXT: retq # sched: [1:1.00]
172 ;
173 ; HASWELL-LABEL: test_haddpd:
174 ; HASWELL: # BB#0:
175 ; HASWELL-NEXT: vhaddpd %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
176 ; HASWELL-NEXT: vhaddpd (%rdi), %xmm0, %xmm0 # sched: [5:2.00]
177 ; HASWELL-NEXT: retq # sched: [2:1.00]
178 ;
179 ; BROADWELL-LABEL: test_haddpd:
180 ; BROADWELL: # BB#0:
181 ; BROADWELL-NEXT: vhaddpd %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
182 ; BROADWELL-NEXT: vhaddpd (%rdi), %xmm0, %xmm0 # sched: [10:2.00]
183 ; BROADWELL-NEXT: retq # sched: [7:1.00]
184 ;
185 ; SKYLAKE-LABEL: test_haddpd:
186 ; SKYLAKE: # BB#0:
187 ; SKYLAKE-NEXT: vhaddpd %xmm1, %xmm0, %xmm0 # sched: [6:2.00]
188 ; SKYLAKE-NEXT: vhaddpd (%rdi), %xmm0, %xmm0 # sched: [12:2.00]
189 ; SKYLAKE-NEXT: retq # sched: [7:1.00]
190 ;
191 ; SKX-LABEL: test_haddpd:
192 ; SKX: # BB#0:
193 ; SKX-NEXT: vhaddpd %xmm1, %xmm0, %xmm0 # sched: [6:2.00]
194 ; SKX-NEXT: vhaddpd (%rdi), %xmm0, %xmm0 # sched: [12:2.00]
195 ; SKX-NEXT: retq # sched: [7:1.00]
196 ;
197 ; BTVER2-LABEL: test_haddpd:
198 ; BTVER2: # BB#0:
199 ; BTVER2-NEXT: vhaddpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
200 ; BTVER2-NEXT: vhaddpd (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
201 ; BTVER2-NEXT: retq # sched: [4:1.00]
202 ;
203 ; ZNVER1-LABEL: test_haddpd:
204 ; ZNVER1: # BB#0:
205 ; ZNVER1-NEXT: vhaddpd %xmm1, %xmm0, %xmm0 # sched: [100:?]
206 ; ZNVER1-NEXT: vhaddpd (%rdi), %xmm0, %xmm0 # sched: [100:?]
207 ; ZNVER1-NEXT: retq # sched: [1:0.50]
208 %1 = call <2 x double> @llvm.x86.sse3.hadd.pd(<2 x double> %a0, <2 x double> %a1)
209 %2 = load <2 x double>, <2 x double> *%a2, align 16
210 %3 = call <2 x double> @llvm.x86.sse3.hadd.pd(<2 x double> %1, <2 x double> %2)
211 ret <2 x double> %3
212 }
213 declare <2 x double> @llvm.x86.sse3.hadd.pd(<2 x double>, <2 x double>) nounwind readnone
215 define <4 x float> @test_haddps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a2) {
216 ; GENERIC-LABEL: test_haddps:
217 ; GENERIC: # BB#0:
218 ; GENERIC-NEXT: haddps %xmm1, %xmm0 # sched: [5:2.00]
219 ; GENERIC-NEXT: haddps (%rdi), %xmm0 # sched: [11:2.00]
220 ; GENERIC-NEXT: retq # sched: [1:1.00]
221 ;
222 ; ATOM-LABEL: test_haddps:
223 ; ATOM: # BB#0:
224 ; ATOM-NEXT: haddps %xmm1, %xmm0 # sched: [8:4.00]
225 ; ATOM-NEXT: haddps (%rdi), %xmm0 # sched: [9:4.50]
226 ; ATOM-NEXT: retq # sched: [79:39.50]
227 ;
228 ; SLM-LABEL: test_haddps:
229 ; SLM: # BB#0:
230 ; SLM-NEXT: haddps %xmm1, %xmm0 # sched: [3:1.00]
231 ; SLM-NEXT: haddps (%rdi), %xmm0 # sched: [6:1.00]
232 ; SLM-NEXT: retq # sched: [4:1.00]
233 ;
234 ; SANDY-LABEL: test_haddps:
235 ; SANDY: # BB#0:
236 ; SANDY-NEXT: vhaddps %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
237 ; SANDY-NEXT: vhaddps (%rdi), %xmm0, %xmm0 # sched: [11:2.00]
238 ; SANDY-NEXT: retq # sched: [1:1.00]
239 ;
240 ; HASWELL-LABEL: test_haddps:
241 ; HASWELL: # BB#0:
242 ; HASWELL-NEXT: vhaddps %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
243 ; HASWELL-NEXT: vhaddps (%rdi), %xmm0, %xmm0 # sched: [5:2.00]
244 ; HASWELL-NEXT: retq # sched: [2:1.00]
245 ;
246 ; BROADWELL-LABEL: test_haddps:
247 ; BROADWELL: # BB#0:
248 ; BROADWELL-NEXT: vhaddps %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
249 ; BROADWELL-NEXT: vhaddps (%rdi), %xmm0, %xmm0 # sched: [10:2.00]
250 ; BROADWELL-NEXT: retq # sched: [7:1.00]
251 ;
252 ; SKYLAKE-LABEL: test_haddps:
253 ; SKYLAKE: # BB#0:
254 ; SKYLAKE-NEXT: vhaddps %xmm1, %xmm0, %xmm0 # sched: [6:2.00]
255 ; SKYLAKE-NEXT: vhaddps (%rdi), %xmm0, %xmm0 # sched: [12:2.00]
256 ; SKYLAKE-NEXT: retq # sched: [7:1.00]
257 ;
258 ; SKX-LABEL: test_haddps:
259 ; SKX: # BB#0:
260 ; SKX-NEXT: vhaddps %xmm1, %xmm0, %xmm0 # sched: [6:2.00]
261 ; SKX-NEXT: vhaddps (%rdi), %xmm0, %xmm0 # sched: [12:2.00]
262 ; SKX-NEXT: retq # sched: [7:1.00]
263 ;
264 ; BTVER2-LABEL: test_haddps:
265 ; BTVER2: # BB#0:
266 ; BTVER2-NEXT: vhaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
267 ; BTVER2-NEXT: vhaddps (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
268 ; BTVER2-NEXT: retq # sched: [4:1.00]
269 ;
270 ; ZNVER1-LABEL: test_haddps:
271 ; ZNVER1: # BB#0:
272 ; ZNVER1-NEXT: vhaddps %xmm1, %xmm0, %xmm0 # sched: [100:?]
273 ; ZNVER1-NEXT: vhaddps (%rdi), %xmm0, %xmm0 # sched: [100:?]
274 ; ZNVER1-NEXT: retq # sched: [1:0.50]
275 %1 = call <4 x float> @llvm.x86.sse3.hadd.ps(<4 x float> %a0, <4 x float> %a1)
276 %2 = load <4 x float>, <4 x float> *%a2, align 16
277 %3 = call <4 x float> @llvm.x86.sse3.hadd.ps(<4 x float> %1, <4 x float> %2)
278 ret <4 x float> %3
279 }
280 declare <4 x float> @llvm.x86.sse3.hadd.ps(<4 x float>, <4 x float>) nounwind readnone
282 define <2 x double> @test_hsubpd(<2 x double> %a0, <2 x double> %a1, <2 x double> *%a2) {
283 ; GENERIC-LABEL: test_hsubpd:
284 ; GENERIC: # BB#0:
285 ; GENERIC-NEXT: hsubpd %xmm1, %xmm0 # sched: [5:2.00]
286 ; GENERIC-NEXT: hsubpd (%rdi), %xmm0 # sched: [11:2.00]
287 ; GENERIC-NEXT: retq # sched: [1:1.00]
288 ;
289 ; ATOM-LABEL: test_hsubpd:
290 ; ATOM: # BB#0:
291 ; ATOM-NEXT: hsubpd %xmm1, %xmm0 # sched: [8:4.00]
292 ; ATOM-NEXT: hsubpd (%rdi), %xmm0 # sched: [9:4.50]
293 ; ATOM-NEXT: retq # sched: [79:39.50]
294 ;
295 ; SLM-LABEL: test_hsubpd:
296 ; SLM: # BB#0:
297 ; SLM-NEXT: hsubpd %xmm1, %xmm0 # sched: [3:1.00]
298 ; SLM-NEXT: hsubpd (%rdi), %xmm0 # sched: [6:1.00]
299 ; SLM-NEXT: retq # sched: [4:1.00]
300 ;
301 ; SANDY-LABEL: test_hsubpd:
302 ; SANDY: # BB#0:
303 ; SANDY-NEXT: vhsubpd %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
304 ; SANDY-NEXT: vhsubpd (%rdi), %xmm0, %xmm0 # sched: [11:2.00]
305 ; SANDY-NEXT: retq # sched: [1:1.00]
306 ;
307 ; HASWELL-LABEL: test_hsubpd:
308 ; HASWELL: # BB#0:
309 ; HASWELL-NEXT: vhsubpd %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
310 ; HASWELL-NEXT: vhsubpd (%rdi), %xmm0, %xmm0 # sched: [5:2.00]
311 ; HASWELL-NEXT: retq # sched: [2:1.00]
312 ;
313 ; BROADWELL-LABEL: test_hsubpd:
314 ; BROADWELL: # BB#0:
315 ; BROADWELL-NEXT: vhsubpd %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
316 ; BROADWELL-NEXT: vhsubpd (%rdi), %xmm0, %xmm0 # sched: [10:2.00]
317 ; BROADWELL-NEXT: retq # sched: [7:1.00]
318 ;
319 ; SKYLAKE-LABEL: test_hsubpd:
320 ; SKYLAKE: # BB#0:
321 ; SKYLAKE-NEXT: vhsubpd %xmm1, %xmm0, %xmm0 # sched: [6:2.00]
322 ; SKYLAKE-NEXT: vhsubpd (%rdi), %xmm0, %xmm0 # sched: [12:2.00]
323 ; SKYLAKE-NEXT: retq # sched: [7:1.00]
324 ;
325 ; SKX-LABEL: test_hsubpd:
326 ; SKX: # BB#0:
327 ; SKX-NEXT: vhsubpd %xmm1, %xmm0, %xmm0 # sched: [6:2.00]
328 ; SKX-NEXT: vhsubpd (%rdi), %xmm0, %xmm0 # sched: [12:2.00]
329 ; SKX-NEXT: retq # sched: [7:1.00]
330 ;
331 ; BTVER2-LABEL: test_hsubpd:
332 ; BTVER2: # BB#0:
333 ; BTVER2-NEXT: vhsubpd %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
334 ; BTVER2-NEXT: vhsubpd (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
335 ; BTVER2-NEXT: retq # sched: [4:1.00]
336 ;
337 ; ZNVER1-LABEL: test_hsubpd:
338 ; ZNVER1: # BB#0:
339 ; ZNVER1-NEXT: vhsubpd %xmm1, %xmm0, %xmm0 # sched: [100:?]
340 ; ZNVER1-NEXT: vhsubpd (%rdi), %xmm0, %xmm0 # sched: [100:?]
341 ; ZNVER1-NEXT: retq # sched: [1:0.50]
342 %1 = call <2 x double> @llvm.x86.sse3.hsub.pd(<2 x double> %a0, <2 x double> %a1)
343 %2 = load <2 x double>, <2 x double> *%a2, align 16
344 %3 = call <2 x double> @llvm.x86.sse3.hsub.pd(<2 x double> %1, <2 x double> %2)
345 ret <2 x double> %3
346 }
347 declare <2 x double> @llvm.x86.sse3.hsub.pd(<2 x double>, <2 x double>) nounwind readnone
349 define <4 x float> @test_hsubps(<4 x float> %a0, <4 x float> %a1, <4 x float> *%a2) {
350 ; GENERIC-LABEL: test_hsubps:
351 ; GENERIC: # BB#0:
352 ; GENERIC-NEXT: hsubps %xmm1, %xmm0 # sched: [5:2.00]
353 ; GENERIC-NEXT: hsubps (%rdi), %xmm0 # sched: [11:2.00]
354 ; GENERIC-NEXT: retq # sched: [1:1.00]
355 ;
356 ; ATOM-LABEL: test_hsubps:
357 ; ATOM: # BB#0:
358 ; ATOM-NEXT: hsubps %xmm1, %xmm0 # sched: [8:4.00]
359 ; ATOM-NEXT: hsubps (%rdi), %xmm0 # sched: [9:4.50]
360 ; ATOM-NEXT: retq # sched: [79:39.50]
361 ;
362 ; SLM-LABEL: test_hsubps:
363 ; SLM: # BB#0:
364 ; SLM-NEXT: hsubps %xmm1, %xmm0 # sched: [3:1.00]
365 ; SLM-NEXT: hsubps (%rdi), %xmm0 # sched: [6:1.00]
366 ; SLM-NEXT: retq # sched: [4:1.00]
367 ;
368 ; SANDY-LABEL: test_hsubps:
369 ; SANDY: # BB#0:
370 ; SANDY-NEXT: vhsubps %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
371 ; SANDY-NEXT: vhsubps (%rdi), %xmm0, %xmm0 # sched: [11:2.00]
372 ; SANDY-NEXT: retq # sched: [1:1.00]
373 ;
374 ; HASWELL-LABEL: test_hsubps:
375 ; HASWELL: # BB#0:
376 ; HASWELL-NEXT: vhsubps %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
377 ; HASWELL-NEXT: vhsubps (%rdi), %xmm0, %xmm0 # sched: [5:2.00]
378 ; HASWELL-NEXT: retq # sched: [2:1.00]
379 ;
380 ; BROADWELL-LABEL: test_hsubps:
381 ; BROADWELL: # BB#0:
382 ; BROADWELL-NEXT: vhsubps %xmm1, %xmm0, %xmm0 # sched: [5:2.00]
383 ; BROADWELL-NEXT: vhsubps (%rdi), %xmm0, %xmm0 # sched: [10:2.00]
384 ; BROADWELL-NEXT: retq # sched: [7:1.00]
385 ;
386 ; SKYLAKE-LABEL: test_hsubps:
387 ; SKYLAKE: # BB#0:
388 ; SKYLAKE-NEXT: vhsubps %xmm1, %xmm0, %xmm0 # sched: [6:2.00]
389 ; SKYLAKE-NEXT: vhsubps (%rdi), %xmm0, %xmm0 # sched: [12:2.00]
390 ; SKYLAKE-NEXT: retq # sched: [7:1.00]
391 ;
392 ; SKX-LABEL: test_hsubps:
393 ; SKX: # BB#0:
394 ; SKX-NEXT: vhsubps %xmm1, %xmm0, %xmm0 # sched: [6:2.00]
395 ; SKX-NEXT: vhsubps (%rdi), %xmm0, %xmm0 # sched: [12:2.00]
396 ; SKX-NEXT: retq # sched: [7:1.00]
397 ;
398 ; BTVER2-LABEL: test_hsubps:
399 ; BTVER2: # BB#0:
400 ; BTVER2-NEXT: vhsubps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
401 ; BTVER2-NEXT: vhsubps (%rdi), %xmm0, %xmm0 # sched: [8:1.00]
402 ; BTVER2-NEXT: retq # sched: [4:1.00]
403 ;
404 ; ZNVER1-LABEL: test_hsubps:
405 ; ZNVER1: # BB#0:
406 ; ZNVER1-NEXT: vhsubps %xmm1, %xmm0, %xmm0 # sched: [100:?]
407 ; ZNVER1-NEXT: vhsubps (%rdi), %xmm0, %xmm0 # sched: [100:?]
408 ; ZNVER1-NEXT: retq # sched: [1:0.50]
409 %1 = call <4 x float> @llvm.x86.sse3.hsub.ps(<4 x float> %a0, <4 x float> %a1)
410 %2 = load <4 x float>, <4 x float> *%a2, align 16
411 %3 = call <4 x float> @llvm.x86.sse3.hsub.ps(<4 x float> %1, <4 x float> %2)
412 ret <4 x float> %3
413 }
414 declare <4 x float> @llvm.x86.sse3.hsub.ps(<4 x float>, <4 x float>) nounwind readnone
416 define <16 x i8> @test_lddqu(i8* %a0) {
417 ; GENERIC-LABEL: test_lddqu:
418 ; GENERIC: # BB#0:
419 ; GENERIC-NEXT: lddqu (%rdi), %xmm0 # sched: [6:0.50]
420 ; GENERIC-NEXT: retq # sched: [1:1.00]
421 ;
422 ; ATOM-LABEL: test_lddqu:
423 ; ATOM: # BB#0:
424 ; ATOM-NEXT: lddqu (%rdi), %xmm0 # sched: [3:1.50]
425 ; ATOM-NEXT: nop # sched: [1:0.50]
426 ; ATOM-NEXT: nop # sched: [1:0.50]
427 ; ATOM-NEXT: retq # sched: [79:39.50]
428 ;
429 ; SLM-LABEL: test_lddqu:
430 ; SLM: # BB#0:
431 ; SLM-NEXT: lddqu (%rdi), %xmm0 # sched: [3:1.00]
432 ; SLM-NEXT: retq # sched: [4:1.00]
433 ;
434 ; SANDY-LABEL: test_lddqu:
435 ; SANDY: # BB#0:
436 ; SANDY-NEXT: vlddqu (%rdi), %xmm0 # sched: [6:0.50]
437 ; SANDY-NEXT: retq # sched: [1:1.00]
438 ;
439 ; HASWELL-LABEL: test_lddqu:
440 ; HASWELL: # BB#0:
441 ; HASWELL-NEXT: vlddqu (%rdi), %xmm0 # sched: [1:0.50]
442 ; HASWELL-NEXT: retq # sched: [2:1.00]
443 ;
444 ; BROADWELL-LABEL: test_lddqu:
445 ; BROADWELL: # BB#0:
446 ; BROADWELL-NEXT: vlddqu (%rdi), %xmm0 # sched: [5:0.50]
447 ; BROADWELL-NEXT: retq # sched: [7:1.00]
448 ;
449 ; SKYLAKE-LABEL: test_lddqu:
450 ; SKYLAKE: # BB#0:
451 ; SKYLAKE-NEXT: vlddqu (%rdi), %xmm0 # sched: [6:0.50]
452 ; SKYLAKE-NEXT: retq # sched: [7:1.00]
453 ;
454 ; SKX-LABEL: test_lddqu:
455 ; SKX: # BB#0:
456 ; SKX-NEXT: vlddqu (%rdi), %xmm0 # sched: [6:0.50]
457 ; SKX-NEXT: retq # sched: [7:1.00]
458 ;
459 ; BTVER2-LABEL: test_lddqu:
460 ; BTVER2: # BB#0:
461 ; BTVER2-NEXT: vlddqu (%rdi), %xmm0 # sched: [5:1.00]
462 ; BTVER2-NEXT: retq # sched: [4:1.00]
463 ;
464 ; ZNVER1-LABEL: test_lddqu:
465 ; ZNVER1: # BB#0:
466 ; ZNVER1-NEXT: vlddqu (%rdi), %xmm0 # sched: [8:0.50]
467 ; ZNVER1-NEXT: retq # sched: [1:0.50]
468 %1 = call <16 x i8> @llvm.x86.sse3.ldu.dq(i8* %a0)
469 ret <16 x i8> %1
470 }
471 declare <16 x i8> @llvm.x86.sse3.ldu.dq(i8*) nounwind readonly
473 define void @test_monitor(i8* %a0, i32 %a1, i32 %a2) {
474 ; GENERIC-LABEL: test_monitor:
475 ; GENERIC: # BB#0:
476 ; GENERIC-NEXT: leaq (%rdi), %rax # sched: [1:0.50]
477 ; GENERIC-NEXT: movl %esi, %ecx # sched: [1:0.33]
478 ; GENERIC-NEXT: monitor # sched: [100:0.33]
479 ; GENERIC-NEXT: retq # sched: [1:1.00]
480 ;
481 ; ATOM-LABEL: test_monitor:
482 ; ATOM: # BB#0:
483 ; ATOM-NEXT: leaq (%rdi), %rax # sched: [1:1.00]
484 ; ATOM-NEXT: movl %esi, %ecx # sched: [1:0.50]
485 ; ATOM-NEXT: monitor # sched: [45:22.50]
486 ; ATOM-NEXT: retq # sched: [79:39.50]
487 ;
488 ; SLM-LABEL: test_monitor:
489 ; SLM: # BB#0:
490 ; SLM-NEXT: leaq (%rdi), %rax # sched: [1:1.00]
491 ; SLM-NEXT: movl %esi, %ecx # sched: [1:0.50]
492 ; SLM-NEXT: monitor # sched: [100:1.00]
493 ; SLM-NEXT: retq # sched: [4:1.00]
494 ;
495 ; SANDY-LABEL: test_monitor:
496 ; SANDY: # BB#0:
497 ; SANDY-NEXT: leaq (%rdi), %rax # sched: [1:0.50]
498 ; SANDY-NEXT: movl %esi, %ecx # sched: [1:0.33]
499 ; SANDY-NEXT: monitor # sched: [100:0.33]
500 ; SANDY-NEXT: retq # sched: [1:1.00]
501 ;
502 ; HASWELL-LABEL: test_monitor:
503 ; HASWELL: # BB#0:
504 ; HASWELL-NEXT: leaq (%rdi), %rax # sched: [1:0.50]
505 ; HASWELL-NEXT: movl %esi, %ecx # sched: [1:0.25]
506 ; HASWELL-NEXT: monitor # sched: [100:0.25]
507 ; HASWELL-NEXT: retq # sched: [2:1.00]
508 ;
509 ; BROADWELL-LABEL: test_monitor:
510 ; BROADWELL: # BB#0:
511 ; BROADWELL-NEXT: leaq (%rdi), %rax # sched: [1:0.50]
512 ; BROADWELL-NEXT: movl %esi, %ecx # sched: [1:0.25]
513 ; BROADWELL-NEXT: monitor # sched: [100:0.25]
514 ; BROADWELL-NEXT: retq # sched: [7:1.00]
515 ;
516 ; SKYLAKE-LABEL: test_monitor:
517 ; SKYLAKE: # BB#0:
518 ; SKYLAKE-NEXT: leaq (%rdi), %rax # sched: [1:0.50]
519 ; SKYLAKE-NEXT: movl %esi, %ecx # sched: [1:0.25]
520 ; SKYLAKE-NEXT: monitor # sched: [100:0.25]
521 ; SKYLAKE-NEXT: retq # sched: [7:1.00]
522 ;
523 ; SKX-LABEL: test_monitor:
524 ; SKX: # BB#0:
525 ; SKX-NEXT: leaq (%rdi), %rax # sched: [1:0.50]
526 ; SKX-NEXT: movl %esi, %ecx # sched: [1:0.25]
527 ; SKX-NEXT: monitor # sched: [100:0.25]
528 ; SKX-NEXT: retq # sched: [7:1.00]
529 ;
530 ; BTVER2-LABEL: test_monitor:
531 ; BTVER2: # BB#0:
532 ; BTVER2-NEXT: leaq (%rdi), %rax # sched: [1:0.50]
533 ; BTVER2-NEXT: movl %esi, %ecx # sched: [1:0.17]
534 ; BTVER2-NEXT: monitor # sched: [100:0.17]
535 ; BTVER2-NEXT: retq # sched: [4:1.00]
536 ;
537 ; ZNVER1-LABEL: test_monitor:
538 ; ZNVER1: # BB#0:
539 ; ZNVER1-NEXT: leaq (%rdi), %rax # sched: [1:0.25]
540 ; ZNVER1-NEXT: movl %esi, %ecx # sched: [1:0.25]
541 ; ZNVER1-NEXT: monitor # sched: [100:?]
542 ; ZNVER1-NEXT: retq # sched: [1:0.50]
543 tail call void @llvm.x86.sse3.monitor(i8* %a0, i32 %a1, i32 %a2)
544 ret void
545 }
546 declare void @llvm.x86.sse3.monitor(i8*, i32, i32)
548 define <2 x double> @test_movddup(<2 x double> %a0, <2 x double> *%a1) {
549 ; GENERIC-LABEL: test_movddup:
550 ; GENERIC: # BB#0:
551 ; GENERIC-NEXT: movddup {{.*#+}} xmm1 = xmm0[0,0] sched: [1:1.00]
552 ; GENERIC-NEXT: movddup {{.*#+}} xmm0 = mem[0,0] sched: [6:0.50]
553 ; GENERIC-NEXT: subpd %xmm1, %xmm0 # sched: [3:1.00]
554 ; GENERIC-NEXT: retq # sched: [1:1.00]
555 ;
556 ; ATOM-LABEL: test_movddup:
557 ; ATOM: # BB#0:
558 ; ATOM-NEXT: movddup {{.*#+}} xmm1 = mem[0,0] sched: [1:1.00]
559 ; ATOM-NEXT: movddup {{.*#+}} xmm0 = xmm0[0,0] sched: [1:1.00]
560 ; ATOM-NEXT: subpd %xmm0, %xmm1 # sched: [6:3.00]
561 ; ATOM-NEXT: movapd %xmm1, %xmm0 # sched: [1:0.50]
562 ; ATOM-NEXT: retq # sched: [79:39.50]
563 ;
564 ; SLM-LABEL: test_movddup:
565 ; SLM: # BB#0:
566 ; SLM-NEXT: movddup {{.*#+}} xmm1 = xmm0[0,0] sched: [1:1.00]
567 ; SLM-NEXT: movddup {{.*#+}} xmm0 = mem[0,0] sched: [3:1.00]
568 ; SLM-NEXT: subpd %xmm1, %xmm0 # sched: [3:1.00]
569 ; SLM-NEXT: retq # sched: [4:1.00]
570 ;
571 ; SANDY-LABEL: test_movddup:
572 ; SANDY: # BB#0:
573 ; SANDY-NEXT: vmovddup {{.*#+}} xmm0 = xmm0[0,0] sched: [1:1.00]
574 ; SANDY-NEXT: vmovddup {{.*#+}} xmm1 = mem[0,0] sched: [6:0.50]
575 ; SANDY-NEXT: vsubpd %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
576 ; SANDY-NEXT: retq # sched: [1:1.00]
577 ;
578 ; HASWELL-LABEL: test_movddup:
579 ; HASWELL: # BB#0:
580 ; HASWELL-NEXT: vmovddup {{.*#+}} xmm0 = xmm0[0,0] sched: [1:1.00]
581 ; HASWELL-NEXT: vmovddup {{.*#+}} xmm1 = mem[0,0] sched: [1:0.50]
582 ; HASWELL-NEXT: vsubpd %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
583 ; HASWELL-NEXT: retq # sched: [2:1.00]
584 ;
585 ; BROADWELL-LABEL: test_movddup:
586 ; BROADWELL: # BB#0:
587 ; BROADWELL-NEXT: vmovddup {{.*#+}} xmm0 = xmm0[0,0] sched: [1:1.00]
588 ; BROADWELL-NEXT: vmovddup {{.*#+}} xmm1 = mem[0,0] sched: [5:0.50]
589 ; BROADWELL-NEXT: vsubpd %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
590 ; BROADWELL-NEXT: retq # sched: [7:1.00]
591 ;
592 ; SKYLAKE-LABEL: test_movddup:
593 ; SKYLAKE: # BB#0:
594 ; SKYLAKE-NEXT: vmovddup {{.*#+}} xmm0 = xmm0[0,0] sched: [1:1.00]
595 ; SKYLAKE-NEXT: vmovddup {{.*#+}} xmm1 = mem[0,0] sched: [5:0.50]
596 ; SKYLAKE-NEXT: vsubpd %xmm0, %xmm1, %xmm0 # sched: [4:0.50]
597 ; SKYLAKE-NEXT: retq # sched: [7:1.00]
598 ;
599 ; SKX-LABEL: test_movddup:
600 ; SKX: # BB#0:
601 ; SKX-NEXT: vmovddup {{.*#+}} xmm0 = xmm0[0,0] sched: [1:1.00]
602 ; SKX-NEXT: vmovddup {{.*#+}} xmm1 = mem[0,0] sched: [5:0.50]
603 ; SKX-NEXT: vsubpd %xmm0, %xmm1, %xmm0 # sched: [4:0.33]
604 ; SKX-NEXT: retq # sched: [7:1.00]
605 ;
606 ; BTVER2-LABEL: test_movddup:
607 ; BTVER2: # BB#0:
608 ; BTVER2-NEXT: vmovddup {{.*#+}} xmm1 = mem[0,0] sched: [5:1.00]
609 ; BTVER2-NEXT: vmovddup {{.*#+}} xmm0 = xmm0[0,0] sched: [1:0.50]
610 ; BTVER2-NEXT: vsubpd %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
611 ; BTVER2-NEXT: retq # sched: [4:1.00]
612 ;
613 ; ZNVER1-LABEL: test_movddup:
614 ; ZNVER1: # BB#0:
615 ; ZNVER1-NEXT: vmovddup {{.*#+}} xmm1 = mem[0,0] sched: [8:0.50]
616 ; ZNVER1-NEXT: vmovddup {{.*#+}} xmm0 = xmm0[0,0] sched: [1:0.50]
617 ; ZNVER1-NEXT: vsubpd %xmm0, %xmm1, %xmm0 # sched: [3:1.00]
618 ; ZNVER1-NEXT: retq # sched: [1:0.50]
619 %1 = shufflevector <2 x double> %a0, <2 x double> undef, <2 x i32> zeroinitializer
620 %2 = load <2 x double>, <2 x double> *%a1, align 16
621 %3 = shufflevector <2 x double> %2, <2 x double> undef, <2 x i32> zeroinitializer
622 %4 = fsub <2 x double> %3, %1 ; Use fsub to stop the movddup from being folded as a broadcast load in avx512vl.
623 ret <2 x double> %4
624 }
626 define <4 x float> @test_movshdup(<4 x float> %a0, <4 x float> *%a1) {
627 ; GENERIC-LABEL: test_movshdup:
628 ; GENERIC: # BB#0:
629 ; GENERIC-NEXT: movshdup {{.*#+}} xmm1 = xmm0[1,1,3,3] sched: [1:1.00]
630 ; GENERIC-NEXT: movshdup {{.*#+}} xmm0 = mem[1,1,3,3] sched: [6:0.50]
631 ; GENERIC-NEXT: addps %xmm1, %xmm0 # sched: [3:1.00]
632 ; GENERIC-NEXT: retq # sched: [1:1.00]
633 ;
634 ; ATOM-LABEL: test_movshdup:
635 ; ATOM: # BB#0:
636 ; ATOM-NEXT: movshdup {{.*#+}} xmm1 = mem[1,1,3,3] sched: [1:1.00]
637 ; ATOM-NEXT: movshdup {{.*#+}} xmm0 = xmm0[1,1,3,3] sched: [1:1.00]
638 ; ATOM-NEXT: addps %xmm0, %xmm1 # sched: [5:5.00]
639 ; ATOM-NEXT: movaps %xmm1, %xmm0 # sched: [1:0.50]
640 ; ATOM-NEXT: retq # sched: [79:39.50]
641 ;
642 ; SLM-LABEL: test_movshdup:
643 ; SLM: # BB#0:
644 ; SLM-NEXT: movshdup {{.*#+}} xmm1 = xmm0[1,1,3,3] sched: [1:1.00]
645 ; SLM-NEXT: movshdup {{.*#+}} xmm0 = mem[1,1,3,3] sched: [3:1.00]
646 ; SLM-NEXT: addps %xmm1, %xmm0 # sched: [3:1.00]
647 ; SLM-NEXT: retq # sched: [4:1.00]
648 ;
649 ; SANDY-LABEL: test_movshdup:
650 ; SANDY: # BB#0:
651 ; SANDY-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3] sched: [1:1.00]
652 ; SANDY-NEXT: vmovshdup {{.*#+}} xmm1 = mem[1,1,3,3] sched: [6:0.50]
653 ; SANDY-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
654 ; SANDY-NEXT: retq # sched: [1:1.00]
655 ;
656 ; HASWELL-LABEL: test_movshdup:
657 ; HASWELL: # BB#0:
658 ; HASWELL-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3] sched: [1:1.00]
659 ; HASWELL-NEXT: vmovshdup {{.*#+}} xmm1 = mem[1,1,3,3] sched: [1:0.50]
660 ; HASWELL-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
661 ; HASWELL-NEXT: retq # sched: [2:1.00]
662 ;
663 ; BROADWELL-LABEL: test_movshdup:
664 ; BROADWELL: # BB#0:
665 ; BROADWELL-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3] sched: [1:1.00]
666 ; BROADWELL-NEXT: vmovshdup {{.*#+}} xmm1 = mem[1,1,3,3] sched: [5:0.50]
667 ; BROADWELL-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
668 ; BROADWELL-NEXT: retq # sched: [7:1.00]
669 ;
670 ; SKYLAKE-LABEL: test_movshdup:
671 ; SKYLAKE: # BB#0:
672 ; SKYLAKE-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3] sched: [1:1.00]
673 ; SKYLAKE-NEXT: vmovshdup {{.*#+}} xmm1 = mem[1,1,3,3] sched: [6:0.50]
674 ; SKYLAKE-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [4:0.50]
675 ; SKYLAKE-NEXT: retq # sched: [7:1.00]
676 ;
677 ; SKX-LABEL: test_movshdup:
678 ; SKX: # BB#0:
679 ; SKX-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3] sched: [1:1.00]
680 ; SKX-NEXT: vmovshdup {{.*#+}} xmm1 = mem[1,1,3,3] sched: [6:0.50]
681 ; SKX-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [4:0.33]
682 ; SKX-NEXT: retq # sched: [7:1.00]
683 ;
684 ; BTVER2-LABEL: test_movshdup:
685 ; BTVER2: # BB#0:
686 ; BTVER2-NEXT: vmovshdup {{.*#+}} xmm1 = mem[1,1,3,3] sched: [5:1.00]
687 ; BTVER2-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3] sched: [1:0.50]
688 ; BTVER2-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
689 ; BTVER2-NEXT: retq # sched: [4:1.00]
690 ;
691 ; ZNVER1-LABEL: test_movshdup:
692 ; ZNVER1: # BB#0:
693 ; ZNVER1-NEXT: vmovshdup {{.*#+}} xmm1 = mem[1,1,3,3] sched: [8:0.50]
694 ; ZNVER1-NEXT: vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3] sched: [1:0.50]
695 ; ZNVER1-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
696 ; ZNVER1-NEXT: retq # sched: [1:0.50]
697 %1 = shufflevector <4 x float> %a0, <4 x float> undef, <4 x i32> <i32 1, i32 1, i32 3, i32 3>
698 %2 = load <4 x float>, <4 x float> *%a1, align 16
699 %3 = shufflevector <4 x float> %2, <4 x float> undef, <4 x i32> <i32 1, i32 1, i32 3, i32 3>
700 %4 = fadd <4 x float> %1, %3
701 ret <4 x float> %4
702 }
704 define <4 x float> @test_movsldup(<4 x float> %a0, <4 x float> *%a1) {
705 ; GENERIC-LABEL: test_movsldup:
706 ; GENERIC: # BB#0:
707 ; GENERIC-NEXT: movsldup {{.*#+}} xmm1 = xmm0[0,0,2,2] sched: [1:1.00]
708 ; GENERIC-NEXT: movsldup {{.*#+}} xmm0 = mem[0,0,2,2] sched: [6:0.50]
709 ; GENERIC-NEXT: addps %xmm1, %xmm0 # sched: [3:1.00]
710 ; GENERIC-NEXT: retq # sched: [1:1.00]
711 ;
712 ; ATOM-LABEL: test_movsldup:
713 ; ATOM: # BB#0:
714 ; ATOM-NEXT: movsldup {{.*#+}} xmm1 = mem[0,0,2,2] sched: [1:1.00]
715 ; ATOM-NEXT: movsldup {{.*#+}} xmm0 = xmm0[0,0,2,2] sched: [1:1.00]
716 ; ATOM-NEXT: addps %xmm0, %xmm1 # sched: [5:5.00]
717 ; ATOM-NEXT: movaps %xmm1, %xmm0 # sched: [1:0.50]
718 ; ATOM-NEXT: retq # sched: [79:39.50]
719 ;
720 ; SLM-LABEL: test_movsldup:
721 ; SLM: # BB#0:
722 ; SLM-NEXT: movsldup {{.*#+}} xmm1 = xmm0[0,0,2,2] sched: [1:1.00]
723 ; SLM-NEXT: movsldup {{.*#+}} xmm0 = mem[0,0,2,2] sched: [3:1.00]
724 ; SLM-NEXT: addps %xmm1, %xmm0 # sched: [3:1.00]
725 ; SLM-NEXT: retq # sched: [4:1.00]
726 ;
727 ; SANDY-LABEL: test_movsldup:
728 ; SANDY: # BB#0:
729 ; SANDY-NEXT: vmovsldup {{.*#+}} xmm0 = xmm0[0,0,2,2] sched: [1:1.00]
730 ; SANDY-NEXT: vmovsldup {{.*#+}} xmm1 = mem[0,0,2,2] sched: [6:0.50]
731 ; SANDY-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
732 ; SANDY-NEXT: retq # sched: [1:1.00]
733 ;
734 ; HASWELL-LABEL: test_movsldup:
735 ; HASWELL: # BB#0:
736 ; HASWELL-NEXT: vmovsldup {{.*#+}} xmm0 = xmm0[0,0,2,2] sched: [1:1.00]
737 ; HASWELL-NEXT: vmovsldup {{.*#+}} xmm1 = mem[0,0,2,2] sched: [1:0.50]
738 ; HASWELL-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
739 ; HASWELL-NEXT: retq # sched: [2:1.00]
740 ;
741 ; BROADWELL-LABEL: test_movsldup:
742 ; BROADWELL: # BB#0:
743 ; BROADWELL-NEXT: vmovsldup {{.*#+}} xmm0 = xmm0[0,0,2,2] sched: [1:1.00]
744 ; BROADWELL-NEXT: vmovsldup {{.*#+}} xmm1 = mem[0,0,2,2] sched: [5:0.50]
745 ; BROADWELL-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
746 ; BROADWELL-NEXT: retq # sched: [7:1.00]
747 ;
748 ; SKYLAKE-LABEL: test_movsldup:
749 ; SKYLAKE: # BB#0:
750 ; SKYLAKE-NEXT: vmovsldup {{.*#+}} xmm0 = xmm0[0,0,2,2] sched: [1:1.00]
751 ; SKYLAKE-NEXT: vmovsldup {{.*#+}} xmm1 = mem[0,0,2,2] sched: [6:0.50]
752 ; SKYLAKE-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [4:0.50]
753 ; SKYLAKE-NEXT: retq # sched: [7:1.00]
754 ;
755 ; SKX-LABEL: test_movsldup:
756 ; SKX: # BB#0:
757 ; SKX-NEXT: vmovsldup {{.*#+}} xmm0 = xmm0[0,0,2,2] sched: [1:1.00]
758 ; SKX-NEXT: vmovsldup {{.*#+}} xmm1 = mem[0,0,2,2] sched: [6:0.50]
759 ; SKX-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [4:0.33]
760 ; SKX-NEXT: retq # sched: [7:1.00]
761 ;
762 ; BTVER2-LABEL: test_movsldup:
763 ; BTVER2: # BB#0:
764 ; BTVER2-NEXT: vmovsldup {{.*#+}} xmm1 = mem[0,0,2,2] sched: [5:1.00]
765 ; BTVER2-NEXT: vmovsldup {{.*#+}} xmm0 = xmm0[0,0,2,2] sched: [1:0.50]
766 ; BTVER2-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
767 ; BTVER2-NEXT: retq # sched: [4:1.00]
768 ;
769 ; ZNVER1-LABEL: test_movsldup:
770 ; ZNVER1: # BB#0:
771 ; ZNVER1-NEXT: vmovsldup {{.*#+}} xmm1 = mem[0,0,2,2] sched: [8:0.50]
772 ; ZNVER1-NEXT: vmovsldup {{.*#+}} xmm0 = xmm0[0,0,2,2] sched: [1:0.50]
773 ; ZNVER1-NEXT: vaddps %xmm1, %xmm0, %xmm0 # sched: [3:1.00]
774 ; ZNVER1-NEXT: retq # sched: [1:0.50]
775 %1 = shufflevector <4 x float> %a0, <4 x float> undef, <4 x i32> <i32 0, i32 0, i32 2, i32 2>
776 %2 = load <4 x float>, <4 x float> *%a1, align 16
777 %3 = shufflevector <4 x float> %2, <4 x float> undef, <4 x i32> <i32 0, i32 0, i32 2, i32 2>
778 %4 = fadd <4 x float> %1, %3
779 ret <4 x float> %4
780 }
782 define void @test_mwait(i32 %a0, i32 %a1) {
783 ; GENERIC-LABEL: test_mwait:
784 ; GENERIC: # BB#0:
785 ; GENERIC-NEXT: movl %edi, %ecx # sched: [1:0.33]
786 ; GENERIC-NEXT: movl %esi, %eax # sched: [1:0.33]
787 ; GENERIC-NEXT: mwait # sched: [100:0.33]
788 ; GENERIC-NEXT: retq # sched: [1:1.00]
789 ;
790 ; ATOM-LABEL: test_mwait:
791 ; ATOM: # BB#0:
792 ; ATOM-NEXT: movl %edi, %ecx # sched: [1:0.50]
793 ; ATOM-NEXT: movl %esi, %eax # sched: [1:0.50]
794 ; ATOM-NEXT: mwait # sched: [46:23.00]
795 ; ATOM-NEXT: retq # sched: [79:39.50]
796 ;
797 ; SLM-LABEL: test_mwait:
798 ; SLM: # BB#0:
799 ; SLM-NEXT: movl %edi, %ecx # sched: [1:0.50]
800 ; SLM-NEXT: movl %esi, %eax # sched: [1:0.50]
801 ; SLM-NEXT: mwait # sched: [100:1.00]
802 ; SLM-NEXT: retq # sched: [4:1.00]
803 ;
804 ; SANDY-LABEL: test_mwait:
805 ; SANDY: # BB#0:
806 ; SANDY-NEXT: movl %edi, %ecx # sched: [1:0.33]
807 ; SANDY-NEXT: movl %esi, %eax # sched: [1:0.33]
808 ; SANDY-NEXT: mwait # sched: [100:0.33]
809 ; SANDY-NEXT: retq # sched: [1:1.00]
810 ;
811 ; HASWELL-LABEL: test_mwait:
812 ; HASWELL: # BB#0:
813 ; HASWELL-NEXT: movl %edi, %ecx # sched: [1:0.25]
814 ; HASWELL-NEXT: movl %esi, %eax # sched: [1:0.25]
815 ; HASWELL-NEXT: mwait # sched: [20:2.50]
816 ; HASWELL-NEXT: retq # sched: [2:1.00]
817 ;
818 ; BROADWELL-LABEL: test_mwait:
819 ; BROADWELL: # BB#0:
820 ; BROADWELL-NEXT: movl %edi, %ecx # sched: [1:0.25]
821 ; BROADWELL-NEXT: movl %esi, %eax # sched: [1:0.25]
822 ; BROADWELL-NEXT: mwait # sched: [100:0.25]
823 ; BROADWELL-NEXT: retq # sched: [7:1.00]
824 ;
825 ; SKYLAKE-LABEL: test_mwait:
826 ; SKYLAKE: # BB#0:
827 ; SKYLAKE-NEXT: movl %edi, %ecx # sched: [1:0.25]
828 ; SKYLAKE-NEXT: movl %esi, %eax # sched: [1:0.25]
829 ; SKYLAKE-NEXT: mwait # sched: [20:2.50]
830 ; SKYLAKE-NEXT: retq # sched: [7:1.00]
831 ;
832 ; SKX-LABEL: test_mwait:
833 ; SKX: # BB#0:
834 ; SKX-NEXT: movl %edi, %ecx # sched: [1:0.25]
835 ; SKX-NEXT: movl %esi, %eax # sched: [1:0.25]
836 ; SKX-NEXT: mwait # sched: [20:2.50]
837 ; SKX-NEXT: retq # sched: [7:1.00]
838 ;
839 ; BTVER2-LABEL: test_mwait:
840 ; BTVER2: # BB#0:
841 ; BTVER2-NEXT: movl %edi, %ecx # sched: [1:0.17]
842 ; BTVER2-NEXT: movl %esi, %eax # sched: [1:0.17]
843 ; BTVER2-NEXT: mwait # sched: [100:0.17]
844 ; BTVER2-NEXT: retq # sched: [4:1.00]
845 ;
846 ; ZNVER1-LABEL: test_mwait:
847 ; ZNVER1: # BB#0:
848 ; ZNVER1-NEXT: movl %edi, %ecx # sched: [1:0.25]
849 ; ZNVER1-NEXT: movl %esi, %eax # sched: [1:0.25]
850 ; ZNVER1-NEXT: mwait # sched: [100:?]
851 ; ZNVER1-NEXT: retq # sched: [1:0.50]
852 tail call void @llvm.x86.sse3.mwait(i32 %a0, i32 %a1)
853 ret void
854 }
855 declare void @llvm.x86.sse3.mwait(i32, i32)