?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
[X86] Use NSW/NUW flags on ISD::TRUNCATE nodes to improve X86 PACKSS/PACKUS lowering...
[llvm-project.git] / llvm / test / tools / llvm-mca / X86 / sqrt-rsqrt-rcp-memop.s
blobabf01cb0b10f955b3a63ccc466475d1a0ec6d284
1 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
2 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=1 -all-views=false -timeline < %s | FileCheck %s -check-prefixes=ALL,BARCELONA
3 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=1 -all-views=false -timeline < %s | FileCheck %s -check-prefixes=ALL,BDVER2
4 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=btver2 -iterations=1 -all-views=false -timeline < %s | FileCheck %s -check-prefixes=ALL,BTVER2
5 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=znver1 -iterations=1 -all-views=false -timeline < %s | FileCheck %s -check-prefixes=ALL,ZNVER1
6 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=znver2 -iterations=1 -all-views=false -timeline < %s | FileCheck %s -check-prefixes=ALL,ZNVER2
7 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=haswell -iterations=1 -all-views=false -timeline < %s | FileCheck %s -check-prefixes=ALL,HASWELL
8 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=broadwell -iterations=1 -all-views=false -timeline < %s | FileCheck %s -check-prefixes=ALL,BROADWELL
9 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=skylake -iterations=1 -all-views=false -timeline < %s | FileCheck %s -check-prefixes=ALL,SKYLAKE
10 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=icelake-server -iterations=1 -all-views=false -timeline < %s | FileCheck %s -check-prefixes=ALL,SKYLAKE
11
12 #LLVM-MCA-BEGIN test_sqrtss
13 leaq 8(%rsp, %rdi, 2), %rax
14 sqrtss (%rax), %xmm1
15 #LLVM-MCA-END
16
17 #LLVM-MCA-BEGIN test_sqrtsd
18 leaq 8(%rsp, %rdi, 2), %rax
19 sqrtsd (%rax), %xmm1
20 #LLVM-MCA-END
21
22 #LLVM-MCA-BEGIN test_rsqrtss
23 leaq 8(%rsp, %rdi, 2), %rax
24 rsqrtss (%rax), %xmm1
25 #LLVM-MCA-END
26
27 #LLVM-MCA-BEGIN test_rcp
28 leaq 8(%rsp, %rdi, 2), %rax
29 rcpss (%rax), %xmm1
30 #LLVM-MCA-END
31
32 # ALL: [0] Code Region - test_sqrtss
33
34 # ALL: Timeline view:
35
36 # BARCELONA-NEXT: 0123456789
37 # BARCELONA-NEXT: Index 0123456789 0123
38
39 # BDVER2-NEXT: 012345678
40 # BDVER2-NEXT: Index 0123456789
41
42 # BROADWELL-NEXT: 0123456789
43 # BROADWELL-NEXT: Index 0123456789
44
45 # BTVER2-NEXT: 0123456789 0
46 # BTVER2-NEXT: Index 0123456789 0123456789
47
48 # HASWELL-NEXT: 0123456789
49 # HASWELL-NEXT: Index 0123456789
50
51 # SKYLAKE-NEXT: 0123456789
52 # SKYLAKE-NEXT: Index 0123456789 0
53
54 # ZNVER1-NEXT: 0123456789
55 # ZNVER1-NEXT: Index 0123456789 01234
56
57 # ZNVER2-NEXT: 0123456789
58 # ZNVER2-NEXT: Index 0123456789 01234
59
60 # BARCELONA: [0,0] DeER . . . . . leaq 8(%rsp,%rdi,2), %rax
61 # BARCELONA-NEXT: [0,1] D=eeeeeeeeeeeeeeeeeeeeER sqrtss (%rax), %xmm1
62
63 # BDVER2: [0,0] DeeER. . . . leaq 8(%rsp,%rdi,2), %rax
64 # BDVER2-NEXT: [0,1] D==eeeeeeeeeeeeeeER sqrtss (%rax), %xmm1
65
66 # BROADWELL: [0,0] DeER . . . . leaq 8(%rsp,%rdi,2), %rax
67 # BROADWELL-NEXT: [0,1] D=eeeeeeeeeeeeeeeeER sqrtss (%rax), %xmm1
68
69 # BTVER2: [0,0] DeeER. . . . . . leaq 8(%rsp,%rdi,2), %rax
70 # BTVER2-NEXT: [0,1] D==eeeeeeeeeeeeeeeeeeeeeeeeeeER sqrtss (%rax), %xmm1
71
72 # HASWELL: [0,0] DeER . . . . leaq 8(%rsp,%rdi,2), %rax
73 # HASWELL-NEXT: [0,1] D=eeeeeeeeeeeeeeeeER sqrtss (%rax), %xmm1
74
75 # SKYLAKE: [0,0] DeER . . . . leaq 8(%rsp,%rdi,2), %rax
76 # SKYLAKE-NEXT: [0,1] D=eeeeeeeeeeeeeeeeeER sqrtss (%rax), %xmm1
77
78 # ZNVER1: [0,0] DeER . . . . . leaq 8(%rsp,%rdi,2), %rax
79 # ZNVER1-NEXT: [0,1] D=eeeeeeeeeeeeeeeeeeeeeER sqrtss (%rax), %xmm1
80
81 # ZNVER2: [0,0] DeER . . . . . leaq 8(%rsp,%rdi,2), %rax
82 # ZNVER2-NEXT: [0,1] D=eeeeeeeeeeeeeeeeeeeeeER sqrtss (%rax), %xmm1
83
84 # ALL: Average Wait times (based on the timeline view):
85 # ALL-NEXT: [0]: Executions
86 # ALL-NEXT: [1]: Average time spent waiting in a scheduler's queue
87 # ALL-NEXT: [2]: Average time spent waiting in a scheduler's queue while ready
88 # ALL-NEXT: [3]: Average time elapsed from WB until retire stage
89
90 # ALL: [0] [1] [2] [3]
91 # ALL-NEXT: 0. 1 1.0 1.0 0.0 leaq 8(%rsp,%rdi,2), %rax
92
93 # BARCELONA-NEXT: 1. 1 2.0 0.0 0.0 sqrtss (%rax), %xmm1
94 # BARCELONA-NEXT: 1 1.5 0.5 0.0 <total>
95
96 # BDVER2-NEXT: 1. 1 3.0 0.0 0.0 sqrtss (%rax), %xmm1
97 # BDVER2-NEXT: 1 2.0 0.5 0.0 <total>
98
99 # BROADWELL-NEXT: 1. 1 2.0 0.0 0.0 sqrtss (%rax), %xmm1
100 # BROADWELL-NEXT: 1 1.5 0.5 0.0 <total>
101
102 # BTVER2-NEXT: 1. 1 3.0 0.0 0.0 sqrtss (%rax), %xmm1
103 # BTVER2-NEXT: 1 2.0 0.5 0.0 <total>
104
105 # HASWELL-NEXT: 1. 1 2.0 0.0 0.0 sqrtss (%rax), %xmm1
106 # HASWELL-NEXT: 1 1.5 0.5 0.0 <total>
107
108 # SKYLAKE-NEXT: 1. 1 2.0 0.0 0.0 sqrtss (%rax), %xmm1
109 # SKYLAKE-NEXT: 1 1.5 0.5 0.0 <total>
110
111 # ZNVER1-NEXT: 1. 1 2.0 0.0 0.0 sqrtss (%rax), %xmm1
112 # ZNVER1-NEXT: 1 1.5 0.5 0.0 <total>
113
114 # ZNVER2-NEXT: 1. 1 2.0 0.0 0.0 sqrtss (%rax), %xmm1
115 # ZNVER2-NEXT: 1 1.5 0.5 0.0 <total>
116
117 # ALL: [1] Code Region - test_sqrtsd
118
119 # ALL: Timeline view:
120
121 # BARCELONA-NEXT: 0123456789 0
122 # BARCELONA-NEXT: Index 0123456789 0123456789
123
124 # BDVER2-NEXT: 012345678
125 # BDVER2-NEXT: Index 0123456789
126
127 # BROADWELL-NEXT: 0123456789
128 # BROADWELL-NEXT: Index 0123456789 01234
129
130 # BTVER2-NEXT: 0123456789 0123456
131 # BTVER2-NEXT: Index 0123456789 0123456789
132
133 # HASWELL-NEXT: 0123456789
134 # HASWELL-NEXT: Index 0123456789 01234
135
136 # SKYLAKE-NEXT: 0123456789
137 # SKYLAKE-NEXT: Index 0123456789 0123456
138
139 # ZNVER1-NEXT: 0123456789 0
140 # ZNVER1-NEXT: Index 0123456789 0123456789
141
142 # ZNVER2-NEXT: 0123456789 0
143 # ZNVER2-NEXT: Index 0123456789 0123456789
144
145 # BARCELONA: [0,0] DeER . . . . . . leaq 8(%rsp,%rdi,2), %rax
146 # BARCELONA-NEXT: [0,1] D=eeeeeeeeeeeeeeeeeeeeeeeeeeeER sqrtsd (%rax), %xmm1
147
148 # BDVER2: [0,0] DeeER. . . . leaq 8(%rsp,%rdi,2), %rax
149 # BDVER2-NEXT: [0,1] D==eeeeeeeeeeeeeeER sqrtsd (%rax), %xmm1
150
151 # BROADWELL: [0,0] DeER . . . . . leaq 8(%rsp,%rdi,2), %rax
152 # BROADWELL-NEXT: [0,1] D=eeeeeeeeeeeeeeeeeeeeeER sqrtsd (%rax), %xmm1
153
154 # BTVER2: [0,0] DeeER. . . . . . .. leaq 8(%rsp,%rdi,2), %rax
155 # BTVER2-NEXT: [0,1] D==eeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeER sqrtsd (%rax), %xmm1
156
157 # HASWELL: [0,0] DeER . . . . . leaq 8(%rsp,%rdi,2), %rax
158 # HASWELL-NEXT: [0,1] D=eeeeeeeeeeeeeeeeeeeeeER sqrtsd (%rax), %xmm1
159
160 # SKYLAKE: [0,0] DeER . . . . .. leaq 8(%rsp,%rdi,2), %rax
161 # SKYLAKE-NEXT: [0,1] D=eeeeeeeeeeeeeeeeeeeeeeeER sqrtsd (%rax), %xmm1
162
163 # ZNVER1: [0,0] DeER . . . . . . leaq 8(%rsp,%rdi,2), %rax
164 # ZNVER1-NEXT: [0,1] D=eeeeeeeeeeeeeeeeeeeeeeeeeeeER sqrtsd (%rax), %xmm1
165
166 # ZNVER2: [0,0] DeER . . . . . . leaq 8(%rsp,%rdi,2), %rax
167 # ZNVER2-NEXT: [0,1] D=eeeeeeeeeeeeeeeeeeeeeeeeeeeER sqrtsd (%rax), %xmm1
168
169 # ALL: Average Wait times (based on the timeline view):
170 # ALL-NEXT: [0]: Executions
171 # ALL-NEXT: [1]: Average time spent waiting in a scheduler's queue
172 # ALL-NEXT: [2]: Average time spent waiting in a scheduler's queue while ready
173 # ALL-NEXT: [3]: Average time elapsed from WB until retire stage
174
175 # ALL: [0] [1] [2] [3]
176 # ALL-NEXT: 0. 1 1.0 1.0 0.0 leaq 8(%rsp,%rdi,2), %rax
177
178 # BARCELONA-NEXT: 1. 1 2.0 0.0 0.0 sqrtsd (%rax), %xmm1
179 # BARCELONA-NEXT: 1 1.5 0.5 0.0 <total>
180
181 # BDVER2-NEXT: 1. 1 3.0 0.0 0.0 sqrtsd (%rax), %xmm1
182 # BDVER2-NEXT: 1 2.0 0.5 0.0 <total>
183
184 # BROADWELL-NEXT: 1. 1 2.0 0.0 0.0 sqrtsd (%rax), %xmm1
185 # BROADWELL-NEXT: 1 1.5 0.5 0.0 <total>
186
187 # BTVER2-NEXT: 1. 1 3.0 0.0 0.0 sqrtsd (%rax), %xmm1
188 # BTVER2-NEXT: 1 2.0 0.5 0.0 <total>
189
190 # HASWELL-NEXT: 1. 1 2.0 0.0 0.0 sqrtsd (%rax), %xmm1
191 # HASWELL-NEXT: 1 1.5 0.5 0.0 <total>
192
193 # SKYLAKE-NEXT: 1. 1 2.0 0.0 0.0 sqrtsd (%rax), %xmm1
194 # SKYLAKE-NEXT: 1 1.5 0.5 0.0 <total>
195
196 # ZNVER1-NEXT: 1. 1 2.0 0.0 0.0 sqrtsd (%rax), %xmm1
197 # ZNVER1-NEXT: 1 1.5 0.5 0.0 <total>
198
199 # ZNVER2-NEXT: 1. 1 2.0 0.0 0.0 sqrtsd (%rax), %xmm1
200 # ZNVER2-NEXT: 1 1.5 0.5 0.0 <total>
201
202 # ALL: [2] Code Region - test_rsqrtss
203
204 # ALL: Timeline view:
205
206 # BARCELONA-NEXT: 01234
207 # BDVER2-NEXT: 01234
208 # BROADWELL-NEXT: 0123
209 # BTVER2-NEXT: 01
210 # HASWELL-NEXT: 0123
211 # SKYLAKE-NEXT: 012
212 # ZNVER1-NEXT: 012345
213 # ZNVER2-NEXT: 012345
214
215 # ALL-NEXT: Index 0123456789
216
217 # BARCELONA: [0,0] DeER . . . leaq 8(%rsp,%rdi,2), %rax
218 # BARCELONA-NEXT: [0,1] D=eeeeeeeeeeeER rsqrtss (%rax), %xmm1
219
220 # BDVER2: [0,0] DeeER. . . leaq 8(%rsp,%rdi,2), %rax
221 # BDVER2-NEXT: [0,1] D==eeeeeeeeeeER rsqrtss (%rax), %xmm1
222
223 # BROADWELL: [0,0] DeER . . . leaq 8(%rsp,%rdi,2), %rax
224 # BROADWELL-NEXT: [0,1] D=eeeeeeeeeeER rsqrtss (%rax), %xmm1
225
226 # BTVER2: [0,0] DeeER. .. leaq 8(%rsp,%rdi,2), %rax
227 # BTVER2-NEXT: [0,1] D==eeeeeeeER rsqrtss (%rax), %xmm1
228
229 # HASWELL: [0,0] DeER . . . leaq 8(%rsp,%rdi,2), %rax
230 # HASWELL-NEXT: [0,1] D=eeeeeeeeeeER rsqrtss (%rax), %xmm1
231
232 # SKYLAKE: [0,0] DeER . . . leaq 8(%rsp,%rdi,2), %rax
233 # SKYLAKE-NEXT: [0,1] D=eeeeeeeeeER rsqrtss (%rax), %xmm1
234
235 # ZNVER1: [0,0] DeER . . . leaq 8(%rsp,%rdi,2), %rax
236 # ZNVER1-NEXT: [0,1] D=eeeeeeeeeeeeER rsqrtss (%rax), %xmm1
237
238 # ZNVER2: [0,0] DeER . . . leaq 8(%rsp,%rdi,2), %rax
239 # ZNVER2-NEXT: [0,1] D=eeeeeeeeeeeeER rsqrtss (%rax), %xmm1
240
241 # ALL: Average Wait times (based on the timeline view):
242 # ALL-NEXT: [0]: Executions
243 # ALL-NEXT: [1]: Average time spent waiting in a scheduler's queue
244 # ALL-NEXT: [2]: Average time spent waiting in a scheduler's queue while ready
245 # ALL-NEXT: [3]: Average time elapsed from WB until retire stage
246
247 # ALL: [0] [1] [2] [3]
248 # ALL-NEXT: 0. 1 1.0 1.0 0.0 leaq 8(%rsp,%rdi,2), %rax
249
250 # BARCELONA-NEXT: 1. 1 2.0 0.0 0.0 rsqrtss (%rax), %xmm1
251 # BARCELONA-NEXT: 1 1.5 0.5 0.0 <total>
252
253 # BDVER2-NEXT: 1. 1 3.0 0.0 0.0 rsqrtss (%rax), %xmm1
254 # BDVER2-NEXT: 1 2.0 0.5 0.0 <total>
255
256 # BROADWELL-NEXT: 1. 1 2.0 0.0 0.0 rsqrtss (%rax), %xmm1
257 # BROADWELL-NEXT: 1 1.5 0.5 0.0 <total>
258
259 # BTVER2-NEXT: 1. 1 3.0 0.0 0.0 rsqrtss (%rax), %xmm1
260 # BTVER2-NEXT: 1 2.0 0.5 0.0 <total>
261
262 # HASWELL-NEXT: 1. 1 2.0 0.0 0.0 rsqrtss (%rax), %xmm1
263 # HASWELL-NEXT: 1 1.5 0.5 0.0 <total>
264
265 # SKYLAKE-NEXT: 1. 1 2.0 0.0 0.0 rsqrtss (%rax), %xmm1
266 # SKYLAKE-NEXT: 1 1.5 0.5 0.0 <total>
267
268 # ZNVER1-NEXT: 1. 1 2.0 0.0 0.0 rsqrtss (%rax), %xmm1
269 # ZNVER1-NEXT: 1 1.5 0.5 0.0 <total>
270
271 # ZNVER2-NEXT: 1. 1 2.0 0.0 0.0 rsqrtss (%rax), %xmm1
272 # ZNVER2-NEXT: 1 1.5 0.5 0.0 <total>
273
274 # ALL: [3] Code Region - test_rcp
275
276 # ALL: Timeline view:
277
278 # BARCELONA-NEXT: 01234
279 # BDVER2-NEXT: 01234
280 # BROADWELL-NEXT: 0123
281 # BTVER2-NEXT: 01
282 # HASWELL-NEXT: 0123
283 # SKYLAKE-NEXT: 012
284 # ZNVER1-NEXT: 012345
285 # ZNVER2-NEXT: 012345
286
287 # ALL-NEXT: Index 0123456789
288
289 # BARCELONA: [0,0] DeER . . . leaq 8(%rsp,%rdi,2), %rax
290 # BARCELONA-NEXT: [0,1] D=eeeeeeeeeeeER rcpss (%rax), %xmm1
291
292 # BDVER2: [0,0] DeeER. . . leaq 8(%rsp,%rdi,2), %rax
293 # BDVER2-NEXT: [0,1] D==eeeeeeeeeeER rcpss (%rax), %xmm1
294
295 # BROADWELL: [0,0] DeER . . . leaq 8(%rsp,%rdi,2), %rax
296 # BROADWELL-NEXT: [0,1] D=eeeeeeeeeeER rcpss (%rax), %xmm1
297
298 # BTVER2: [0,0] DeeER. .. leaq 8(%rsp,%rdi,2), %rax
299 # BTVER2-NEXT: [0,1] D==eeeeeeeER rcpss (%rax), %xmm1
300
301 # HASWELL: [0,0] DeER . . . leaq 8(%rsp,%rdi,2), %rax
302 # HASWELL-NEXT: [0,1] D=eeeeeeeeeeER rcpss (%rax), %xmm1
303
304 # SKYLAKE: [0,0] DeER . . . leaq 8(%rsp,%rdi,2), %rax
305 # SKYLAKE-NEXT: [0,1] D=eeeeeeeeeER rcpss (%rax), %xmm1
306
307 # ZNVER1: [0,0] DeER . . . leaq 8(%rsp,%rdi,2), %rax
308 # ZNVER1-NEXT: [0,1] D=eeeeeeeeeeeeER rcpss (%rax), %xmm1
309
310 # ZNVER2: [0,0] DeER . . . leaq 8(%rsp,%rdi,2), %rax
311 # ZNVER2-NEXT: [0,1] D=eeeeeeeeeeeeER rcpss (%rax), %xmm1
312
313 # ALL: Average Wait times (based on the timeline view):
314 # ALL-NEXT: [0]: Executions
315 # ALL-NEXT: [1]: Average time spent waiting in a scheduler's queue
316 # ALL-NEXT: [2]: Average time spent waiting in a scheduler's queue while ready
317 # ALL-NEXT: [3]: Average time elapsed from WB until retire stage
318
319 # ALL: [0] [1] [2] [3]
320 # ALL-NEXT: 0. 1 1.0 1.0 0.0 leaq 8(%rsp,%rdi,2), %rax
321
322 # BARCELONA-NEXT: 1. 1 2.0 0.0 0.0 rcpss (%rax), %xmm1
323 # BARCELONA-NEXT: 1 1.5 0.5 0.0 <total>
324
325 # BDVER2-NEXT: 1. 1 3.0 0.0 0.0 rcpss (%rax), %xmm1
326 # BDVER2-NEXT: 1 2.0 0.5 0.0 <total>
327
328 # BROADWELL-NEXT: 1. 1 2.0 0.0 0.0 rcpss (%rax), %xmm1
329 # BROADWELL-NEXT: 1 1.5 0.5 0.0 <total>
330
331 # BTVER2-NEXT: 1. 1 3.0 0.0 0.0 rcpss (%rax), %xmm1
332 # BTVER2-NEXT: 1 2.0 0.5 0.0 <total>
333
334 # HASWELL-NEXT: 1. 1 2.0 0.0 0.0 rcpss (%rax), %xmm1
335 # HASWELL-NEXT: 1 1.5 0.5 0.0 <total>
336
337 # SKYLAKE-NEXT: 1. 1 2.0 0.0 0.0 rcpss (%rax), %xmm1
338 # SKYLAKE-NEXT: 1 1.5 0.5 0.0 <total>
339
340 # ZNVER1-NEXT: 1. 1 2.0 0.0 0.0 rcpss (%rax), %xmm1
341 # ZNVER1-NEXT: 1 1.5 0.5 0.0 <total>
342
343 # ZNVER2-NEXT: 1. 1 2.0 0.0 0.0 rcpss (%rax), %xmm1
344 # ZNVER2-NEXT: 1 1.5 0.5 0.0 <total>
LLVM monorepo