Mercurial > hg > CbC > CbC_llvm

comparison llvm/test/CodeGen/AMDGPU/sdwa-ops.mir @ 223:5f17cb93ff66 llvm-original

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
LLVM13 (2021/7/18)
author Shinji KONO <kono@ie.u-ryukyu.ac.jp>
date Sun, 18 Jul 2021 22:43:00 +0900
parents 79ff65ed7e25
children 1f2b6ac9f198
comparison
equal deleted inserted replaced
222:81f6424ef0e3 223:5f17cb93ff66
27 %23:vgpr_32 = V_AND_B32_e32 %22, %0, implicit $exec 27 %23:vgpr_32 = V_AND_B32_e32 %22, %0, implicit $exec
28 %30:vreg_64 = COPY $sgpr0_sgpr1 28 %30:vreg_64 = COPY $sgpr0_sgpr1
29 %63:vgpr_32, %65:sreg_64_xexec = nsw V_ADD_CO_U32_e64 %30.sub0, %23, 0, implicit $exec 29 %63:vgpr_32, %65:sreg_64_xexec = nsw V_ADD_CO_U32_e64 %30.sub0, %23, 0, implicit $exec
30 %64:vgpr_32, dead %66:sreg_64_xexec = nuw V_ADDC_U32_e64 %30.sub1, %0, killed %65, 0, implicit $exec 30 %64:vgpr_32, dead %66:sreg_64_xexec = nuw V_ADDC_U32_e64 %30.sub1, %0, killed %65, 0, implicit $exec
31 %62:vreg_64 = REG_SEQUENCE %63, %subreg.sub0, %64, %subreg.sub1 31 %62:vreg_64 = REG_SEQUENCE %63, %subreg.sub0, %64, %subreg.sub1
32 GLOBAL_STORE_DWORDX2_SADDR %30.sub0, %62, %1, 0, 0, implicit $exec, implicit $exec :: (store 8) 32 GLOBAL_STORE_DWORDX2_SADDR %30.sub0, %62, %1, 0, 0, implicit $exec, implicit $exec :: (store (s64))
33 33
34 %161:vgpr_32 = V_AND_B32_e32 %22, %0, implicit $exec 34 %161:vgpr_32 = V_AND_B32_e32 %22, %0, implicit $exec
35 %163:vgpr_32, %165:sreg_64_xexec = V_ADD_CO_U32_e64 %30.sub0, %161, 0, implicit $exec 35 %163:vgpr_32, %165:sreg_64_xexec = V_ADD_CO_U32_e64 %30.sub0, %161, 0, implicit $exec
36 %164:vgpr_32, dead %166:sreg_64_xexec = V_ADDC_U32_e64 %30.sub1, %0, killed %165, 0, implicit $exec 36 %164:vgpr_32, dead %166:sreg_64_xexec = V_ADDC_U32_e64 %30.sub1, %0, killed %165, 0, implicit $exec
37 %162:vreg_64 = REG_SEQUENCE %163, %subreg.sub0, %164, %subreg.sub1 37 %162:vreg_64 = REG_SEQUENCE %163, %subreg.sub0, %164, %subreg.sub1
38 GLOBAL_STORE_DWORDX2_SADDR %30.sub0, %162, %1, 0, 0, implicit $exec, implicit $exec :: (store 8) 38 GLOBAL_STORE_DWORDX2_SADDR %30.sub0, %162, %1, 0, 0, implicit $exec, implicit $exec :: (store (s64))
39 39
40 %171:vgpr_32 = V_AND_B32_e32 %22, %0, implicit $exec 40 %171:vgpr_32 = V_AND_B32_e32 %22, %0, implicit $exec
41 %173:vgpr_32, %175:sreg_64_xexec = V_ADD_CO_U32_e64 %30.sub0, %171, 0, implicit $exec 41 %173:vgpr_32, %175:sreg_64_xexec = V_ADD_CO_U32_e64 %30.sub0, %171, 0, implicit $exec
42 %174:vgpr_32, dead %176:sreg_64_xexec = V_ADDC_U32_e64 %30.sub1, %0, killed %175, 0, implicit $exec 42 %174:vgpr_32, dead %176:sreg_64_xexec = V_ADDC_U32_e64 %30.sub1, %0, killed %175, 0, implicit $exec
43 %172:vreg_64 = REG_SEQUENCE %173, %subreg.sub0, %174, %subreg.sub1 43 %172:vreg_64 = REG_SEQUENCE %173, %subreg.sub0, %174, %subreg.sub1
44 GLOBAL_STORE_DWORDX2_SADDR %30.sub0, %172, %1, 0, 0, implicit $exec, implicit $exec :: (store 8) 44 GLOBAL_STORE_DWORDX2_SADDR %30.sub0, %172, %1, 0, 0, implicit $exec, implicit $exec :: (store (s64))
45 45
46 ... 46 ...
47 47
48 # test for VCC interference on sdwa, should generate 1 xform only 48 # test for VCC interference on sdwa, should generate 1 xform only
49 # GFX9-LABEL: name: test2_add_co_sdwa 49 # GFX9-LABEL: name: test2_add_co_sdwa
75 %164:vgpr_32, dead %166:sreg_64_xexec = V_ADDC_U32_e64 %30.sub1, %0, killed %165, 0, implicit $exec 75 %164:vgpr_32, dead %166:sreg_64_xexec = V_ADDC_U32_e64 %30.sub1, %0, killed %165, 0, implicit $exec
76 %162:vreg_64 = REG_SEQUENCE %163, %subreg.sub0, %164, %subreg.sub1 76 %162:vreg_64 = REG_SEQUENCE %163, %subreg.sub0, %164, %subreg.sub1
77 77
78 %64:vgpr_32, dead %66:sreg_64_xexec = V_ADDC_U32_e64 %30.sub1, %0, killed %65, 0, implicit $exec 78 %64:vgpr_32, dead %66:sreg_64_xexec = V_ADDC_U32_e64 %30.sub1, %0, killed %65, 0, implicit $exec
79 %62:vreg_64 = REG_SEQUENCE %63, %subreg.sub0, %64, %subreg.sub1 79 %62:vreg_64 = REG_SEQUENCE %63, %subreg.sub0, %64, %subreg.sub1
80 GLOBAL_STORE_DWORDX2_SADDR %30.sub0, %62, %1, 0, 0, implicit $exec, implicit $exec :: (store 8) 80 GLOBAL_STORE_DWORDX2_SADDR %30.sub0, %62, %1, 0, 0, implicit $exec, implicit $exec :: (store (s64))
81 81
82 %161:vgpr_32 = V_AND_B32_e32 %22, %0, implicit $exec 82 %161:vgpr_32 = V_AND_B32_e32 %22, %0, implicit $exec
83 %163:vgpr_32, %165:sreg_64_xexec = V_ADD_CO_U32_e64 %30.sub0, %161, 0, implicit $exec 83 %163:vgpr_32, %165:sreg_64_xexec = V_ADD_CO_U32_e64 %30.sub0, %161, 0, implicit $exec
84 %164:vgpr_32, dead %166:sreg_64_xexec = V_ADDC_U32_e64 %30.sub1, %0, killed %165, 0, implicit $exec 84 %164:vgpr_32, dead %166:sreg_64_xexec = V_ADDC_U32_e64 %30.sub1, %0, killed %165, 0, implicit $exec
85 %162:vreg_64 = REG_SEQUENCE %163, %subreg.sub0, %164, %subreg.sub1 85 %162:vreg_64 = REG_SEQUENCE %163, %subreg.sub0, %164, %subreg.sub1
86 GLOBAL_STORE_DWORDX2_SADDR %30.sub0, %162, %1, 0, 0, implicit $exec, implicit $exec :: (store 8) 86 GLOBAL_STORE_DWORDX2_SADDR %30.sub0, %162, %1, 0, 0, implicit $exec, implicit $exec :: (store (s64))
87 87
88 ... 88 ...
89 89
90 # test for CarryOut used, should reject 90 # test for CarryOut used, should reject
91 # GFX9-LABEL: name: test3_add_co_sdwa 91 # GFX9-LABEL: name: test3_add_co_sdwa
111 %23:vgpr_32 = V_AND_B32_e32 %22, %0, implicit $exec 111 %23:vgpr_32 = V_AND_B32_e32 %22, %0, implicit $exec
112 %30:vreg_64 = COPY $sgpr0_sgpr1 112 %30:vreg_64 = COPY $sgpr0_sgpr1
113 %63:vgpr_32, %65:sreg_64_xexec = V_ADD_CO_U32_e64 %30.sub0, %23, 0, implicit $exec 113 %63:vgpr_32, %65:sreg_64_xexec = V_ADD_CO_U32_e64 %30.sub0, %23, 0, implicit $exec
114 %64:vgpr_32, %66:sreg_64_xexec = V_ADDC_U32_e64 %30.sub1, %0, killed %65, 0, implicit $exec 114 %64:vgpr_32, %66:sreg_64_xexec = V_ADDC_U32_e64 %30.sub1, %0, killed %65, 0, implicit $exec
115 %62:vreg_64 = REG_SEQUENCE %63, %subreg.sub0, %66, %subreg.sub1 115 %62:vreg_64 = REG_SEQUENCE %63, %subreg.sub0, %66, %subreg.sub1
116 GLOBAL_STORE_DWORDX2_SADDR %30.sub0, %62, %1, 0, 0, implicit $exec, implicit $exec :: (store 8) 116 GLOBAL_STORE_DWORDX2_SADDR %30.sub0, %62, %1, 0, 0, implicit $exec, implicit $exec :: (store (s64))
117 117
118 ... 118 ...
119 119
120 # test for CarryIn used more than once, should reject 120 # test for CarryIn used more than once, should reject
121 # GFX9-LABEL: name: test4_add_co_sdwa 121 # GFX9-LABEL: name: test4_add_co_sdwa
141 %23:vgpr_32 = V_AND_B32_e32 %22, %0, implicit $exec 141 %23:vgpr_32 = V_AND_B32_e32 %22, %0, implicit $exec
142 %30:vreg_64 = COPY $sgpr0_sgpr1 142 %30:vreg_64 = COPY $sgpr0_sgpr1
143 %63:vgpr_32, %65:sreg_64_xexec = V_ADD_CO_U32_e64 %30.sub0, %23, 0, implicit $exec 143 %63:vgpr_32, %65:sreg_64_xexec = V_ADD_CO_U32_e64 %30.sub0, %23, 0, implicit $exec
144 %64:vgpr_32, %66:sreg_64_xexec = V_ADDC_U32_e64 %30.sub1, %0, %65, 0, implicit $exec 144 %64:vgpr_32, %66:sreg_64_xexec = V_ADDC_U32_e64 %30.sub1, %0, %65, 0, implicit $exec
145 %62:vreg_64 = REG_SEQUENCE %63, %subreg.sub0, %65, %subreg.sub1 145 %62:vreg_64 = REG_SEQUENCE %63, %subreg.sub0, %65, %subreg.sub1
146 GLOBAL_STORE_DWORDX2_SADDR %30.sub0, %62, %1, 0, 0, implicit $exec, implicit $exec :: (store 8) 146 GLOBAL_STORE_DWORDX2_SADDR %30.sub0, %62, %1, 0, 0, implicit $exec, implicit $exec :: (store (s64))
147 147
148 148
149 ... 149 ...
150 150
151 # test for simple example, should generate sdwa 151 # test for simple example, should generate sdwa
170 %23:vgpr_32 = V_AND_B32_e32 %22, %0, implicit $exec 170 %23:vgpr_32 = V_AND_B32_e32 %22, %0, implicit $exec
171 %30:vreg_64 = COPY $sgpr0_sgpr1 171 %30:vreg_64 = COPY $sgpr0_sgpr1
172 %63:vgpr_32, %65:sreg_64_xexec = V_ADD_CO_U32_e64 %30.sub0, %23, 0, implicit $exec 172 %63:vgpr_32, %65:sreg_64_xexec = V_ADD_CO_U32_e64 %30.sub0, %23, 0, implicit $exec
173 %64:vgpr_32, %66:sreg_64_xexec = V_ADDC_U32_e64 %30.sub1, %0, %65, 0, implicit $exec 173 %64:vgpr_32, %66:sreg_64_xexec = V_ADDC_U32_e64 %30.sub1, %0, %65, 0, implicit $exec
174 %62:vreg_64 = REG_SEQUENCE %63, %subreg.sub0, %64, %subreg.sub1 174 %62:vreg_64 = REG_SEQUENCE %63, %subreg.sub0, %64, %subreg.sub1
175 GLOBAL_STORE_DWORDX2_SADDR %30.sub0, %62, %1, 0, 0, implicit $exec, implicit $exec :: (store 8) 175 GLOBAL_STORE_DWORDX2_SADDR %30.sub0, %62, %1, 0, 0, implicit $exec, implicit $exec :: (store (s64))
176 176
177 177
178 ... 178 ...
179 179
180 # test for V_ADD_CO_U32_e64 only, should reject 180 # test for V_ADD_CO_U32_e64 only, should reject
199 %22:sreg_32_xm0 = S_MOV_B32 255 199 %22:sreg_32_xm0 = S_MOV_B32 255
200 %23:vgpr_32 = V_AND_B32_e32 %22, %0, implicit $exec 200 %23:vgpr_32 = V_AND_B32_e32 %22, %0, implicit $exec
201 %30:vreg_64 = COPY $sgpr0_sgpr1 201 %30:vreg_64 = COPY $sgpr0_sgpr1
202 %63:vgpr_32, %65:sreg_64_xexec = V_ADD_CO_U32_e64 %30.sub0, %23, 0, implicit $exec 202 %63:vgpr_32, %65:sreg_64_xexec = V_ADD_CO_U32_e64 %30.sub0, %23, 0, implicit $exec
203 %62:vreg_64 = REG_SEQUENCE %63, %subreg.sub0, %23, %subreg.sub1 203 %62:vreg_64 = REG_SEQUENCE %63, %subreg.sub0, %23, %subreg.sub1
204 GLOBAL_STORE_DWORDX2_SADDR %30.sub0, %62, %1, 0, 0, implicit $exec, implicit $exec :: (store 8) 204 GLOBAL_STORE_DWORDX2_SADDR %30.sub0, %62, %1, 0, 0, implicit $exec, implicit $exec :: (store (s64))
205 205
206 206
207 ... 207 ...
208 208
209 # test for V_ADDC_U32_e64 only, should reject 209 # test for V_ADDC_U32_e64 only, should reject
230 %24:sreg_64_xexec = COPY $sgpr0_sgpr1 230 %24:sreg_64_xexec = COPY $sgpr0_sgpr1
231 231
232 %30:vreg_64 = COPY $sgpr0_sgpr1 232 %30:vreg_64 = COPY $sgpr0_sgpr1
233 %64:vgpr_32, %66:sreg_64_xexec = V_ADDC_U32_e64 %30.sub1, %0, %24, 0, implicit $exec 233 %64:vgpr_32, %66:sreg_64_xexec = V_ADDC_U32_e64 %30.sub1, %0, %24, 0, implicit $exec
234 %62:vreg_64 = REG_SEQUENCE %23, %subreg.sub0, %23, %subreg.sub1 234 %62:vreg_64 = REG_SEQUENCE %23, %subreg.sub0, %23, %subreg.sub1
235 GLOBAL_STORE_DWORDX2_SADDR %30.sub0, %62, %1, 0, 0, implicit $exec, implicit $exec :: (store 8) 235 GLOBAL_STORE_DWORDX2_SADDR %30.sub0, %62, %1, 0, 0, implicit $exec, implicit $exec :: (store (s64))
236 236
237 237
238 ... 238 ...
239 239
240 # test for $vcc defined between two adds, should not generate 240 # test for $vcc defined between two adds, should not generate
261 %63:vgpr_32, %65:sreg_64_xexec = V_ADD_CO_U32_e64 %30.sub0, %23, 0, implicit $exec 261 %63:vgpr_32, %65:sreg_64_xexec = V_ADD_CO_U32_e64 %30.sub0, %23, 0, implicit $exec
262 $vcc = COPY %30 262 $vcc = COPY %30
263 %64:vgpr_32, %66:sreg_64_xexec = V_ADDC_U32_e64 %30.sub1, %0, %65, 0, implicit $exec 263 %64:vgpr_32, %66:sreg_64_xexec = V_ADDC_U32_e64 %30.sub1, %0, %65, 0, implicit $exec
264 %31:vreg_64 = COPY $vcc 264 %31:vreg_64 = COPY $vcc
265 %62:vreg_64 = REG_SEQUENCE %63, %subreg.sub0, %64, %subreg.sub1 265 %62:vreg_64 = REG_SEQUENCE %63, %subreg.sub0, %64, %subreg.sub1
266 GLOBAL_STORE_DWORDX2_SADDR %31.sub0, %62, %1, 0, 0, implicit $exec, implicit $exec :: (store 8) 266 GLOBAL_STORE_DWORDX2_SADDR %31.sub0, %62, %1, 0, 0, implicit $exec, implicit $exec :: (store (s64))
267 267
268 268
269 ... 269 ...
270 270
271 # test for non dead $vcc, should not generate 271 # test for non dead $vcc, should not generate
292 %23:vgpr_32 = V_AND_B32_e32 %22, %0, implicit $exec 292 %23:vgpr_32 = V_AND_B32_e32 %22, %0, implicit $exec
293 %63:vgpr_32, %65:sreg_64_xexec = V_ADD_CO_U32_e64 %30.sub0, %23, 0, implicit $exec 293 %63:vgpr_32, %65:sreg_64_xexec = V_ADD_CO_U32_e64 %30.sub0, %23, 0, implicit $exec
294 %64:vgpr_32, %66:sreg_64_xexec = V_ADDC_U32_e64 %30.sub1, %0, %65, 0, implicit $exec 294 %64:vgpr_32, %66:sreg_64_xexec = V_ADDC_U32_e64 %30.sub1, %0, %65, 0, implicit $exec
295 %31:vreg_64 = COPY $vcc 295 %31:vreg_64 = COPY $vcc
296 %62:vreg_64 = REG_SEQUENCE %63, %subreg.sub0, %64, %subreg.sub1 296 %62:vreg_64 = REG_SEQUENCE %63, %subreg.sub0, %64, %subreg.sub1
297 GLOBAL_STORE_DWORDX2_SADDR %31.sub0, %62, %1, 0, 0, implicit $exec, implicit $exec :: (store 8) 297 GLOBAL_STORE_DWORDX2_SADDR %31.sub0, %62, %1, 0, 0, implicit $exec, implicit $exec :: (store (s64))
298 298
299 299
300 ... 300 ...
301 301
302 # test for def $vcc_lo, should not generate 302 # test for def $vcc_lo, should not generate
324 %63:vgpr_32, %65:sreg_64_xexec = V_ADD_CO_U32_e64 %30.sub0, %23, 0, implicit $exec 324 %63:vgpr_32, %65:sreg_64_xexec = V_ADD_CO_U32_e64 %30.sub0, %23, 0, implicit $exec
325 %31:vgpr_32 = COPY $vcc_lo 325 %31:vgpr_32 = COPY $vcc_lo
326 %32:vreg_64 = REG_SEQUENCE %31, %subreg.sub0, %23, %subreg.sub1 326 %32:vreg_64 = REG_SEQUENCE %31, %subreg.sub0, %23, %subreg.sub1
327 %64:vgpr_32, %66:sreg_64_xexec = V_ADDC_U32_e64 %30.sub1, %0, %65, 0, implicit $exec 327 %64:vgpr_32, %66:sreg_64_xexec = V_ADDC_U32_e64 %30.sub1, %0, %65, 0, implicit $exec
328 %62:vreg_64 = REG_SEQUENCE %63, %subreg.sub0, %64, %subreg.sub1 328 %62:vreg_64 = REG_SEQUENCE %63, %subreg.sub0, %64, %subreg.sub1
329 GLOBAL_STORE_DWORDX2_SADDR %32.sub0, %62, %1, 0, 0, implicit $exec, implicit $exec :: (store 8) 329 GLOBAL_STORE_DWORDX2_SADDR %32.sub0, %62, %1, 0, 0, implicit $exec, implicit $exec :: (store (s64))
330 330
331 ... 331 ...
332 332
333 # test for read $vcc_hi, should not generate 333 # test for read $vcc_hi, should not generate
334 # GFX9-LABEL: name: test11_add_co_sdwa 334 # GFX9-LABEL: name: test11_add_co_sdwa
355 %63:vgpr_32, %65:sreg_64_xexec = V_ADD_CO_U32_e64 %30.sub0, %23, 0, implicit $exec 355 %63:vgpr_32, %65:sreg_64_xexec = V_ADD_CO_U32_e64 %30.sub0, %23, 0, implicit $exec
356 %31:vgpr_32 = COPY $vcc_hi 356 %31:vgpr_32 = COPY $vcc_hi
357 %32:vreg_64 = REG_SEQUENCE %31, %subreg.sub0, %23, %subreg.sub1 357 %32:vreg_64 = REG_SEQUENCE %31, %subreg.sub0, %23, %subreg.sub1
358 %64:vgpr_32, %66:sreg_64_xexec = V_ADDC_U32_e64 %30.sub1, %0, %65, 0, implicit $exec 358 %64:vgpr_32, %66:sreg_64_xexec = V_ADDC_U32_e64 %30.sub1, %0, %65, 0, implicit $exec
359 %62:vreg_64 = REG_SEQUENCE %63, %subreg.sub0, %64, %subreg.sub1 359 %62:vreg_64 = REG_SEQUENCE %63, %subreg.sub0, %64, %subreg.sub1
360 GLOBAL_STORE_DWORDX2_SADDR %32.sub0, %62, %1, 0, 0, implicit $exec, implicit $exec :: (store 8) 360 GLOBAL_STORE_DWORDX2_SADDR %32.sub0, %62, %1, 0, 0, implicit $exec, implicit $exec :: (store (s64))
361 361
362 ... 362 ...
363 363
364 # test for $vcc defined and used between adds, should not generate 364 # test for $vcc defined and used between adds, should not generate
365 # GFX9-LABEL: name: test12_add_co_sdwa 365 # GFX9-LABEL: name: test12_add_co_sdwa
385 %63:vgpr_32, %65:sreg_64_xexec = V_ADD_CO_U32_e64 %30.sub0, %23, 0, implicit $exec 385 %63:vgpr_32, %65:sreg_64_xexec = V_ADD_CO_U32_e64 %30.sub0, %23, 0, implicit $exec
386 $vcc = COPY %30 386 $vcc = COPY %30
387 %31:vreg_64 = COPY killed $vcc 387 %31:vreg_64 = COPY killed $vcc
388 %64:vgpr_32, %66:sreg_64_xexec = V_ADDC_U32_e64 %30.sub1, %0, %65, 0, implicit $exec 388 %64:vgpr_32, %66:sreg_64_xexec = V_ADDC_U32_e64 %30.sub1, %0, %65, 0, implicit $exec
389 %62:vreg_64 = REG_SEQUENCE %63, %subreg.sub0, %64, %subreg.sub1 389 %62:vreg_64 = REG_SEQUENCE %63, %subreg.sub0, %64, %subreg.sub1
390 GLOBAL_STORE_DWORDX2_SADDR %31.sub0, %62, %1, 0, 0, implicit $exec, implicit $exec :: (store 8) 390 GLOBAL_STORE_DWORDX2_SADDR %31.sub0, %62, %1, 0, 0, implicit $exec, implicit $exec :: (store (s64))
391 391