ttsem 0.1.0__py3-none-any.whl
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- ttsem/__init__.py +137 -0
- ttsem/__main__.py +92 -0
- ttsem/_compile_runner.py +65 -0
- ttsem/_examples/e15_11519_main_e1f944a7a.json +635 -0
- ttsem/_examples/e15_repro.py +48 -0
- ttsem/_fakedriver.py +129 -0
- ttsem/_guard.py +1045 -0
- ttsem/_install.py +26 -0
- ttsem/_interp_runner.py +85 -0
- ttsem/_judge_runner.py +791 -0
- ttsem/_tritonbench_runner.py +234 -0
- ttsem/bot.py +143 -0
- ttsem/check_fixtures.py +91 -0
- ttsem/defuse.py +109 -0
- ttsem/demo.py +334 -0
- ttsem/dump_corpus.py +60 -0
- ttsem/harness.py +1671 -0
- ttsem/inductor_trace.py +88 -0
- ttsem/interp.py +336 -0
- ttsem/interp2.py +146 -0
- ttsem/ir_types.py +93 -0
- ttsem/judge.py +1068 -0
- ttsem/judge_io.py +537 -0
- ttsem/judge_shapes.py +287 -0
- ttsem/judge_tritonbench.py +724 -0
- ttsem/layout_checks.py +285 -0
- ttsem/layouts.py +976 -0
- ttsem/linear_layout.py +676 -0
- ttsem/memory.py +372 -0
- ttsem/minimize.py +523 -0
- ttsem/mlir.py +1028 -0
- ttsem/ops.py +2634 -0
- ttsem/pass_stress.py +83 -0
- ttsem/pidraces.py +221 -0
- ttsem/py.typed +0 -0
- ttsem/pytest_ttsem.py +264 -0
- ttsem/races.py +1081 -0
- ttsem/races_corpus.py +132 -0
- ttsem/sanitize.py +1036 -0
- ttsem/torch_demo.py +130 -0
- ttsem/validate.py +425 -0
- ttsem/validate_corpus.py +118 -0
- ttsem/values.py +430 -0
- ttsem-0.1.0.dist-info/METADATA +386 -0
- ttsem-0.1.0.dist-info/RECORD +48 -0
- ttsem-0.1.0.dist-info/WHEEL +4 -0
- ttsem-0.1.0.dist-info/entry_points.txt +2 -0
- ttsem-0.1.0.dist-info/licenses/LICENSE +21 -0
|
@@ -0,0 +1,635 @@
|
|
|
1
|
+
{
|
|
2
|
+
"program": "e15_repro.py",
|
|
3
|
+
"launches": [
|
|
4
|
+
{
|
|
5
|
+
"launch_id": 0,
|
|
6
|
+
"fn_name": "kernel",
|
|
7
|
+
"passes": [
|
|
8
|
+
{
|
|
9
|
+
"pass_name": "Before InlinerPass: inline{default-pipeline=canonicalize inlining-threshold=4294967295 max-iterations=4 }",
|
|
10
|
+
"verdict": "mismatch",
|
|
11
|
+
"n_diff": 320,
|
|
12
|
+
"unsupported": [],
|
|
13
|
+
"message": ""
|
|
14
|
+
},
|
|
15
|
+
{
|
|
16
|
+
"pass_name": "Before CanonicalizerPass: canonicalize{cse-between-iterations=false max-iterations=10 max-num-rewrites=-1 region-simplify=normal test-convergence=false top-down=true}",
|
|
17
|
+
"verdict": "mismatch",
|
|
18
|
+
"n_diff": 320,
|
|
19
|
+
"unsupported": [],
|
|
20
|
+
"message": ""
|
|
21
|
+
},
|
|
22
|
+
{
|
|
23
|
+
"pass_name": "Before TritonCombineOps: triton-combine",
|
|
24
|
+
"verdict": "mismatch",
|
|
25
|
+
"n_diff": 320,
|
|
26
|
+
"unsupported": [],
|
|
27
|
+
"message": ""
|
|
28
|
+
},
|
|
29
|
+
{
|
|
30
|
+
"pass_name": "Before TritonReorderBroadcast: triton-reorder-broadcast",
|
|
31
|
+
"verdict": "mismatch",
|
|
32
|
+
"n_diff": 320,
|
|
33
|
+
"unsupported": [],
|
|
34
|
+
"message": ""
|
|
35
|
+
},
|
|
36
|
+
{
|
|
37
|
+
"pass_name": "Before CSEPass: cse",
|
|
38
|
+
"verdict": "mismatch",
|
|
39
|
+
"n_diff": 320,
|
|
40
|
+
"unsupported": [],
|
|
41
|
+
"message": ""
|
|
42
|
+
},
|
|
43
|
+
{
|
|
44
|
+
"pass_name": "Before SymbolDCEPass: symbol-dce",
|
|
45
|
+
"verdict": "mismatch",
|
|
46
|
+
"n_diff": 320,
|
|
47
|
+
"unsupported": [],
|
|
48
|
+
"message": ""
|
|
49
|
+
},
|
|
50
|
+
{
|
|
51
|
+
"pass_name": "Before TritonLoopUnroll: triton-loop-unroll",
|
|
52
|
+
"verdict": "mismatch",
|
|
53
|
+
"n_diff": 320,
|
|
54
|
+
"unsupported": [],
|
|
55
|
+
"message": ""
|
|
56
|
+
},
|
|
57
|
+
{
|
|
58
|
+
"pass_name": "Before ConvertTritonToTritonGPU: convert-triton-to-tritongpu{enable-source-remat=false num-ctas=1 num-warps=4 target=cuda:120 threads-per-warp=32}",
|
|
59
|
+
"verdict": "mismatch",
|
|
60
|
+
"n_diff": 320,
|
|
61
|
+
"unsupported": [],
|
|
62
|
+
"message": ""
|
|
63
|
+
},
|
|
64
|
+
{
|
|
65
|
+
"pass_name": "Before TritonGPUCoalesce: tritongpu-coalesce",
|
|
66
|
+
"verdict": "mismatch",
|
|
67
|
+
"n_diff": 320,
|
|
68
|
+
"unsupported": [],
|
|
69
|
+
"message": ""
|
|
70
|
+
},
|
|
71
|
+
{
|
|
72
|
+
"pass_name": "Before TritonGPUF32DotTC: tritongpu-F32DotTC{emu-tf32=true}",
|
|
73
|
+
"verdict": "mismatch",
|
|
74
|
+
"n_diff": 320,
|
|
75
|
+
"unsupported": [],
|
|
76
|
+
"message": ""
|
|
77
|
+
},
|
|
78
|
+
{
|
|
79
|
+
"pass_name": "Before TritonNvidiaGPUAssignCGALayoutsPass: triton-nvidia-gpu-assign-cga-layouts",
|
|
80
|
+
"verdict": "mismatch",
|
|
81
|
+
"n_diff": 320,
|
|
82
|
+
"unsupported": [],
|
|
83
|
+
"message": ""
|
|
84
|
+
},
|
|
85
|
+
{
|
|
86
|
+
"pass_name": "Before TritonGPURemoveLayoutConversions: tritongpu-remove-layout-conversions{disable-remat-splitting=false}",
|
|
87
|
+
"verdict": "mismatch",
|
|
88
|
+
"n_diff": 320,
|
|
89
|
+
"unsupported": [],
|
|
90
|
+
"message": ""
|
|
91
|
+
},
|
|
92
|
+
{
|
|
93
|
+
"pass_name": "Before TritonNvidiaGPUOptimizeCTALocalityPass: triton-nvidia-gpu-optimize-cta-locality",
|
|
94
|
+
"verdict": "mismatch",
|
|
95
|
+
"n_diff": 320,
|
|
96
|
+
"unsupported": [],
|
|
97
|
+
"message": ""
|
|
98
|
+
},
|
|
99
|
+
{
|
|
100
|
+
"pass_name": "Before TritonGPURemoveLayoutConversions: tritongpu-remove-layout-conversions{disable-remat-splitting=false}",
|
|
101
|
+
"verdict": "mismatch",
|
|
102
|
+
"n_diff": 320,
|
|
103
|
+
"unsupported": [],
|
|
104
|
+
"message": ""
|
|
105
|
+
},
|
|
106
|
+
{
|
|
107
|
+
"pass_name": "Before TritonGPUOptimizeThreadLocality: tritongpu-optimize-thread-locality",
|
|
108
|
+
"verdict": "mismatch",
|
|
109
|
+
"n_diff": 320,
|
|
110
|
+
"unsupported": [],
|
|
111
|
+
"message": ""
|
|
112
|
+
},
|
|
113
|
+
{
|
|
114
|
+
"pass_name": "Before TritonGPUAccelerateMatmul: tritongpu-accelerate-matmul",
|
|
115
|
+
"verdict": "mismatch",
|
|
116
|
+
"n_diff": 320,
|
|
117
|
+
"unsupported": [],
|
|
118
|
+
"message": ""
|
|
119
|
+
},
|
|
120
|
+
{
|
|
121
|
+
"pass_name": "Before TritonNvidiaGPUCheckMatmulTwoCTAPass: triton-nvidia-check-matmul-two-cta",
|
|
122
|
+
"verdict": "mismatch",
|
|
123
|
+
"n_diff": 320,
|
|
124
|
+
"unsupported": [],
|
|
125
|
+
"message": ""
|
|
126
|
+
},
|
|
127
|
+
{
|
|
128
|
+
"pass_name": "Before TritonGPURemoveLayoutConversions: tritongpu-remove-layout-conversions{disable-remat-splitting=false}",
|
|
129
|
+
"verdict": "mismatch",
|
|
130
|
+
"n_diff": 320,
|
|
131
|
+
"unsupported": [],
|
|
132
|
+
"message": ""
|
|
133
|
+
},
|
|
134
|
+
{
|
|
135
|
+
"pass_name": "Before TritonGPUOptimizeDotOperands: tritongpu-optimize-dot-operands{hoist-layout-conversion=true}",
|
|
136
|
+
"verdict": "mismatch",
|
|
137
|
+
"n_diff": 320,
|
|
138
|
+
"unsupported": [],
|
|
139
|
+
"message": ""
|
|
140
|
+
},
|
|
141
|
+
{
|
|
142
|
+
"pass_name": "Before CanonicalizerPass: canonicalize{cse-between-iterations=false max-iterations=10 max-num-rewrites=-1 region-simplify=normal test-convergence=false top-down=true}",
|
|
143
|
+
"verdict": "mismatch",
|
|
144
|
+
"n_diff": 320,
|
|
145
|
+
"unsupported": [],
|
|
146
|
+
"message": ""
|
|
147
|
+
},
|
|
148
|
+
{
|
|
149
|
+
"pass_name": "Before TritonNvidiaGPUOptimizeDescriptorEncodingPass: triton-nvidia-optimize-descriptor-encoding",
|
|
150
|
+
"verdict": "mismatch",
|
|
151
|
+
"n_diff": 320,
|
|
152
|
+
"unsupported": [],
|
|
153
|
+
"message": ""
|
|
154
|
+
},
|
|
155
|
+
{
|
|
156
|
+
"pass_name": "Before TritonLoopAwareCSE: triton-loop-aware-cse",
|
|
157
|
+
"verdict": "mismatch",
|
|
158
|
+
"n_diff": 320,
|
|
159
|
+
"unsupported": [],
|
|
160
|
+
"message": ""
|
|
161
|
+
},
|
|
162
|
+
{
|
|
163
|
+
"pass_name": "Before TritonGPUFuseNestedLoops: tritongpu-fuse-nested-loops",
|
|
164
|
+
"verdict": "mismatch",
|
|
165
|
+
"n_diff": 320,
|
|
166
|
+
"unsupported": [],
|
|
167
|
+
"message": ""
|
|
168
|
+
},
|
|
169
|
+
{
|
|
170
|
+
"pass_name": "Before CanonicalizerPass: canonicalize{cse-between-iterations=false max-iterations=10 max-num-rewrites=-1 region-simplify=normal test-convergence=false top-down=true}",
|
|
171
|
+
"verdict": "unsupported",
|
|
172
|
+
"n_diff": -1,
|
|
173
|
+
"unsupported": [
|
|
174
|
+
"arith.ceildivsi"
|
|
175
|
+
],
|
|
176
|
+
"message": "arith.ceildivsi: %11 = \"arith.ceildivsi\"(%10, %1) : (i32, i32) -> i32"
|
|
177
|
+
},
|
|
178
|
+
{
|
|
179
|
+
"pass_name": "Before TritonLoopInvariantCodeMotion: triton-licm",
|
|
180
|
+
"verdict": "match",
|
|
181
|
+
"n_diff": 0,
|
|
182
|
+
"unsupported": [],
|
|
183
|
+
"message": ""
|
|
184
|
+
},
|
|
185
|
+
{
|
|
186
|
+
"pass_name": "Before TritonGPUOptimizeAccumulatorInit: tritongpu-optimize-accumulator-init",
|
|
187
|
+
"verdict": "match",
|
|
188
|
+
"n_diff": 0,
|
|
189
|
+
"unsupported": [],
|
|
190
|
+
"message": ""
|
|
191
|
+
},
|
|
192
|
+
{
|
|
193
|
+
"pass_name": "Before TritonGPUHoistTMEMAlloc: tritongpu-hoist-tmem-alloc{post-pipeline=false}",
|
|
194
|
+
"verdict": "match",
|
|
195
|
+
"n_diff": 0,
|
|
196
|
+
"unsupported": [],
|
|
197
|
+
"message": ""
|
|
198
|
+
},
|
|
199
|
+
{
|
|
200
|
+
"pass_name": "Before TritonNvidiaGPUPromoteLHSToTMemPass: tritongpu-promote-lhs-to-tmem",
|
|
201
|
+
"verdict": "match",
|
|
202
|
+
"n_diff": 0,
|
|
203
|
+
"unsupported": [],
|
|
204
|
+
"message": ""
|
|
205
|
+
},
|
|
206
|
+
{
|
|
207
|
+
"pass_name": "Before TritonGPUAssignLatencies: tritongpu-assign-latencies{num-stages=3}",
|
|
208
|
+
"verdict": "match",
|
|
209
|
+
"n_diff": 0,
|
|
210
|
+
"unsupported": [],
|
|
211
|
+
"message": ""
|
|
212
|
+
},
|
|
213
|
+
{
|
|
214
|
+
"pass_name": "Before TritonGPUScheduleLoops: tritongpu-schedule-loops",
|
|
215
|
+
"verdict": "match",
|
|
216
|
+
"n_diff": 0,
|
|
217
|
+
"unsupported": [],
|
|
218
|
+
"message": ""
|
|
219
|
+
},
|
|
220
|
+
{
|
|
221
|
+
"pass_name": "Before TritonGPUAutomaticWarpSpecialization: tritongpu-automatic-warp-specialization{num-stages=3}",
|
|
222
|
+
"verdict": "match",
|
|
223
|
+
"n_diff": 0,
|
|
224
|
+
"unsupported": [],
|
|
225
|
+
"message": ""
|
|
226
|
+
},
|
|
227
|
+
{
|
|
228
|
+
"pass_name": "Before TritonGPUNormalizeWSWhileLoops: tritongpu-normalize-ws-while-loops",
|
|
229
|
+
"verdict": "match",
|
|
230
|
+
"n_diff": 0,
|
|
231
|
+
"unsupported": [],
|
|
232
|
+
"message": ""
|
|
233
|
+
},
|
|
234
|
+
{
|
|
235
|
+
"pass_name": "Before TritonGPUPartitionScheduling: tritongpu-partition-scheduling",
|
|
236
|
+
"verdict": "match",
|
|
237
|
+
"n_diff": 0,
|
|
238
|
+
"unsupported": [],
|
|
239
|
+
"message": ""
|
|
240
|
+
},
|
|
241
|
+
{
|
|
242
|
+
"pass_name": "Before {anonymous}::VerifyWarpSpecializationPartitions: unknown<{anonymous}::VerifyWarpSpecializationPartitions>",
|
|
243
|
+
"verdict": "match",
|
|
244
|
+
"n_diff": 0,
|
|
245
|
+
"unsupported": [],
|
|
246
|
+
"message": ""
|
|
247
|
+
},
|
|
248
|
+
{
|
|
249
|
+
"pass_name": "Before NVWSHoistTmemStore: nvws-hoist-tmem-store",
|
|
250
|
+
"verdict": "match",
|
|
251
|
+
"n_diff": 0,
|
|
252
|
+
"unsupported": [],
|
|
253
|
+
"message": ""
|
|
254
|
+
},
|
|
255
|
+
{
|
|
256
|
+
"pass_name": "Before {anonymous}::VerifyWarpSpecializationPartitions: unknown<{anonymous}::VerifyWarpSpecializationPartitions>",
|
|
257
|
+
"verdict": "match",
|
|
258
|
+
"n_diff": 0,
|
|
259
|
+
"unsupported": [],
|
|
260
|
+
"message": ""
|
|
261
|
+
},
|
|
262
|
+
{
|
|
263
|
+
"pass_name": "Before NVWSInsertAref: nvws-insert-aref",
|
|
264
|
+
"verdict": "match",
|
|
265
|
+
"n_diff": 0,
|
|
266
|
+
"unsupported": [],
|
|
267
|
+
"message": ""
|
|
268
|
+
},
|
|
269
|
+
{
|
|
270
|
+
"pass_name": "Before {anonymous}::VerifyWarpSpecializationPartitions: unknown<{anonymous}::VerifyWarpSpecializationPartitions>",
|
|
271
|
+
"verdict": "match",
|
|
272
|
+
"n_diff": 0,
|
|
273
|
+
"unsupported": [],
|
|
274
|
+
"message": ""
|
|
275
|
+
},
|
|
276
|
+
{
|
|
277
|
+
"pass_name": "Before NVWSInsertTmemAref: nvws-insert-tmem-aref",
|
|
278
|
+
"verdict": "match",
|
|
279
|
+
"n_diff": 0,
|
|
280
|
+
"unsupported": [],
|
|
281
|
+
"message": ""
|
|
282
|
+
},
|
|
283
|
+
{
|
|
284
|
+
"pass_name": "Before {anonymous}::VerifyWarpSpecializationPartitions: unknown<{anonymous}::VerifyWarpSpecializationPartitions>",
|
|
285
|
+
"verdict": "match",
|
|
286
|
+
"n_diff": 0,
|
|
287
|
+
"unsupported": [],
|
|
288
|
+
"message": ""
|
|
289
|
+
},
|
|
290
|
+
{
|
|
291
|
+
"pass_name": "Before SCCPPass: sccp",
|
|
292
|
+
"verdict": "match",
|
|
293
|
+
"n_diff": 0,
|
|
294
|
+
"unsupported": [],
|
|
295
|
+
"message": ""
|
|
296
|
+
},
|
|
297
|
+
{
|
|
298
|
+
"pass_name": "Before {anonymous}::VerifyWarpSpecializationPartitions: unknown<{anonymous}::VerifyWarpSpecializationPartitions>",
|
|
299
|
+
"verdict": "match",
|
|
300
|
+
"n_diff": 0,
|
|
301
|
+
"unsupported": [],
|
|
302
|
+
"message": ""
|
|
303
|
+
},
|
|
304
|
+
{
|
|
305
|
+
"pass_name": "Before CSEPass: cse",
|
|
306
|
+
"verdict": "match",
|
|
307
|
+
"n_diff": 0,
|
|
308
|
+
"unsupported": [],
|
|
309
|
+
"message": ""
|
|
310
|
+
},
|
|
311
|
+
{
|
|
312
|
+
"pass_name": "Before {anonymous}::VerifyWarpSpecializationPartitions: unknown<{anonymous}::VerifyWarpSpecializationPartitions>",
|
|
313
|
+
"verdict": "match",
|
|
314
|
+
"n_diff": 0,
|
|
315
|
+
"unsupported": [],
|
|
316
|
+
"message": ""
|
|
317
|
+
},
|
|
318
|
+
{
|
|
319
|
+
"pass_name": "Before NVWSLowerAref: nvws-lower-aref{num-stages=3}",
|
|
320
|
+
"verdict": "match",
|
|
321
|
+
"n_diff": 0,
|
|
322
|
+
"unsupported": [],
|
|
323
|
+
"message": ""
|
|
324
|
+
},
|
|
325
|
+
{
|
|
326
|
+
"pass_name": "Before NVWSAssignStagePhase: nvws-assign-stage-phase",
|
|
327
|
+
"verdict": "match",
|
|
328
|
+
"n_diff": 0,
|
|
329
|
+
"unsupported": [],
|
|
330
|
+
"message": ""
|
|
331
|
+
},
|
|
332
|
+
{
|
|
333
|
+
"pass_name": "Before {anonymous}::VerifyWarpSpecializationPartitions: unknown<{anonymous}::VerifyWarpSpecializationPartitions>",
|
|
334
|
+
"verdict": "match",
|
|
335
|
+
"n_diff": 0,
|
|
336
|
+
"unsupported": [],
|
|
337
|
+
"message": ""
|
|
338
|
+
},
|
|
339
|
+
{
|
|
340
|
+
"pass_name": "Before TritonGPUPartitionLoops: tritongpu-partition-loops",
|
|
341
|
+
"verdict": "match",
|
|
342
|
+
"n_diff": 0,
|
|
343
|
+
"unsupported": [],
|
|
344
|
+
"message": ""
|
|
345
|
+
},
|
|
346
|
+
{
|
|
347
|
+
"pass_name": "Before NVWSLowerWarpGroup: nvws-lower-warp-group",
|
|
348
|
+
"verdict": "match",
|
|
349
|
+
"n_diff": 0,
|
|
350
|
+
"unsupported": [],
|
|
351
|
+
"message": ""
|
|
352
|
+
},
|
|
353
|
+
{
|
|
354
|
+
"pass_name": "Before TritonGPUScheduleLoops: tritongpu-schedule-loops",
|
|
355
|
+
"verdict": "match",
|
|
356
|
+
"n_diff": 0,
|
|
357
|
+
"unsupported": [],
|
|
358
|
+
"message": ""
|
|
359
|
+
},
|
|
360
|
+
{
|
|
361
|
+
"pass_name": "Before TritonGPUPipeline: tritongpu-pipeline{dump-intermediate-steps=true num-stages=3}",
|
|
362
|
+
"verdict": "match",
|
|
363
|
+
"n_diff": 0,
|
|
364
|
+
"unsupported": [],
|
|
365
|
+
"message": ""
|
|
366
|
+
},
|
|
367
|
+
{
|
|
368
|
+
"pass_name": "Before TritonGPUPipeline: tritongpu-pipeline{dump-intermediate-steps=true num-stages=3} (step 1)",
|
|
369
|
+
"verdict": "match",
|
|
370
|
+
"n_diff": 0,
|
|
371
|
+
"unsupported": [],
|
|
372
|
+
"message": ""
|
|
373
|
+
},
|
|
374
|
+
{
|
|
375
|
+
"pass_name": "Before TritonGPUPipeline: tritongpu-pipeline{dump-intermediate-steps=true num-stages=3} (step 2)",
|
|
376
|
+
"verdict": "match",
|
|
377
|
+
"n_diff": 0,
|
|
378
|
+
"unsupported": [],
|
|
379
|
+
"message": ""
|
|
380
|
+
},
|
|
381
|
+
{
|
|
382
|
+
"pass_name": "Before TritonGPUOptimizePartitionWarps: tritongpu-optimize-partition-warps",
|
|
383
|
+
"verdict": "match",
|
|
384
|
+
"n_diff": 0,
|
|
385
|
+
"unsupported": [],
|
|
386
|
+
"message": ""
|
|
387
|
+
},
|
|
388
|
+
{
|
|
389
|
+
"pass_name": "Before TritonGPUCombineTensorSelectAndIf: tritongpu-combine-tensor-select-and-if",
|
|
390
|
+
"verdict": "match",
|
|
391
|
+
"n_diff": 0,
|
|
392
|
+
"unsupported": [],
|
|
393
|
+
"message": ""
|
|
394
|
+
},
|
|
395
|
+
{
|
|
396
|
+
"pass_name": "Before TritonGPUHoistTMEMAlloc: tritongpu-hoist-tmem-alloc{post-pipeline=true}",
|
|
397
|
+
"verdict": "match",
|
|
398
|
+
"n_diff": 0,
|
|
399
|
+
"unsupported": [],
|
|
400
|
+
"message": ""
|
|
401
|
+
},
|
|
402
|
+
{
|
|
403
|
+
"pass_name": "Before TritonNvidiaGPURemoveTMEMTokensPass: triton-nvidia-gpu-remove-tmem-tokens",
|
|
404
|
+
"verdict": "match",
|
|
405
|
+
"n_diff": 0,
|
|
406
|
+
"unsupported": [],
|
|
407
|
+
"message": ""
|
|
408
|
+
},
|
|
409
|
+
{
|
|
410
|
+
"pass_name": "Before CanonicalizerPass: canonicalize{cse-between-iterations=false max-iterations=10 max-num-rewrites=-1 region-simplify=normal test-convergence=false top-down=true}",
|
|
411
|
+
"verdict": "unsupported",
|
|
412
|
+
"n_diff": -1,
|
|
413
|
+
"unsupported": [
|
|
414
|
+
"type: !ttg.async.token"
|
|
415
|
+
],
|
|
416
|
+
"message": "type: !ttg.async.token"
|
|
417
|
+
},
|
|
418
|
+
{
|
|
419
|
+
"pass_name": "Before TritonLoopAwareCSE: triton-loop-aware-cse",
|
|
420
|
+
"verdict": "match",
|
|
421
|
+
"n_diff": 0,
|
|
422
|
+
"unsupported": [],
|
|
423
|
+
"message": ""
|
|
424
|
+
},
|
|
425
|
+
{
|
|
426
|
+
"pass_name": "Before TritonGPUOptimizeDotOperands: tritongpu-optimize-dot-operands{hoist-layout-conversion=true}",
|
|
427
|
+
"verdict": "match",
|
|
428
|
+
"n_diff": 0,
|
|
429
|
+
"unsupported": [],
|
|
430
|
+
"message": ""
|
|
431
|
+
},
|
|
432
|
+
{
|
|
433
|
+
"pass_name": "Before CanonicalizerPass: canonicalize{cse-between-iterations=false max-iterations=10 max-num-rewrites=-1 region-simplify=normal test-convergence=false top-down=true}",
|
|
434
|
+
"verdict": "match",
|
|
435
|
+
"n_diff": 0,
|
|
436
|
+
"unsupported": [],
|
|
437
|
+
"message": ""
|
|
438
|
+
},
|
|
439
|
+
{
|
|
440
|
+
"pass_name": "Before TritonGPUCoalesceAsyncCopy: tritongpu-coalesce-async-copy",
|
|
441
|
+
"verdict": "match",
|
|
442
|
+
"n_diff": 0,
|
|
443
|
+
"unsupported": [],
|
|
444
|
+
"message": ""
|
|
445
|
+
},
|
|
446
|
+
{
|
|
447
|
+
"pass_name": "Before TritonNvidiaGPUOptimizeTMemLayoutsPass: triton-nvidia-optimize-tmem-layouts",
|
|
448
|
+
"verdict": "match",
|
|
449
|
+
"n_diff": 0,
|
|
450
|
+
"unsupported": [],
|
|
451
|
+
"message": ""
|
|
452
|
+
},
|
|
453
|
+
{
|
|
454
|
+
"pass_name": "Before TritonNvidiaGPUFuseTMEMLoadReducePass: triton-nvidia-tmem-load-reduce",
|
|
455
|
+
"verdict": "match",
|
|
456
|
+
"n_diff": 0,
|
|
457
|
+
"unsupported": [],
|
|
458
|
+
"message": ""
|
|
459
|
+
},
|
|
460
|
+
{
|
|
461
|
+
"pass_name": "Before TritonNvidiaGPUTMALoweringPass: triton-nvidia-tma-lowering",
|
|
462
|
+
"verdict": "match",
|
|
463
|
+
"n_diff": 0,
|
|
464
|
+
"unsupported": [],
|
|
465
|
+
"message": ""
|
|
466
|
+
},
|
|
467
|
+
{
|
|
468
|
+
"pass_name": "Before TritonNvidiaGPULowerCLCPass: triton-nvidia-gpu-lower-clc",
|
|
469
|
+
"verdict": "match",
|
|
470
|
+
"n_diff": 0,
|
|
471
|
+
"unsupported": [],
|
|
472
|
+
"message": ""
|
|
473
|
+
},
|
|
474
|
+
{
|
|
475
|
+
"pass_name": "Before TritonGPURemoveLayoutConversions: tritongpu-remove-layout-conversions{disable-remat-splitting=false}",
|
|
476
|
+
"verdict": "match",
|
|
477
|
+
"n_diff": 0,
|
|
478
|
+
"unsupported": [],
|
|
479
|
+
"message": ""
|
|
480
|
+
},
|
|
481
|
+
{
|
|
482
|
+
"pass_name": "Before TritonNvidiaGPUInterleaveTMemPass: triton-nvidia-interleave-tmem",
|
|
483
|
+
"verdict": "match",
|
|
484
|
+
"n_diff": 0,
|
|
485
|
+
"unsupported": [],
|
|
486
|
+
"message": ""
|
|
487
|
+
},
|
|
488
|
+
{
|
|
489
|
+
"pass_name": "Before TritonGPUReduceDataDuplication: tritongpu-reduce-data-duplication",
|
|
490
|
+
"verdict": "match",
|
|
491
|
+
"n_diff": 0,
|
|
492
|
+
"unsupported": [],
|
|
493
|
+
"message": ""
|
|
494
|
+
},
|
|
495
|
+
{
|
|
496
|
+
"pass_name": "Before TritonGPUReorderInstructions: tritongpu-reorder-instructions",
|
|
497
|
+
"verdict": "match",
|
|
498
|
+
"n_diff": 0,
|
|
499
|
+
"unsupported": [],
|
|
500
|
+
"message": ""
|
|
501
|
+
},
|
|
502
|
+
{
|
|
503
|
+
"pass_name": "Before TritonLoopAwareCSE: triton-loop-aware-cse",
|
|
504
|
+
"verdict": "match",
|
|
505
|
+
"n_diff": 0,
|
|
506
|
+
"unsupported": [],
|
|
507
|
+
"message": ""
|
|
508
|
+
},
|
|
509
|
+
{
|
|
510
|
+
"pass_name": "Before SymbolDCEPass: symbol-dce",
|
|
511
|
+
"verdict": "match",
|
|
512
|
+
"n_diff": 0,
|
|
513
|
+
"unsupported": [],
|
|
514
|
+
"message": ""
|
|
515
|
+
},
|
|
516
|
+
{
|
|
517
|
+
"pass_name": "Before TritonGPUFenceInsertion: triton-nvidia-gpu-fence-insertion{compute-capability=120}",
|
|
518
|
+
"verdict": "match",
|
|
519
|
+
"n_diff": 0,
|
|
520
|
+
"unsupported": [],
|
|
521
|
+
"message": ""
|
|
522
|
+
},
|
|
523
|
+
{
|
|
524
|
+
"pass_name": "Before TritonNvidiaGPUMMALoweringPass: triton-nvidia-mma-lowering",
|
|
525
|
+
"verdict": "match",
|
|
526
|
+
"n_diff": 0,
|
|
527
|
+
"unsupported": [],
|
|
528
|
+
"message": ""
|
|
529
|
+
},
|
|
530
|
+
{
|
|
531
|
+
"pass_name": "Before TritonNvidiaGPUHoistMBarrierLifecyclePass: triton-nvidia-gpu-hoist-mbarrier-lifecycle{compute-capability=120}",
|
|
532
|
+
"verdict": "match",
|
|
533
|
+
"n_diff": 0,
|
|
534
|
+
"unsupported": [],
|
|
535
|
+
"message": ""
|
|
536
|
+
},
|
|
537
|
+
{
|
|
538
|
+
"pass_name": "Before SCCPPass: sccp",
|
|
539
|
+
"verdict": "match",
|
|
540
|
+
"n_diff": 0,
|
|
541
|
+
"unsupported": [],
|
|
542
|
+
"message": ""
|
|
543
|
+
},
|
|
544
|
+
{
|
|
545
|
+
"pass_name": "Before CSEPass: cse",
|
|
546
|
+
"verdict": "match",
|
|
547
|
+
"n_diff": 0,
|
|
548
|
+
"unsupported": [],
|
|
549
|
+
"message": ""
|
|
550
|
+
},
|
|
551
|
+
{
|
|
552
|
+
"pass_name": "Before CanonicalizerPass: canonicalize{cse-between-iterations=false max-iterations=10 max-num-rewrites=-1 region-simplify=normal test-convergence=false top-down=true}",
|
|
553
|
+
"verdict": "match",
|
|
554
|
+
"n_diff": 0,
|
|
555
|
+
"unsupported": [],
|
|
556
|
+
"message": ""
|
|
557
|
+
},
|
|
558
|
+
{
|
|
559
|
+
"pass_name": "Before TritonGPUCombineTensorSelectAndIf: tritongpu-combine-tensor-select-and-if",
|
|
560
|
+
"verdict": "match",
|
|
561
|
+
"n_diff": 0,
|
|
562
|
+
"unsupported": [],
|
|
563
|
+
"message": ""
|
|
564
|
+
},
|
|
565
|
+
{
|
|
566
|
+
"pass_name": "Before TritonGPUAllocateWarpGroups: tritongpu-allocate-warp-groups{instrumented=false}",
|
|
567
|
+
"verdict": "match",
|
|
568
|
+
"n_diff": 0,
|
|
569
|
+
"unsupported": [],
|
|
570
|
+
"message": ""
|
|
571
|
+
},
|
|
572
|
+
{
|
|
573
|
+
"pass_name": "Before SCFToControlFlowPass: convert-scf-to-cf{allow-pattern-rollback=true}",
|
|
574
|
+
"verdict": "match",
|
|
575
|
+
"n_diff": 0,
|
|
576
|
+
"unsupported": [],
|
|
577
|
+
"message": ""
|
|
578
|
+
},
|
|
579
|
+
{
|
|
580
|
+
"pass_name": "Before GluonInline: gluon-inline",
|
|
581
|
+
"verdict": "match",
|
|
582
|
+
"n_diff": 0,
|
|
583
|
+
"unsupported": [],
|
|
584
|
+
"message": ""
|
|
585
|
+
},
|
|
586
|
+
{
|
|
587
|
+
"pass_name": "Before AllocateSharedMemoryNv: allocate-shared-memory-nv{compute-capability=120 ptx-version=93}",
|
|
588
|
+
"verdict": "match",
|
|
589
|
+
"n_diff": 0,
|
|
590
|
+
"unsupported": [],
|
|
591
|
+
"message": ""
|
|
592
|
+
},
|
|
593
|
+
{
|
|
594
|
+
"pass_name": "Before TritonTensorMemoryAllocationPass: triton-tensor-memory-allocation",
|
|
595
|
+
"verdict": "match",
|
|
596
|
+
"n_diff": 0,
|
|
597
|
+
"unsupported": [],
|
|
598
|
+
"message": ""
|
|
599
|
+
},
|
|
600
|
+
{
|
|
601
|
+
"pass_name": "Before TritonGPUProxyFenceInsertion: triton-nvidia-gpu-proxy-fence-insertion{compute-capability=120}",
|
|
602
|
+
"verdict": "match",
|
|
603
|
+
"n_diff": 0,
|
|
604
|
+
"unsupported": [],
|
|
605
|
+
"message": ""
|
|
606
|
+
},
|
|
607
|
+
{
|
|
608
|
+
"pass_name": "Before TritonNvidiaGPUTMemBarrierInsertionPass: triton-nvidia-gpu-tmem-barrier-insertion",
|
|
609
|
+
"verdict": "match",
|
|
610
|
+
"n_diff": 0,
|
|
611
|
+
"unsupported": [],
|
|
612
|
+
"message": ""
|
|
613
|
+
},
|
|
614
|
+
{
|
|
615
|
+
"pass_name": "Before ConvertTritonGPUToLLVM: convert-triton-gpu-to-llvm{compute-capability=120 enable-concurrency-sanitizer=false ptx-version=93}",
|
|
616
|
+
"verdict": "match",
|
|
617
|
+
"n_diff": 0,
|
|
618
|
+
"unsupported": [],
|
|
619
|
+
"message": ""
|
|
620
|
+
},
|
|
621
|
+
{
|
|
622
|
+
"pass_name": "Before InitializeWSClusterBarriers: initialize-ws-cluster-barriers{compute-capability=120 ptx-version=93}",
|
|
623
|
+
"verdict": "unsupported",
|
|
624
|
+
"n_diff": -1,
|
|
625
|
+
"unsupported": [
|
|
626
|
+
"<llvm dialect>"
|
|
627
|
+
],
|
|
628
|
+
"message": "past lowering"
|
|
629
|
+
}
|
|
630
|
+
],
|
|
631
|
+
"first_bad_pass": "Before InlinerPass: inline{default-pipeline=canonicalize inlining-threshold=4294967295 max-iterations=4 }",
|
|
632
|
+
"culprit": "Before TritonGPUFuseNestedLoops: tritongpu-fuse-nested-loops"
|
|
633
|
+
}
|
|
634
|
+
]
|
|
635
|
+
}
|
|
@@ -0,0 +1,48 @@
|
|
|
1
|
+
"""triton#11519: FuseNestedLoops trusts an llvm.assume on the branch the loop nest cannot reach.
|
|
2
|
+
|
|
3
|
+
m=4, n=0, flatten=True: the inner loop never runs, so out must stay 0 and sink untouched.
|
|
4
|
+
"""
|
|
5
|
+
|
|
6
|
+
import argparse
|
|
7
|
+
|
|
8
|
+
import numpy as np
|
|
9
|
+
import torch
|
|
10
|
+
import triton
|
|
11
|
+
import triton.language as tl
|
|
12
|
+
|
|
13
|
+
BLOCK = 64
|
|
14
|
+
SENTINEL = -7777.0
|
|
15
|
+
|
|
16
|
+
|
|
17
|
+
@triton.jit
|
|
18
|
+
def kernel(p, out, sink, m, n, BLOCK: tl.constexpr, FLAT: tl.constexpr):
|
|
19
|
+
offs = tl.arange(0, BLOCK)
|
|
20
|
+
acc = tl.zeros([BLOCK], dtype=tl.float32)
|
|
21
|
+
if m == 0:
|
|
22
|
+
tl.assume(n > 0)
|
|
23
|
+
acc += tl.load(p + offs)
|
|
24
|
+
else:
|
|
25
|
+
for i in tl.range(0, m, flatten=FLAT):
|
|
26
|
+
for j in range(0, n):
|
|
27
|
+
acc += tl.load(p + j * BLOCK + offs)
|
|
28
|
+
tl.store(sink + (i * 8 + j) * BLOCK + offs, acc)
|
|
29
|
+
tl.store(out + offs, acc)
|
|
30
|
+
|
|
31
|
+
|
|
32
|
+
def main() -> None:
|
|
33
|
+
ap = argparse.ArgumentParser()
|
|
34
|
+
ap.add_argument("--device", default="cuda")
|
|
35
|
+
ap.add_argument("--out", required=True)
|
|
36
|
+
args = ap.parse_args()
|
|
37
|
+
m, n = 4, 0
|
|
38
|
+
sink_len = max(m, 1) * 8 * BLOCK
|
|
39
|
+
p_host = (np.arange(BLOCK * max(n, 1), dtype=np.float32) % 7.0 + 1.0).astype(np.float32)
|
|
40
|
+
p = torch.from_numpy(p_host).to(args.device)
|
|
41
|
+
out = torch.full((BLOCK,), SENTINEL, dtype=torch.float32, device=args.device)
|
|
42
|
+
sink = torch.full((sink_len,), SENTINEL, dtype=torch.float32, device=args.device)
|
|
43
|
+
kernel[(1,)](p, out, sink, m, n, BLOCK=BLOCK, FLAT=1, num_warps=4)
|
|
44
|
+
np.save(args.out, np.concatenate([out.cpu().numpy(), sink.cpu().numpy()]))
|
|
45
|
+
|
|
46
|
+
|
|
47
|
+
if __name__ == "__main__":
|
|
48
|
+
main()
|