forked from microsoft/onnxruntime
-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathmodel_tests.cc
1213 lines (1148 loc) · 69.7 KB
/
model_tests.cc
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
// Copyright (c) Microsoft Corporation. All rights reserved.
// Licensed under the MIT License.
#include <iterator>
#include <gtest/gtest.h>
#include "core/session/onnxruntime_c_api.h"
#include "core/session/onnxruntime_cxx_api.h"
#include "core/session/ort_apis.h"
#include "core/session/inference_session.h"
#include "core/session/ort_env.h"
#include "core/providers/tensorrt/tensorrt_provider_options.h"
#include "asserts.h"
#include <core/platform/path_lib.h>
#include "default_providers.h"
#include <string>
#include <codecvt>
#include <locale>
#ifdef USE_DNNL
#include "core/providers/dnnl/dnnl_provider_factory.h"
#endif
#ifdef USE_NNAPI
#include "core/providers/nnapi/nnapi_provider_factory.h"
#endif
#ifdef USE_RKNPU
#include "core/providers/rknpu/rknpu_provider_factory.h"
#endif
#ifdef USE_ACL
#include "core/providers/acl/acl_provider_factory.h"
#endif
#ifdef USE_ARMNN
#include "core/providers/armnn/armnn_provider_factory.h"
#endif
// test infrastructure
#include "test/onnx/testenv.h"
#include "test/onnx/TestCase.h"
#include "test/compare_ortvalue.h"
#include "test/onnx/heap_buffer.h"
#include "test/onnx/onnx_model_info.h"
#include "test/onnx/callback.h"
#include "test/onnx/testcase_request.h"
extern std::unique_ptr<Ort::Env> ort_env;
// asserts that the OrtStatus* result of `status_expr` does not indicate an error
// note: this takes ownership of the OrtStatus* result
#define ASSERT_ORT_STATUS_OK(status_expr) \
do { \
if (OrtStatus* _status = (status_expr); _status != nullptr) { \
std::unique_ptr<OrtStatus, decltype(&OrtApis::ReleaseStatus)> _rel_status{ \
_status, &OrtApis::ReleaseStatus}; \
FAIL() << "OrtStatus error: " << OrtApis::GetErrorMessage(_rel_status.get()); \
} \
} while (false)
using namespace onnxruntime::common;
namespace onnxruntime {
namespace test {
// parameter is provider_name + "_" + model_path
class ModelTest : public testing::TestWithParam<std::basic_string<ORTCHAR_T>> {};
namespace {
struct BrokenTest {
std::string test_name_;
std::string reason_;
std::set<std::string> broken_opset_versions_ = {}; // apply to all versions if empty
BrokenTest(std::string name, std::string reason) : test_name_(std::move(name)), reason_(std::move(reason)) {
}
BrokenTest(std::string name, std::string reason, const std::initializer_list<std::string>& opversions)
: test_name_(std::move(name)), reason_(std::move(reason)), broken_opset_versions_(opversions) {
}
bool operator<(const struct BrokenTest& test) const {
return strcmp(test_name_.c_str(), test.test_name_.c_str()) < 0;
}
};
} // namespace
#ifdef GTEST_ALLOW_UNINSTANTIATED_PARAMETERIZED_TEST
GTEST_ALLOW_UNINSTANTIATED_PARAMETERIZED_TEST(ModelTest);
#endif
void SkipTest(const std::string& reason = "") {
GTEST_SKIP() << "Skipping single test " << reason;
}
TEST_P(ModelTest, Run) {
std::basic_string<ORTCHAR_T> param = GetParam();
size_t pos = param.find(ORT_TSTR("_"));
ASSERT_NE(pos, std::string::npos);
std::string provider_name = ToUTF8String(param.substr(0, pos));
std::basic_string<ORTCHAR_T> model_path = param.substr(pos + 1);
double per_sample_tolerance = 1e-3;
double relative_per_sample_tolerance = 1e-3;
// when cuda or openvino is enabled, set it to a larger value for resolving random MNIST test failure
if (model_path.find(ORT_TSTR("_MNIST")) > 0) {
if (provider_name == "cuda" || provider_name == "openvino") {
relative_per_sample_tolerance = 1e-2;
}
}
std::unique_ptr<OnnxModelInfo> model_info = std::make_unique<OnnxModelInfo>(model_path.c_str());
if (model_info->GetONNXOpSetVersion() != 14 && model_info->GetONNXOpSetVersion() != 15 &&
provider_name == "tensorrt") {
// TensorRT can run most of the model tests, but only part of
// them is enabled here to save CI build time.
// Besides saving CI build time, TRT isn’t able to support full ONNX ops spec and therefore some testcases will
// fail. That's one of reasons we skip those testcases and only test latest ONNX opsets.
SkipTest(" tensorrt only support opset 14 or 15");
return;
}
if (model_info->GetONNXOpSetVersion() == 10 && provider_name == "dnnl") {
// DNNL can run most of the model tests, but only part of
// them is enabled here to save CI build time.
SkipTest(" dnnl doesn't support opset 10");
return;
}
if (model_info->HasDomain(ONNX_NAMESPACE::AI_ONNX_TRAINING_DOMAIN) ||
model_info->HasDomain(ONNX_NAMESPACE::AI_ONNX_PREVIEW_TRAINING_DOMAIN)) {
SkipTest("it has the training domain. No pipeline should need to run these tests.");
return;
}
std::set<BrokenTest> broken_tests = {
{"slice_neg_steps",
"Type parameter (Tind) bound to different types (tensor(int64) and tensor(int32) in node ()."},
{"cast_BFLOAT16_to_FLOAT", "Unexpected input data type"},
{"loop13_seq", "Creation of empty sequences is currently not supported in the test runner"},
{"sequence_insert_at_front", "shape mismatch, expect {4} got {3}"},
{"cast_FLOAT_to_BFLOAT16", "expect uint16 got bfloat16"},
{"mnist", "Input data isn't in valid range"},
{"BERT_Squad", "test data bug"},
{"constantofshape_float_ones", "test data bug", {"opset9", "opset10"}},
{"constantofshape_int_zeros", "test data bug", {"opset9", "opset10"}},
{"cast_STRING_to_FLOAT", "Linux CI has old ONNX python package with bad test data", {"opset9", "opset10"}},
// Numpy float to string has unexpected rounding for some results given numpy default precision is meant to be 8.
// "e.g. 0.296140194 -> '0.2961402' not '0.29614019'. ORT produces the latter with precision set to 8,
// which doesn't match the expected output that was generated with numpy.
{"cast_FLOAT_to_STRING", "Numpy float to string has unexpected rounding for some results."},
{"tf_nasnet_large", "disable temporarily"},
{"tf_nasnet_mobile", "disable temporarily"},
{"tf_pnasnet_large", "disable temporarily"},
{"shrink", "test case is wrong", {"opset9"}},
{"maxpool_with_argmax_2d_precomputed_strides", "ShapeInferenceError"},
{"tf_inception_v2", "result mismatch"},
{"tf_resnet_v1_50", "result mismatch when Conv BN Fusion is applied"},
{"tf_resnet_v1_101", "result mismatch when Conv BN Fusion is applied"},
{"tf_resnet_v1_152", "result mismatch when Conv BN Fusion is applied"},
{"mxnet_arcface", "Model is an invalid ONNX model"},
{"unique_not_sorted_without_axis", "Expected data for 'Y' is incorrect and in sorted order."},
{"cumsum_1d_reverse_exclusive", "only failing linux GPU CI. Likely build error."},
{"resize_downsample_scales_cubic_align_corners", "results mismatch with onnx tests"},
{"resize_downsample_scales_linear_align_corners", "results mismatch with onnx tests"},
{"resize_tf_crop_and_resize", "Bad onnx test output. Needs test fix."},
{"resize_upsample_sizes_nearest_ceil_half_pixel", "Bad onnx test output. Needs test fix."},
{"resize_upsample_sizes_nearest_floor_align_corners", "Bad onnx test output. Needs test fix."},
{"resize_upsample_sizes_nearest_round_prefer_ceil_asymmetric", "Bad onnx test output. Needs test fix."},
{"bitshift_right_uint16", "BitShift(11) uint16 support not enabled currently"},
{"bitshift_left_uint16", "BitShift(11) uint16 support not enabled currently"},
{"maxunpool_export_with_output_shape",
"Invalid output in ONNX test. See https://github.com/onnx/onnx/issues/2398"},
{"cntk_simple_seg", "Bad onnx test output caused by wrong SAME_UPPER/SAME_LOWER for ConvTranspose"},
{"training_dropout", "result differs", {}}, // Temporary, subsequent PR will remove this.
{"training_dropout_default", "result differs", {}}, // Temporary, subsequent PR will remove this.
{"training_dropout_default_mask", "result differs", {}}, // Temporary, subsequent PR will remove this.
{"training_dropout_mask", "result differs", {}}, // Temporary, subsequent PR will remove this.
{"batchnorm_epsilon_training_mode", "training only", {}},
{"batchnorm_example_training_mode", "training only", {}},
{"bernoulli", "type error", {}},
{"bernoulli_double", "type error", {}},
{"bernoulli_double_expanded", "type error", {}},
{"bernoulli_expanded", "type error", {}},
{"bernoulli_seed", "type error", {}},
{"bernoulli_seed_expanded", "type error", {}},
{"castlike_BFLOAT16_to_FLOAT", "type error", {}},
{"castlike_BFLOAT16_to_FLOAT_expanded", "type error", {}},
{"castlike_FLOAT_to_BFLOAT16", "type error", {}},
{"castlike_FLOAT_to_BFLOAT16_expanded", "type error", {}},
{"castlike_FLOAT_to_STRING", "type error", {}},
{"castlike_FLOAT_to_STRING_expanded", "type error", {}},
{"convtranspose_autopad_same", "Test data has been corrected in ONNX 1.10.", {"opset13", "opset14"}},
{"gru_batchwise", "type error", {}},
{"lstm_batchwise", "type error", {}},
{"optional_get_element", "type error", {}},
{"optional_get_element_sequence", "type error", {}},
{"optional_has_element", "type error", {}},
{"optional_has_element_empty", "type error", {}},
{"shape_end_1", "type error", {}},
{"shape_end_negative_1", "type error", {}},
{"shape_start_1", "type error", {}},
{"shape_start_1_end_2", "type error", {}},
{"shape_start_1_end_negative_1", "type error", {}},
{"shape_start_negative_1", "type error", {}},
{"simple_rnn_batchwise", "type error", {}},
{"mod_float_mixed_sign_example", "fmod attribute must be true for floating point types", {}},
{"col2im_pads", "result mismatch", {"opset18"}},
#ifdef ENABLE_TRAINING_CORE
{"adagrad", "not a registered function/op", {}}, // Op not registered.
{"adagrad_multiple", "not a registered function/op", {}}, // Op not registered.
{"adam", "not a registered function/op", {}}, // Op not registered.
{"adam_multiple", "not a registered function/op", {}}, // Op not registered.
{"gradient_of_add", "not a registered function/op", {}}, // Op not registered.
{"gradient_of_add_and_mul", "not a registered function/op", {}}, // Op not registered.
{"momentum", "not a registered function/op", {}}, // Op not registered.
{"momentum_multiple", "not a registered function/op", {}}, // Op not registered.
{"nesterov_momentum", "not a registered function/op", {}}, // Op not registered.
{"softmax_cross_entropy_input_shape_is_NCd1d2d3d4d5_none_no_weight", "type error", {"opset12"}},
{"softmax_cross_entropy_mean_weight_ignore_index_log_prob", "type error", {"opset12"}},
{"softmax_cross_entropy_input_shape_is_NCd1_mean_weight_negative_ignore_index_log_prob",
"type error",
{"opset12"}},
{"softmax_cross_entropy_mean_weight_log_prob", "type error", {"opset12"}},
{"softmax_cross_entropy_input_shape_is_NCd1d2d3d4d5_mean_weight_log_prob", "type error", {"opset12"}},
{"softmax_cross_entropy_mean_weight_ignore_index_3d", "type error", {"opset12"}},
{"softmax_cross_entropy_mean_weight_ignore_index_4d_log_prob", "type error", {"opset12"}},
{"softmax_cross_entropy_input_shape_is_NCd1d2d3d4d5_none_no_weight_log_prob", "type error", {"opset12"}},
{"softmax_cross_entropy_mean_weight_ignore_index_4d", "type error", {"opset12"}},
{"softmax_cross_entropy_mean", "type error", {"opset12"}},
{"softmax_cross_entropy_mean_log_prob", "type error", {"opset12"}},
{"softmax_cross_entropy_mean_no_weight_ignore_index", "type error", {"opset12"}},
{"softmax_cross_entropy_input_shape_is_NCd1d2d3_sum_weight_high_ignore_index_log_prob",
"type error",
{"opset12"}},
{"softmax_cross_entropy_mean_3d_log_prob", "type error", {"opset12"}},
{"softmax_cross_entropy_none_log_prob", "type error", {"opset12"}},
{"softmax_cross_entropy_mean_3d", "type error", {"opset12"}},
{"softmax_cross_entropy_input_shape_is_NCd1d2d3_none_no_weight_negative_ignore_index", "type error", {"opset12"}},
{"softmax_cross_entropy_mean_weight_ignore_index_3d_log_prob", "type error", {"opset12"}},
{"softmax_cross_entropy_mean_no_weight_ignore_index_3d_log_prob", "type error", {"opset12"}},
{"softmax_cross_entropy_none_weights_log_prob", "type error", {"opset12"}},
{"softmax_cross_entropy_sum_log_prob", "type error", {"opset12"}},
{"softmax_cross_entropy_mean_weight_ignore_index", "type error", {"opset12"}},
{"softmax_cross_entropy_mean_no_weight_ignore_index_log_prob", "type error", {"opset12"}},
{"softmax_cross_entropy_mean_no_weight_ignore_index_3d", "type error", {"opset12"}},
{"softmax_cross_entropy_input_shape_is_NCd1d2d3_sum_weight_high_ignore_index", "type error", {"opset12"}},
{"softmax_cross_entropy_sum", "type error", {"opset12"}},
{"softmax_cross_entropy_input_shape_is_NCd1d2d3_none_no_weight_negative_ignore_index_log_prob",
"type error",
{"opset12"}},
{"softmax_cross_entropy_none_weights", "type error", {"opset12"}},
{"softmax_cross_entropy_mean_no_weight_ignore_index_4d_log_prob", "type error", {"opset12"}},
{"softmax_cross_entropy_none", "type error", {"opset12"}},
{"softmax_cross_entropy_input_shape_is_NCd1_mean_weight_negative_ignore_index", "type error", {"opset12"}},
{"softmax_cross_entropy_input_shape_is_NCd1d2d3d4d5_mean_weight", "type error", {"opset12"}},
{"softmax_cross_entropy_mean_weight", "type error", {"opset12"}},
{"softmax_cross_entropy_mean_no_weight_ignore_index_4d", "type error", {"opset12"}},
#endif
{"mask_rcnn_keras", "this model currently has an invalid contrib op version set to 10", {}}};
// Some EPs may fail to pass some specific testcases.
// For example TenosrRT EP may fail on FLOAT16 related testcases if GPU doesn't support float16.
// Instead of list all these testcases, we can use following keyword set to filter out testcases wchich contain
// specific keyword.
std::set<std::string> broken_tests_keyword_set = {};
if (provider_name == "cuda") {
#ifdef _WIN32
broken_tests.insert({"LSTM_Seq_lens_unpacked", "this test fails with new image since Aug 25."});
broken_tests.insert({"bidaf", "this test fails with new image since Aug 25."});
#endif
}
if (provider_name == "nnapi") {
broken_tests.insert({"scan9_sum", "Error with the extra graph"});
broken_tests.insert({"scan_sum", "Error with the extra graph"});
broken_tests.insert({"mvn_expanded", "Failed to find kernel for MemcpyFromHost(1) (node Memcpy_1)"});
broken_tests.insert({"dynamicquantizelinear_expanded", "Temporarily disabled pending investigation"});
broken_tests.insert({"dynamicquantizelinear_max_adjusted_expanded", "Temporarily disabled pending investigation"});
broken_tests.insert({"dynamicquantizelinear_min_adjusted_expanded", "Temporarily disabled pending investigation"});
broken_tests.insert({"gemm_transposeB", "Temporarily disabled pending investigation"});
broken_tests.insert({"range_float_type_positive_delta_expanded", "Temporarily disabled pending investigation"});
broken_tests.insert({"range_int32_type_negative_delta_expanded", "Temporarily disabled pending investigation"});
broken_tests.insert({"convtranspose_1d", "1d convtranspose not supported yet"});
broken_tests.insert({"convtranspose_3d", "3d convtranspose not supported yet"});
broken_tests.insert({"maxpool_2d_uint8", "result mismatch"});
broken_tests.insert({"negative_log_likelihood_loss_input_shape_is_NC_expanded", "shape mismatch"});
broken_tests.insert({"negative_log_likelihood_loss_input_shape_is_NCd1d2_expanded", "shape mismatch"});
broken_tests.insert(
{"negative_log_likelihood_loss_input_shape_is_NCd1d2_reduction_mean_expanded", "shape mismatch"});
broken_tests.insert(
{"negative_log_likelihood_loss_input_shape_is_NCd1d2_reduction_sum_expanded", "shape mismatch"});
broken_tests.insert({"negative_log_likelihood_loss_input_shape_is_NCd1d2_with_weight_expanded", "shape mismatch"});
broken_tests.insert(
{"negative_log_likelihood_loss_input_shape_is_NCd1d2_with_weight_reduction_mean_expanded", "shape mismatch"});
broken_tests.insert(
{"negative_log_likelihood_loss_input_shape_is_NCd1d2_with_weight_reduction_sum_expanded", "shape mismatch"});
// Disable based on George Wu's recommendation.
broken_tests.insert(
{"negative_log_likelihood_loss_input_shape_is_NCd1d2_with_weight_reduction_sum_ignore_index_expanded",
"shape mismatch"});
broken_tests.insert({"negative_log_likelihood_loss_iinput_shape_is_NCd1_weight_ignore_index", "Shape mismatch"});
broken_tests.insert(
{"negative_log_likelihood_loss_iinput_shape_is_NCd1_weight_ignore_index_expanded", "Shape mismatch"});
broken_tests.insert({"negative_log_likelihood_loss_input_shape_is_NC", "Shape mismatch"});
broken_tests.insert({"negative_log_likelihood_loss_input_shape_is_NCd1", "Shape mismatch"});
broken_tests.insert({"negative_log_likelihood_loss_input_shape_is_NCd1_expanded", "Shape mismatch"});
broken_tests.insert({"negative_log_likelihood_loss_input_shape_is_NCd1_ignore_index", "Shape mismatch"});
broken_tests.insert({"negative_log_likelihood_loss_input_shape_is_NCd1_ignore_index_expanded", "Shape mismatch"});
broken_tests.insert(
{"negative_log_likelihood_loss_input_shape_is_NCd1_mean_weight_negative_ignore_index", "Shape mismatch"});
broken_tests.insert({"negative_log_likelihood_loss_input_shape_is_NCd1_mean_weight_negative_ignore_index_expanded",
"Shape mismatch"});
broken_tests.insert({"negative_log_likelihood_loss_input_shape_is_NCd1_weight", "Shape mismatch"});
broken_tests.insert({"negative_log_likelihood_loss_input_shape_is_NCd1_weight_expanded", "Shape mismatch"});
broken_tests.insert({"negative_log_likelihood_loss_input_shape_is_NCd1d2", "Shape mismatch"});
broken_tests.insert(
{"negative_log_likelihood_loss_input_shape_is_NCd1d2_no_weight_reduction_mean_ignore_index", "Shape mismatch"});
broken_tests.insert(
{"negative_log_likelihood_loss_input_shape_is_NCd1d2_no_weight_reduction_mean_ignore_index_expanded",
"Shape mismatch"});
broken_tests.insert({"negative_log_likelihood_loss_input_shape_is_NCd1d2_reduction_mean", "Shape mismatch"});
broken_tests.insert({"negative_log_likelihood_loss_input_shape_is_NCd1d2_reduction_sum", "Shape mismatch"});
broken_tests.insert({"negative_log_likelihood_loss_input_shape_is_NCd1d2_with_weight", "Shape mismatch"});
broken_tests.insert(
{"negative_log_likelihood_loss_input_shape_is_NCd1d2_with_weight_reduction_mean", "Shape mismatch"});
broken_tests.insert(
{"negative_log_likelihood_loss_input_shape_is_NCd1d2_with_weight_reduction_sum", "Shape mismatch"});
broken_tests.insert({"negative_log_likelihood_loss_input_shape_is_NCd1d2_with_weight_reduction_sum_ignore_index",
"Shape mismatch"});
broken_tests.insert({"negative_log_likelihood_loss_input_shape_is_NCd1d2d3_none_no_weight_negative_ignore_index",
"Shape mismatch"});
broken_tests.insert(
{"negative_log_likelihood_loss_input_shape_is_NCd1d2d3_none_no_weight_negative_ignore_index_expanded",
"Shape mismatch"});
broken_tests.insert(
{"negative_log_likelihood_loss_input_shape_is_NCd1d2d3_sum_weight_high_ignore_index", "Shape mismatch"});
broken_tests.insert({"negative_log_likelihood_loss_input_shape_is_NCd1d2d3_sum_weight_high_ignore_index_expanded",
"Shape mismatch"});
broken_tests.insert({"negative_log_likelihood_loss_input_shape_is_NCd1d2d3d4d5_mean_weight", "Shape mismatch"});
broken_tests.insert(
{"negative_log_likelihood_loss_input_shape_is_NCd1d2d3d4d5_mean_weight_expanded", "Shape mismatch"});
broken_tests.insert({"negative_log_likelihood_loss_input_shape_is_NCd1d2d3d4d5_none_no_weight", "Shape mismatch"});
broken_tests.insert(
{"negative_log_likelihood_loss_input_shape_is_NCd1d2d3d4d5_none_no_weight_expanded", "Shape mismatch"});
broken_tests.insert(
{"softmax_cross_entropy_input_shape_is_NCd1_mean_weight_negative_ignore_index", "Shape mismatch"});
broken_tests.insert(
{"softmax_cross_entropy_input_shape_is_NCd1_mean_weight_negative_ignore_index_expanded", "Shape mismatch"});
broken_tests.insert(
{"softmax_cross_entropy_input_shape_is_NCd1_mean_weight_negative_ignore_index_log_prob", "Shape mismatch"});
broken_tests.insert(
{"softmax_cross_entropy_input_shape_is_NCd1_mean_weight_negative_ignore_index_log_prob_expanded",
"Shape mismatch"});
broken_tests.insert(
{"softmax_cross_entropy_input_shape_is_NCd1d2d3_none_no_weight_negative_ignore_index", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_input_shape_is_NCd1d2d3_none_no_weight_negative_ignore_index_expanded",
"Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_input_shape_is_NCd1d2d3_none_no_weight_negative_ignore_index_log_prob",
"Shape mismatch"});
broken_tests.insert(
{"softmax_cross_entropy_input_shape_is_NCd1d2d3_none_no_weight_negative_ignore_index_log_prob_expanded",
"Shape mismatch"});
broken_tests.insert(
{"softmax_cross_entropy_input_shape_is_NCd1d2d3_sum_weight_high_ignore_index", "Shape mismatch"});
broken_tests.insert(
{"softmax_cross_entropy_input_shape_is_NCd1d2d3_sum_weight_high_ignore_index_expanded", "Shape mismatch"});
broken_tests.insert(
{"softmax_cross_entropy_input_shape_is_NCd1d2d3_sum_weight_high_ignore_index_log_prob", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_input_shape_is_NCd1d2d3_sum_weight_high_ignore_index_log_prob_expanded",
"Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_input_shape_is_NCd1d2d3d4d5_mean_weight", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_input_shape_is_NCd1d2d3d4d5_mean_weight_expanded", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_input_shape_is_NCd1d2d3d4d5_mean_weight_log_prob", "Shape mismatch"});
broken_tests.insert(
{"softmax_cross_entropy_input_shape_is_NCd1d2d3d4d5_mean_weight_log_prob_expanded", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_input_shape_is_NCd1d2d3d4d5_none_no_weight", "Shape mismatch"});
broken_tests.insert(
{"softmax_cross_entropy_input_shape_is_NCd1d2d3d4d5_none_no_weight_expanded", "Shape mismatch"});
broken_tests.insert(
{"softmax_cross_entropy_input_shape_is_NCd1d2d3d4d5_none_no_weight_log_prob", "Shape mismatch"});
broken_tests.insert(
{"softmax_cross_entropy_input_shape_is_NCd1d2d3d4d5_none_no_weight_log_prob_expanded", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_mean", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_mean_3d", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_mean_3d_expanded", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_mean_3d_log_prob", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_mean_3d_log_prob_expanded", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_mean_expanded", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_mean_log_prob", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_mean_log_prob_expanded", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_mean_no_weight_ignore_index", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_mean_no_weight_ignore_index_3d", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_mean_no_weight_ignore_index_3d_expanded", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_mean_no_weight_ignore_index_3d_log_prob", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_mean_no_weight_ignore_index_3d_log_prob_expanded", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_mean_no_weight_ignore_index_4d", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_mean_no_weight_ignore_index_4d_expanded", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_mean_no_weight_ignore_index_4d_log_prob", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_mean_no_weight_ignore_index_4d_log_prob_expanded", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_mean_no_weight_ignore_index_expanded", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_mean_no_weight_ignore_index_log_prob", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_mean_no_weight_ignore_index_log_prob_expanded", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_mean_weight", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_mean_weight_expanded", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_mean_weight_ignore_index", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_mean_weight_ignore_index_3d", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_mean_weight_ignore_index_3d_expanded", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_mean_weight_ignore_index_3d_log_prob", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_mean_weight_ignore_index_3d_log_prob_expanded", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_mean_weight_ignore_index_4d", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_mean_weight_ignore_index_4d_expanded", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_mean_weight_ignore_index_4d_log_prob", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_mean_weight_ignore_index_4d_log_prob_expanded", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_mean_weight_ignore_index_expanded", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_mean_weight_ignore_index_log_prob", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_mean_weight_ignore_index_log_prob_expanded", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_mean_weight_log_prob", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_mean_weight_log_prob_expanded", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_none", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_none_expanded", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_none_log_prob", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_none_log_prob_expanded", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_none_weights", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_none_weights_expanded", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_none_weights_log_prob", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_none_weights_log_prob_expanded", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_sum", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_sum_expanded", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_sum_log_prob", "Shape mismatch"});
broken_tests.insert({"softmax_cross_entropy_sum_log_prob_expanded", "Shape mismatch"});
}
if (provider_name == "tensorrt") {
broken_tests.insert({"convtranspose_with_kernel", "It causes segmentation fault"});
broken_tests.insert({"convtranspose_pad", "It causes segmentation fault"});
broken_tests.insert({"convtranspose_kernel_shape", "It causes segmentation fault"});
broken_tests.insert({"dynamicquantizelinear_expanded", "It causes segmentation fault"});
broken_tests.insert({"dynamicquantizelinear_min_adjusted_expanded", "It causes segmentation fault"});
broken_tests.insert({"dynamicquantizelinear_max_adjusted_expanded", "It causes segmentation fault"});
broken_tests.insert({"basic_conv_with_padding",
"Cannot set more than one input unless network has Q/DQ layers. TensorRT EP could not build "
"engine for fused node"});
broken_tests.insert({"basic_conv_without_padding",
"Cannot set more than one input unless network has Q/DQ layers. TensorRT EP could not build "
"engine for fused node"});
broken_tests.insert({"conv_with_strides_no_padding",
"Cannot set more than one input unless network has Q/DQ layers. TensorRT EP could not build "
"engine for fused node"});
broken_tests.insert({"conv_with_autopad_same",
"Internal Error (node_of_y: Cannot set more than one input unless network has Q/DQ layers.)"});
// sce op is not supported
broken_tests_keyword_set.insert({"sce"});
// TensorRT EP CI uses Nvidia Tesla M60 which doesn't support fp16.
broken_tests_keyword_set.insert({"FLOAT16"});
}
if (provider_name == "dml") {
broken_tests.insert({"tinyyolov3", "The parameter is incorrect"});
broken_tests.insert({"PixelShuffle", "Test requires 6D Reshape, which isn't supported by DirectML"});
broken_tests.insert({"operator_permute2", "Test requires 6D Transpose, which isn't supported by DirectML"});
broken_tests.insert({"resize_downsample_linear",
"ORT 0.4 uses asymmetric but will conform to half_pixel in the next ONNX version."});
broken_tests.insert(
{"resize_upsample_linear", "ORT 0.4 uses asymmetric but will conform to half_pixel in the next ONNX version."});
broken_tests.insert(
{"resize_upsample_linear", "ORT 0.4 uses asymmetric but will conform to half_pixel in the next ONNX version."});
// These tests are temporarily disabled pending investigation
broken_tests.insert({"dynamicquantizelinear_expanded", "Temporarily disabled pending investigation"});
broken_tests.insert({"dynamicquantizelinear_max_adjusted_expanded", "Temporarily disabled pending investigation"});
broken_tests.insert({"dynamicquantizelinear_min_adjusted_expanded", "Temporarily disabled pending investigation"});
broken_tests.insert({"mxnet_arcface", "Temporarily disabled pending investigation"});
broken_tests.insert({"yolov3", "Temporarily disabled pending investigation"});
broken_tests.insert({"tf_inception_v2", "Temporarily disabled pending investigation"});
broken_tests.insert({"fp16_inception_v1", "Temporarily disabled pending investigation"});
broken_tests.insert({"candy", "Temporarily disabled pending investigation"});
broken_tests.insert({"BERT_Squad", "Temporarily disabled pending investigation"});
broken_tests.insert({"LSTM_Seq_lens_unpacked", "The parameter is incorrect"});
broken_tests.insert({"resize_downsample_scales_linear",
"DML uses half_pixel and this test assumed \"asymmetric\" but does not include \"mode\""});
broken_tests.insert({"resize_downsample_sizes_linear_pytorch_half_pixel",
"DML does not support downsampling by such a large factor - skips input pixels"});
broken_tests.insert({"resize_downsample_sizes_nearest",
"DML uses pixel centers for nearest, rounding 1 value off for the middle column"});
broken_tests.insert({"resize_upsample_sizes_nearest",
"DML uses pixel centers for nearest, which makes more sense (the 3rd row mismatches)"});
broken_tests.insert({"unsqueeze_three_axes", "DML does not support 6D tensors"});
broken_tests.insert({"unsqueeze_unsorted_axes", "DMLdoes not support 6D tensors"});
broken_tests.insert({"negative_log_likelihood_loss_input_shape_is_NCd1d2d3_none_no_weight_negative_ignore_index",
"DML does not support 5D+ tensors"});
broken_tests.insert(
{"negative_log_likelihood_loss_input_shape_is_NCd1d2d3_none_no_weight_negative_ignore_index_expanded",
"DML does not support 5D+ tensors"});
broken_tests.insert(
{"negative_log_likelihood_loss_input_shape_is_NCd1d2d3d4d5_mean_weight", "DML does not support 5D+ tensors"});
broken_tests.insert({"negative_log_likelihood_loss_input_shape_is_NCd1d2d3d4d5_mean_weight_expanded",
"DML does not support 5D+ tensors"});
broken_tests.insert({"negative_log_likelihood_loss_input_shape_is_NCd1d2d3d4d5_none_no_weight",
"DML does not support 5D+ tensors"});
broken_tests.insert({"negative_log_likelihood_loss_input_shape_is_NCd1d2d3d4d5_none_no_weight_expanded",
"DML does not support 5D+ tensors"});
broken_tests.insert({"softmax_cross_entropy_input_shape_is_NCd1d2d3_none_no_weight_negative_ignore_index",
"DML does not support 5D+ tensors"});
broken_tests.insert({"softmax_cross_entropy_input_shape_is_NCd1d2d3_none_no_weight_negative_ignore_index_expanded",
"DML does not support 5D+ tensors"});
broken_tests.insert({"softmax_cross_entropy_input_shape_is_NCd1d2d3_none_no_weight_negative_ignore_index_log_prob",
"DML does not support 5D+ tensors"});
broken_tests.insert(
{"softmax_cross_entropy_input_shape_is_NCd1d2d3_none_no_weight_negative_ignore_index_log_prob_expanded",
"DML does not support 5D+ tensors"});
broken_tests.insert(
{"softmax_cross_entropy_input_shape_is_NCd1d2d3d4d5_mean_weight", "DML does not support 5D+ tensors"});
broken_tests.insert(
{"softmax_cross_entropy_input_shape_is_NCd1d2d3d4d5_mean_weight_expanded", "DML does not support 5D+ tensors"});
broken_tests.insert(
{"softmax_cross_entropy_input_shape_is_NCd1d2d3d4d5_mean_weight_log_prob", "DML does not support 5D+ tensors"});
broken_tests.insert({"softmax_cross_entropy_input_shape_is_NCd1d2d3d4d5_mean_weight_log_prob_expanded",
"DML does not support 5D+ tensors"});
broken_tests.insert(
{"softmax_cross_entropy_input_shape_is_NCd1d2d3d4d5_none_no_weight", "DML does not support 5D+ tensors"});
broken_tests.insert({"softmax_cross_entropy_input_shape_is_NCd1d2d3d4d5_none_no_weight_expanded",
"DML does not support 5D+ tensors"});
broken_tests.insert({"softmax_cross_entropy_input_shape_is_NCd1d2d3d4d5_none_no_weight_log_prob",
"DML does not support 5D+ tensors"});
broken_tests.insert({"softmax_cross_entropy_input_shape_is_NCd1d2d3d4d5_none_no_weight_log_prob_expanded",
"DML does not support 5D+ tensors"});
}
#ifdef DISABLE_CONTRIB_OPS
broken_tests.insert({"coreml_SqueezeNet_ImageNet", "This model uses contrib ops."});
broken_tests.insert({"keras2coreml_Permute_ImageNet", "This model uses contrib ops."});
broken_tests.insert({"keras2coreml_ReLU_ImageNet", "This model uses contrib ops."});
broken_tests.insert({"keras2coreml_Padding-Upsampling-Normalizer_ImageNet", "This model uses contrib ops."});
broken_tests.insert({"tiny_yolov2", "This model uses contrib ops."});
broken_tests.insert({"fp16_tiny_yolov2", "This model uses contrib ops."});
broken_tests.insert({"keras2coreml_Pooling_ImageNet", "This model uses contrib ops."});
broken_tests.insert({"keras2coreml_Padding_ImageNet", "This model uses contrib ops."});
broken_tests.insert({"keras2coreml_Normalizer_ImageNet", "This model uses contrib ops."});
broken_tests.insert({"keras2coreml_linear_sklearn_load_breast_cancer", "This model uses contrib ops."});
broken_tests.insert({"keras2coreml_linear_ImageNet_small", "This model uses contrib ops."});
broken_tests.insert({"keras2coreml_linear_ImageNet_large", "This model uses contrib ops."});
broken_tests.insert({"keras2coreml_linear_ImageNet", "This model uses contrib ops."});
broken_tests.insert({"keras2coreml_leakyrelu_ImageNet", "This model uses contrib ops."});
broken_tests.insert({"keras2coreml_hard_sigmoid_ImageNet", "This model uses contrib ops."});
broken_tests.insert({"keras2coreml_elu_ImageNet", "This model uses contrib ops."});
broken_tests.insert({"keras2coreml_Dense_ImageNet", "This model uses contrib ops."});
broken_tests.insert({"keras2coreml_Conv2D_ImageNet", "This model uses contrib ops."});
broken_tests.insert({"coreml_VGG16_ImageNet", "This model uses contrib ops."});
broken_tests.insert({"coreml_Resnet50_ImageNet", "This model uses contrib ops."});
broken_tests.insert({"coreml_Inceptionv3_ImageNet", "This model uses contrib ops."});
broken_tests.insert({"coreml_FNS-Candy_ImageNet", "This model uses contrib ops."});
broken_tests.insert({"coreml_AgeNet_ImageNet", "This model uses contrib ops."});
broken_tests.insert({"keras2coreml_thresholdedrelu_ImageNet_large", "This model uses contrib ops."});
broken_tests.insert({"keras2coreml_thresholdedrelu_ImageNet_small", "This model uses contrib ops."});
broken_tests.insert({"keras2coreml_thresholdedrelu_sklearn_load_breast_cancer", "This model uses contrib ops."});
broken_tests.insert({"thresholdedrelu", "This model uses contrib ops."});
broken_tests.insert({"thresholdedrelu_default", "This model uses contrib ops."});
broken_tests.insert({"dynamic_slice_default_axes", "This model uses contrib ops."});
broken_tests.insert({"thresholdedrelu_example", "This model uses contrib ops."});
broken_tests.insert({"dynamic_slice_neg failed", "This model uses contrib ops."});
broken_tests.insert({"dynamic_slice_start_out_of_bounds", "This model uses contrib ops."});
broken_tests.insert({"dynamic_slice", "This model uses contrib ops."});
broken_tests.insert({"dynamic_slice_end_out_of_bounds", "This model uses contrib ops."});
broken_tests.insert({"dynamic_slice_neg", "This model uses contrib ops."});
broken_tests.insert({"mvn", "This model uses contrib ops.", {"onnx130"}});
broken_tests.insert({"cdist_float32_euclidean_1000_2000_1", "This model uses contrib ops."});
broken_tests.insert({"cdist_float32_euclidean_1000_2000_500", "This model uses contrib ops."});
broken_tests.insert({"cdist_float32_euclidean_1_1_1", "This model uses contrib ops."});
broken_tests.insert({"cdist_float32_sqeuclidean_1000_2000_1", "This model uses contrib ops."});
broken_tests.insert({"cdist_float32_sqeuclidean_1000_2000_500", "This model uses contrib ops."});
broken_tests.insert({"cdist_float32_sqeuclidean_1_1_1", "This model uses contrib ops."});
broken_tests.insert({"cdist_float64_euclidean_1000_2000_1", "This model uses contrib ops."});
broken_tests.insert({"cdist_float64_euclidean_1000_2000_500", "This model uses contrib ops."});
broken_tests.insert({"cdist_float64_euclidean_1_1_1", "This model uses contrib ops."});
broken_tests.insert({"cdist_float64_sqeuclidean_1000_2000_1", "This model uses contrib ops."});
broken_tests.insert({"cdist_float64_sqeuclidean_1000_2000_500", "This model uses contrib ops."});
broken_tests.insert({"cdist_float64_sqeuclidean_1_1_1", "This model uses contrib ops."});
broken_tests.insert({"keras2coreml_Average_ImageNet", "This model uses contrib ops."});
broken_tests.insert({"bidaf", "This model uses contrib ops."});
broken_tests.insert({"fp16_test_tiny_yolov2", "This model uses contrib ops."});
broken_tests.insert({"fp16_coreml_FNS-Candy", "This model uses contrib ops."});
broken_tests.insert({"keras2coreml_Repeat_ImageNet", "This model uses contrib ops."});
broken_tests.insert({"keras2coreml_BiDirectional_ImageNet", "This model uses contrib ops."});
broken_tests.insert({"fp16_coreml_LinearRegression_NYCTaxi", "This model uses contrib ops."});
broken_tests.insert({"keras2coreml_Average_ImageNet", "This model uses contrib ops."});
broken_tests.insert({"keras2coreml_GRU_ImageNet", "This model uses contrib ops."});
broken_tests.insert({"keras2coreml_SimpleRNN_ImageNet", "This model uses contrib ops."});
broken_tests.insert({"keras2coreml_Dot_imageNet", "This model uses contrib ops."});
#endif
std::basic_string<ORTCHAR_T> model_dir;
(void)GetDirNameFromFilePath(model_path, model_dir);
std::basic_string<PATH_CHAR_TYPE> test_case_name = GetLastComponent(model_dir);
if (test_case_name.compare(0, 5, ORT_TSTR("test_")) == 0)
test_case_name = test_case_name.substr(5);
{
BrokenTest t = {ToUTF8String(test_case_name), ""};
auto iter = broken_tests.find(t);
auto opset_version = model_info->GetNominalOpsetVersion();
if (iter != broken_tests.end() &&
(opset_version == TestModelInfo::unknown_version || iter->broken_opset_versions_.empty() ||
iter->broken_opset_versions_.find(opset_version) != iter->broken_opset_versions_.end())) {
SkipTest("It's in broken_tests");
return;
}
for (auto iter2 = broken_tests_keyword_set.begin(); iter2 != broken_tests_keyword_set.end(); ++iter2) {
std::string keyword = *iter2;
if (ToUTF8String(test_case_name).find(keyword) != std::string::npos) {
SkipTest("It's in broken_tests_keyword");
return;
}
}
}
// TODO(leca): move the parallel run test list to a config file and load it in GetParameterStrings() to make the load process run only once
std::set<std::string> tests_run_parallel = {"test_resnet18v2",
"test_resnet34v2",
"test_resnet50",
"test_resnet50v2",
"test_resnet101v2",
"test_resnet152v2",
"keras_lotus_resnet3D",
"coreml_Resnet50_ImageNet",
"mlperf_mobilenet",
"mlperf_resnet",
"mlperf_ssd_mobilenet_300",
"mlperf_ssd_resnet34_1200"};
bool is_single_node = !model_info->GetNodeName().empty();
std::vector<ExecutionMode> execution_modes = {ExecutionMode::ORT_SEQUENTIAL};
if (provider_name == "cpu" && !is_single_node)
execution_modes.push_back(ExecutionMode::ORT_PARALLEL);
std::vector<bool> use_single_thread{false};
// Test the model with intra op threadpool disabled
if (provider_name == "cpu" && is_single_node)
use_single_thread.push_back(true);
std::unique_ptr<ITestCase> l = CreateOnnxTestCase(ToUTF8String(test_case_name), std::move(model_info),
per_sample_tolerance, relative_per_sample_tolerance);
#ifndef USE_DNNL
auto tp = TestEnv::CreateThreadPool(Env::Default());
#endif
for (bool is_single_thread : use_single_thread) {
for (ExecutionMode execution_mode : execution_modes) {
Ort::SessionOptions ortso{};
if (!is_single_thread) {
ortso.DisablePerSessionThreads();
} else {
ortso.SetIntraOpNumThreads(1);
}
ortso.SetExecutionMode(execution_mode);
ortso.SetLogId(ToUTF8String(test_case_name).c_str());
ortso.SetLogSeverityLevel(ORT_LOGGING_LEVEL_ERROR);
if (provider_name == "cuda") {
OrtCUDAProviderOptionsV2* cuda_options = nullptr;
ASSERT_ORT_STATUS_OK(OrtApis::CreateCUDAProviderOptions(&cuda_options));
std::unique_ptr<OrtCUDAProviderOptionsV2, decltype(&OrtApis::ReleaseCUDAProviderOptions)> rel_cuda_options(
cuda_options, &OrtApis::ReleaseCUDAProviderOptions);
ortso.AppendExecutionProvider_CUDA_V2(*cuda_options);
} else if (provider_name == "rocm") {
OrtROCMProviderOptions ep_options;
ortso.AppendExecutionProvider_ROCM(ep_options);
}
#ifdef USE_DNNL
else if (provider_name == "dnnl") {
OrtDnnlProviderOptions* ep_option;
ASSERT_ORT_STATUS_OK(OrtApis::CreateDnnlProviderOptions(&ep_option));
std::unique_ptr<OrtDnnlProviderOptions, decltype(&OrtApis::ReleaseDnnlProviderOptions)>
rel_dnnl_options(ep_option, &OrtApis::ReleaseDnnlProviderOptions);
ep_option->use_arena = 0;
ASSERT_ORT_STATUS_OK(OrtApis::SessionOptionsAppendExecutionProvider_Dnnl(ortso, ep_option));
}
#endif
else if (provider_name == "tensorrt") {
if (test_case_name.find(ORT_TSTR("FLOAT16")) != std::string::npos) {
OrtTensorRTProviderOptionsV2 params{0, 0, nullptr, 1000, 1, 1 << 30,
1, // enable fp16
0, nullptr, 0, 0, 0, 0, 0, nullptr, 0, nullptr, 0, 0, 0, 0, 0, 0};
ortso.AppendExecutionProvider_TensorRT_V2(params);
} else {
OrtTensorRTProviderOptionsV2* ep_option = nullptr;
ASSERT_ORT_STATUS_OK(OrtApis::CreateTensorRTProviderOptions(&ep_option));
std::unique_ptr<OrtTensorRTProviderOptionsV2, decltype(&OrtApis::ReleaseTensorRTProviderOptions)>
rel_cuda_options(ep_option, &OrtApis::ReleaseTensorRTProviderOptions);
ortso.AppendExecutionProvider_TensorRT_V2(*ep_option);
}
// Enable CUDA fallback
OrtCUDAProviderOptionsV2* cuda_options = nullptr;
ASSERT_ORT_STATUS_OK(OrtApis::CreateCUDAProviderOptions(&cuda_options));
std::unique_ptr<OrtCUDAProviderOptionsV2, decltype(&OrtApis::ReleaseCUDAProviderOptions)> rel_cuda_options(
cuda_options, &OrtApis::ReleaseCUDAProviderOptions);
ortso.AppendExecutionProvider_CUDA_V2(*cuda_options);
} else if (provider_name == "migraphx") {
OrtMIGraphXProviderOptions ep_options;
ortso.AppendExecutionProvider_MIGraphX(ep_options);
} else if (provider_name == "openvino") {
OrtOpenVINOProviderOptions ep_options;
ortso.AppendExecutionProvider_OpenVINO(ep_options);
}
#ifdef USE_NNAPI
else if (provider_name == "nnapi") {
ASSERT_ORT_STATUS_OK(OrtSessionOptionsAppendExecutionProvider_Nnapi(ortso, 0));
}
#endif
#ifdef USE_RKNPU
else if (provider_name == "rknpu") {
ASSERT_ORT_STATUS_OK(OrtSessionOptionsAppendExecutionProvider_Rknpu(ortso));
}
#endif
#ifdef USE_ACL
else if (provider_name == "acl") {
ASSERT_ORT_STATUS_OK(OrtSessionOptionsAppendExecutionProvider_ACL(ortso, 0));
}
#endif
#ifdef USE_ARMNN
else if (provider_name == "armnn") {
ASSERT_ORT_STATUS_OK(OrtSessionOptionsAppendExecutionProvider_ArmNN(ortso));
}
#endif
OrtSession* ort_session;
OrtStatus* ort_st = OrtApis::CreateSession(*ort_env, model_path.c_str(), ortso, &ort_session);
if (ort_st != nullptr) {
OrtErrorCode error_code = OrtApis::GetErrorCode(ort_st);
if (error_code == ORT_NOT_IMPLEMENTED) {
OrtApis::ReleaseStatus(ort_st);
continue;
}
FAIL() << OrtApis::GetErrorMessage(ort_st);
}
std::unique_ptr<OrtSession, decltype(&OrtApis::ReleaseSession)> rel_ort_session(ort_session,
&OrtApis::ReleaseSession);
const size_t data_count = l->GetDataCount();
#ifndef USE_DNNL // potential crash for DNNL pipeline
if (data_count > 1 && tests_run_parallel.find(l->GetTestCaseName()) != tests_run_parallel.end()) {
LOGS_DEFAULT(ERROR) << "Parallel test for " << l->GetTestCaseName(); // TODO(leca): change level to INFO or even delete the log once verified parallel test working
std::shared_ptr<TestCaseResult> results = TestCaseRequestContext::Run(tp.get(), *l, *ort_env, ortso, data_count, 1 /*repeat_count*/);
for (EXECUTE_RESULT res : results->GetExcutionResult()) {
EXPECT_EQ(res, EXECUTE_RESULT::SUCCESS) << "is_single_thread:" << is_single_thread << ", execution_mode:" << execution_mode << ", provider_name:"
<< provider_name << ", test name:" << results->GetName() << ", result: " << res;
}
continue;
}
#endif // !USE_DNNL
// TODO(leca): leverage TestCaseRequestContext::Run() to make it short
auto default_allocator = std::make_unique<MockedOrtAllocator>();
for (size_t task_id = 0; task_id != data_count; ++task_id) {
onnxruntime::test::HeapBuffer holder;
std::unordered_map<std::string, Ort::Value> feeds;
l->LoadTestData(task_id, holder, feeds, true);
size_t output_count;
ASSERT_ORT_STATUS_OK(OrtApis::SessionGetOutputCount(ort_session, &output_count));
// Create output feed
std::vector<char*> output_names(output_count);
for (size_t i = 0; i != output_count; ++i) {
ASSERT_ORT_STATUS_OK(
OrtApis::SessionGetOutputName(ort_session, i, default_allocator.get(), &output_names[i]));
}
std::vector<const char*> input_names;
std::vector<OrtValue*> input_values;
std::vector<OrtValue*> output_values(output_count);
{
for (auto& p : feeds) {
input_names.push_back(p.first.c_str());
input_values.push_back(p.second);
}
ort_st = OrtApis::Run(ort_session, nullptr, input_names.data(), input_values.data(), input_values.size(),
output_names.data(), output_names.size(), output_values.data());
if (ort_st != nullptr) {
OrtErrorCode error_code = OrtApis::GetErrorCode(ort_st);
if (error_code == ORT_NOT_IMPLEMENTED) {
OrtApis::ReleaseStatus(ort_st);
for (char* p : output_names) {
default_allocator->Free(p);
}
for (OrtValue* v : output_values) {
OrtApis::ReleaseValue(v);
}
}
FAIL() << OrtApis::GetErrorMessage(ort_st);
}
}
bool post_procesing = false;
Status status;
l->GetPerSampleTolerance(&per_sample_tolerance);
l->GetRelativePerSampleTolerance(&relative_per_sample_tolerance);
l->GetPostProcessing(&post_procesing);
// TODO: if there are no output value files, just skip the validation
std::unordered_map<std::string, Ort::Value> expected_output_values;
l->LoadTestData(task_id, holder, expected_output_values, false);
std::unordered_map<std::string, OrtValue*> name_fetch_output_map;
std::unordered_map<std::string, const ONNX_NAMESPACE::ValueInfoProto*> name_output_value_info_proto;
size_t i = 0;
for (auto& output_name : output_names) {
// p_fetches is filled in the order of output_names.
name_fetch_output_map[output_name] = output_values[i];
const ONNX_NAMESPACE::ValueInfoProto* infoProto = l->GetOutputInfoFromModel(i);
if (infoProto != nullptr)
name_output_value_info_proto.insert(std::make_pair(infoProto->name(), infoProto));
i++;
}
for (auto& output : expected_output_values) {
const OrtValue* expected_output_value = output.second;
const std::string& output_name = output.first;
auto iter = name_fetch_output_map.find(output_name);
ASSERT_NE(iter, name_fetch_output_map.end());
OrtValue* actual_output_value = iter->second;
std::pair<COMPARE_RESULT, std::string> ret =
CompareOrtValue(*actual_output_value, *expected_output_value, per_sample_tolerance,
relative_per_sample_tolerance, post_procesing);
COMPARE_RESULT compare_result = ret.first;
ASSERT_EQ(COMPARE_RESULT::SUCCESS, ret.first) << ret.second;
const ONNX_NAMESPACE::ValueInfoProto* v = name_output_value_info_proto[output_name];
if (v == nullptr)
continue;
ret = VerifyValueInfo(*v, actual_output_value);
compare_result = ret.first;
ASSERT_EQ(COMPARE_RESULT::SUCCESS, ret.first) << ret.second;
if (compare_result != COMPARE_RESULT::SUCCESS) {
break;
}
}
for (char* p : output_names) {
default_allocator->Free(p);
}
for (OrtValue* v : output_values) {
OrtApis::ReleaseValue(v);
}
}
}
}
}
// TODO: all providers
::std::vector<::std::basic_string<ORTCHAR_T>> GetParameterStrings() {
std::vector<const ORTCHAR_T*> provider_names;
provider_names.push_back(ORT_TSTR("cpu"));
#ifdef USE_TENSORRT
provider_names.push_back(ORT_TSTR("tensorrt"));
#endif
#ifdef USE_MIGRAPHX
provider_names.push_back(ORT_TSTR("migraphx"));
#endif
#ifdef USE_OPENVINO
provider_names.push_back(ORT_TSTR("openvino"));
#endif
#ifdef USE_CUDA
provider_names.push_back(ORT_TSTR("cuda"));
#endif
#ifdef USE_ROCM
provider_names.push_back(ORT_TSTR("rocm"));
#endif
#ifdef USE_DNNL
provider_names.push_back(ORT_TSTR("dnnl"));
#endif
// For any non-Android system, NNAPI will only be used for ort model converter
#if defined(USE_NNAPI) && defined(__ANDROID__)
provider_names.push_back(ORT_TSTR("nnapi"));
#endif
#ifdef USE_RKNPU
provider_names.push_back(ORT_TSTR("rknpu"));
#endif
#ifdef USE_ACL
provider_names.push_back(ORT_TSTR("acl"));
#endif
#ifdef USE_ARMNN
provider_names.push_back(ORT_TSTR("armnn"));
#endif
#ifdef USE_DML
provider_names.push_back(ORT_TSTR("dml"));
#endif
std::vector<std::basic_string<ORTCHAR_T>> v;
// Permanently exclude following tests because ORT support only opset starting from 7,
// Please make no more changes to the list
static const ORTCHAR_T* immutable_broken_tests[] = {
ORT_TSTR("AvgPool1d"),
ORT_TSTR("AvgPool1d_stride"),
ORT_TSTR("AvgPool2d"),
ORT_TSTR("AvgPool2d_stride"),
ORT_TSTR("AvgPool3d"),
ORT_TSTR("AvgPool3d_stride"),
ORT_TSTR("AvgPool3d_stride1_pad0_gpu_input"),
ORT_TSTR("BatchNorm1d_3d_input_eval"),
ORT_TSTR("BatchNorm2d_eval"),
ORT_TSTR("BatchNorm2d_momentum_eval"),
ORT_TSTR("BatchNorm3d_eval"),
ORT_TSTR("BatchNorm3d_momentum_eval"),
ORT_TSTR("GLU"),
ORT_TSTR("GLU_dim"),
ORT_TSTR("Linear"),
ORT_TSTR("PReLU_1d"),
ORT_TSTR("PReLU_1d_multiparam"),
ORT_TSTR("PReLU_2d"),
ORT_TSTR("PReLU_2d_multiparam"),
ORT_TSTR("PReLU_3d"),
ORT_TSTR("PReLU_3d_multiparam"),
ORT_TSTR("PoissonNLLLLoss_no_reduce"),
ORT_TSTR("Softsign"),
ORT_TSTR("operator_add_broadcast"),
ORT_TSTR("operator_add_size1_broadcast"),
ORT_TSTR("operator_add_size1_right_broadcast"),
ORT_TSTR("operator_add_size1_singleton_broadcast"),
ORT_TSTR("operator_addconstant"),
ORT_TSTR("operator_addmm"),
ORT_TSTR("operator_basic"),
ORT_TSTR("operator_mm"),
ORT_TSTR("operator_non_float_params"),
ORT_TSTR("operator_params"),
ORT_TSTR("operator_pow"),
};
static const ORTCHAR_T* cuda_flaky_tests[] = {ORT_TSTR("fp16_inception_v1"),
ORT_TSTR("fp16_shufflenet"),
ORT_TSTR("fp16_tiny_yolov2"),
ORT_TSTR("candy"),
ORT_TSTR("tinyyolov3"),
ORT_TSTR("mlperf_ssd_mobilenet_300"),
ORT_TSTR("mlperf_ssd_resnet34_1200"),
ORT_TSTR("tf_inception_v1"),
ORT_TSTR("faster_rcnn"),
ORT_TSTR("split_zero_size_splits"),
ORT_TSTR("convtranspose_3d"),
ORT_TSTR("fp16_test_tiny_yolov2-Candy"),
ORT_TSTR("fp16_coreml_FNS-Candy"),
ORT_TSTR("fp16_test_tiny_yolov2"),
ORT_TSTR("fp16_test_shufflenet"),
ORT_TSTR("keras2coreml_SimpleRNN_ImageNet")};
static const ORTCHAR_T* openvino_disabled_tests[] = {
ORT_TSTR("tf_mobilenet_v1_1.0_224"),
ORT_TSTR("bertsquad"),
ORT_TSTR("yolov3"),
ORT_TSTR("LSTM_Seq_lens_unpacked"),
ORT_TSTR("tinyyolov3"),
ORT_TSTR("faster_rcnn"),
ORT_TSTR("mask_rcnn"),
ORT_TSTR("coreml_FNS-Candy_ImageNet"),
ORT_TSTR("tf_mobilenet_v2_1.0_224"),
ORT_TSTR("tf_mobilenet_v2_1.4_224"),
ORT_TSTR("operator_permute2"),
ORT_TSTR("operator_repeat"),
ORT_TSTR("operator_repeat_dim_overflow"),
ORT_TSTR("mlperf_ssd_resnet34_1200"),
ORT_TSTR("candy"),
ORT_TSTR("cntk_simple_seg"),
ORT_TSTR("GPT2_LM_HEAD"),
ORT_TSTR("mlperf_ssd_mobilenet_300"),
ORT_TSTR("fp16_coreml_FNS-Candy"),
ORT_TSTR("fp16_test_tiny_yolov2"),
ORT_TSTR("negative_log_likelihood_loss_input_shape_is_NCd1d2d3d4d5_mean_weight"),
ORT_TSTR("negative_log_likelihood_loss_input_shape_is_NCd1d2d3d4d5_mean_weight_expanded"),
ORT_TSTR("negative_log_likelihood_loss_input_shape_is_NCd1d2d3d4d5_none_no_weight"),
ORT_TSTR("negative_log_likelihood_loss_input_shape_is_NCd1d2d3d4d5_none_no_weight_expanded"),
ORT_TSTR("softmax_cross_entropy_input_shape_is_NCd1d2d3d4d5_mean_weight"),
ORT_TSTR("softmax_cross_entropy_input_shape_is_NCd1d2d3d4d5_mean_weight_expanded"),
ORT_TSTR("softmax_cross_entropy_input_shape_is_NCd1d2d3d4d5_mean_weight_log_prob"),
ORT_TSTR("softmax_cross_entropy_input_shape_is_NCd1d2d3d4d5_mean_weight_log_prob_expanded"),
ORT_TSTR("softmax_cross_entropy_input_shape_is_NCd1d2d3d4d5_none_no_weight"),
ORT_TSTR("softmax_cross_entropy_input_shape_is_NCd1d2d3d4d5_none_no_weight_expanded"),
ORT_TSTR("softmax_cross_entropy_input_shape_is_NCd1d2d3d4d5_none_no_weight_log_prob"),
ORT_TSTR("softmax_cross_entropy_input_shape_is_NCd1d2d3d4d5_none_no_weight_log_prob_expanded"),
// models from model zoo
ORT_TSTR("Tiny YOLOv3"),
ORT_TSTR("BERT-Squad"),
ORT_TSTR("YOLOv3"),
ORT_TSTR("Candy"),
ORT_TSTR("SSD"),
ORT_TSTR("ResNet101_DUC_HDC-12"),
ORT_TSTR("YOLOv3-12")};
static const ORTCHAR_T* dml_disabled_tests[] = {ORT_TSTR("mlperf_ssd_resnet34_1200"),
ORT_TSTR("mlperf_ssd_mobilenet_300"),
ORT_TSTR("mask_rcnn"),
ORT_TSTR("faster_rcnn"),
ORT_TSTR("tf_pnasnet_large"),
ORT_TSTR("zfnet512"),
ORT_TSTR("keras2coreml_Dense_ImageNet")};
static const ORTCHAR_T* dnnl_disabled_tests[] = {ORT_TSTR("densenet121"),
ORT_TSTR("resnet18v2"),
ORT_TSTR("resnet34v2"),
ORT_TSTR("resnet50v2"),
ORT_TSTR("resnet101v2"),
ORT_TSTR("resnet101v2"),
ORT_TSTR("vgg19"),
ORT_TSTR("tf_inception_resnet_v2"),