-
Notifications
You must be signed in to change notification settings - Fork 1.6k
/
auto_scan_base.py
941 lines (860 loc) · 38 KB
/
auto_scan_base.py
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
# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import numpy as np
import math
import unittest
import abc
import os
import sys
import platform
import enum
import time
import logging
import shutil
import paddle
import paddle.fluid as fluid
import global_var_model as gl
from paddle.fluid.initializer import NumpyArrayInitializer
from paddle.fluid.core import PassVersionChecker
import paddle.fluid.core as core
from paddle import compat as cpt
import paddle.inference as paddle_infer
from typing import Optional, List, Callable, Dict, Any, Set
from program_config import TensorConfig, OpConfig, ProgramConfig, create_fake_model, create_quant_model
from itertools import product
from program_config import CxxConfig, TargetType, PrecisionType, DataLayoutType, Place
import hypothesis
from hypothesis import given, settings, seed, Verbosity
import hypothesis.strategies as st
import argparse
parser = argparse.ArgumentParser()
parser.add_argument(
"--target",
choices=[
'Host', 'X86', 'CUDA', 'ARM', 'OpenCL', 'FPGA', 'NPU', 'XPU', 'BM',
'MLU', 'RKNPU', 'APU', 'HUAWEI_ASCEND_NPU', 'IMAGINATION_NNA',
'INTEL_FPGA', 'Metal', 'NNAdapter'
],
required=True)
parser.add_argument(
"--enforce_rpc", default='off', type=str, help="whther rpc is enforced")
parser.add_argument(
"--server_ip",
default="localhost",
type=str,
help="when rpc is used , the ip address of the server")
parser.add_argument(
"--url",
type=str,
help="Address of model download in model test", )
parser.add_argument(
"--file_name",
type=str,
help="File name of the compressed model package downloaded in the model test",
)
parser.add_argument(
"--model_name",
type=str,
help="Model name(That is, the prefix of the compressed package) in model test",
)
parser.add_argument(
"--input_shapes", help="The tested model's input_shapes", action="append")
parser.add_argument(
"--nnadapter_device_names",
default="",
type=str,
help="Set nnadapter device names")
parser.add_argument(
"--nnadapter_context_properties",
default="",
type=str,
help="Set nnadapter context properties")
parser.add_argument(
"--nnadapter_model_cache_dir",
default="",
type=str,
help="Set nnadapter model cache dir")
parser.add_argument(
"--nnadapter_subgraph_partition_config_path",
default="",
type=str,
help="Set nnadapter subgraph partition config path")
parser.add_argument(
"--nnadapter_mixed_precision_quantization_config_path",
default="",
type=str,
help="Set nnadapter mixed precision quantization config path")
args = parser.parse_args()
logging.basicConfig(level=logging.INFO, format="%(message)s")
logger = logging.getLogger(__name__)
logger.setLevel(
logging.INFO) # for Mac, logging only print waring/fatal, can't print info
settings.register_profile(
"ci",
max_examples=200,
suppress_health_check=hypothesis.HealthCheck.all(),
deadline=None,
print_blob=True,
derandomize=True,
report_multiple_bugs=False)
settings.register_profile(
"ce",
max_examples=1000,
suppress_health_check=hypothesis.HealthCheck.all(),
deadline=None,
print_blob=True,
derandomize=True,
report_multiple_bugs=False)
class IgnoreReasonsBase(enum.Enum):
# Paddle cannot predict normally (an error is reported in the prediction process) -- Both paddle and Lite do not predict
PADDLE_NOT_SUPPORT = 0
# Lite does not have a corresponding operator or Lite predicts an error -- Paddle predicts but Lite does not.
PADDLELITE_NOT_SUPPORT = 1
# When diff exists in the calculation results of Paddle and Lite -- Both Paddle and Lite predict,
# but do not compare the results.
ACCURACY_ERROR = 2
#For ignore op fusion
OP_FUSION_ERROR = 3
class AutoScanBaseTest(unittest.TestCase):
def __init__(self, *args, **kwargs):
self.valid_places = []
self.thread_num = [1]
np.random.seed(1024)
paddle.enable_static()
super(AutoScanBaseTest, self).__init__(*args, **kwargs)
self.ignore_cases = []
abs_dir = os.path.abspath(os.path.dirname(__file__))
self.cache_dir = os.path.join(abs_dir,
str(self.__module__) + '_cache_dir')
self.available_passes_in_framework = set()
args = parser.parse_args()
self.args = args
self.vaild_nnadapter_device_names = []
@abc.abstractmethod
def sample_program_configs(self, draw):
'''
Generate all config with the combination of different Input tensor shape and
different Attr values.
'''
raise NotImplementedError
@abc.abstractmethod
def prepare_input_data(self, draw):
raise NotImplementedError
@abc.abstractmethod
def sample_predictor_configs(self):
raise NotImplementedError
@abc.abstractmethod
def add_ignore_check_case(
self,
teller: [Callable[[ProgramConfig, CxxConfig], bool]],
reason: IgnoreReasonsBase,
note: str):
self.ignore_cases.append((teller, reason, note))
@abc.abstractmethod
def is_program_valid(self,
program_config: ProgramConfig,
predictor_config: CxxConfig) -> bool:
return True
def run_test_config(self, model, params, pred_config,
feed_data) -> Dict[str, np.ndarray]:
'''
Test a single case.
'''
pred_config.set_model_buffer(model, len(model), params, len(params))
predictor = paddle_infer.create_predictor(pred_config)
self.available_passes_in_framework = self.available_passes_in_framework | set(
pred_config.pass_builder().all_passes())
for name in feed_data:
input_tensor = predictor.get_input_handle(name)
input_tensor.copy_from_cpu(feed_data[name]['data'])
if feed_data[name]['lod'] is not None:
input_tensor.set_lod(feed_data[name]['lod'])
predictor.run()
result = {}
for o_name in predictor.get_output_names():
result[o_name] = predictor.get_output_handle(o_name).copy_to_cpu()
return result
# count FP16 precision diff
def count_fp16_diff(self, arr, base, atol, rtol) -> bool:
diff = abs(arr - base)
if len(diff) <= 0:
pass
max_diff = max(diff)
check = False
max_val = 0.0
max_index = 0
if max_diff > atol:
print("max_diff: ", max_diff)
size = len(arr)
count = 0
check = False
for i in range(size):
rel_val = diff[i] / max(arr[i], base[i])
if (diff[i] > 1e-1 and abs(rel_val > rtol)):
if max_val < rel_val:
max_val = rel_val
max_index = i
check = True
if check:
print("max_val and max_index: ", max_val, max_index)
print("value: ", base[max_index], arr[max_index],
diff[max_index])
print("FP16 Output has diff. ")
return False
else:
return True
return True
# count shape diff and data diff
def count_shape_and_diff(self, base, arr, atol, rtol, flag_precision_fp16):
base_shape = base.shape
arr_shape = arr.shape
base_len = len(base_shape)
arr_len = len(arr_shape)
diff_len = abs(base_len - arr_len)
if (base_len == 0):
# 0D-tensor
self.assertTrue(
arr_shape == base_shape,
"The output shapes are not equal, the baseline shape is " +
str(base_shape) + ', but got ' + str(arr.shape))
if flag_precision_fp16:
# count diff
arr_value = arr.flatten()
base_value = base.flatten()
# return true: has diff
res = self.count_fp16_diff(arr_value, base_value, atol, rtol)
self.assertTrue(res, "Output has diff. ")
else:
diff = abs(base - arr)
self.assertTrue(
np.allclose(
base.flatten(), arr.flatten(), atol=atol, rtol=rtol),
"Output has diff, max_diff : {}, index : {}.\nbase={}, \narr={}".
format(diff.max(), diff.argmax(), base, arr))
elif diff_len == 1:
# base=[1, K], arr=[k]
if base_len > arr_len and (base_shape[0] == 1 or
base_shape[-1] == 1):
new_shape = base_shape[0:base_len - 1]
if base_shape[0] == 1:
for i in range(1, base_len):
if i != 0:
new_shape[i - 1] = base_shape[i]
self.assertTrue(
new_shape == arr_shape,
"The output shapes are not equal, the baseline shape is " +
str(new_shape) + ', but got ' + str(arr.shape))
if flag_precision_fp16:
# count diff
arr_value = arr.flatten()
base_value = base.flatten()
# return true: has diff
res = self.count_fp16_diff(arr_value, base_value, atol,
rtol)
self.assertTrue(res, "Output has diff. ")
else:
diff = abs(base - arr)
self.assertTrue(
np.allclose(
base.flatten(),
arr.flatten(),
atol=atol,
rtol=rtol),
"Output has diff, max_diff : {}, index : {}.\nbase={}, \narr={}".
format(diff.max(), diff.argmax(), base, arr))
# arr=[1, K], base=[k]
elif base_len < arr_len and (arr_shape[0] == 1 or
arr_shape[-1] == 1):
new_shape = arr_shape[0:arr_len - 1]
if arr_shape[0] == 1:
for i in range(1, arr_len):
new_shape[i - 1] = arr_shape[i]
self.assertTrue(
new_shape == base_shape,
"The output shapes are not equal, the baseline shape is " +
str(base.shape) + ', but got ' + str(new_shape))
if flag_precision_fp16:
# count diff
arr_value = arr.flatten()
base_value = base.flatten()
# return true: has diff
res = self.count_fp16_diff(arr_value, base_value, atol,
rtol)
self.assertTrue(res, "Output has diff. ")
else:
diff = abs(base - arr)
self.assertTrue(
np.allclose(
base.flatten(),
arr.flatten(),
atol=atol,
rtol=rtol),
"Output has diff, max_diff : {}, index : {}.\nbase={}, \narr={}".
format(diff.max(), diff.argmax(), base, arr))
else:
self.assertTrue(
base.shape == arr.shape,
"The output shapes are not equal, the baseline shape is " +
str(base.shape) + ', but got ' + str(arr.shape))
else:
self.assertTrue(
base.shape == arr.shape,
"The output shapes are not equal, the baseline shape is " +
str(base.shape) + ', but got ' + str(arr.shape))
if flag_precision_fp16:
# count diff
arr_value = arr.flatten()
base_value = base.flatten()
# return False: has diff
res = self.count_fp16_diff(arr_value, base_value, atol, rtol)
self.assertTrue(res, "Output has diff. ")
else:
diff = abs(base - arr)
if diff.size != 0:
self.assertTrue(
np.allclose(
base, arr, atol=atol, rtol=rtol),
"Output has diff, max_diff : {}, index : {}.\nbase={}, \narr={}".
format(diff.max(), diff.argmax(), base, arr))
else:
self.assertTrue(
np.allclose(
base, arr, atol=atol, rtol=rtol),
"Output has diff,\nbase={}, \narr={}".format(base,
arr))
@abc.abstractmethod
def assert_tensors_near(self,
atol: float,
rtol: float,
tensor: Dict[str, np.array],
baseline: Dict[str, np.array],
flag_precision_fp16: False):
if len(tensor) == 0 and len(baseline) == 0:
return
if len(tensor) == 1 and len(baseline) == 1:
tensor_key = list(tensor.keys())
arr = np.array(tensor[tensor_key[0]])
base_key = list(baseline.keys())
base = np.array(baseline[base_key[0]])
self.count_shape_and_diff(base, arr, atol, rtol,
flag_precision_fp16)
else:
for key in tensor:
suffix_str = [
"/target_trans", "__Mangled_1", "/precision_trans"
]
paddlekey = key
for s_str in suffix_str:
index = key.rfind(s_str)
if index > 0:
paddlekey = key[0:index]
if (paddlekey == "saved_mean" or
paddlekey == "saved_variance" or
paddlekey == "mean_data" or
paddlekey == "variance_data"):
# training using data
continue
arr = np.array(tensor[key])
self.count_shape_and_diff(baseline[paddlekey], arr, atol, rtol,
flag_precision_fp16)
def generate_op_config(self,
ops_config: List[Dict[str, Any]]) -> List[OpConfig]:
ops = []
for i in range(len(ops_config)):
op_config = ops_config[i]
ops.append(
OpConfig(
type=op_config['op_type'],
inputs=op_config['op_inputs'],
outputs=op_config['op_outputs'],
attrs=op_config['op_attrs']))
return ops
@abc.abstractmethod
def ignore_log(self, msg: str):
logger.warning("SKIP: " + msg)
@abc.abstractmethod
def fail_log(self, msg: str):
logger.fatal("FAILE: " + msg)
@abc.abstractmethod
def success_log(self, msg: str):
logger.info("SUCCESS: " + msg)
@abc.abstractmethod
def is_model_test(self) -> bool:
return False
@abc.abstractmethod
def get_model(self, draw):
raise NotImplementedError
@abc.abstractmethod
def create_inference_config(self,
passes: Optional[List[str]]=None,
use_gpu: bool=False,
use_mkldnn: bool=False,
ir_optim: Optional[bool]=None):
config = paddle_infer.Config()
config.switch_ir_debug(True)
config.disable_glog_info()
if ir_optim is not None:
config.switch_ir_optim(ir_optim)
if use_gpu:
config.enable_use_gpu(100, 0)
if use_mkldnn:
config.enable_mkldnn()
if passes is not None:
config.pass_builder().set_passes(passes)
return config
@abc.abstractmethod
def insert_leaky_relu_op(self, prog_configs=None):
alpha_data = 0.01
leaky_relu_op = OpConfig(
type="leaky_relu",
inputs={"X": []},
outputs={"Out": ["output_act_data"]},
attrs={"alpha": alpha_data})
leaky_relu_op.inputs["X"].append(prog_configs.outputs[0])
prog_configs.ops.append(leaky_relu_op)
prog_configs.outputs[0] = "output_act_data"
def run_test(self, quant=False, prog_configs=None):
status = True
paddlelite_configs, op_list_, (atol_,
rtol_) = self.sample_predictor_configs()
for prog_config in prog_configs:
predictor_idx = -1
cnt = 0
for paddlelite_config in paddlelite_configs:
flag_precision_fp16 = False
if paddlelite_config.precision(
) == PrecisionType.FP16 and paddlelite_config.target(
) == TargetType.ARM:
flag_precision_fp16 = True
if paddlelite_config.precision(
) == PrecisionType.INT8 and paddlelite_config.target(
) == TargetType.ARM:
quant = True
else:
quant = False
predictor_idx += 1
# judge validity of program
if not self.is_program_valid(prog_config, paddlelite_config):
self.num_invalid_programs_list[predictor_idx] += 1
continue
self.num_ran_programs_list[predictor_idx] += 1
if flag_precision_fp16:
if platform.system() == 'Linux':
# only run in M1
continue
# creat model and prepare feed data
if flag_precision_fp16:
atol_ = 1e-1
rtol_ = 5.3e-2
if quant:
if platform.system() == 'Darwin' or platform.processor(
) == 'x86_64':
# only run in linux
continue
atol_ = 8e-3
rtol_ = 8e-3
if cnt == 0:
self.insert_leaky_relu_op(prog_config)
cnt = cnt + 1
model, params = create_fake_model(prog_config)
model, params = create_quant_model(
model, params, self.cache_dir, prog_config)
else:
model, params = create_fake_model(prog_config)
feed_data = {}
for name, tensor_config in prog_config.inputs.items():
feed_data[name] = {
'data': tensor_config.data,
'lod': tensor_config.lod
}
results: List[Dict[str, np.ndarray]] = []
# ignore info
accuracy_error_flag = False
paddle_not_support_flag = False
paddle_lite_not_support_flag = False
op_fusion_error_flag = False
pred_config = paddlelite_config.value()
for ignore_info in self.ignore_cases:
if ignore_info[0](prog_config, paddlelite_config):
self.num_ignore_tests_list[predictor_idx] += 1
if ignore_info[1] == IgnoreReasonsBase.ACCURACY_ERROR:
accuracy_error_flag = True
self.ignore_log("[ACCURACY_ERROR] " + ignore_info[
2] + ' ' + ' vs ' + self.paddlelite_config_str(
pred_config))
gl.set_out_diff_ops(
self.get_target(),
self.get_nnadapter_device_name(), sys.argv[0])
elif ignore_info[
1] == IgnoreReasonsBase.PADDLELITE_NOT_SUPPORT:
paddle_lite_not_support_flag = True
self.ignore_log("[PADDLELITE_NOT_SUPPORT ERROR] " +
ignore_info[2] + ' ' + ' vs ' +
self.paddlelite_config_str(
pred_config))
elif ignore_info[
1] == IgnoreReasonsBase.PADDLE_NOT_SUPPORT:
paddle_not_support_flag = True
self.ignore_log("[PADDLE_NOT_SUPPORT ERROR] " +
ignore_info[2] + ' ' + ' vs ' +
self.paddlelite_config_str(
pred_config))
elif ignore_info[
1] == IgnoreReasonsBase.OP_FUSION_ERROR:
op_fusion_error_flag = True
self.ignore_log("[OP_FUSION ERROR] " + ignore_info[
2] + ' ' + ' vs ' + self.paddlelite_config_str(
pred_config))
else:
raise NotImplementedError
if paddle_not_support_flag:
gl.set_paddle_not_supported_ops(
self.get_target(),
self.get_nnadapter_device_name(), sys.argv[0])
continue
# baseline: cpu no ir_optim run
base_config = self.create_inference_config(ir_optim=False)
logger.info('[ProgramConfig]: ' + str(prog_config))
results.append(
self.run_test_config(model, params, base_config,
feed_data))
if paddle_lite_not_support_flag:
gl.set_lite_not_supported_ops(
self.get_target(),
self.get_nnadapter_device_name(), sys.argv[0])
continue
if os.path.exists(self.cache_dir):
shutil.rmtree(self.cache_dir)
if not os.path.exists(self.cache_dir):
os.mkdir(self.cache_dir)
try:
result, opt_model_bytes = self.run_lite_config(
model, params, feed_data, pred_config, args.server_ip)
results.append(result)
# add ignore methods
if self.passes is not None: # pass check
if not accuracy_error_flag:
self.assert_tensors_near(atol_, rtol_, results[-1],
results[0],
flag_precision_fp16)
if not op_fusion_error_flag:
self.assert_op_list(opt_model_bytes, op_list_)
else: # op check
op_white_list = ["empty", "gaussian_random"]
self.assert_kernel_type(opt_model_bytes, op_list_,
paddlelite_config)
if not accuracy_error_flag:
if str(prog_config.ops[0]
.type) not in op_white_list:
self.assert_tensors_near(
atol_, rtol_, results[-1], results[0],
flag_precision_fp16)
else:
logger.info('[AccuracyWarning] ' + str(
prog_config.ops[0].type) +
' skip accuracy test.')
except Exception as e:
self.fail_log(
self.paddlelite_config_str(pred_config) +
'\033[1;31m \nERROR INFO: {}\033[0m'.format(str(e)))
status = False
break
self.success_log('PredictorConfig: ' +
self.paddlelite_config_str(pred_config))
self.assertTrue(status)
gl.set_success_ops(self.get_target(),
self.get_nnadapter_device_name(), sys.argv[0])
def run_model_test(self, inputs_configs=None, model=None, params=None):
status = True
paddlelite_configs, _, (atol_, rtol_) = self.sample_predictor_configs()
for inputs_config in inputs_configs:
feed_data = {}
for name, tensor_config in inputs_config.items():
feed_data[name] = {
'data': tensor_config.data,
'lod': tensor_config.lod
}
results: List[Dict[str, np.ndarray]] = []
# baseline: cpu no ir_optim run
base_config = self.create_inference_config(ir_optim=False)
results.append(
self.run_test_config(model, params, base_config, feed_data))
flag_precision_fp16 = False
for paddlelite_config in paddlelite_configs:
pred_config = paddlelite_config.value()
try:
result, opt_model_bytes = self.run_lite_config(
model, params, feed_data, pred_config, args.server_ip)
results.append(result)
self.assert_tensors_near(atol_, rtol_, results[-1],
results[0], flag_precision_fp16)
except Exception as e:
self.fail_log(
self.paddlelite_config_str(pred_config) +
'\033[1;31m \nERROR INFO: {}\033[0m'.format(str(e)))
status = False
break
self.success_log('PredictorConfig: ' +
self.paddlelite_config_str(pred_config))
self.assertTrue(status)
def inference_config_str(self, config) -> bool:
dic = {}
enable_mkldnn = config.mkldnn_enabled()
dic['use_mkldnn'] = enable_mkldnn
enable_gpu = config.use_gpu()
return str(dic)
def paddlelite_config_str(self, config) -> bool:
return str(config)
# method for ignoring
def add_ignore_pass_case(self):
return
# judge if program contain op_list
def assert_op_list(self, model_bytes, op_list_after_fusion):
if not self.passes:
raise ValueError(
"In PassAutoScan you should give a valid pass name.")
pg = paddle.static.deserialize_program(model_bytes)
main_block = pg.desc.block(0)
after_op_list = list()
for i in range(main_block.op_size()):
if main_block.op(i).type(
) in ["feed", "fetch", "layout", "io_copy"]:
continue
after_op_list.append(main_block.op(i).type())
self.assertTrue(
op_list_after_fusion == after_op_list,
"Expected operator list after fusion is {}, but now it's {}".
format(op_list_after_fusion, after_op_list), )
# judge if correct kernel is picked
def assert_kernel_type(self, model_bytes, op_list, paddlelite_config):
pg = paddle.static.deserialize_program(model_bytes)
main_block = pg.desc.block(0)
after_op_list = list()
target_ = paddlelite_config.target()
precision_ = paddlelite_config.precision()
layout_ = paddlelite_config.layout()
for i in range(main_block.op_size()):
if main_block.op(i).type() in op_list:
kernel_type_info = main_block.op(i).attr(
"__@kernel_type_attr@__").split("/")
self.assertTrue(
len(kernel_type_info) == 5,
"Incompleted kernel info of {}:{}".format(
main_block.op(i).type(),
main_block.op(i).attr("__@kernel_type_attr@__")))
current_target_ = TargetType(int(kernel_type_info[2]))
current_precision_ = PrecisionType(int(kernel_type_info[3]))
current_layout_ = DataLayoutType(int(kernel_type_info[4]))
correct_kernel_flag_ = (target_ == current_target_) and (
precision_ == current_precision_ or
current_precision_ == PrecisionType.Any) and (
layout_ == current_layout_ or
current_layout_ == DataLayoutType.Any)
self.assertTrue(
correct_kernel_flag_ == True,
"Expected kernel_type of op {} is ({},{},{}), but now it's ({},{},{})".
format(
main_block.op(i).type(), target_, precision_, layout_,
current_target_, current_precision_, current_layout_))
def run_and_statis(
self,
quant=False,
max_examples=100,
reproduce=None,
min_success_num=25,
passes=None,
model=None,
params=None, ):
self.init_statistical_parameters()
settings.register_profile(
"dev",
max_examples=max_examples,
suppress_health_check=hypothesis.HealthCheck.all(),
deadline=None,
print_blob=True,
derandomize=True,
report_multiple_bugs=False,
verbosity=Verbosity.verbose)
if os.getenv('HYPOTHESIS_TEST_PROFILE') == "ci":
settings.load_profile("ci")
elif os.getenv('HYPOTHESIS_TEST_PROFILE') == "ce":
settings.load_profile("ce")
else:
settings.load_profile("dev")
self.passes = passes
self.add_ignore_pass_case()
def program_generator(draw):
return self.sample_program_configs(draw)
def inputs_generator(draw):
return self.prepare_input_data(draw)
def run_test(prog_config):
return self.run_test(quant=quant, prog_configs=[prog_config])
def run_model_test(inputs_configs):
return self.run_model_test(
inputs_configs=[inputs_configs], model=model, params=params)
# if current unittest is not active on the input targ paddlelite_not_support_flag = Trueet, we will exit directly.
gl.set_all_test_ops(self.get_target(),
self.get_nnadapter_device_name(), sys.argv[0])
if not self.is_actived():
logger.info("Error: This test is not actived on " +
self.get_target())
return
if not self.is_nnadapter_device_actived():
logger.info("Error: This test is not actived on " +
self.get_nnadapter_device_name())
return
if self.get_target().upper() == "NNADAPTER":
self.assertTrue(
self.args.nnadapter_device_names != "",
"Args Error: Please set nnadapter_device_names when target=nnadapter!"
)
if self.is_model_test():
generator = st.composite(inputs_generator)
loop_func = given(generator())(run_model_test)
else:
generator = st.composite(program_generator)
loop_func = given(generator())(run_test)
if reproduce is not None:
loop_func = reproduce(loop_func)
logger.info("Start to running test of {}".format(type(self)))
loop_func()
if self.is_model_test():
logger.info(
"===================Statistical Information===================")
logger.info("Number of Input Configs: {}".format(max_examples))
logger.info("Number of Predictor Kinds: {}".format(
int(self.num_predictor_kinds)))
else:
logger.info(
"===================Statistical Information===================")
logger.info("Number of Generated Programs: {}".format(
max_examples))
logger.info("Number of Predictor Kinds: {}".format(
int(self.num_predictor_kinds)))
self.assertTrue(self.num_predictor_kinds > 0,
"Number of Predictor Kinds must be greater than 0")
logger.info("Number of Ran Programs: {}".format(
self.num_ran_programs_list))
logger.info("Number of Invalid Programs: {}".format(
self.num_invalid_programs_list))
logger.info("Number of Ignored Tests: {}".format(
self.num_ignore_tests_list))
successful_ran_programs = int(
(sum(self.num_ran_programs_list) +
sum(self.num_ignore_tests_list)) / self.num_predictor_kinds)
logger.info(
"Number of successfully ran programs approximately equal to {}".
format(successful_ran_programs))
if successful_ran_programs < min_success_num:
logger.fatal(
"At least {} programs need to ran successfully, but now only about {} programs satisfied.".
format(min_success_num, successful_ran_programs))
assert False
@abc.abstractmethod
def run_lite_config(self,
model,
params,
feed_data,
pred_config,
server_ip="localhost") -> Dict[str, np.ndarray]:
raise NotImplementedError
# enable a predictor config
# configs will be generated automatically according to inputs
def enable_testing_on_place(self,
target=None,
precision=None,
layout=None,
thread=None,
places=None) -> None:
# set thread_num
if isinstance(thread, list):
self.thread_num = list(set(self.thread_num + thread))
if isinstance(thread, int):
self.thread_num.append(thread)
self.thread_num = list(self.thread_num)
# arm basic places:
arm_basic_places = [
Place(TargetType.ARM, PrecisionType.INT32),
Place(TargetType.ARM, PrecisionType.INT64)
]
# if list[Place] is inputed, this will be used directly
if places is not None:
assert isinstance(places, list)
self.valid_places.append(places)
return
# otherwise we will generate a list[Place] from the inputed[target\precision\layout]
assert (target is not None)
target_ = target if isinstance(target, list) else [target]
self.target = target_
precision_ = precision if isinstance(precision, list) else [precision]
layout_ = layout if isinstance(layout, list) else [layout]
for tar_, pre_, lay_ in product(target_, precision_, layout_):
if (tar_ == TargetType.ARM):
self.valid_places.append([Place(tar_, pre_, lay_)] +
arm_basic_places)
else:
self.valid_places.append([Place(tar_, pre_, lay_)])
def enable_devices_on_nnadapter(self, device_names=list) -> None:
self.vaild_nnadapter_device_names = device_names
def get_target(self) -> str:
return self.args.target
def is_actived(self) -> bool:
for valid_place_ in self.valid_places:
if self.get_target() in valid_place_[0]:
return True
return False
def is_nnadapter_device_actived(self) -> bool:
if self.get_target().upper() != "NNADAPTER":
return True
if self.get_nnadapter_device_name(
) in self.vaild_nnadapter_device_names:
return True
return False
def get_nnadapter_device_name(self) -> str:
nnadapter_device_name_list = self.args.nnadapter_device_names.split(
",")
return nnadapter_device_name_list[0]
def get_predictor_configs(self) -> List[CxxConfig]:
return self.target_to_predictor_configs(self, self.get_target())
def init_statistical_parameters(self):
self.num_predictor_kinds = len(self.get_predictor_configs())
self.num_invalid_programs_list = [0] * self.num_predictor_kinds
self.num_ran_programs_list = [0] * self.num_predictor_kinds
self.num_ignore_tests_list = [0] * self.num_predictor_kinds
@staticmethod
def nnadapter_config_set(self, config: CxxConfig):
config.set_nnadapter_device_names(
self.args.nnadapter_device_names.split(","))
config.set_nnadapter_context_properties(
self.args.nnadapter_context_properties)
config.set_nnadapter_model_cache_dir(
self.args.nnadapter_model_cache_dir)
config.set_nnadapter_subgraph_partition_config_path(
self.args.nnadapter_subgraph_partition_config_path)
config.set_nnadapter_mixed_precision_quantization_config_path(
self.args.nnadapter_mixed_precision_quantization_config_path)
# get valid test configs
@staticmethod
def target_to_predictor_configs(self, target: str) -> List[CxxConfig]:
configs_ = []
for elem_ in self.valid_places:
if target in elem_[0]:
for thread_ in self.thread_num:
config_ = CxxConfig()
config_.set_valid_places(elem_)
config_.set_threads(thread_)
if target.upper() == "NNADAPTER":
self.nnadapter_config_set(self, config_)
configs_.append(config_)
return configs_