-
Notifications
You must be signed in to change notification settings - Fork 534
/
importerUtils.cpp
2207 lines (1984 loc) · 86.6 KB
/
importerUtils.cpp
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
/*
* SPDX-License-Identifier: Apache-2.0
*/
#include "importerUtils.hpp"
#include "NvInferSafeRuntime.h"
#include "OnnxAttrs.hpp"
#include "bfloat16.hpp"
#include <ctype.h>
#include <regex>
#include <set>
namespace onnx2trt
{
void PluginDeleter::operator()(nvinfer1::IPluginV2* t)
{
t->destroy();
}
Status notInvalidType(TensorOrWeights const& input, std::vector<std::string> const& invalidTypes,
::ONNX_NAMESPACE::NodeProto const& node, size_t const nodeIdx)
{
bool invalid = std::any_of(invalidTypes.begin(), invalidTypes.end(),
[&](std::string invalidType) { return input.getType() == invalidType; });
if (invalid)
{
ASSERT_NODE(
false, "Found unsupported input type of " << input.getType(), node, nodeIdx, ErrorCode::kUNSUPPORTED_NODE);
}
return Status::success();
}
NodeImportResult activationHelper(ImporterContext* ctx, const ::ONNX_NAMESPACE::NodeProto& node, size_t const nodeIdx,
std::vector<TensorOrWeights>& inputs, nvinfer1::ActivationType op, float* alpha, float* beta)
{
CHECK_STATUS(notInvalidType(inputs.at(0), {"INT32", "BOOL", "UINT8"}, node, nodeIdx));
nvinfer1::ITensor& input = convertToTensor(inputs.at(0), ctx);
nvinfer1::IActivationLayer* layer = N_CHECK(ctx->network()->addActivation(input, op));
ASSERT_NODE(layer, "Failed to add activation layer!", node, nodeIdx, ErrorCode::kINTERNAL_ERROR);
if (alpha)
{
layer->setAlpha(*alpha);
}
if (beta)
{
layer->setBeta(*beta);
}
ctx->registerLayer(layer, node);
auto* output = N_CHECK(layer->getOutput(0));
return {{output}};
}
nvinfer1::ITensor* addClip(ImporterContext* ctx, nvinfer1::ITensor* input, float clip)
{
if (clip >= 0.f)
{
nvinfer1::IActivationLayer* layer = N_CHECK(ctx->network()->addActivation(*input, nvinfer1::ActivationType::kCLIP));
layer->setAlpha(-clip);
layer->setBeta(clip);
return N_CHECK(layer->getOutput(0));
}
return input;
};
NodeImportResult argMinMaxHelper(ImporterContext* ctx, const ::ONNX_NAMESPACE::NodeProto& node, size_t const nodeIdx,
std::vector<TensorOrWeights>& inputs, nvinfer1::TopKOperation op)
{
CHECK_STATUS(notInvalidType(inputs.at(0), {"UINT8"}, node, nodeIdx));
nvinfer1::ITensor* tensor = &convertToTensor(inputs.at(0), ctx);
// Get attributes.
OnnxAttrs attrs(node, ctx);
int32_t keepdims = attrs.get("keepdims", 1);
int32_t axis = attrs.get("axis", 0);
int32_t selectLastIndex = attrs.get<int32_t>("select_last_index", 0);
// Insert a TopK layer with k set to 1.
int32_t nbDims = tensor->getDimensions().nbDims;
CHECK_STATUS(convertAxis(axis, nbDims, node, nodeIdx));
uint32_t axisMask = 1 << axis;
nvinfer1::ITopKLayer* layer;
// New attribute added to Opset-12
// Whether to select the last index or the first index if the {name} appears in multiple indices, default is False
// (first index).
if (selectLastIndex)
{
// Need to flip the data input along the given axis using the Slice operator
auto const dims = shapeOf(*tensor);
ShapeTensor starts = shapeVector(-1);
ShapeTensor ends = shapeVector(static_cast<int64_t>(INT_MIN));
ShapeTensor axes = shapeVector(axis);
ShapeTensor steps = shapeVector(-1);
if (axes.size() < dims.size())
{
// axes specify a subset of the dimensions, or out of order.
// Convert starts/ends/steps to complete in-order form.
ShapeTensor const subscripts{axesToInterlaceSubscripts(axes, dims.size())};
starts = interlace(ctx, similar(ctx, dims, 0), starts, subscripts);
ends = interlace(ctx, dims, ends, subscripts);
steps = interlace(ctx, similar(ctx, dims, 1), steps, subscripts);
}
decodeOnnxStartsAndEnds(ctx, dims, steps, starts, ends);
// TensorRT uses sizes of the output dimensions instead of ends.
ShapeTensor const sizes = computeSliceSizes(ctx, starts, ends, steps, dims);
nvinfer1::ISliceLayer* slice = addSlice(ctx, *tensor, starts, sizes, steps);
auto flippedTensor = N_CHECK(slice->getOutput(0));
layer = N_CHECK(ctx->network()->addTopK(*flippedTensor, op, 1, axisMask));
}
else
{
layer = N_CHECK(ctx->network()->addTopK(*tensor, op, 1, axisMask));
}
ctx->registerLayer(layer, node);
// We don't care about the TopK values, just the indices.
nvinfer1::ITensor* indices = N_CHECK(layer->getOutput(1));
indices = castHelper(ctx, indices, nvinfer1::DataType::kINT64);
// If selectLastIndex is true, the TopK operation was performed on reversed data on the provided axis.
// Convert reversed indices back to forward indices by calculating the following:
// indices = shape(tensor)[axis] - indices - 1
if (selectLastIndex)
{
// Use shapeTensor semantics to support dynamic shapes
auto const dims = shapeOf(*tensor);
auto const indicesDims = shapeOf(*indices);
auto const axisTensor = shapeVector(axis);
auto const dimOnAxis = gather(ctx, dims, axisTensor);
// Create constant of shape indicesDims with values tensor.shape[axis]
auto const tensorDimOnAxis = constantOfShape(ctx, node, &dimOnAxis.tensor(ctx), &indicesDims.tensor(ctx));
// Create constant of shape indicesDims with values of 1
auto const ones = constantOfShape(ctx, node, &shapeVector(1).tensor(ctx), &indicesDims.tensor(ctx));
std::vector<TensorOrWeights> newInputs{tensorDimOnAxis, indices, ones};
std::vector<TensorOrWeights> indicesUpdate;
GET_VALUE(
elementwiseHelper(ctx, node, nodeIdx, newInputs, nvinfer1::ElementWiseOperation::kSUB), &indicesUpdate);
indices = &convertToTensor(indicesUpdate.at(0), ctx);
}
// The default behavior of the TopK layer is to keepdims.
// Otherwise, we need to squeeze the axis dimension.
if (!keepdims)
{
std::vector<int32_t> axes{axis};
indices = squeezeTensor(ctx, node, *indices, axes);
}
// TensorRT doesn't support int64 for TopK indices
indices = castHelper(ctx, indices, nvinfer1::DataType::kINT64);
return {{indices}};
}
Status broadcastTensor(ImporterContext* ctx, nvinfer1::ITensor*& t, int const nbDims)
{
ASSERT(ctx->getOpsetVersion() >= 7 && "Pre-opset 7 broadcasting is unsupported in this version of the ONNX parser",
ErrorCode::kUNSUPPORTED_NODE);
auto const inputDims = shapeOf(*t);
int const nbInputDims = inputDims.size();
ASSERT((nbInputDims <= nbDims) && "Cannot broadcast a higher rank tensor to a lower rank tensor.",
ErrorCode::kUNSUPPORTED_NODE);
if (nbInputDims < nbDims)
{
nvinfer1::IShuffleLayer* reshape
= addShuffle(ctx, *t, concat(ctx, fillShapeVector(ctx, 1, shapeVector(nbDims - nbInputDims)), shapeOf(*t)));
ctx->registerLayer(reshape, "ONNXTRT_Broadcast", nullptr);
t = N_CHECK(reshape->getOutput(0));
}
return Status::success();
}
Status broadcastTensors(ImporterContext* ctx, nvinfer1::ITensor*& t1, nvinfer1::ITensor*& t2)
{
int const t1Dims = t1->getDimensions().nbDims;
int const t2Dims = t2->getDimensions().nbDims;
if (t1Dims == t2Dims)
{
return Status::success();
}
if (t1Dims > t2Dims)
{
return broadcastTensor(ctx, t2, t1Dims);
}
return broadcastTensor(ctx, t1, t2Dims);
}
Status broadcastTensors(ImporterContext* ctx, nvinfer1::ITensor*& t1, nvinfer1::ITensor*& t2, nvinfer1::ITensor*& t3)
{
int const maxDims = std::max({t1->getDimensions().nbDims, t2->getDimensions().nbDims, t3->getDimensions().nbDims});
CHECK_STATUS(broadcastTensor(ctx, t1, maxDims));
CHECK_STATUS(broadcastTensor(ctx, t2, maxDims));
CHECK_STATUS(broadcastTensor(ctx, t3, maxDims));
return Status::success();
}
// Helper functions for calculateBias:
int32_t getBias(std::vector<int32_t> const& dimension_count, std::vector<int32_t> const& pitches, int32_t axis)
{
int32_t result{0};
for (int32_t i = 0; i < static_cast<int32_t>(dimension_count.size()); i++)
{
if (i != axis)
{
result += dimension_count[i] * pitches[i];
}
}
return result;
}
void incrementOuterDimension(std::vector<int32_t>& dimensionCount, nvinfer1::Dims idxDims)
{
// Start at [x,x,0]. Increment starting from the outer dimension.
int32_t rank = dimensionCount.size();
for (int32_t i = rank - 1; i >= 0; i--)
{
int dimLimit = idxDims.d[i];
// If we're not at the limit, increment current axis and return
if (++dimensionCount[i] != dimLimit)
{
break;
}
// Else, we increment on the next dimension and reset current one
dimensionCount[i] = 0;
}
}
std::vector<int32_t> calculateBias(
nvinfer1::Dims const& daDims, nvinfer1::Dims const& idxDims, std::vector<int32_t> const& pitches, int32_t axis)
{
std::vector<int32_t> biasVector;
std::vector<int32_t> dimensionCount(daDims.nbDims, 0);
int64_t total = volume(idxDims);
for (int64_t i = 0; i < total; i++)
{
int32_t bias = getBias(dimensionCount, pitches, axis);
biasVector.push_back(bias);
incrementOuterDimension(dimensionCount, idxDims);
}
return biasVector;
}
std::vector<int32_t> calculatePitches(nvinfer1::Dims const& inputDims)
{
int32_t pitch = 1;
int32_t nbDims = inputDims.nbDims;
std::vector<int32_t> pitches(nbDims);
pitches[nbDims - 1] = pitch;
for (int32_t i = nbDims - 2; i >= 0; i--)
{
pitch *= inputDims.d[i + 1];
pitches[i] = pitch;
}
return pitches;
}
bool canUseNDResize(size_t const scaleSize, float const* scaleFactors, size_t const n)
{
// Linear resize supports up to 3D resize on the outermost dimensions (n = 3).
if (scaleSize > n)
{
for (size_t i = 0; i < scaleSize - n; i++)
{
if (scaleFactors[i] != 1)
{
return false;
}
}
}
return true;
}
nvinfer1::ITensor* castHelper(ImporterContext* ctx, nvinfer1::ITensor* input, nvinfer1::DataType dtype)
{
nvinfer1::ICastLayer* cast = N_CHECK(ctx->network()->addCast(*input, dtype));
ctx->registerLayer(cast, "ONNXTRT_castHelper", nullptr);
return N_CHECK(cast->getOutput(0));
}
nvinfer1::ITensor* constantOfShape(ImporterContext* ctx, const ::ONNX_NAMESPACE::NodeProto& node,
nvinfer1::ITensor* constant, nvinfer1::ITensor* shape)
{
ShapeTensor shapeT{*shape};
ShapeTensor zeros = similar(ctx, shapeT, 0);
// `constant` must be broadcasted to the same rank as `shape`.
ShapeTensor broadcastedShape = similar(ctx, shapeT, 1);
constant = &reshape(ctx, *constant, broadcastedShape);
auto* l = N_CHECK(addSlice(ctx, *constant, zeros, shapeT, zeros));
return N_CHECK(l->getOutput(0));
}
Status convertAxis(int32_t& axis, int32_t const nbDims, ::ONNX_NAMESPACE::NodeProto const& node, size_t const nodeIdx)
{
// Support negative indexing
if (axis < 0)
{
axis += nbDims;
}
// Support nbDims as a valid axis for QuantDequantLinearHelper
ASSERT_NODE((axis >= 0 && axis <= nbDims),
"Axis must be in the range [0, nbDims (" << nbDims << ")]. Provided axis is: " << axis, node, nodeIdx,
ErrorCode::kUNSUPPORTED_NODE);
return Status::success();
}
bool convertDtype(int32_t onnx_dtype, nvinfer1::DataType* trt_dtype)
{
switch (onnx_dtype)
{
case ::ONNX_NAMESPACE::TensorProto::DOUBLE: *trt_dtype = nvinfer1::DataType::kFLOAT; break;
case ::ONNX_NAMESPACE::TensorProto::FLOAT: *trt_dtype = nvinfer1::DataType::kFLOAT; break;
case ::ONNX_NAMESPACE::TensorProto::INT8: *trt_dtype = nvinfer1::DataType::kINT8; break;
case ::ONNX_NAMESPACE::TensorProto::UINT8: *trt_dtype = nvinfer1::DataType::kUINT8; break;
case ::ONNX_NAMESPACE::TensorProto::FLOAT16: *trt_dtype = nvinfer1::DataType::kHALF; break;
case ::ONNX_NAMESPACE::TensorProto::BFLOAT16: *trt_dtype = nvinfer1::DataType::kBF16; break;
case ::ONNX_NAMESPACE::TensorProto::BOOL: *trt_dtype = nvinfer1::DataType::kBOOL; break;
case ::ONNX_NAMESPACE::TensorProto::INT32: *trt_dtype = nvinfer1::DataType::kINT32; break;
case ::ONNX_NAMESPACE::TensorProto::INT64: *trt_dtype = nvinfer1::DataType::kINT64; break;
case ::ONNX_NAMESPACE::TensorProto::FLOAT8E4M3FN: *trt_dtype = nvinfer1::DataType::kFP8; break;
case ::ONNX_NAMESPACE::TensorProto::INT4: *trt_dtype = nvinfer1::DataType::kINT4; break;
default:
std::cerr << "Unsupported ONNX data type: " << getDtypeName(onnx_dtype) << " (" << std::to_string(onnx_dtype)
<< ")" << std::endl;
return false;
}
return true;
}
bool convertOnnxPadding(ImporterContext* ctx, int32_t nbInputDims, std::vector<int64_t> const& onnxPadding,
nvinfer1::ITensor*& startTensor, nvinfer1::ITensor*& totalPaddingTensor)
{
std::vector<int64_t> start;
std::vector<int64_t> totalPadding;
if (onnxPadding.size() % 2U != 0)
{
return false;
}
auto const diff = nbInputDims - static_cast<int32_t>(onnxPadding.size() / 2U);
if (diff < 0)
{
return false;
}
start.resize(nbInputDims, 0);
totalPadding.resize(nbInputDims, 0);
for (int32_t i = diff; i < nbInputDims; i++)
{
auto const idx = i - diff;
auto const pre = onnxPadding[idx];
auto const post = onnxPadding[onnxPadding.size() / 2U + idx];
if (pre < 0 || post < 0)
{
return false;
}
start[i] = -pre;
totalPadding[i] = pre + post;
}
auto* startLayer = N_CHECK(addConstant(ctx, start, ::ONNX_NAMESPACE::TensorProto::INT64, nvinfer1::Dims{1, {nbInputDims}}));
startTensor = N_CHECK(startLayer->getOutput(0));
auto* totalPaddingLayer = N_CHECK(addConstant(ctx, totalPadding, ::ONNX_NAMESPACE::TensorProto::INT64, nvinfer1::Dims{1, {nbInputDims}}));
totalPaddingTensor = N_CHECK(totalPaddingLayer->getOutput(0));
return startTensor && totalPaddingTensor;
}
bool shiftIsAllZeros(ShapedWeights const& shift)
{
// Check if all of the values in the shift tensor are zeros. Shift dtype is one of [INT8, UINT8, INT4, UINT4]
auto const* v = static_cast<int8_t const*>(shift.values);
size_t const count = shift.size_bytes();
auto allZeros = std::all_of(v, v + count, [](int8_t x) { return x == 0; });
return allZeros;
}
onnx2trt::ShapedWeights createZeroShifts(onnx2trt::ShapedWeights const& shiftInt, int32_t type, ImporterContext* ctx)
{
if (!shiftIsAllZeros(shiftInt))
{
LOG_WARNING("TensorRT currenly supports only zero shifts values for QuatizeLinear/DequantizeLinear ops");
}
auto shift = ctx->createNamedTempWeights(type, shiftInt.shape);
float* sh = static_cast<float*>(shift.values);
for (int i = 0, n = shift.count(); i < n; i++)
{
sh[i] = 0.0f;
}
return shift;
}
nvinfer1::ITensor* createZeroTensor(ImporterContext* ctx, nvinfer1::ITensor* data)
{
auto* zeroLayer = N_CHECK(addConstant(ctx, std::vector<float>{0.f}, ::ONNX_NAMESPACE::TensorProto::FLOAT, {0, {1}}));
auto* zeroTensor = N_CHECK(zeroLayer->getOutput(0));
zeroTensor = castHelper(ctx, zeroTensor, data->getType());
auto result = broadcastTensors(ctx, zeroTensor, data);
if (result.is_error())
{
return nullptr;
}
auto* zeroBroadcastLayer = N_CHECK(ctx->network()->addElementWise(*data, *zeroTensor, nvinfer1::ElementWiseOperation::kPROD));
ctx->registerLayer(zeroBroadcastLayer, "ONNXTRT_createZeroTensor", nullptr);
return N_CHECK(zeroBroadcastLayer->getOutput(0));
}
nvinfer1::ITensor* convertToScalar(ImporterContext* ctx, nvinfer1::ITensor* inpTensor)
{
if (inpTensor->getDimensions().nbDims == 0)
{
return inpTensor;
}
auto const tensorVolume = volume(inpTensor->getDimensions());
if (tensorVolume != 1)
{
LOG_VERBOSE("Cannot convert tensor to scalar. Note: Tensor dimensions were: "
<< inpTensor->getDimensions() << ", with volume: " << tensorVolume);
return nullptr;
}
nvinfer1::IShuffleLayer* reshape = N_CHECK(ctx->network()->addShuffle(*inpTensor));
reshape->setReshapeDimensions(nvinfer1::Dims{0});
ctx->registerLayer(reshape, "ONNXTRT_convertToScalar", nullptr);
// Do not need to call setZeroIsPlaceholder, since reshape dimensions are empty.
return N_CHECK(reshape->getOutput(0));
}
nvinfer1::ITensor& convertToTensor(TensorOrWeights& input, ImporterContext* ctx)
{
if (input.is_tensor())
{
return input.tensor();
}
// Handle non-tensor indices input by adding a new constant layer to the network.
ShapedWeights& weights = input.weights();
auto const existingConstantLayer = ctx->getConstantLayer(weights.getName());
if (existingConstantLayer != nullptr)
{
return *N_CHECK(existingConstantLayer->getOutput(0));
}
auto* constantLayer = N_CHECK(ctx->network()->addConstant(weights.shape, weights));
// Register layer and constant name (if set) into RefitMap:
if (weights.getName())
{
ctx->registerLayer(constantLayer, weights.getName(), nullptr);
ctx->network()->setWeightsName(weights, weights.getName());
}
auto* output = N_CHECK(constantLayer->getOutput(0));
return *output;
}
nvinfer1::ITensor* convertToScalar(TensorOrWeights& input, ImporterContext* ctx)
{
if (input.is_tensor())
{
return convertToScalar(ctx, &input.tensor());
}
ShapedWeights& weights = input.weights();
if (volume(weights.shape) != 1)
{
LOG_VERBOSE("Cannot convert weights to scalar. Note: Tensor dimensions were: "
<< weights.shape << ", with volume: " << volume(weights.shape));
return nullptr;
}
auto* scalarLayer = N_CHECK(ctx->network()->addConstant(nvinfer1::Dims{0, {0}}, weights));
return N_CHECK(scalarLayer->getOutput(0));
}
int divCeil(int n, int d)
{
return (n - 1) / d + 1;
}
std::string getTrtDtypeName(nvinfer1::DataType TrtDtype)
{
switch (TrtDtype)
{
case nvinfer1::DataType::kFLOAT: return "FLOAT";
case nvinfer1::DataType::kHALF: return "HALF";
case nvinfer1::DataType::kINT8: return "INT8";
case nvinfer1::DataType::kINT32: return "INT32";
case nvinfer1::DataType::kBOOL: return "BOOL";
case nvinfer1::DataType::kUINT8: return "UINT8";
case nvinfer1::DataType::kFP8: return "FP8";
case nvinfer1::DataType::kBF16: return "BF16";
case nvinfer1::DataType::kINT64: return "INT64";
default: return "<UNKNOWN>";
}
}
std::string getElementWiseOpName(nvinfer1::ElementWiseOperation op)
{
switch (op)
{
case nvinfer1::ElementWiseOperation::kSUM: return "SUM";
case nvinfer1::ElementWiseOperation::kPROD: return "PROD";
case nvinfer1::ElementWiseOperation::kMAX: return "MAX";
case nvinfer1::ElementWiseOperation::kMIN: return "MIN";
case nvinfer1::ElementWiseOperation::kSUB: return "SUB";
case nvinfer1::ElementWiseOperation::kDIV: return "DIV";
case nvinfer1::ElementWiseOperation::kPOW: return "POW";
case nvinfer1::ElementWiseOperation::kFLOOR_DIV: return "FLOOR_DIV";
case nvinfer1::ElementWiseOperation::kAND: return "AND";
case nvinfer1::ElementWiseOperation::kOR: return "OR";
case nvinfer1::ElementWiseOperation::kXOR: return "XOR";
case nvinfer1::ElementWiseOperation::kEQUAL: return "EQUAL";
case nvinfer1::ElementWiseOperation::kGREATER: return "GREATER";
case nvinfer1::ElementWiseOperation::kLESS: return "LESS";
default: return "<UNKNOWN>";
}
}
Status elementwiseCheck(std::vector<TensorOrWeights> const& inputs, const nvinfer1::ElementWiseOperation op,
::ONNX_NAMESPACE::NodeProto const& node, size_t const nodeIdx)
{
switch (op)
{
// These operations only support boolean inputs
case nvinfer1::ElementWiseOperation::kAND:
case nvinfer1::ElementWiseOperation::kOR:
case nvinfer1::ElementWiseOperation::kXOR:
ASSERT_NODE(
std::all_of(inputs.begin(), inputs.end(), [](TensorOrWeights const& input) { return input.isBool(); }),
"Elementwise layer only supports operator " + getElementWiseOpName(op)
+ " and the given inputs with type BOOL.",
node, nodeIdx, ErrorCode::kUNSUPPORTED_NODE);
break;
// These operations do not support boolean types
case nvinfer1::ElementWiseOperation::kDIV:
case nvinfer1::ElementWiseOperation::kFLOOR_DIV:
case nvinfer1::ElementWiseOperation::kGREATER:
case nvinfer1::ElementWiseOperation::kLESS:
case nvinfer1::ElementWiseOperation::kMAX:
case nvinfer1::ElementWiseOperation::kMIN:
case nvinfer1::ElementWiseOperation::kPROD:
case nvinfer1::ElementWiseOperation::kSUB:
case nvinfer1::ElementWiseOperation::kSUM:
ASSERT_NODE(
!std::any_of(inputs.begin(), inputs.end(), [](TensorOrWeights const& input) { return input.isBool(); }),
"Elementwise layer does not support operator " + getElementWiseOpName(op)
+ " and the given inputs with type BOOL.",
node, nodeIdx, ErrorCode::kUNSUPPORTED_NODE);
break;
// Pow does not support bool or integer types
case nvinfer1::ElementWiseOperation::kPOW:
ASSERT_NODE(!std::any_of(inputs.begin(), inputs.end(),
[](TensorOrWeights const& input) { return input.isBool() || input.isInt32() || input.isInt64(); }),
"Elementwise layer does not support operator POW with boolean or integer types.", node,
nodeIdx, ErrorCode::kUNSUPPORTED_NODE);
break;
// Equal supports all types.
case nvinfer1::ElementWiseOperation::kEQUAL: break;
}
return Status::success();
}
NodeImportResult elementwiseHelper(ImporterContext* ctx, ::ONNX_NAMESPACE::NodeProto const& node, size_t const nodeIdx,
std::vector<TensorOrWeights> const& inputs, nvinfer1::ElementWiseOperation binary_op)
{
ASSERT_NODE((!inputs.empty()), "Inputs vector is empty.", node, nodeIdx, ErrorCode::kINVALID_NODE);
std::vector<nvinfer1::ITensor*> inputTensors;
int maxNbDims = -1;
for (auto input : inputs)
{
maxNbDims = std::max(maxNbDims, input.shape().nbDims);
}
for (auto input : inputs)
{
auto* tensor_ptr = &convertToTensor(input, ctx);
// Broadcast all input tensors to size of maxNbDims
CHECK_STATUS(broadcastTensor(ctx, tensor_ptr, maxNbDims));
ASSERT_NODE(tensor_ptr->getDimensions().nbDims == maxNbDims,
"The number of dimensions should remain the same adding inputs: " << tensor_ptr->getDimensions().nbDims
<< " != " << maxNbDims << ".",
node, nodeIdx, ErrorCode::kUNSUPPORTED_NODE);
inputTensors.push_back(tensor_ptr);
}
CHECK_STATUS(elementwiseCheck(inputs, binary_op, node, nodeIdx));
// Use the first tensor input as the base for the elementwise operation
nvinfer1::ITensor* combined = inputTensors.at(0);
if (inputTensors.size() == 1)
{
// Note: Single input must be wrapped in identity to avoid messing up network outputs
return {{identity(ctx, combined)}};
}
for (size_t i = 1; i < inputTensors.size(); ++i)
{
nvinfer1::ITensor* tensor = inputTensors.at(i);
ASSERT_NODE((tensor->getDimensions().nbDims == combined->getDimensions().nbDims),
"The number of dimensions should remain the same adding inputs: "
<< tensor->getDimensions().nbDims << " != " << combined->getDimensions().nbDims << ".",
node, nodeIdx, ErrorCode::kUNSUPPORTED_NODE);
auto* layer = N_CHECK(ctx->network()->addElementWise(*combined, *tensor, binary_op));
ctx->registerLayer(layer, node);
combined = N_CHECK(layer->getOutput(0));
}
return {{combined}};
}
nvinfer1::ITensor* flattenTensor(
ImporterContext* ctx, ::ONNX_NAMESPACE::NodeProto const& node, nvinfer1::ITensor& tensor, int axis, bool regLayer)
{
auto const dims = shapeOf(tensor);
auto const d0 = product(ctx, dims, 0, axis, 1);
auto const d1 = product(ctx, dims, axis, dims.size(), 1);
// ShuffleLayer here interprets dim extent 0 as empty dim to support empty tensor
nvinfer1::IShuffleLayer* flattenLayer = N_CHECK(addShuffle(ctx, tensor, concat(ctx, d0, d1), /*zeroIsPlaceholder=*/false));
if (regLayer)
{
ctx->registerLayer(flattenLayer, node);
}
else
{
ctx->registerLayer(flattenLayer, "ONNXTRT_flattenTensor", nullptr);
}
return N_CHECK(flattenLayer->getOutput(0));
}
nvinfer1::ITensor* extractDimension(ImporterContext* ctx, nvinfer1::ITensor* shapeTensor, int dim, nvinfer1::Dims shape)
{
// Comparing with gather, slice is more flexible. It does not need to convert dim into a constant.
// It is important for refit as when add an additional constant, this gather may not be optimized out.
auto* slice = N_CHECK(ctx->network()->addSlice(
*shapeTensor, nvinfer1::Dims{1, {dim}}, nvinfer1::Dims{1, {1}}, nvinfer1::Dims{1, {1}}));
ctx->registerLayer(slice, "ONNXTRT_extractDimension", nullptr);
if (shape != nvinfer1::Dims{1, {1}})
{
auto* reshape = N_CHECK(ctx->network()->addShuffle(*slice->getOutput(0)));
reshape->setReshapeDimensions(shape);
ctx->registerLayer(reshape, "ONNXTRT_extractDimensionReshape", nullptr);
return N_CHECK(reshape->getOutput(0));
}
return N_CHECK(slice->getOutput(0));
}
// Helper function to generate padding values for convTranspose
void generatePadding(nvinfer1::Dims inputShape, nvinfer1::Dims outputShape, nvinfer1::Dims kernelSize,
nvinfer1::Dims strides, nvinfer1::Dims dilations, int const nbSpatialDims, nvinfer1::Dims& begPadding,
nvinfer1::Dims& endPadding, nvinfer1::Dims& outputPadding, nvinfer1::PaddingMode paddingMode)
{
nvinfer1::Dims totalPadding{nbSpatialDims, {}};
// Pre and post padding calculated as per https://github.com/onnx/onnx/blob/master/docs/Operators.md#ConvTranspose
// Note that output shape is inconsistent in the spec - can either be in full dimensions form (i.e. NCHW) or just
// spatial dimensions form (i.e. HW). Calculate potential offset here.
auto const outputOffset = outputShape.nbDims - nbSpatialDims;
for (int32_t i = 0; i < nbSpatialDims; i++)
{
totalPadding.d[i] = strides.d[i] * (inputShape.d[2 + i] - 1) + outputPadding.d[i]
+ ((kernelSize.d[i] - 1) * dilations.d[i] + 1) - outputShape.d[outputOffset + i];
// Same upper is calculated differently
if (paddingMode != nvinfer1::PaddingMode::kSAME_UPPER)
{
begPadding.d[i] = totalPadding.d[i] / 2;
endPadding.d[i] = totalPadding.d[i] - (totalPadding.d[i] / 2);
}
else
{
begPadding.d[i] = totalPadding.d[i] - (totalPadding.d[i] / 2);
endPadding.d[i] = (totalPadding.d[i] / 2);
}
}
}
float getActivationDefaultAlpha(nvinfer1::ActivationType type)
{
switch (type)
{
case nvinfer1::ActivationType::kCLIP: return 0.f;
case nvinfer1::ActivationType::kELU: return 1.0f;
case nvinfer1::ActivationType::kGELU_ERF: return 0.f;
case nvinfer1::ActivationType::kGELU_TANH: return 0.f;
case nvinfer1::ActivationType::kHARD_SIGMOID: return 0.2f;
case nvinfer1::ActivationType::kLEAKY_RELU: return 0.01f;
case nvinfer1::ActivationType::kRELU: return 0.f;
case nvinfer1::ActivationType::kSCALED_TANH: return 1.0f;
case nvinfer1::ActivationType::kSELU: return 1.67326319217681884765625f;
case nvinfer1::ActivationType::kSIGMOID: return 0.f;
case nvinfer1::ActivationType::kSOFTPLUS: return 0.f;
case nvinfer1::ActivationType::kSOFTSIGN: return 0.f;
case nvinfer1::ActivationType::kTANH: return 0.f;
case nvinfer1::ActivationType::kTHRESHOLDED_RELU: return 1.0f;
}
throw std::runtime_error{"Unrecognized activation type"};
}
float getActivationDefaultBeta(nvinfer1::ActivationType type)
{
switch (type)
{
case nvinfer1::ActivationType::kCLIP: return 0.f;
case nvinfer1::ActivationType::kELU: return 0.f;
case nvinfer1::ActivationType::kGELU_ERF: return 0.f;
case nvinfer1::ActivationType::kGELU_TANH: return 0.f;
case nvinfer1::ActivationType::kHARD_SIGMOID: return 0.5f;
case nvinfer1::ActivationType::kLEAKY_RELU: return 0.f;
case nvinfer1::ActivationType::kRELU: return 0.f;
case nvinfer1::ActivationType::kSCALED_TANH: return 1.0f;
case nvinfer1::ActivationType::kSELU: return 1.05070102214813232421875f;
case nvinfer1::ActivationType::kSIGMOID: return 0.f;
case nvinfer1::ActivationType::kSOFTPLUS: return 0.f;
case nvinfer1::ActivationType::kSOFTSIGN: return 0.f;
case nvinfer1::ActivationType::kTANH: return 0.f;
case nvinfer1::ActivationType::kTHRESHOLDED_RELU: return 0.f;
}
throw std::runtime_error{"Unrecognized activation type"};
}
nvinfer1::ITensor* getAxisLength(ImporterContext* ctx, nvinfer1::ITensor* inpTensor, int32_t axis, nvinfer1::Dims shape)
{
// Let TRT handle the shape tensor optimization.
auto* shapeLayer = N_CHECK(ctx->network()->addShape(*inpTensor));
nvinfer1::ITensor* inpShape = N_CHECK(shapeLayer->getOutput(0));
// TRT-22536 - remove the cast and fix clients of getAxisLength to use 64-bit lengths.
auto* castLayer = N_CHECK(ctx->network()->addCast(*inpShape, nvinfer1::DataType::kINT32));
inpShape = N_CHECK(castLayer->getOutput(0));
return extractDimension(ctx, inpShape, axis, shape);
}
nvinfer1::ITensor* getAxisLengthInt64(
ImporterContext* ctx, nvinfer1::ITensor* inpTensor, int axis, nvinfer1::Dims shape)
{
auto* shapeLayer = N_CHECK(ctx->network()->addShape(*inpTensor));
nvinfer1::ITensor* inpShape = N_CHECK(shapeLayer->getOutput(0));
return extractDimension(ctx, inpShape, axis, shape);
}
nvinfer1::ITensor* getElementWiseResult(ImporterContext* ctx, nvinfer1::ITensor& lhs, nvinfer1::ITensor& rhs, nvinfer1::ElementWiseOperation op)
{
auto* elemLayer = N_CHECK(ctx->network()->addElementWise(lhs, rhs, op));
return N_CHECK(elemLayer->getOutput(0));
}
nvinfer1::ITensor* getUnaryResult(ImporterContext* ctx, nvinfer1::ITensor& input, nvinfer1::UnaryOperation op)
{
auto* unaryLayer = N_CHECK(ctx->network()->addUnary(input, op));
return N_CHECK(unaryLayer->getOutput(0));
}
Status getKernelParams(ImporterContext* ctx, ::ONNX_NAMESPACE::NodeProto const& node, nvinfer1::Dims* kernelSize,
nvinfer1::Dims* strides, nvinfer1::Dims* begPadding, nvinfer1::Dims* endPadding,
nvinfer1::PaddingMode& paddingMode, bool& countExcludePadding, nvinfer1::Dims* dilations,
nvinfer1::Dims* outputPadding, bool const poolingCeilMode)
{
int32_t const nbSpatialDims = kernelSize->nbDims;
OnnxAttrs attrs(node, ctx);
if (attrs.count("kernel_shape"))
{
auto const* onnxKernelSize = attrs.at("kernel_shape");
setAttr(kernelSize, onnxKernelSize, nbSpatialDims, 1);
}
if (attrs.count("strides"))
{
auto const* onnxStrides = attrs.at("strides");
setAttr(strides, onnxStrides, nbSpatialDims, 1);
}
if (dilations && attrs.count("dilations"))
{
auto const* onnxDilations = attrs.at("dilations");
setAttr(dilations, onnxDilations, nbSpatialDims, 1);
}
if (attrs.count("count_include_pad"))
{
auto const* includePad = attrs.at("count_include_pad");
int32_t val = includePad->i();
val == 1 ? countExcludePadding = false : countExcludePadding = true;
}
// For ConvTranspose Layer
if (attrs.count("output_padding"))
{
auto const* onnxOutputPadding = attrs.at("output_padding");
setAttr(outputPadding, onnxOutputPadding, nbSpatialDims, 0);
}
paddingMode
= poolingCeilMode ? nvinfer1::PaddingMode::kEXPLICIT_ROUND_UP : nvinfer1::PaddingMode::kEXPLICIT_ROUND_DOWN;
auto onnxAutoPad = attrs.get("auto_pad", std::string("NOTSET"));
if (onnxAutoPad != "SAME_LOWER" && onnxAutoPad != "SAME_UPPER")
{
if (attrs.count("pads"))
{
auto onnxPadding = attrs.get<std::vector<int32_t>>("pads");
int32_t ndim = onnxPadding.size() / 2;
for (int32_t i = 0; i < nbSpatialDims; ++i)
{
if (i < ndim)
{
begPadding->d[i] = onnxPadding.at(i);
endPadding->d[i] = onnxPadding.at(i + ndim);
}
else
{
begPadding->d[i] = 0;
endPadding->d[i] = 0;
}
}
}
if (onnxAutoPad == "EXPLICIT_ROUND_UP")
{
paddingMode = nvinfer1::PaddingMode::kEXPLICIT_ROUND_UP;
}
}
else
{
// If auto_pad is SAME_LOWER or SAME_UPPER, input padding should be calculated
// "pads" attribute should not be specified
ASSERT(!attrs.count("pads")
&& "Pads attribute should not be specified with SAME_LOWER or SAME_UPPER auto padding!",
ErrorCode::kINVALID_NODE);
// Note: ONNX is always NCHW ordering
if (onnxAutoPad == "SAME_LOWER")
{
paddingMode = nvinfer1::PaddingMode::kSAME_LOWER;
}
else if (onnxAutoPad == "SAME_UPPER")
{
paddingMode = nvinfer1::PaddingMode::kSAME_UPPER;
}
else
{
ASSERT(false && "invalid autopad attribute was set!", ErrorCode::kINVALID_NODE);
}
}
return Status::success();
}
float getSingleValueAsFloat(ShapedWeights const& weights)
{
assert(weights.count() == 1 && "Expected weights to contain only a single value");
switch (weights.type)
{
case ::ONNX_NAMESPACE::TensorProto::FLOAT: return static_cast<float const*>(weights.values)[0];
case ::ONNX_NAMESPACE::TensorProto::FLOAT16:
return static_cast<float>(static_cast<half_float::half const*>(weights.values)[0]);
case ::ONNX_NAMESPACE::TensorProto::BFLOAT16:
return static_cast<float>(static_cast<BFloat16 const*>(weights.values)[0]);
default: assert(false && "Unsupported type!"); return 0.F;
}
return 0.F;
}
nvinfer1::ITensor* globalPoolingHelper(ImporterContext* ctx, ::ONNX_NAMESPACE::NodeProto const& node,
nvinfer1::ITensor& tensor, nvinfer1::ReduceOperation op)
{
nvinfer1::Dims dims = tensor.getDimensions();
// Generate a bitmask of all 1s except the last 2 bits (N and C axes)
uint32_t reduceAxes = ((1 << dims.nbDims) - 1) & ~0b11;
auto* layer = N_CHECK(ctx->network()->addReduce(tensor, op, reduceAxes, /*keepDimensions=*/true));
ctx->registerLayer(layer, node);
return N_CHECK(layer->getOutput(0));
}
NodeImportResult greaterLessOrEqual(ImporterContext* ctx, const ::ONNX_NAMESPACE::NodeProto& node,
size_t const nodeIdx, nvinfer1::ITensor* inputA, nvinfer1::ITensor* inputB, bool greater)
{
nvinfer1::ElementWiseOperation op
= greater ? nvinfer1::ElementWiseOperation::kGREATER : nvinfer1::ElementWiseOperation::kLESS;
std::vector<TensorOrWeights> firstOpResults;
GET_VALUE(elementwiseHelper(ctx, node, nodeIdx, {inputA, inputB}, op), &firstOpResults);
std::vector<TensorOrWeights> equalsResult;
GET_VALUE(elementwiseHelper(ctx, node, nodeIdx, {inputA, inputB}, nvinfer1::ElementWiseOperation::kEQUAL), &equalsResult);
return elementwiseHelper(ctx, node, nodeIdx, {firstOpResults.at(0), equalsResult.at(0)}, nvinfer1::ElementWiseOperation::kOR);
}
nvinfer1::IPluginCreatorInterface* importPluginCreator(ImporterContext* ctx, std::string const& pluginName,
std::string const& pluginVersion, std::string const& pluginNamespace)
{
nvinfer1::IPluginCreatorInterface* creator = nullptr;
#if ENABLE_STD_PLUGIN
auto& pluginRegistry = ctx->network()->getBuilder().getPluginRegistry();
creator = pluginRegistry.getCreator(pluginName.c_str(), pluginVersion.c_str(), pluginNamespace.c_str());
#endif // ENABLE_STD_PLUGIN
#if ENABLE_SAFE_PLUGIN
auto safetyPluginRegistry = nvinfer1::getBuilderSafePluginRegistry(nvinfer1::EngineCapability::kSAFETY);
if (creator == nullptr && safetyPluginRegistry != nullptr)
{
creator = safetyPluginRegistry->getPluginCreator(
pluginName.c_str(), pluginVersion.c_str(), pluginNamespace.c_str());
}
#endif // ENABLE_SAFE_PLUGIN
// Do not perform a N_CHECK here as a plugin not being found is a valid case. It is up to the callers to handle the
// nullptr correctly.
return creator;
}
std::unique_ptr<nvinfer1::IPluginV2, PluginDeleter> createPlugin(std::string const& name,
nvinfer1::IPluginCreator* pluginCreator, std::vector<nvinfer1::PluginField> const& pluginFields)
{
if (!pluginCreator)
{
return nullptr;
}
nvinfer1::PluginFieldCollection fc;
fc.nbFields = pluginFields.size();
fc.fields = pluginFields.data();
return std::unique_ptr<nvinfer1::IPluginV2, PluginDeleter>{pluginCreator->createPlugin(name.c_str(), &fc)};
}
std::unique_ptr<nvinfer1::IPluginV3> createPlugin(std::string const& name, nvinfer1::IPluginCreatorV3One* pluginCreator,
std::vector<nvinfer1::PluginField> const& pluginFields)
{
if (!pluginCreator)
{
return nullptr;
}
nvinfer1::PluginFieldCollection fc;
fc.nbFields = pluginFields.size();
fc.fields = pluginFields.data();
return std::unique_ptr<nvinfer1::IPluginV3>{
pluginCreator->createPlugin(name.c_str(), &fc, nvinfer1::TensorRTPhase::kBUILD)};
}
NodeImportResult staticSliceImporter(ImporterContext* ctx, ::ONNX_NAMESPACE::NodeProto const& node,
size_t const nodeIdx, std::vector<TensorOrWeights>& inputs, nvinfer1::ITensor& data)
{
auto const nbInputs = inputs.size();
nvinfer1::Dims inputDims = data.getDimensions();
auto const nbDims = inputDims.nbDims;
// Create start, sizes, and steps with default values
nvinfer1::Dims starts{nbDims, {}};
nvinfer1::Dims sizes{nbDims, {}};
nvinfer1::Dims steps{nbDims, {}};
for (int32_t i = 0; i < nbDims; i++)
{
starts.d[i] = 0;
sizes.d[i] = inputDims.d[i];
steps.d[i] = 1;
}
// Default axes / steps values
std::vector<int32_t> defaultAxes(nbDims);
std::iota(defaultAxes.begin(), defaultAxes.end(), 0);
std::vector<int32_t> defaultSteps(nbDims, 1);
// Get int32 pointer representation of ONNX provided values
int32_t* startVals = static_cast<int32_t*>(inputs.at(1).weights().values);
int32_t* endVals = static_cast<int32_t*>(inputs.at(2).weights().values);
int32_t* axesVals = nbInputs > 3 ? static_cast<int32_t*>(inputs.at(3).weights().values) : defaultAxes.data();
int32_t* stepVals = nbInputs > 4 ? static_cast<int32_t*>(inputs.at(4).weights().values) : defaultSteps.data();
// Handle non-standard values for slice
// Start values must in range of [0, dims.d[i]] for + steps, [0, dims.d[i] - 1] for - steps
auto convertStarts = [](int32_t start, int32_t upper, int32_t stepSign) {
int32_t newStarts = start < 0 ? start + upper : start;
newStarts = std::min(std::max(newStarts, 0), upper + stepSign);
return newStarts;
};
// End values must in range of [0, dims.d[i]] for + steps, [-1, dims.d[i]] for - steps
auto convertEnds = [](int32_t end, int32_t upper, int32_t stepSign) {
int32_t newEnds = end < 0 ? end + upper : end;
newEnds = std::min(std::max(newEnds, stepSign), upper);
return newEnds;
};
// Axes values must in range of [0, nbDims]
auto convertAxes = [&nbDims](int32_t axis) { return axis < 0 ? axis + nbDims : axis; };
// Since axes can be sparse, get the expected number of provided values
auto const nbValues = inputs.at(1).shape().d[0];
for (int32_t i = 0; i < nbValues; i++)
{
auto axesIndex = convertAxes(axesVals[i]);
// Modify starts
int32_t stepSign = stepVals[i] < 0 ? -1 : 0;
starts.d[axesIndex] = convertStarts(startVals[i], inputDims.d[axesIndex], stepSign);
// Modify ends
int32_t modifiedEnds = convertEnds(endVals[i], inputDims.d[axesIndex], stepSign);
steps.d[axesIndex] = stepVals[i];
// Perform ceil integer division of (ends - starts) / steps to compute sizes.
// Note ceil(x/y) = (x+y-1) / y for postive x & y, and ceil(x/y) = (x+y+1)/y for negative x&y
// Negative sizes indicates an empty slice, so clamp to 0
sizes.d[axesIndex]
= std::max<int64_t>((modifiedEnds - starts.d[axesIndex] + steps.d[axesIndex] - (steps.d[axesIndex] > 0 ? 1 : -1))
/ steps.d[axesIndex],
0);
}