-
Notifications
You must be signed in to change notification settings - Fork 1.1k
/
Copy pathgemm_permute.cu
1224 lines (996 loc) · 47.4 KB
/
gemm_permute.cu
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) 2017 - 2023 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
* SPDX-License-Identifier: BSD-3-Clause
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions are met:
*
* 1. Redistributions of source code must retain the above copyright notice, this
* list of conditions and the following disclaimer.
*
* 2. Redistributions in binary form must reproduce the above copyright notice,
* this list of conditions and the following disclaimer in the documentation
* and/or other materials provided with the distribution.
*
* 3. Neither the name of the copyright holder nor the names of its
* contributors may be used to endorse or promote products derived from
* this software without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
* AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
* DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
* SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
* CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
* OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*
**************************************************************************************************/
/*! \file
\brief GEMM Permute Example.
This example computes batched GEMM operations with output results permuted as reshaped tensors.
We provide layout plugin as a flexible tool for users to add any customized input/output tensor permute operation,
or any other generalized global memory writeout address computation. To add a customized layout, add new class
in include/cutlass/layout/permute.h
In this example we use several permute operations (permute([0, 2, 1, 3]))
In this example, we used Tensor4DPermuteBMM0213 layout to perform Batched GEMM with permute([0, 2, 1, 3]) on BMM
whole output tensor, and used Tensor5DPermute20314 layout to perform Normal GEMM with permute([2, 0, 3, 1, 4]) on
output matrix. The address computations are performed in compute(col_init, row_init, stride_init,
BMM_batch_idx) with {col_permute, row_permute and stride_permute} as new addresses after permute op.
(check include/cutlass/layout/permute.h)
Tips:
1) Make sure to set batch_stride to zero for BMM permute; also the BMM GEMM should be in mode
cutlass::gemm::GemmUniversalMode::kBatched instead of kArray.
2) When the contiguous dimension is touched in permute op (for example [0, 2, 3, 1] for row-major matrix
or [1, 0, 2, 3] for column-major), Alignment should be set to 1 for the corresponding matrix.
If the last dimension is untouched, one can set Alignment to be larger like 8 in our example.
As a result, permute op without touching the unit stride dimension is recommended to obtain the best performance.
Examples:
# Runs a batched GEMM with 96 batches
$ ./examples/39_gemm_permute/39_gemm_permute --problem-count=96
# Runs a batched GEMM with 96 batches (with GEMM-K dimension equal to 1024)
$ ./examples/39_gemm_permute/39_gemm_permute --problem-count=96 --k=1024 --verbose=true
# Execute batched GEMM and profile with NSight
$ nv-nsight-cu-cli ./examples/39_gemm_permute/39_gemm_permute --m=256 --n=192 --k=256 --verbose=true --iterations=1 --reference-check=false
*/
/////////////////////////////////////////////////////////////////////////////////////////////////
#include <iostream>
#include <fstream>
#include <sstream>
#include <vector>
#include <map>
#include <unordered_map>
#include "cutlass/cutlass.h"
#include "cutlass/gemm/gemm.h"
#include "cutlass/gemm/device/gemm_universal.h"
#include "cutlass/util/command_line.h"
#include "cutlass/util/distribution.h"
#include "cutlass/util/device_memory.h"
#include "cutlass/util/tensor_view_io.h"
#include "cutlass/util/host_tensor.h"
#include "cutlass/util/reference/host/gemm_complex.h"
#include "cutlass/util/reference/device/gemm_complex.h"
#include "cutlass/util/reference/host/tensor_compare.h"
#include "cutlass/util/reference/device/tensor_compare.h"
#include "cutlass/util/reference/host/tensor_copy.h"
#include "cutlass/util/reference/device/tensor_fill.h"
#include "cutlass/util/reference/host/tensor_norm.h"
#include "cutlass/layout/permute.h"
#include "layouts.h"
#include "permute_info.h"
/// Tensor4DPermuteBMM0213 --->
/// Permute layout function for 4-D permuted tensors for BMM with BMM tensor (dimension as [B, M, N]) reshaped
/// as [B/D1, D1, M, N]. Then perform permute([0, 2, 1, 3]) on the corresponding whole BMM tensor.
int constexpr D1 = 12;
/// Tensor5DPermute20314 --->
/// Permute layout function for 5-D permuted tensors with matrix (dimension as [M, N]) reshaped
/// as [M/T1, T1, T2, T3, N/T2/T3]. Then perform permute([2, 0, 3, 1, 4]) on the corresponding tensor.
int constexpr T1 = 16;
int constexpr T2 = 3;
int constexpr T3 = 8;
/// Tensor4DPermute0213 --->
/// Permute layout function for 4-D permuted tensors with matrix (dimension as [M, N]) reshaped
/// as [M/S1, S1, S2, N/S2]. Then perform permute([0, 2, 1, 3]) on the corresponding tensor.
int constexpr S1 = 8;
int constexpr S2 = 4;
// // // Alignments
int constexpr AlignmentA = 8;
int constexpr AlignmentB = 8;
int constexpr AlignmentC = 8;
/// GEMM element types
using ElementInput = cutlass::half_t;
using ElementOutput = cutlass::half_t;
using ElementAccumulator = float;
/////////////////////////////////////////////////////////////////////////////////////////////////
/// Useful macros
#define CHECK_CUDA_CALL(call, handler) \
do { \
cudaError_t __err = (call); \
if (__err != cudaSuccess) { \
std::cerr << #call " failed: " << cudaGetErrorString(__err) << std::endl; \
handler; \
} \
} while(0)
#define CHECK_CUTLASS_CALL(call, handler) \
do { \
cutlass::Status __status = (call); \
if (__status != cutlass::Status::kSuccess) { \
std::cerr << #call " failed: " << cutlass::cutlassGetStatusString(__status) << std::endl; \
handler; \
} \
} while(0)
/////////////////////////////////////////////////////////////////////////////////////////////////
// Command line options parsing
struct Options {
bool help;
bool error;
bool reference_check;
cutlass::gemm::GemmCoord problem_each;
int batch_count;
int iterations;
int cuda_streams;
bool verbose;
float alpha;
float beta;
//
// Methods
//
Options():
help(false),
error(false),
reference_check(true),
batch_count(-1),
iterations(20),
cuda_streams(0),
verbose(false),
alpha(1),
beta()
{ }
// Parses the command line
void parse(int argc, char const **args) {
cutlass::CommandLine cmd(argc, args);
if (cmd.check_cmd_line_flag("help")) {
help = true;
return;
}
cmd.get_cmd_line_argument("alpha", alpha, 1.0f);
cmd.get_cmd_line_argument("beta", beta, 0.0f);
cmd.get_cmd_line_argument("iterations", iterations, 20);
cmd.get_cmd_line_argument("streams", cuda_streams, 0);
cmd.get_cmd_line_argument("verbose", verbose, false);
cmd.get_cmd_line_argument("reference-check", reference_check, true);
int m, n, k;
cmd.get_cmd_line_argument("m", m, 384);
cmd.get_cmd_line_argument("n", n, 192);
cmd.get_cmd_line_argument("k", k, 384);
cmd.get_cmd_line_argument("batch-count", batch_count, 96);
problem_each = cutlass::gemm::GemmCoord(m, n, k);
}
/// Prints the usage statement.
std::ostream & print_usage(std::ostream &out) const {
out <<
"39_gemm_permute\n"
"\n"
" This example tests and profiles the performance of normal GEMM and batched GEMM with different"
" combinations of fused permutations of input and output tensors."
"\n"
" Permutations considered in this example:\n"
"\n"
" Normal GEMM:\n"
" 1) Tensor4DPermute0213: matrix of shape [X, Y] is reshaped as [X/S1, S1, S2, Y/S2] and has its dimensions"
" permuted as [0, 2, 1, 3], resulting in shape [X/S1, S2, S1, Y/S2] viewed as matrix of shape [X*S2/S1, Y*S1/S2].\n"
" 2) Tensor5DPermute20314: matrix of shape [X, Y] is reshaped as [X/T1, T1, T2, T3, Y/T2/T3] and has its dimensions"
" permuted as [2, 0, 3, 1, 4], resulting in shape [T2, X/T1, T3, T1, Y/T2/T3] viewed as matrix of shape [X*T2/T1, Y*T1/T2].\n"
"\n"
" Batched GEMM:\n"
" 3) Tensor4DPermuteBMM0213: batched tensor of 3D shape [B, X, Y] is reshaped as 4D shape [B/D1, D1, X, Y]"
" and has its dimensions permuted as [0, 2, 1, 3], resulting in shape [B/D1, X, D1, Y] viewed as"
" a matrix of shape [B/D1, X, Y*D1] for batched GEMM purposes.\n"
"\n"
" Note: S1, S2, D1, D2, T1, T2, T3 are compile-time constants defined in gemm_permute.cu."
" Runtime specification of these values is not supported."
" These values along with alignment requirements place constraints on supported matrix sizes.\n"
"\n"
" Note: X, Y above may refer to M, N or K dimensions of GEMM problem, depending on the tensor considered (A, B or D)."
" For the output tensor D the values correspond directly to dimensions of D, whereas for A and B the original dimensions"
" X', Y' are inferred from the ones supplied to the GEMM, taking into account the permute operation.\n"
"\n"
"Options:\n"
"\n"
" --help If specified, displays this usage statement.\n\n"
" --batch-count=<int> Sets the number of batches in batched GEMM (batch number for BMM). (default: --batch-count=768)\n"
" --m=<int> Sets the M dimension for both batched GEMM and normal GEMM problems. (default: --m=128)\n"
" --n=<int> Sets the N dimension for both batched GEMM and normal GEMM problems. (default: --n=192)\n"
" --k=<int> Sets the K dimension for both batched GEMM and normal GEMM problems. (default: --k=384)\n"
" --alpha=<f32> Epilogue scalar alpha (real part)\n"
" --beta=<f32> Epilogue scalar beta (real part)\n\n"
" --iterations=<int> Number of profiling iterations to perform.\n"
" --reference-check=<bool> If true, performs reference check.\n"
" --verbose=<bool> If true, prints problem sizes and batching structure.\n"
"\n"
"Examples:\n"
"\n"
"# Runs a batched GEMM with 96 batches\n"
"$ ./examples/39_gemm_permute/39_gemm_permute --batch-count=96\n"
"\n"
"# Runs a batched GEMM with 96 batches (with GEMM-K dimension equal to 1024)\n"
"$ ./examples/39_gemm_permute/39_gemm_permute --batch-count=96 --k=1024 --verbose=true\n"
"\n"
"# Execute batched GEMM and profile with NSight\n"
"$ nv-nsight-cu-cli ./examples/39_gemm_permute/39_gemm_permute --m=256 --n=192 --k=256 --verbose=true --iterations=1 --reference-check=false\n"
"\n";
return out;
}
/// Compute performance in GFLOP/s
double gflops(double runtime_s, bool batched) const {
// Number of real-valued multiply-adds
int64_t fmas = int64_t();
fmas += problem_each.product() * (batched ? batch_count : 1);
// Two flops per multiply-add
return 2.0 * double(fmas) / double(1.0e9) / runtime_s;
}
};
///////////////////////////////////////////////////////////////////////////////////////////////////
namespace detail
{
/// Dimension-generic permutation loop
template<int I, typename Element, typename Layout, typename PermuteOp, typename Coord>
void permute_host_impl(
cutlass::TensorView<Element const, Layout> const & input,
cutlass::TensorView<Element, Layout> const & output,
PermuteOp && permute,
Coord & coord
) {
static_assert(Layout::kRank == Coord::kRank, "Incompatible Layout and Coord types");
if constexpr (I == Coord::kRank) {
output.at(permute(coord)) = input.at(coord);
}
else {
for (coord[I] = 0; coord[I] < input.extent(I); ++coord[I]) {
permute_host_impl<I+1>(input, output, std::forward<PermuteOp>(permute), coord);
}
}
}
} // namespace detail
/// Perform a reference (host-based) permutation of an input tensor
template<typename PermuteLayout, typename Element, typename Layout>
void permute_host(
cutlass::TensorView<Element const, Layout> const &input,
cutlass::TensorView<Element, Layout> const &output,
int batch_count) {
Layout layout = input.layout();
cutlass::MatrixCoord extent = input.extent();
std::size_t num_elems = layout.capacity(extent) * batch_count;
std::vector<Element> h_input(num_elems);
cutlass::device_memory::copy_to_host(h_input.data(), input.data(), num_elems);
std::vector<Element> h_output(num_elems);
using Info = PermuteInfo<PermuteLayout>;
using TensorLayout = typename Info::Layout;
auto shape_orig = Info::original_shape(extent, batch_count);
auto shape_perm = Info::permute(shape_orig);
cutlass::TensorView<Element const, TensorLayout> view_input(h_input.data(), TensorLayout::packed(shape_orig), shape_orig);
cutlass::TensorView<Element, TensorLayout> view_output(h_output.data(), TensorLayout::packed(shape_perm), shape_perm);
decltype(shape_orig) coord;
detail::permute_host_impl<0>(view_input, view_output, Info::permute, coord);
cutlass::device_memory::copy_to_device(output.data(), h_output.data(), num_elems);
}
template<typename Layout>
struct LayoutInfo;
template<>
struct LayoutInfo<cutlass::layout::RowMajor> {
static std::string name() { return "RowMajor"; }
};
template<>
struct LayoutInfo<cutlass::layout::ColumnMajor> {
static std::string name() { return "ColumnMajor"; }
};
///////////////////////////////////////////////////////////////////////////////////////////////////
template <typename ElementA, typename ElementB, typename ElementC>
class Testbed {
private:
//
// Data members
//
Options & options;
/// Initialization
cutlass::Distribution::Kind init_A;
cutlass::Distribution::Kind init_B;
cutlass::Distribution::Kind init_C;
uint32_t seed;
cutlass::DeviceAllocation<ElementA> block_A;
cutlass::DeviceAllocation<ElementB> block_B;
cutlass::DeviceAllocation<ElementC> block_C;
cutlass::DeviceAllocation<ElementC> block_D;
public:
//
// Methods
//
Testbed(
Options &options_,
cutlass::Distribution::Kind init_A_ = cutlass::Distribution::Uniform,
cutlass::Distribution::Kind init_B_ = cutlass::Distribution::Uniform,
cutlass::Distribution::Kind init_C_ = cutlass::Distribution::Uniform,
uint32_t seed_ = 3090
):
options(options_), init_A(init_A_), init_B(init_B_), init_C(init_C_), seed(seed_) { }
private:
/// Print permutation info for one tensor
template<typename PermuteLayout>
void print_tensor_info(
std::ostream & os,
std::string const &tensor_name,
int row_dim,
int col_dim) {
cutlass::MatrixCoord extent(options.problem_each.at(row_dim), options.problem_each.at(col_dim));
using Info = PermuteInfo<PermuteLayout>;
os << "tensor " << tensor_name << ": " << Info::desc() << "\n";
os << " extent: [" << extent.row() << ", " << extent.column() << "]";
if (Info::kBatched) {
os << ", batch count: " << options.batch_count;
}
os << "\n";
if (!cutlass::layout::is_trivial_permute<PermuteLayout>) {
auto shape_orig = Info::original_shape(extent, options.batch_count);
auto shape_perm = Info::permute(shape_orig);
os << " original: [" << shape_orig << "]\n";
os << " permuted: [" << shape_perm << "]\n";
}
}
/// Check shape compatibility for one tensor
template<typename Layout, typename PermuteLayout, int Alignment>
bool check_tensor_shape(
std::string const &tensor_name,
int row_dim,
int col_dim) {
cutlass::MatrixCoord extent(options.problem_each.at(row_dim), options.problem_each.at(col_dim));
using Info = PermuteInfo<PermuteLayout>;
auto rowAlign = cutlass::platform::is_same<Layout, cutlass::layout::ColumnMajor>::value ? Alignment : 1;
auto colAlign = cutlass::platform::is_same<Layout, cutlass::layout::RowMajor>::value ? Alignment : 1;
auto rowFactor = Info::kRowFactor * rowAlign;
auto colFactor = Info::kColumnFactor * colAlign;
// Assumes row-major layout
bool const valid_row = extent.row() % rowFactor == 0;
if (!valid_row) {
std::cerr << "Invalid tensor " << tensor_name << " row size = " << extent.row() << ", "
"must be divisible by " << rowFactor << ", "
"required by " << Info::name() <<
(rowAlign > 1 ? (" and alignment of " + std::to_string(rowAlign)) : "") << std::endl;
}
bool const valid_col = extent.column() % colFactor == 0;
if (!valid_col) {
std::cerr << "Invalid tensor " << tensor_name << " column size = " << extent.column() << ", "
"must be divisible by " << colFactor << ", "
"required by " << Info::name() <<
(colAlign > 1 ? (" and alignment of " + std::to_string(colAlign)) : "") << std::endl;
}
bool const valid_bsz = options.batch_count % Info::kBatchFactor == 0;
if (!valid_bsz) {
std::cerr << "Invalid batch count = " << options.batch_count << ", "
"must be divisible by " << Info::kBatchFactor << ", "
"required by " << Info::name() << std::endl;
}
return valid_row && valid_col && valid_bsz;
}
/// Helper to initialize a tensor view
template <typename Element>
void initialize_tensor_(
Element *ptr,
size_t capacity,
cutlass::Distribution::Kind dist_kind,
uint32_t seed) {
if (dist_kind == cutlass::Distribution::Uniform) {
Element scope_max, scope_min;
int bits_input = cutlass::sizeof_bits<Element>::value;
int bits_output = cutlass::sizeof_bits<ElementC>::value;
if (bits_input == 1) {
scope_max = 2;
scope_min = 0;
} else if (bits_input <= 8) {
scope_max = 2;
scope_min = -2;
} else if (bits_output == 16) {
if (cutlass::sizeof_bits<ElementAccumulator>::value <= 16) {
scope_max = 5;
scope_min = -5;
}
else {
scope_max = 8;
scope_min = -8;
}
} else {
scope_max = 8;
scope_min = -8;
}
cutlass::reference::device::BlockFillRandomUniform(
ptr, capacity, seed, scope_max, scope_min, 0);
}
else if (dist_kind == cutlass::Distribution::Gaussian) {
cutlass::reference::device::BlockFillRandomGaussian(
ptr, capacity, seed, Element(), Element(0.5f));
}
else if (dist_kind == cutlass::Distribution::Sequential) {
// Fill with increasing elements
cutlass::reference::device::BlockFillSequential(
ptr, capacity, Element(1), Element());
}
else {
// Fill with all 1s
cutlass::reference::device::BlockFillSequential(
ptr, capacity, Element(), Element(1));
}
}
/// Initializes data structures
void initialize(int batch_count) {
srand(seed);
int64_t total_elements_A = options.problem_each.m() * options.problem_each.k() * batch_count;
int64_t total_elements_B = options.problem_each.n() * options.problem_each.k() * batch_count;
int64_t total_elements_C = options.problem_each.m() * options.problem_each.n() * batch_count;
int64_t total_elements_D = options.problem_each.m() * options.problem_each.n() * batch_count;
// Allocate space
block_A.reset(total_elements_A);
block_B.reset(total_elements_B);
block_C.reset(total_elements_C);
block_D.reset(total_elements_D);
// Initialize input tensors
initialize_tensor_(block_A.get(), total_elements_A, init_A, seed * 2021);
initialize_tensor_(block_B.get(), total_elements_B, init_B, seed * 2022);
initialize_tensor_(block_C.get(), total_elements_C, init_C, seed * 2023);
cutlass::reference::device::BlockFillSequential(
block_D.get(), total_elements_D, ElementC(), ElementC());
}
/// Check device GEMM results against a reference implementation with separate host-based permutation
template<typename Gemm>
bool validate(Gemm const &gemm) {
bool constexpr kBatched = PermuteInfo<typename Gemm::PermuteALayout>::kBatched
|| PermuteInfo<typename Gemm::PermuteBLayout>::kBatched
|| PermuteInfo<typename Gemm::PermuteDLayout>::kBatched;
int const batch_count = kBatched ? options.batch_count : 1;
cutlass::gemm::GemmCoord problem = options.problem_each;
cutlass::MatrixCoord extent_A{problem.m(), problem.k()};
cutlass::MatrixCoord extent_B{problem.k(), problem.n()};
cutlass::MatrixCoord extent_C{problem.m(), problem.n()};
using LayoutA = typename Gemm::LayoutA;
using LayoutB = typename Gemm::LayoutB;
using LayoutC = typename Gemm::LayoutC;
LayoutA layout_A(LayoutA::packed(extent_A));
LayoutB layout_B(LayoutB::packed(extent_B));
LayoutC layout_C(LayoutC::packed(extent_C));
auto size_A = layout_A.capacity(extent_A) * batch_count;
auto size_B = layout_B.capacity(extent_B) * batch_count;
auto size_C = layout_C.capacity(extent_C) * batch_count;
cutlass::TensorView<ElementA, LayoutA> view_A(block_A.get(), layout_A, extent_A);
cutlass::TensorView<ElementB, LayoutB> view_B(block_B.get(), layout_B, extent_B);
cutlass::TensorView<ElementC, LayoutC> view_C(block_C.get(), layout_C, extent_C);
cutlass::TensorView<ElementC, LayoutC> view_D(block_D.get(), layout_C, extent_C);
cutlass::DeviceAllocation<ElementA> block_A_perm(size_A);
cutlass::DeviceAllocation<ElementA> block_B_perm(size_B);
cutlass::TensorView<ElementA, LayoutA> view_A_perm(block_A_perm.get(), layout_A, extent_A);
cutlass::TensorView<ElementB, LayoutB> view_B_perm(block_B_perm.get(), layout_B, extent_B);
permute_host<typename Gemm::PermuteALayout>(view_A.const_view(), view_A_perm, batch_count);
permute_host<typename Gemm::PermuteBLayout>(view_B.const_view(), view_B_perm, batch_count);
cutlass::DeviceAllocation<ElementC> block_D_ref(size_C);
cutlass::TensorView<ElementC, LayoutC> view_D_ref(block_D_ref.get(), layout_C, extent_C);
using EpilogueOutputOp = typename Gemm::GemmKernel::Epilogue::OutputOp;
// Reference GEMM
cutlass::reference::device::GemmComplex<
ElementA, LayoutA,
ElementB, LayoutB,
ElementC, LayoutC,
typename EpilogueOutputOp::ElementCompute,
typename Gemm::ElementAccumulator
>(
problem,
options.alpha,
view_A_perm,
Gemm::kTransformA,
view_B_perm,
Gemm::kTransformB,
options.beta,
view_C,
view_D_ref,
ElementAccumulator(0),
batch_count,
options.problem_each.m() * options.problem_each.k(),
options.problem_each.n() * options.problem_each.k(),
options.problem_each.m() * options.problem_each.n(),
options.problem_each.m() * options.problem_each.n()
);
cutlass::DeviceAllocation<ElementC> block_D_perm(size_C);
cutlass::TensorView<ElementC, LayoutC> view_D_perm(block_D_perm.get(), layout_C, extent_C);
permute_host<typename Gemm::PermuteDLayout>(view_D_ref.const_view(), view_D_perm, batch_count);
// Reference check
return cutlass::reference::device::BlockCompareEqual(view_D_perm.data(), view_D.data(), size_C);
}
public:
template<typename Gemm>
bool profile_GEMM_permute() {
using LayoutA = typename Gemm::LayoutA;
using LayoutB = typename Gemm::LayoutB;
using LayoutC = typename Gemm::LayoutC;
using PermuteALayout = typename Gemm::PermuteALayout;
using PermuteBLayout = typename Gemm::PermuteBLayout;
using PermuteDLayout = typename Gemm::PermuteDLayout;
bool constexpr kBatched = PermuteInfo<PermuteALayout>::kBatched
|| PermuteInfo<PermuteBLayout>::kBatched
|| PermuteInfo<PermuteDLayout>::kBatched;
std::cout << "\n"
"====================================================\n"
<< (kBatched ? "Batched" : "Normal") << " GEMM:"
<< "\n A=" << LayoutInfo<LayoutA>::name() << "," << PermuteInfo<PermuteALayout>::name()
<< "\n B=" << LayoutInfo<LayoutB>::name() << "," << PermuteInfo<PermuteBLayout>::name()
<< "\n D=" << LayoutInfo<LayoutC>::name() << "," << PermuteInfo<PermuteDLayout>::name()
<< "\n"
"====================================================\n";
if (options.verbose) {
print_tensor_info<PermuteALayout>(std::cout, "A", 0, 2);
print_tensor_info<PermuteBLayout>(std::cout, "B", 2, 1);
print_tensor_info<PermuteDLayout>(std::cout, "D", 0, 1);
}
std::cout << std::endl;
bool valid = true;
valid &= check_tensor_shape<LayoutA, PermuteALayout, Gemm::kAlignmentA>("A", 0, 2);
valid &= check_tensor_shape<LayoutB, PermuteBLayout, Gemm::kAlignmentB>("B", 2, 1);
valid &= check_tensor_shape<LayoutC, PermuteDLayout, Gemm::kAlignmentC>("D", 0, 1);
if (!valid)
{
std::cout << "Skipped test" << std::endl;
return true;
}
int const batch_count = kBatched ? options.batch_count : 1;
// Initialize the problem
initialize(batch_count);
// Configure the GEMM arguments
using EpilogueOutputOp = typename Gemm::GemmKernel::Epilogue::OutputOp;
typename EpilogueOutputOp::Params epilogue_op(options.alpha, options.beta);
// Please make sure all problem_sizes are the same for kBatched mode
auto problem = options.problem_each;
cutlass::MatrixCoord extent_A{problem.m(), problem.k()};
cutlass::MatrixCoord extent_B{problem.k(), problem.n()};
cutlass::MatrixCoord extent_C{problem.m(), problem.n()};
LayoutA layout_A(LayoutA::packed(extent_A));
LayoutB layout_B(LayoutB::packed(extent_B));
LayoutC layout_C(LayoutC::packed(extent_C));
// Configure GEMM arguments
typename Gemm::Arguments arguments{
kBatched ? cutlass::gemm::GemmUniversalMode::kBatched : cutlass::gemm::GemmUniversalMode::kGemm,
problem,
batch_count,
epilogue_op,
(void*)block_A.get(),
(void*)block_B.get(),
(void*)block_C.get(),
(void*)block_D.get(),
// For any non-trivial permute the batch stride must be set to 0
cutlass::layout::is_trivial_permute<PermuteALayout> ? layout_A.capacity(extent_A) : 0,
cutlass::layout::is_trivial_permute<PermuteBLayout> ? layout_B.capacity(extent_B) : 0,
layout_C.capacity(extent_C),
cutlass::layout::is_trivial_permute<PermuteDLayout> ? layout_C.capacity(extent_C) : 0,
layout_A.stride(0),
layout_B.stride(0),
layout_C.stride(0),
layout_C.stride(0),
};
// Initialize the GEMM object
Gemm gemm_normal;
CHECK_CUTLASS_CALL(gemm_normal.initialize(arguments, nullptr), return false);
// Run the normal GEMM object
CHECK_CUTLASS_CALL(gemm_normal.run(), return false);
// Wait for completion
CHECK_CUDA_CALL(cudaDeviceSynchronize(), return false);
//
// Verify correctness
//
if (options.reference_check) {
if (validate(gemm_normal)) {
std::cout << "\nPassed verification\n" << std::endl;
}
else {
std::cerr << "\n*** Error - problem failed the QA check ***\n" << std::endl;
return false;
}
}
// Warm-up run of the normal GEMM object
CHECK_CUTLASS_CALL(gemm_normal.run(), return false);
// Construct events
cudaEvent_t events[2];
for (auto & event : events) {
CHECK_CUDA_CALL(cudaEventCreate(&event), return false);
}
// Record an event at the start of a series of GEMM operations
CHECK_CUDA_CALL(cudaEventRecord(events[0]), return false);
// Run profiling loop
for (int iter = 0; iter < options.iterations; ++iter) {
gemm_normal();
}
// Record an event when the GEMM operations have been launched.
CHECK_CUDA_CALL(cudaEventRecord(events[1]), return false);
// Wait for work on the device to complete.
CHECK_CUDA_CALL(cudaEventSynchronize(events[1]), return false);
// Measure elapsed runtime
float runtime_total_ms = 0;
CHECK_CUDA_CALL(cudaEventElapsedTime(&runtime_total_ms, events[0], events[1]), return false);
// Compute average runtime and GFLOPs.
double runtime_avg_ms = double(runtime_total_ms) / double(options.iterations);
double gflops = options.gflops(runtime_avg_ms / 1000.0, kBatched);
// Cleanup
for (auto event : events) {
CHECK_CUDA_CALL(cudaEventDestroy(event), return false);
}
std::cout << " Runtime: " << runtime_avg_ms << " ms\n"
" GFLOPs: " << gflops << std::endl;
return true;
}
};
/// Shorthand alist for GEMM instantiations
template<typename LayoutA, typename PermuteALayout,
typename LayoutB, typename PermuteBLayout,
typename LayoutC, typename PermuteDLayout>
using GemmPermute = cutlass::gemm::device::GemmUniversal<
ElementInput, LayoutA,
ElementInput, LayoutB,
ElementOutput, LayoutC,
ElementAccumulator,
cutlass::arch::OpClassTensorOp,
cutlass::arch::Sm80,
cutlass::gemm::GemmShape<128, 128, 32>,
cutlass::gemm::GemmShape<64, 64, 32>,
cutlass::gemm::GemmShape<16, 8, 16>,
cutlass::epilogue::thread::LinearCombination<
ElementOutput,
AlignmentC, //128 / cutlass::sizeof_bits<ElementOutput>::value,
ElementAccumulator,
ElementAccumulator
>,
cutlass::gemm::threadblock::GemmIdentityThreadblockSwizzle<8>,
4, /*kStages*/
AlignmentA, /*AlignmentA*/
AlignmentB, /*AlignmentB*/
cutlass::arch::OpMultiplyAdd,
cutlass::ComplexTransform::kNone,
cutlass::ComplexTransform::kNone,
false, /*GatherA*/
false, /*GatherB*/
false, /*ScatterD*/
PermuteDLayout, /*PermuteDLayout*/
typename cutlass::layout::InversePermute<PermuteALayout>::type, /*PermuteALayout*/
typename cutlass::layout::InversePermute<PermuteBLayout>::type /*PermuteBLayout*/
>;
///////////////////////////////////////////////////////////////////////////////////////////////////
int main(int argc, char const **args) {
//
// This example uses mma.sync to directly access Tensor Cores to achieve peak performance.
//
cudaDeviceProp props;
CHECK_CUDA_CALL(cudaGetDeviceProperties(&props, 0), return EXIT_FAILURE);
if (__CUDACC_VER_MAJOR__ < 11 || props.major < 8) {
//
// This example requires an NVIDIA Ampere-architecture GPU.
//
std::cout << "CUTLASS's GEMM+Permute example requires a GPU of NVIDIA's Ampere Architecture "
"or later (compute capability 80 or greater).\n";
return EXIT_SUCCESS;
}
//
// Parse options
//
Options options;
options.parse(argc, args);
if (options.help) {
options.print_usage(std::cout) << std::endl;
return EXIT_SUCCESS;
}
if (options.error) {
std::cerr << "Aborting execution." << std::endl;
return EXIT_FAILURE;
}
//
// Define GEMM types to test
//
//
// TTT (Row-major) GEMMs
//
using TTTGemmNormalPermuteNone = GemmPermute<
cutlass::layout::RowMajor, cutlass::layout::NoPermute,
cutlass::layout::RowMajor, cutlass::layout::NoPermute,
cutlass::layout::RowMajor, cutlass::layout::NoPermute
>;
using TTTGemmNormalPermuteA = GemmPermute<
cutlass::layout::RowMajor, cutlass::layout::Tensor4DPermute0213RowMajor<S1, S2>,
cutlass::layout::RowMajor, cutlass::layout::NoPermute,
cutlass::layout::RowMajor, cutlass::layout::NoPermute
>;
using TTTGemmNormalPermuteAD = GemmPermute<
cutlass::layout::RowMajor, cutlass::layout::Tensor4DPermute0213RowMajor<S1, S2>,
cutlass::layout::RowMajor, cutlass::layout::NoPermute,
cutlass::layout::RowMajor, cutlass::layout::Tensor5DPermute20314RowMajor<T1, T2, T3>
>;
using TTTGemmNormalPermuteB = GemmPermute<
cutlass::layout::RowMajor, cutlass::layout::NoPermute,
cutlass::layout::RowMajor, cutlass::layout::Tensor4DPermute0213RowMajor<S1, S2>,
cutlass::layout::RowMajor, cutlass::layout::NoPermute
>;
using TTTGemmNormalPermuteBD = GemmPermute<
cutlass::layout::RowMajor, cutlass::layout::NoPermute,
cutlass::layout::RowMajor, cutlass::layout::Tensor4DPermute0213RowMajor<S1, S2>,
cutlass::layout::RowMajor, cutlass::layout::Tensor5DPermute20314RowMajor<T1, T2, T3>
>;
using TTTGemmNormalPermuteD = GemmPermute<
cutlass::layout::RowMajor, cutlass::layout::NoPermute,
cutlass::layout::RowMajor, cutlass::layout::NoPermute,
cutlass::layout::RowMajor, cutlass::layout::Tensor5DPermute20314RowMajor<T1, T2, T3>
>;
using TTTGemmNormalPermuteAB = GemmPermute<
cutlass::layout::RowMajor, cutlass::layout::Tensor4DPermute0213RowMajor<S1, S2>,
cutlass::layout::RowMajor, cutlass::layout::Tensor4DPermute0213RowMajor<S1, S2>,
cutlass::layout::RowMajor, cutlass::layout::NoPermute
>;
using TTTGemmNormalPermuteABD = GemmPermute<
cutlass::layout::RowMajor, cutlass::layout::Tensor4DPermute0213RowMajor<S1, S2>,
cutlass::layout::RowMajor, cutlass::layout::Tensor4DPermute0213RowMajor<S1, S2>,
cutlass::layout::RowMajor, cutlass::layout::Tensor5DPermute20314RowMajor<T1, T2, T3>
>;
//
// NNN (Col-major) GEMMs
//
using NNNGemmNormalPermuteNone = GemmPermute<
cutlass::layout::ColumnMajor, cutlass::layout::NoPermute,
cutlass::layout::ColumnMajor, cutlass::layout::NoPermute,
cutlass::layout::ColumnMajor, cutlass::layout::NoPermute
>;
using NNNGemmNormalPermuteA = GemmPermute<
cutlass::layout::ColumnMajor, cutlass::layout::Tensor5DPermute02413ColumnMajor<T1, T2, T3>,
cutlass::layout::ColumnMajor, cutlass::layout::NoPermute,
cutlass::layout::ColumnMajor, cutlass::layout::NoPermute
>;
using NNNGemmNormalPermuteAD = GemmPermute<
cutlass::layout::ColumnMajor, cutlass::layout::Tensor5DPermute02413ColumnMajor<T1, T2, T3>,
cutlass::layout::ColumnMajor, cutlass::layout::NoPermute,
cutlass::layout::ColumnMajor, cutlass::layout::Tensor5DPermute02413ColumnMajor<T1, T2, T3>
>;
using NNNGemmNormalPermuteB = GemmPermute<
cutlass::layout::ColumnMajor, cutlass::layout::NoPermute,
cutlass::layout::ColumnMajor, cutlass::layout::Tensor4DPermute0213ColumnMajor<S1, S2>,
cutlass::layout::ColumnMajor, cutlass::layout::NoPermute
>;
using NNNGemmNormalPermuteBD = GemmPermute<
cutlass::layout::ColumnMajor, cutlass::layout::NoPermute,
cutlass::layout::ColumnMajor, cutlass::layout::Tensor4DPermute0213ColumnMajor<S1, S2>,
cutlass::layout::ColumnMajor, cutlass::layout::Tensor5DPermute02413ColumnMajor<T1, T2, T3>
>;
using NNNGemmNormalPermuteD = GemmPermute<
cutlass::layout::ColumnMajor, cutlass::layout::NoPermute,
cutlass::layout::ColumnMajor, cutlass::layout::NoPermute,
cutlass::layout::ColumnMajor, cutlass::layout::Tensor5DPermute02413ColumnMajor<T1, T2, T3>
>;
using NNNGemmNormalPermuteAB = GemmPermute<
cutlass::layout::ColumnMajor, cutlass::layout::Tensor5DPermute02413ColumnMajor<T1, T2, T3>,
cutlass::layout::ColumnMajor, cutlass::layout::Tensor4DPermute0213ColumnMajor<S1, S2>,
cutlass::layout::ColumnMajor, cutlass::layout::NoPermute
>;
using NNNGemmNormalPermuteABD = GemmPermute<
cutlass::layout::ColumnMajor, cutlass::layout::Tensor5DPermute02413ColumnMajor<T1, T2, T3>,
cutlass::layout::ColumnMajor, cutlass::layout::Tensor4DPermute0213ColumnMajor<S1, S2>,
cutlass::layout::ColumnMajor, cutlass::layout::Tensor5DPermute02413ColumnMajor<T1, T2, T3>
>;
//
// NNT (Col-major inputs, row-major output) GEMMs
//
using NNTGemmNormalPermuteNone = GemmPermute<
cutlass::layout::ColumnMajor, cutlass::layout::NoPermute,
cutlass::layout::ColumnMajor, cutlass::layout::NoPermute,
cutlass::layout::RowMajor, cutlass::layout::NoPermute
>;
using NNTGemmNormalPermuteA = GemmPermute<
cutlass::layout::ColumnMajor, cutlass::layout::Tensor4DPermute0213RowMajor<S1, S2>,
cutlass::layout::ColumnMajor, cutlass::layout::NoPermute,
cutlass::layout::RowMajor, cutlass::layout::NoPermute
>;
using NNTGemmNormalPermuteAD = GemmPermute<
cutlass::layout::ColumnMajor, cutlass::layout::Tensor4DPermute0213RowMajor<S1, S2>,
cutlass::layout::ColumnMajor, cutlass::layout::NoPermute,
cutlass::layout::RowMajor, cutlass::layout::Tensor5DPermute20314RowMajor<T1, T2, T3>
>;
using NNTGemmNormalPermuteB = GemmPermute<
cutlass::layout::ColumnMajor, cutlass::layout::NoPermute,
cutlass::layout::ColumnMajor, cutlass::layout::Tensor4DPermute0213ColumnMajor<S1, S2>,
cutlass::layout::RowMajor, cutlass::layout::NoPermute
>;
using NNTGemmNormalPermuteBD = GemmPermute<
cutlass::layout::ColumnMajor, cutlass::layout::NoPermute,
cutlass::layout::ColumnMajor, cutlass::layout::Tensor4DPermute0213ColumnMajor<S1, S2>,
cutlass::layout::RowMajor, cutlass::layout::Tensor5DPermute20314RowMajor<T1, T2, T3>
>;
using NNTGemmNormalPermuteD = GemmPermute<
cutlass::layout::ColumnMajor, cutlass::layout::NoPermute,
cutlass::layout::ColumnMajor, cutlass::layout::NoPermute,
cutlass::layout::RowMajor, cutlass::layout::Tensor5DPermute20314RowMajor<T1, T2, T3>
>;
using NNTGemmNormalPermuteAB = GemmPermute<