-
Notifications
You must be signed in to change notification settings - Fork 1
/
torch-2.2.0-mac-with-tensorpipe-cuda10.1-10.2-support-memory-mpi-enabling.patch
1432 lines (1311 loc) · 58.1 KB
/
torch-2.2.0-mac-with-tensorpipe-cuda10.1-10.2-support-memory-mpi-enabling.patch
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
From 39798de17b24a19ee22bb74b40c2a57ab8718c65 Mon Sep 17 00:00:00 2001
From: Orlando Ding <xiandao.airs@gmail.com>
Date: Mon, 12 Feb 2024 22:28:43 -0800
Subject: [PATCH 1/8] orlando - for updates of settings
---
third_party/cutlass | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/third_party/cutlass b/third_party/cutlass
index 5a586c30b8..63fc6f05ff 160000
--- a/third_party/cutlass
+++ b/third_party/cutlass
@@ -1 +1 @@
-Subproject commit 5a586c30b81629fcf391c16f4314bb85dc5f23ff
+Subproject commit 63fc6f05ffbfa66ca9e5548a041517bb6100e52c
--
2.17.2 (Apple Git-113)
From 294eccdd7cdd9d2ac8c9758290c423fedf8dd277 Mon Sep 17 00:00:00 2001
From: Orlando Ding <xiandao.airs@gmail.com>
Date: Tue, 13 Feb 2024 10:06:23 -0800
Subject: [PATCH 2/8] orlando - for updates of tensorpipe settings
---
migration_note.md | 21 ++++++++++++++++++---
1 file changed, 18 insertions(+), 3 deletions(-)
diff --git a/migration_note.md b/migration_note.md
index f063e72d4a..d0cf1e1d10 100644
--- a/migration_note.md
+++ b/migration_note.md
@@ -1,5 +1,7 @@
# Migration note
+Preparation of building library:
+
```bash
export CXXFLAGS=-D_LIBCPP_DISABLE_AVAILABILITY
export CMAKE_PREFIX_PATH=${CONDA_PREFIX:-"$(dirname $(which conda))/../"}
@@ -9,11 +11,14 @@ MAGMA_HOME="/usr/local/lib/magma2.6.1-cu101" MACOSX_DEPLOYMENT_TARGET=10.9 CC=cl
## 1, Missing ATen cuda
+```bash
/usr/local/bin/ccache /Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/clang++ -DHAVE_MMAP=1 -DHAVE_SHM_OPEN=1 -DHAVE_SHM_UNLINK=1 -DIDEEP_USE_MKL -DMINIZ_DISABLE_ZIP_READER_CRC32_CHECKS -DONNXIFI_ENABLE_EXT=1 -DONNX_ML=1 -DONNX_NAMESPACE=onnx_torch -DUSE_CUDA_MPI=1 -DUSE_EXTERNAL_MZCRC -D_FILE_OFFSET_BITS=64 -Dcaffe2_nvrtc_EXPORTS -I/Users/llv23/Documents/05_machine_learning/dl_gpu_mac/pytorch-2.2.0-tensorpipe/build/aten/src -I/Users/llv23/Documents/05_machine_learning/dl_gpu_mac/pytorch-2.2.0-tensorpipe/aten/src -I/Users/llv23/Documents/05_machine_learning/dl_gpu_mac/pytorch-2.2.0-tensorpipe/build -I/Users/llv23/Documents/05_machine_learning/dl_gpu_mac/pytorch-2.2.0-tensorpipe -I/Users/llv23/Documents/05_machine_learning/dl_gpu_mac/pytorch-2.2.0-tensorpipe/cmake/../third_party/benchmark/include -I/Users/llv23/Documents/05_machine_learning/dl_gpu_mac/pytorch-2.2.0-tensorpipe/third_party/onnx -I/Users/llv23/Documents/05_machine_learning/dl_gpu_mac/pytorch-2.2.0-tensorpipe/build/third_party/onnx -I/Users/llv23/Documents/05_machine_learning/dl_gpu_mac/pytorch-2.2.0-tensorpipe/third_party/foxi -I/Users/llv23/Documents/05_machine_learning/dl_gpu_mac/pytorch-2.2.0-tensorpipe/build/third_party/foxi -isystem /Users/llv23/Documents/05_machine_learning/dl_gpu_mac/pytorch-2.2.0-tensorpipe/build/third_party/gloo -isystem /Users/llv23/Documents/05_machine_learning/dl_gpu_mac/pytorch-2.2.0-tensorpipe/cmake/../third_party/gloo -isystem /Users/llv23/Documents/05_machine_learning/dl_gpu_mac/pytorch-2.2.0-tensorpipe/cmake/../third_party/tensorpipe/third_party/libuv/include -isystem /Users/llv23/Documents/05_machine_learning/dl_gpu_mac/pytorch-2.2.0-tensorpipe/cmake/../third_party/googletest/googlemock/include -isystem /Users/llv23/Documents/05_machine_learning/dl_gpu_mac/pytorch-2.2.0-tensorpipe/cmake/../third_party/googletest/googletest/include -isystem /Users/llv23/Documents/05_machine_learning/dl_gpu_mac/pytorch-2.2.0-tensorpipe/third_party/protobuf/src -isystem /Users/llv23/opt/miniconda3/include -isystem /Users/llv23/Documents/05_machine_learning/dl_gpu_mac/pytorch-2.2.0-tensorpipe/third_party/gemmlowp -isystem /Users/llv23/Documents/05_machine_learning/dl_gpu_mac/pytorch-2.2.0-tensorpipe/third_party/neon2sse -isystem /Users/llv23/Documents/05_machine_learning/dl_gpu_mac/pytorch-2.2.0-tensorpipe/third_party/XNNPACK/include -isystem /Users/llv23/Documents/05_machine_learning/dl_gpu_mac/pytorch-2.2.0-tensorpipe/third_party/ittapi/include -isystem /Users/llv23/Documents/05_machine_learning/dl_gpu_mac/pytorch-2.2.0-tensorpipe/cmake/../third_party/eigen -isystem /Users/llv23/Documents/05_machine_learning/dl_gpu_mac/pytorch-2.2.0-tensorpipe/cmake/../third_party/cub -isystem /usr/local/include -isystem /Users/llv23/Documents/05_machine_learning/dl_gpu_mac/pytorch-2.2.0-tensorpipe/third_party/ideep/include -D_LIBCPP_DISABLE_AVAILABILITY -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -DNDEBUG -DUSE_KINETO -DLIBKINETO_NOCUPTI -DLIBKINETO_NOROCTRACER -DUSE_FBGEMM -DUSE_QNNPACK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -O2 -fPIC -Wall -Wextra -Werror=return-type -Werror=non-virtual-dtor -Werror=braced-scalar-init -Wnarrowing -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wno-unused-parameter -Wno-unused-function -Wno-unused-result -Wno-strict-overflow -Wno-strict-aliasing -Wvla-extension -Wnewline-eof -Winconsistent-missing-override -Winconsistent-missing-destructor-override -Wno-pass-failed -Wno-error=pedantic -Wno-error=old-style-cast -Wno-error=inconsistent-missing-override -Wno-error=inconsistent-missing-destructor-override -Wconstant-conversion -Wno-invalid-partial-specialization -Wno-aligned-allocation-unavailable -Wno-missing-braces -Qunused-arguments -fcolor-diagnostics -faligned-new -fno-math-errno -fno-trapping-math -Werror=format -Wno-unused-private-field -Wno-missing-braces -DHAVE_AVX2_CPU_DEFINITION -O3 -DNDEBUG -DNDEBUG -std=gnu++14 -isysroot /Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX10.14.sdk -mmacosx-version-min=10.9 -fPIC -DMKL_HAS_SBGEMM -DTORCH_USE_LIBUV -DCAFFE2_USE_GLOO -MD -MT caffe2/CMakeFiles/caffe2_nvrtc.dir/__/aten/src/ATen/cuda/nvrtc_stub/ATenNVRTC.cpp.o -MF caffe2/CMakeFiles/caffe2_nvrtc.dir/__/aten/src/ATen/cuda/nvrtc_stub/ATenNVRTC.cpp.o.d -o caffe2/CMakeFiles/caffe2_nvrtc.dir/__/aten/src/ATen/cuda/nvrtc_stub/ATenNVRTC.cpp.o -c /Users/llv23/Documents/05_machine_learning/dl_gpu_mac/pytorch-2.2.0-tensorpipe/aten/src/ATen/cuda/nvrtc_stub/ATenNVRTC.cpp
+```
## 2, Migrating from c10 to std
-#if defined(__APPLE__) && defined(__MACH__)
+```c++
+# if defined(__APPLE__) && defined(__MACH__)
#include <c10/util/variant.h>
namespace std {
using ::c10::variant;
@@ -25,8 +30,9 @@ namespace std {
#else
#include <variant>
#endif
+```
-
+```c++
#if defined(__APPLE__) && defined(__MACH__)
#include <c10/util/Optional.h>
namespace std {
@@ -35,22 +41,30 @@ namespace std {
#else
#include <optional>
#endif
+```
+```c++
#if defined(__APPLE__) && defined(__MACH__)
#include <c10/util/variant.h>
#endif
+```
+```c++
#if defined(__APPLE__) && defined(__MACH__)
#include <c10/util/variant.h>
#else
#include <variant>
#endif
+```
+```c++
#if defined(__APPLE__) && defined(__MACH__)
c10::visit
#else
#endif
+```
+```c++
MetadataShape compute_variant_shape(const at::Tensor& input) {
if (input.is_nested() && !input.unsafeGetTensorImpl()->is_python_dispatch()) {
auto nested_size = input._nested_tensor_size();
@@ -66,6 +80,7 @@ MetadataShape compute_variant_shape(const at::Tensor& input) {
return MetadataShape{std::in_place_type<SymIntSmallVec>, input.sym_sizes()};
#endif
}
+```
## 3, Issue of loading include headers
@@ -84,7 +99,7 @@ FAILED: caffe2/CMakeFiles/torch_cuda.dir/__/aten/src/ATen/native/nested/cuda/Nes
#include <cutlass/gemm/device/default_gemm_configuration.h>
```
-solution: correct the caffe2/CMakeLists.txt in Line 96 by
+Solution: correct the caffe2/CMakeLists.txt in Line 96 and switch cutlass to 2.11.0, a prior version to 3.0.0 for CUDA 11.x
```cmake
list(APPEND Caffe2_GPU_INCLUDE ${ATen_CUDA_INCLUDE} /usr/local/cuda/include ${PROJECT_SOURCE_DIR}/third_party/cutlass/include)
--
2.17.2 (Apple Git-113)
From 49f18e626e3c2a1c7e18fdca0dece3bf92b04d03 Mon Sep 17 00:00:00 2001
From: Orlando Ding <xiandao.airs@gmail.com>
Date: Fri, 16 Feb 2024 23:01:56 -0800
Subject: [PATCH 3/8] orlando - for updates of torch 2.2.0, but meeting with
issues
---
aten/src/ATen/cuda/CUDABlas.cpp | 177 ++++++++++++++++--
.../sparse/cuda/SparseSemiStructuredLinear.cu | 4 +-
c10/util/Optional.cpp | 17 ++
c10/util/Optional.h | 6 +-
migration_note.md | 59 +++++-
third_party/cutlass | 2 +-
torch/csrc/distributed/c10d/init.cpp | 10 +-
torch/csrc/distributed/rpc/init.cpp | 6 +-
8 files changed, 256 insertions(+), 25 deletions(-)
diff --git a/aten/src/ATen/cuda/CUDABlas.cpp b/aten/src/ATen/cuda/CUDABlas.cpp
index a161786074..c58a987680 100644
--- a/aten/src/ATen/cuda/CUDABlas.cpp
+++ b/aten/src/ATen/cuda/CUDABlas.cpp
@@ -15,6 +15,27 @@
// added bf16 support
#if !defined(USE_ROCM) && !defined(_MSC_VER)
#include <cublasLt.h>
+
+#if defined(__APPLE__) && defined(__MACH__)
+/** Semi-opaque descriptor for cublasLtMatmul() operation details
+ */
+typedef struct {
+ uint64_t data[32];
+} cublasLtMatmulDescOpaque_t;
+
+/** Semi-opaque descriptor for matrix memory layout
+ */
+typedef struct {
+ uint64_t data[8];
+} cublasLtMatrixLayoutOpaque_t;
+
+/** Semi-opaque descriptor for cublasLtMatmulPreference() operation details
+ */
+typedef struct {
+ uint64_t data[8];
+} cublasLtMatmulPreferenceOpaque_t;
+#endif
+
#endif
// refer to http://www.jcuda.org/jcuda/jcublas/doc/constant-values.html#jcuda.jcublas.cublasMath.CUBLAS_MATH_DISALLOW_REDUCED_PRECISION_REDUCTION
@@ -205,10 +226,60 @@ static size_t _getWorkspaceSize() {
} // anonymous namespace
-namespace at::cuda::blas {
+namespace at{ namespace cuda{ namespace blas {
/* LEVEL 3 BLAS FUNCTIONS */
+#ifndef USE_ROCM
+#if defined(CUDA_VERSION) && CUDA_VERSION >= 11020
+#define cublasGemmStridedBatchedExFix cublasGemmStridedBatchedEx
+#else
+// Workaround for https://github.com/pytorch/pytorch/issues/45724
+cublasStatus_t cublasGemmStridedBatchedExFix(cublasHandle_t &handle,
+ cublasOperation_t transa,
+ cublasOperation_t transb,
+ int m,
+ int n,
+ int k,
+ const void *alpha,
+ const void *A,
+ cudaDataType Atype,
+ int lda,
+ long long int strideA,
+ const void *B,
+ cudaDataType Btype,
+ int ldb,
+ long long int strideB,
+ const void *beta,
+ void *C,
+ cudaDataType Ctype,
+ int ldc,
+ long long int strideC,
+ int64_t batchCount,
+ cudaDataType computeType,
+ cublasGemmAlgo_t algo)
+{
+ cudaDeviceProp* prop = at::cuda::getCurrentDeviceProperties();
+ if (prop->major != 7) {
+ return cublasGemmStridedBatchedEx(handle, transa, transb, m, n, k, alpha, A, Atype, lda, strideA, B, Btype, ldb, strideB, beta, C, Ctype, ldc, strideC, batchCount, computeType, algo);
+ }
+ cublasStatus_t result;
+ constexpr int64_t split = 63 * 1024;
+ for(int64_t i = 0; i < batchCount; i += split) {
+ int64_t count = std::min<int64_t>(split, batchCount - i);
+ result = cublasGemmStridedBatchedEx(handle, transa, transb, m, n, k, alpha,
+ (char *)A + i * strideA * 2, Atype, lda, strideA,
+ (char *)B + i * strideB * 2, Btype, ldb, strideB,
+ beta,
+ (char *)C + i * strideC * 2, Ctype, ldc, strideC,
+ (int)count, computeType, algo);
+ TORCH_CUDABLAS_CHECK(result);
+ }
+ return result;
+}
+#endif
+#endif
+
#define GEMM_CHECK_ARGVALUES(Dtype) \
do { \
CUDABLAS_NONNEGINT_CHECK(gemm<Dtype>, m); \
@@ -527,7 +598,43 @@ void gemm<at::Half>(CUDABLAS_GEMM_ARGTYPES(at::Half)) {
#endif
}
-#if !defined(USE_ROCM)
+#ifdef defined(USE_ROCM)
+template <>
+void gemm<at::BFloat16>(CUDABLAS_GEMM_ARGTYPES(at::BFloat16)) {
+ cublasHandle_t handle = at::cuda::getCurrentCUDABlasHandle();
+ cublasOperation_t opa = _cublasOpFromChar(transa);
+ cublasOperation_t opb = _cublasOpFromChar(transb);
+ float falpha = alpha;
+ float fbeta = beta;
+ _cublasAdjustLdLevel3(transa, transb, m, n, k, &lda, &ldb, &ldc);
+ GEMM_CHECK_ARGVALUES(at::BFloat16);
+ TORCH_CUDABLAS_CHECK(rocblas_gemm_ex(
+ handle,
+ opa,
+ opb,
+ m,
+ n,
+ k,
+ &falpha,
+ a,
+ rocblas_datatype_bf16_r,
+ lda,
+ b,
+ rocblas_datatype_bf16_r,
+ ldb,
+ &fbeta,
+ c,
+ rocblas_datatype_bf16_r,
+ ldc,
+ c,
+ rocblas_datatype_bf16_r,
+ ldc,
+ rocblas_datatype_f32_r,
+ rocblas_gemm_algo_standard,
+ 0,
+ 0));
+}
+#else
template <>
void gemm<at::BFloat16>(CUDABLAS_GEMM_ARGTYPES(at::BFloat16)) {
globalContext().alertCuBLASConfigNotDeterministic();
@@ -567,7 +674,7 @@ void gemm<at::BFloat16>(CUDABLAS_GEMM_ARGTYPES(at::BFloat16)) {
}
#endif // !defined(USE_ROCM)
-#if !defined(USE_ROCM) && !defined(_MSC_VER) && defined(CUDA_VERSION) && CUDA_VERSION >= 11000
+#if !defined(USE_ROCM) && !defined(_MSC_VER)
namespace {
// Following the pattern of CuSparseDescriptor
@@ -597,6 +704,24 @@ class CuBlasLtDescriptor {
std::unique_ptr<T, CuBlasLtDeleter<T, destructor>> descriptor_;
};
+#if defined(__APPLE__) && defined(__MACH__)
+class CuBlasLtMatmulDescriptor : public CuBlasLtDescriptor<
+ cublasLtMatmulDescStruct,
+ &cublasLtMatmulDescDestroy> {
+ public:
+ CuBlasLtMatmulDescriptor(
+ cudaDataType_t scale_type) {
+ cublasLtMatmulDesc_t raw_descriptor = nullptr;
+ TORCH_CUDABLAS_CHECK(
+ cublasLtMatmulDescCreate(&raw_descriptor, scale_type));
+ descriptor_.reset(raw_descriptor);
+ }
+ template <typename T>
+ inline void setAttribute(cublasLtMatmulDescAttributes_t attr, const T value) {
+ TORCH_CUDABLAS_CHECK(::cublasLtMatmulDescSetAttribute(descriptor(), attr, &value, sizeof(T)));
+ }
+};
+#else
class CuBlasLtMatmulDescriptor : public CuBlasLtDescriptor<
cublasLtMatmulDescOpaque_t,
&cublasLtMatmulDescDestroy> {
@@ -614,9 +739,10 @@ class CuBlasLtMatmulDescriptor : public CuBlasLtDescriptor<
TORCH_CUDABLAS_CHECK(::cublasLtMatmulDescSetAttribute(descriptor(), attr, &value, sizeof(T)));
}
};
+#endif
class CuBlasLtMatrixLayout : public CuBlasLtDescriptor<
- cublasLtMatrixLayoutOpaque_t,
+ cublasLtMatrixLayoutStruct,
&cublasLtMatrixLayoutDestroy> {
public:
CuBlasLtMatrixLayout(
@@ -633,7 +759,7 @@ class CuBlasLtMatrixLayout : public CuBlasLtDescriptor<
};
class CuBlasLtMatmulPreference : public CuBlasLtDescriptor<
- cublasLtMatmulPreferenceOpaque_t,
+ cublasLtMatmulPreferenceStruct,
&cublasLtMatmulPreferenceDestroy> {
public:
CuBlasLtMatmulPreference() {
@@ -648,8 +774,6 @@ class CuBlasLtMatmulPreference : public CuBlasLtDescriptor<
};
} // namespace
-
-#if !defined(USE_ROCM) && CUDA_VERSION >= 11000
template <typename Dtype>
void gemm_and_bias(
bool transpose_mat1,
@@ -670,24 +794,38 @@ void gemm_and_bias(
opmath_t beta_val = 0; // bias is added in epilogue
cudaDataType_t abcType = CUDA_R_32F;
+#if !defined(__APPLE__) && !defined(__MACH__)
cublasComputeType_t computeType = CUBLAS_COMPUTE_32F;
+#endif
cudaDataType_t scaleType = CUDA_R_32F;
- if constexpr (std::is_same_v<Dtype, double>) {
+ if constexpr (std::is_same<Dtype, double>::value) {
abcType = CUDA_R_64F;
+#if !defined(__APPLE__) && !defined(__MACH__)
computeType = CUBLAS_COMPUTE_64F;
+#endif
scaleType = CUDA_R_64F;
- } else if constexpr (std::is_same_v<Dtype, float>) {
+ } else if constexpr (std::is_same<Dtype, float>::value) {
+#if !defined(__APPLE__) && !defined(__MACH__)
if (at::globalContext().allowTF32CuBLAS()) {
computeType = CUBLAS_COMPUTE_32F_FAST_TF32;
}
+#endif
abcType = CUDA_R_32F;
- } else if constexpr (std::is_same_v<Dtype, at::Half>) {
+ } else if constexpr (std::is_same<Dtype, at::Half>::value) {
abcType = CUDA_R_16F;
- } else if constexpr (std::is_same_v<Dtype, at::BFloat16>) {
+ } else if constexpr (std::is_same<Dtype, at::BFloat16>::value) {
+#if !defined(__APPLE__) && !defined(__MACH__)
abcType = CUDA_R_16BF;
+#else
+ abcType = CUDA_R_16F;
+#endif
}
+#if defined(__APPLE__) && defined(__MACH__)
+ CuBlasLtMatmulDescriptor computeDesc(scaleType);
+#else
CuBlasLtMatmulDescriptor computeDesc(computeType, scaleType);
+#endif
cublasOperation_t transa = transpose_mat1 ? CUBLAS_OP_T : CUBLAS_OP_N;
computeDesc.setAttribute(CUBLASLT_MATMUL_DESC_TRANSA, transa);
cublasOperation_t transb = transpose_mat2 ? CUBLAS_OP_T : CUBLAS_OP_N;
@@ -783,8 +921,10 @@ void gemm_and_bias(
result_ld,
" abcType ",
abcType,
+#if !defined(__APPLE__) && !defined(__MACH__)
" computeType ",
computeType,
+#endif
" scaleType ",
scaleType);
}
@@ -852,7 +992,6 @@ template void gemm_and_bias(
at::BFloat16* result_ptr,
int64_t result_ld,
GEMMAndBiasActivationEpilogue activation);
-#endif
void scaled_gemm(
char transa,
@@ -880,7 +1019,11 @@ void scaled_gemm(
const auto computeType = CUBLAS_COMPUTE_32F;
const auto scaleType = CUDA_R_32F;
const int8_t fastAccuMode = use_fast_accum ? 1 : 0;
+#if defined(__APPLE__) && defined(__MACH__)
+ CuBlasLtMatmulDescriptor computeDesc(scaleType);
+#else
CuBlasLtMatmulDescriptor computeDesc(computeType, scaleType);
+#endif
computeDesc.setAttribute(CUBLASLT_MATMUL_DESC_TRANSA, _cublasOpFromChar(transa));
computeDesc.setAttribute(CUBLASLT_MATMUL_DESC_TRANSB, _cublasOpFromChar(transb));
computeDesc.setAttribute(CUBLASLT_MATMUL_DESC_A_SCALE_POINTER, mat1_scale_ptr);
@@ -982,13 +1125,19 @@ void int8_gemm(
int32_t* result_ptr,
int64_t result_ld) {
+#if !defined(__APPLE__) && !defined(__MACH__)
cublasComputeType_t computeType = CUBLAS_COMPUTE_32I;
+#endif
cudaDataType_t scaleType = CUDA_R_32I;
cudaDataType_t abType = CUDA_R_8I;
cudaDataType_t cType = CUDA_R_32I;
+#if defined(__APPLE__) && defined(__MACH__)
+ CuBlasLtMatmulDescriptor computeDesc(scaleType);
+#else
CuBlasLtMatmulDescriptor computeDesc(computeType, scaleType);
+#endif
cublasOperation_t transa = transpose_mat1 ? CUBLAS_OP_T : CUBLAS_OP_N;
computeDesc.setAttribute(CUBLASLT_MATMUL_DESC_TRANSA, transa);
cublasOperation_t transb = transpose_mat2 ? CUBLAS_OP_T : CUBLAS_OP_N;
@@ -1047,8 +1196,10 @@ void int8_gemm(
abType,
" cType ",
cType,
+#if !defined(__APPLE__) && !defined(__MACH__)
" computeType ",
computeType,
+#endif
" scaleType ",
scaleType);
}
@@ -1591,4 +1742,4 @@ void gelsBatched<c10::complex<float>>(CUDABLAS_GELS_BATCHED_ARGTYPES(c10::comple
batchSize));
}
-} // namespace at::cuda::blas
+}}} // namespace at::cuda::blas
diff --git a/aten/src/ATen/native/sparse/cuda/SparseSemiStructuredLinear.cu b/aten/src/ATen/native/sparse/cuda/SparseSemiStructuredLinear.cu
index 3ea75cc84d..03d1c4319e 100644
--- a/aten/src/ATen/native/sparse/cuda/SparseSemiStructuredLinear.cu
+++ b/aten/src/ATen/native/sparse/cuda/SparseSemiStructuredLinear.cu
@@ -3,7 +3,7 @@
#include <ATen/cuda/CUDAUtils.h>
#include <ATen/Dispatch.h>
-#if !defined(USE_ROCM) && !defined(__APPLE__) && !defined(__MACH__)
+#if !defined(USE_ROCM)
#include <cuda_runtime.h>
#include <cutlass/cutlass.h>
#include <cutlass/layout/layout.h>
@@ -12,8 +12,10 @@
#include <cutlass/epilogue/thread/linear_combination_relu.h>
#include <cutlass/epilogue/thread/linear_combination_silu.h>
#include <cutlass/gemm/gemm.h>
+#if !defined(__APPLE__) && !defined(__MACH__)
#include <cutlass/gemm/device/gemm_sparse_row_broadcast.h>
#endif
+#endif
#include <type_traits>
#if defined(__APPLE__) && defined(__MACH__)
diff --git a/c10/util/Optional.cpp b/c10/util/Optional.cpp
index 7389393e66..c83614d448 100644
--- a/c10/util/Optional.cpp
+++ b/c10/util/Optional.cpp
@@ -1 +1,18 @@
+#include <c10/util/ArrayRef.h>
#include <c10/util/Optional.h>
+
+#include <type_traits>
+
+static_assert(
+ C10_IS_TRIVIALLY_COPYABLE(c10::optional<int>),
+ "c10::optional<int> should be trivially copyable");
+static_assert(
+ C10_IS_TRIVIALLY_COPYABLE(c10::optional<bool>),
+ "c10::optional<bool> should be trivially copyable");
+static_assert(
+ C10_IS_TRIVIALLY_COPYABLE(c10::optional<c10::IntArrayRef>),
+ "c10::optional<IntArrayRef> should be trivially copyable");
+static_assert(
+ sizeof(c10::optional<c10::IntArrayRef>) == sizeof(c10::IntArrayRef),
+ "c10::optional<IntArrayRef> should be size-optimized");
+
diff --git a/c10/util/Optional.h b/c10/util/Optional.h
index 45d58282e3..23eac9e0ec 100644
--- a/c10/util/Optional.h
+++ b/c10/util/Optional.h
@@ -1,7 +1,7 @@
#ifndef C10_UTIL_OPTIONAL_H_
#define C10_UTIL_OPTIONAL_H_
-#if defined(__APPLE__) && defined(__MACH__)
+// #if defined(__APPLE__) && defined(__MACH__)
#include <c10/macros/Macros.h>
#include <c10/util/ArrayRef.h>
@@ -1235,7 +1235,7 @@ struct hash<c10::optional<T&>> {
C10_CLANG_DIAGNOSTIC_POP()
-#else
+#if !defined(__APPLE__) && !defined(__MACH__)
#include <optional>
#include <type_traits>
@@ -1281,6 +1281,6 @@ constexpr T value_or_else(optional<T>&& v, F&& func) {
}
} // namespace c10
-#endif // defined(__APPLE__) && defined(__MACH__)
+#endif // !defined(__APPLE__) && !defined(__MACH__)
#endif // C10_UTIL_OPTIONAL_H_
diff --git a/migration_note.md b/migration_note.md
index d0cf1e1d10..4ea0691f13 100644
--- a/migration_note.md
+++ b/migration_note.md
@@ -6,7 +6,9 @@ Preparation of building library:
export CXXFLAGS=-D_LIBCPP_DISABLE_AVAILABILITY
export CMAKE_PREFIX_PATH=${CONDA_PREFIX:-"$(dirname $(which conda))/../"}
MAGMA_HOME="/usr/local/lib/magma2.6.1-cu101" MACOSX_DEPLOYMENT_TARGET=10.9 CC=clang CXX=clang++ USE_LIBUV=1 USE_DISTRIBUTED=ON USE_MPI=ON USE_TENSORPIPE=ON USE_GLOO=ON USE_CUDA_MPI=ON python setup.py clean # prepare
-MAGMA_HOME="/usr/local/lib/magma2.6.1-cu101" MACOSX_DEPLOYMENT_TARGET=10.9 CC=clang CXX=clang++ USE_LIBUV=1 USE_DISTRIBUTED=ON USE_MPI=ON USE_TENSORPIPE=ON USE_GLOO=ON USE_CUDA_MPI=ON python setup.py bdist_wheel
+MAGMA_HOME="/usr/local/lib/magma2.6.1-cu101" MACOSX_DEPLOYMENT_TARGET=10.9 CC=clang CXX=clang++ CMAKE_BUILD_TYPE=1 USE_LIBUV=1 USE_DISTRIBUTED=ON USE_MPI=ON USE_TENSORPIPE=ON USE_GLOO=ON USE_CUDA_MPI=ON python setup.py bdist_wheel
+MAGMA_HOME="/usr/local/lib/magma2.6.1-cu101" MACOSX_DEPLOYMENT_TARGET=10.9 CC=clang CXX=clang++ CMAKE_BUILD_TYPE=1 USE_LIBUV=1 USE_CUSPARSELT=1 USE_DISTRIBUTED=ON USE_MPI=ON USE_TENSORPIPE=ON USE_GLOO=ON USE_CUDA_MPI=ON python setup.py bdist_wheel
+MAGMA_HOME="/usr/local/lib/magma2.6.1-cu101" MACOSX_DEPLOYMENT_TARGET=10.9 CC=clang CXX=clang++ USE_LIBUV=1 USE_CUSPARSELT=1 USE_DISTRIBUTED=ON USE_MPI=OFF USE_TENSORPIPE=ON USE_GLOO=ON USE_CUDA_MPI=ON python setup.py develop
```
## 1, Missing ATen cuda
@@ -104,3 +106,58 @@ Solution: correct the caffe2/CMakeLists.txt in Line 96 and switch cutlass to 2.1
```cmake
list(APPEND Caffe2_GPU_INCLUDE ${ATen_CUDA_INCLUDE} /usr/local/cuda/include ${PROJECT_SOURCE_DIR}/third_party/cutlass/include)
```
+
+## 4. Runtime issue
+
+torch 2.2.0
+
+```bash
+(base) Orlando:gpu-magma2.6.1-distributed-all-2.2.0-py3.10 llv23$ otool -L /Users/llv23/opt/miniconda3/lib/python3.10/site-packages/torch/lib/libtorch_python.dylib
+/Users/llv23/opt/miniconda3/lib/python3.10/site-packages/torch/lib/libtorch_python.dylib:
+ @rpath/libtorch_python.dylib (compatibility version 0.0.0, current version 0.0.0)
+ @rpath/libshm.dylib (compatibility version 0.0.0, current version 0.0.0)
+ @rpath/libtorch.dylib (compatibility version 0.0.0, current version 0.0.0)
+ @rpath/libtorch_cuda.dylib (compatibility version 0.0.0, current version 0.0.0)
+ @rpath/libnvToolsExt.1.dylib (compatibility version 0.0.0, current version 1.0.0)
+ @rpath/libtorch_cpu.dylib (compatibility version 0.0.0, current version 0.0.0)
+ @rpath/libmkl_intel_lp64.2.dylib (compatibility version 0.0.0, current version 0.0.0)
+ @rpath/libmkl_intel_thread.2.dylib (compatibility version 0.0.0, current version 0.0.0)
+ @rpath/libmkl_core.2.dylib (compatibility version 0.0.0, current version 0.0.0)
+ @rpath/libomp.dylib (compatibility version 5.0.0, current version 5.0.0)
+ /usr/lib/libSystem.B.dylib (compatibility version 1.0.0, current version 1252.200.5)
+ @rpath/libc10_cuda.dylib (compatibility version 0.0.0, current version 0.0.0)
+ @rpath/libc10.dylib (compatibility version 0.0.0, current version 0.0.0)
+ @rpath/libcudart.10.2.dylib (compatibility version 0.0.0, current version 10.2.89)
+ @rpath/libcudnn.7.dylib (compatibility version 0.0.0, current version 7.6.5)
+ /usr/local/opt/open-mpi/lib/libmpi.40.dylib (compatibility version 71.0.0, current version 71.1.0)
+ /usr/lib/libc++.1.dylib (compatibility version 1.0.0, current version 400.9.4)
+```
+
+torch 2.0.0
+
+```bash
+(base) Orlando:lib llv23$ otool -L /Users/llv23/opt/miniconda3/lib/python3.10/site-packages/torch/lib/libtorch_python.dylib
+/Users/llv23/opt/miniconda3/lib/python3.10/site-packages/torch/lib/libtorch_python.dylib:
+ @rpath/libtorch_python.dylib (compatibility version 0.0.0, current version 0.0.0)
+ @rpath/libshm.dylib (compatibility version 0.0.0, current version 0.0.0)
+ /usr/local/opt/open-mpi/lib/libmpi.40.dylib (compatibility version 71.0.0, current version 71.1.0)
+ @rpath/libtorch.dylib (compatibility version 0.0.0, current version 0.0.0)
+ @rpath/libtorch_cuda.dylib (compatibility version 0.0.0, current version 0.0.0)
+ @rpath/libnvrtc.10.1.dylib (compatibility version 0.0.0, current version 10.1.243)
+ @rpath/libnvToolsExt.1.dylib (compatibility version 0.0.0, current version 1.0.0)
+ @rpath/libtorch_cpu.dylib (compatibility version 0.0.0, current version 0.0.0)
+ @rpath/libmkl_intel_lp64.2.dylib (compatibility version 0.0.0, current version 0.0.0)
+ @rpath/libmkl_intel_thread.2.dylib (compatibility version 0.0.0, current version 0.0.0)
+ @rpath/libmkl_core.2.dylib (compatibility version 0.0.0, current version 0.0.0)
+ @rpath/libomp.dylib (compatibility version 5.0.0, current version 5.0.0)
+ /usr/lib/libSystem.B.dylib (compatibility version 1.0.0, current version 1252.200.5)
+ @rpath/libc10_cuda.dylib (compatibility version 0.0.0, current version 0.0.0)
+ @rpath/libc10.dylib (compatibility version 0.0.0, current version 0.0.0)
+ @rpath/libcudart.10.1.dylib (compatibility version 0.0.0, current version 10.1.243)
+ @rpath/libcufft.10.dylib (compatibility version 0.0.0, current version 10.1.1)
+ @rpath/libcurand.10.dylib (compatibility version 0.0.0, current version 10.1.1)
+ @rpath/libcublas.10.dylib (compatibility version 0.0.0, current version 10.2.1)
+ @rpath/libcublasLt.10.dylib (compatibility version 0.0.0, current version 10.2.1)
+ @rpath/libcudnn.7.dylib (compatibility version 0.0.0, current version 7.6.5)
+ /usr/lib/libc++.1.dylib (compatibility version 1.0.0, current version 400.9.4)
+```
diff --git a/third_party/cutlass b/third_party/cutlass
index 63fc6f05ff..b72cbf957d 160000
--- a/third_party/cutlass
+++ b/third_party/cutlass
@@ -1 +1 @@
-Subproject commit 63fc6f05ffbfa66ca9e5548a041517bb6100e52c
+Subproject commit b72cbf957df8cf84a6d0ff91c190ad51a9c1d24a
diff --git a/torch/csrc/distributed/c10d/init.cpp b/torch/csrc/distributed/c10d/init.cpp
index 3296bd3754..0206be063d 100644
--- a/torch/csrc/distributed/c10d/init.cpp
+++ b/torch/csrc/distributed/c10d/init.cpp
@@ -1726,8 +1726,8 @@ Arguments:
},
py::arg("device"),
py::arg("backend_type"),
- py::arg("backend") =
- c10::optional<c10::intrusive_ptr<::c10d::Backend>>(),
+ // py::arg("backend") = c10::optional<c10::intrusive_ptr<::c10d::Backend>>(),
+ py::arg("backend"),
py::call_guard<py::gil_scoped_release>())
.def(
"_get_backend",
@@ -2589,7 +2589,8 @@ Example::
py::arg("bucket_size"),
py::arg("expect_sparse_gradient") = std::vector<bool>(),
py::arg("tensor_indices") = std::vector<int64_t>(),
- py::arg("logger") = c10::optional<std::shared_ptr<::c10d::Logger>>{},
+ // py::arg("logger") = c10::optional<std::shared_ptr<::c10d::Logger>>{},
+ py::arg("logger"),
py::call_guard<py::gil_scoped_release>());
module.def(
@@ -2607,7 +2608,8 @@ Example::
},
py::arg("process_group"),
py::arg("params"),
- py::arg("logger") = c10::optional<std::shared_ptr<::c10d::Logger>>{},
+ // py::arg("logger") = c10::optional<std::shared_ptr<::c10d::Logger>>{},
+ py::arg("logger"),
py::call_guard<py::gil_scoped_release>());
module.def(
diff --git a/torch/csrc/distributed/rpc/init.cpp b/torch/csrc/distributed/rpc/init.cpp
index 7b8a2d1f18..69ac2a13ce 100644
--- a/torch/csrc/distributed/rpc/init.cpp
+++ b/torch/csrc/distributed/rpc/init.cpp
@@ -544,8 +544,10 @@ PyObject* rpc_init(PyObject* _unused, PyObject* noargs) {
std::unordered_map<std::string, DeviceMap>,
std::vector<c10::Device>>(),
py::arg("num_worker_threads") = kDefaultNumWorkerThreads,
- py::arg("_transports") = optional<std::vector<std::string>>(),
- py::arg("_channels") = optional<std::vector<std::string>>(),
+ // py::arg("_transports") = optional<std::vector<std::string>>(),
+ py::arg("_transports"),
+ // py::arg("_channels") = optional<std::vector<std::string>>(),
+ py::arg("_channels"),
py::arg("rpc_timeout") = kDefaultRpcTimeoutSeconds,
py::arg("init_method") = kDefaultInitMethod,
py::arg("device_maps") = std::unordered_map<std::string, DeviceMap>(),
--
2.17.2 (Apple Git-113)
From 46b15b281dabf8ea5974ceb12670d113bdc94cf5 Mon Sep 17 00:00:00 2001
From: orlando <xiandao.airs@gmail.com>
Date: Sun, 18 Feb 2024 21:07:47 -0800
Subject: [PATCH 4/8] Update intrusive_ptr.h
updates of headers
---
c10/util/intrusive_ptr.h | 3 +++
1 file changed, 3 insertions(+)
diff --git a/c10/util/intrusive_ptr.h b/c10/util/intrusive_ptr.h
index 8e43dbd876..704cc486bb 100644
--- a/c10/util/intrusive_ptr.h
+++ b/c10/util/intrusive_ptr.h
@@ -1,10 +1,13 @@
#pragma once
+#include <c10/util/C++17.h>
#include <c10/util/Exception.h>
+#include <c10/util/ExclusivelyOwned.h>
#include <c10/util/MaybeOwned.h>
#include <atomic>
#include <climits>
#include <memory>
+#include <stdexcept>
namespace pybind11 {
template <typename, typename...>
--
2.17.2 (Apple Git-113)
From 9c9075760717f51df205bc16623abee398131651 Mon Sep 17 00:00:00 2001
From: Orlando Ding <xiandao.airs@gmail.com>
Date: Thu, 22 Feb 2024 14:20:26 -0800
Subject: [PATCH 5/8] orlando - for fixing the issue of pocketfft invalid url
---
migration_note.md | 4 ++--
third_party/pocketfft | 2 +-
torch/csrc/distributed/c10d/init.cpp | 12 ++++++------
torch/csrc/distributed/rpc/init.cpp | 8 ++++----
4 files changed, 13 insertions(+), 13 deletions(-)
diff --git a/migration_note.md b/migration_note.md
index 4ea0691f13..6907bf5c79 100644
--- a/migration_note.md
+++ b/migration_note.md
@@ -5,9 +5,9 @@ Preparation of building library:
```bash
export CXXFLAGS=-D_LIBCPP_DISABLE_AVAILABILITY
export CMAKE_PREFIX_PATH=${CONDA_PREFIX:-"$(dirname $(which conda))/../"}
-MAGMA_HOME="/usr/local/lib/magma2.6.1-cu101" MACOSX_DEPLOYMENT_TARGET=10.9 CC=clang CXX=clang++ USE_LIBUV=1 USE_DISTRIBUTED=ON USE_MPI=ON USE_TENSORPIPE=ON USE_GLOO=ON USE_CUDA_MPI=ON python setup.py clean # prepare
+MAGMA_HOME="/usr/local/lib/magma2.6.1-cu101" MACOSX_DEPLOYMENT_TARGET=10.9 CC=clang CXX=clang++ USE_LIBUV=1 USE_DISTRIBUTED=ON USE_MPI=ON USE_TENSORPIPE=ON USE_GLOO=ON USE_CUDA_MPI=ON python setup.py clean
MAGMA_HOME="/usr/local/lib/magma2.6.1-cu101" MACOSX_DEPLOYMENT_TARGET=10.9 CC=clang CXX=clang++ CMAKE_BUILD_TYPE=1 USE_LIBUV=1 USE_DISTRIBUTED=ON USE_MPI=ON USE_TENSORPIPE=ON USE_GLOO=ON USE_CUDA_MPI=ON python setup.py bdist_wheel
-MAGMA_HOME="/usr/local/lib/magma2.6.1-cu101" MACOSX_DEPLOYMENT_TARGET=10.9 CC=clang CXX=clang++ CMAKE_BUILD_TYPE=1 USE_LIBUV=1 USE_CUSPARSELT=1 USE_DISTRIBUTED=ON USE_MPI=ON USE_TENSORPIPE=ON USE_GLOO=ON USE_CUDA_MPI=ON python setup.py bdist_wheel
+MAGMA_HOME="/usr/local/lib/magma2.6.1-cu101" MACOSX_DEPLOYMENT_TARGET=10.9 CC=clang CXX=clang++ CMAKE_BUILD_TYPE=1 DEBUG=1 USE_LIBUV=1 USE_CUSPARSELT=1 USE_DISTRIBUTED=ON USE_MPI=ON USE_TENSORPIPE=ON USE_GLOO=ON USE_CUDA_MPI=ON python setup.py bdist_wheel # current running
MAGMA_HOME="/usr/local/lib/magma2.6.1-cu101" MACOSX_DEPLOYMENT_TARGET=10.9 CC=clang CXX=clang++ USE_LIBUV=1 USE_CUSPARSELT=1 USE_DISTRIBUTED=ON USE_MPI=OFF USE_TENSORPIPE=ON USE_GLOO=ON USE_CUDA_MPI=ON python setup.py develop
```
diff --git a/third_party/pocketfft b/third_party/pocketfft
index ad1eec0fb2..81d171a6d5 160000
--- a/third_party/pocketfft
+++ b/third_party/pocketfft
@@ -1 +1 @@
-Subproject commit ad1eec0fb2f8bfb28e287c559a29bc16d059abf0
+Subproject commit 81d171a6d5562e3aaa2c73489b70f564c633ff81
diff --git a/torch/csrc/distributed/c10d/init.cpp b/torch/csrc/distributed/c10d/init.cpp
index 0206be063d..a9662a975d 100644
--- a/torch/csrc/distributed/c10d/init.cpp
+++ b/torch/csrc/distributed/c10d/init.cpp
@@ -1726,8 +1726,8 @@ Arguments:
},
py::arg("device"),
py::arg("backend_type"),
- // py::arg("backend") = c10::optional<c10::intrusive_ptr<::c10d::Backend>>(),
- py::arg("backend"),
+ py::arg("backend") = c10::optional<c10::intrusive_ptr<::c10d::Backend>>(),
+ // py::arg("backend"),
py::call_guard<py::gil_scoped_release>())
.def(
"_get_backend",
@@ -2589,8 +2589,8 @@ Example::
py::arg("bucket_size"),
py::arg("expect_sparse_gradient") = std::vector<bool>(),
py::arg("tensor_indices") = std::vector<int64_t>(),
- // py::arg("logger") = c10::optional<std::shared_ptr<::c10d::Logger>>{},
- py::arg("logger"),
+ py::arg("logger") = c10::optional<std::shared_ptr<::c10d::Logger>>{},
+ // py::arg("logger"),
py::call_guard<py::gil_scoped_release>());
module.def(
@@ -2608,8 +2608,8 @@ Example::
},
py::arg("process_group"),
py::arg("params"),
- // py::arg("logger") = c10::optional<std::shared_ptr<::c10d::Logger>>{},
- py::arg("logger"),
+ py::arg("logger") = c10::optional<std::shared_ptr<::c10d::Logger>>{},
+ // py::arg("logger"),
py::call_guard<py::gil_scoped_release>());
module.def(
diff --git a/torch/csrc/distributed/rpc/init.cpp b/torch/csrc/distributed/rpc/init.cpp
index 69ac2a13ce..aa8f0d7a87 100644
--- a/torch/csrc/distributed/rpc/init.cpp
+++ b/torch/csrc/distributed/rpc/init.cpp
@@ -544,10 +544,10 @@ PyObject* rpc_init(PyObject* _unused, PyObject* noargs) {
std::unordered_map<std::string, DeviceMap>,
std::vector<c10::Device>>(),
py::arg("num_worker_threads") = kDefaultNumWorkerThreads,
- // py::arg("_transports") = optional<std::vector<std::string>>(),
- py::arg("_transports"),
- // py::arg("_channels") = optional<std::vector<std::string>>(),
- py::arg("_channels"),
+ py::arg("_transports") = optional<std::vector<std::string>>(),
+ // py::arg("_transports"),
+ py::arg("_channels") = optional<std::vector<std::string>>(),
+ // py::arg("_channels"),
py::arg("rpc_timeout") = kDefaultRpcTimeoutSeconds,
py::arg("init_method") = kDefaultInitMethod,
py::arg("device_maps") = std::unordered_map<std::string, DeviceMap>(),
--
2.17.2 (Apple Git-113)
From 43ad1043be66454df9c5fc9eb3ce7679a2ee8baa Mon Sep 17 00:00:00 2001
From: Orlando Ding <xiandao.airs@gmail.com>
Date: Sat, 24 Feb 2024 22:58:08 -0800
Subject: [PATCH 6/8] orlando - for fixing issues of init.cpp and avoid issue
---
c10/util/Optional.h | 7 +++----
caffe2/serialize/inline_container.cc | 4 +++-
caffe2/serialize/inline_container.h | 4 ++--
caffe2/serialize/inline_container_test.cc | 4 ++--
torch/csrc/distributed/c10d/init.cpp | 5 ++++-
torch/csrc/distributed/rpc/init.cpp | 10 +++++-----
6 files changed, 19 insertions(+), 15 deletions(-)
diff --git a/c10/util/Optional.h b/c10/util/Optional.h
index 23eac9e0ec..e2ae1f81e5 100644
--- a/c10/util/Optional.h
+++ b/c10/util/Optional.h
@@ -1,7 +1,7 @@
#ifndef C10_UTIL_OPTIONAL_H_
#define C10_UTIL_OPTIONAL_H_
-// #if defined(__APPLE__) && defined(__MACH__)
+#if defined(__APPLE__) && defined(__MACH__)
#include <c10/macros/Macros.h>
#include <c10/util/ArrayRef.h>
@@ -1235,7 +1235,7 @@ struct hash<c10::optional<T&>> {
C10_CLANG_DIAGNOSTIC_POP()
-#if !defined(__APPLE__) && !defined(__MACH__)
+#else // !defined(__APPLE__) && !defined(__MACH__)
#include <optional>
#include <type_traits>
@@ -1250,7 +1250,6 @@ namespace c10 {
using std::bad_optional_access;
using std::in_place;
using std::in_place_t;
-using std::make_optional;
using std::nullopt;
using std::nullopt_t;
using std::optional;
@@ -1281,6 +1280,6 @@ constexpr T value_or_else(optional<T>&& v, F&& func) {
}
} // namespace c10
-#endif // !defined(__APPLE__) && !defined(__MACH__)
+#endif // defined(__APPLE__) && defined(__MACH__)
#endif // C10_UTIL_OPTIONAL_H_
diff --git a/caffe2/serialize/inline_container.cc b/caffe2/serialize/inline_container.cc
index 533fd42a04..20ea4e6923 100644
--- a/caffe2/serialize/inline_container.cc
+++ b/caffe2/serialize/inline_container.cc
@@ -34,12 +34,14 @@ constexpr c10::string_view kDebugPklSuffix(".debug_pkl");
struct MzZipReaderIterWrapper {
MzZipReaderIterWrapper(mz_zip_reader_extract_iter_state* iter) : impl(iter) {}
mz_zip_reader_extract_iter_state* impl;
+ // Disable the move constructor
+ MzZipReaderIterWrapper(MzZipReaderIterWrapper&& other) = delete;
};
ChunkRecordIterator::ChunkRecordIterator(
size_t recordSize,
size_t chunkSize,
- std::unique_ptr<MzZipReaderIterWrapper> iter)
+ std::shared_ptr<MzZipReaderIterWrapper> iter)
: recordSize_(recordSize),
chunkSize_(chunkSize),
offset_(0),
diff --git a/caffe2/serialize/inline_container.h b/caffe2/serialize/inline_container.h
index aa0cb8e043..d4b98b41a6 100644
--- a/caffe2/serialize/inline_container.h
+++ b/caffe2/serialize/inline_container.h
@@ -109,12 +109,12 @@ class TORCH_API ChunkRecordIterator {
ChunkRecordIterator(
size_t recordSize,
size_t chunkSize,
- std::unique_ptr<MzZipReaderIterWrapper> iter);
+ std::shared_ptr<MzZipReaderIterWrapper> iter);
const size_t recordSize_;
const size_t chunkSize_;
size_t offset_;
- std::unique_ptr<MzZipReaderIterWrapper> iter_;
+ std::shared_ptr<MzZipReaderIterWrapper> iter_;
friend class PyTorchStreamReader;
};
diff --git a/caffe2/serialize/inline_container_test.cc b/caffe2/serialize/inline_container_test.cc
index 4fe2c236e0..2e597a01fc 100644
--- a/caffe2/serialize/inline_container_test.cc
+++ b/caffe2/serialize/inline_container_test.cc
@@ -464,7 +464,7 @@ TEST_P(ChunkRecordIteratorTest, ChunkRead) {
LOG(INFO) << "Testing chunk size " << chunkSize;
PyTorchStreamReader reader(fileName);
ASSERT_TRUE(reader.hasRecord(recordName));
- #if !defined(__APPLE__) && !defined(__MACH__)
+ // #if !defined(__APPLE__) && !defined(__MACH__)
//see: to avoid "error: call to implicitly-deleted copy constructor of 'caffe2::serialize::ChunkRecordIterator'"
caffe2::serialize::ChunkRecordIterator chunkIterator = reader.createChunkReaderIter(
recordName, tensorDataSizeInBytes, chunkSize);
@@ -476,7 +476,7 @@ TEST_P(ChunkRecordIteratorTest, ChunkRead) {
totalReadSize += readSize;
}
ASSERT_EQ(totalReadSize, tensorDataSizeInBytes);
- #endif
+ // #endif
// clean up
remove(fileName);
}
diff --git a/torch/csrc/distributed/c10d/init.cpp b/torch/csrc/distributed/c10d/init.cpp
index a9662a975d..d81f7c2087 100644
--- a/torch/csrc/distributed/c10d/init.cpp
+++ b/torch/csrc/distributed/c10d/init.cpp
@@ -107,6 +107,9 @@ namespace c10d {
namespace {
+using ::c10::in_place;
+using ::c10::in_place_t;
+
template <typename T>
using shared_ptr_class_ = py::class_<T, std::shared_ptr<T>>;
@@ -1726,8 +1729,8 @@ Arguments:
},
py::arg("device"),
py::arg("backend_type"),
- py::arg("backend") = c10::optional<c10::intrusive_ptr<::c10d::Backend>>(),
// py::arg("backend"),
+ py::arg("backend") = c10::optional<c10::intrusive_ptr<::c10d::Backend>>(),
py::call_guard<py::gil_scoped_release>())
.def(
"_get_backend",
diff --git a/torch/csrc/distributed/rpc/init.cpp b/torch/csrc/distributed/rpc/init.cpp
index aa8f0d7a87..b90fe6c387 100644
--- a/torch/csrc/distributed/rpc/init.cpp
+++ b/torch/csrc/distributed/rpc/init.cpp
@@ -537,16 +537,16 @@ PyObject* rpc_init(PyObject* _unused, PyObject* noargs) {
.def(
py::init<
int,
- optional<std::vector<std::string>>,
- optional<std::vector<std::string>>,
+ c10::optional<std::vector<std::string>>,
+ c10::optional<std::vector<std::string>>,
float,
std::string,
std::unordered_map<std::string, DeviceMap>,
std::vector<c10::Device>>(),
py::arg("num_worker_threads") = kDefaultNumWorkerThreads,
- py::arg("_transports") = optional<std::vector<std::string>>(),
+ py::arg("_transports") = c10::optional<std::vector<std::string>>(),
// py::arg("_transports"),
- py::arg("_channels") = optional<std::vector<std::string>>(),
+ py::arg("_channels") = c10::optional<std::vector<std::string>>(),
// py::arg("_channels"),
py::arg("rpc_timeout") = kDefaultRpcTimeoutSeconds,
py::arg("init_method") = kDefaultInitMethod,
@@ -579,7 +579,7 @@ PyObject* rpc_init(PyObject* _unused, PyObject* noargs) {
[](const c10::intrusive_ptr<::c10d::Store>& store,
std::string selfName,
worker_id_t selfId,
- optional<int> worldSize,
+ c10::optional<int> worldSize,
TensorPipeRpcBackendOptions opts,
std::unordered_map<std::string, DeviceMap> reverseDeviceMaps,
std::vector<c10::Device> devices) {
--
2.17.2 (Apple Git-113)
From 3322cd3fa1d8189275f6e4b96fdee2526f9358d5 Mon Sep 17 00:00:00 2001
From: Orlando Ding <xiandao.airs@gmail.com>
Date: Sun, 25 Feb 2024 22:24:58 -0800
Subject: [PATCH 7/8] orlando - for updates of torch init.cpp and library.h
---
aten/src/ATen/functorch/Interpreter.h | 2 +-
aten/src/ATen/native/LinearAlgebra.cpp | 1 +
c10/util/Exception.h | 4 ---
migration_note.md | 10 ++++++-
.../include/torch/nn/functional/upsampling.h | 1 -
torch/csrc/api/include/torch/nn/init.h | 17 ------------
.../csrc/api/include/torch/nn/modules/conv.h | 1 -
.../torch/nn/options/transformerlayer.h | 8 +-----
.../api/include/torch/nn/options/upsampling.h | 26 +++----------------
torch/csrc/api/src/nn/modules/conv.cpp | 1 -
torch/csrc/autograd/profiler_kineto.cpp | 1 -
torch/csrc/distributed/c10d/init.cpp | 6 ++---
torch/csrc/distributed/rpc/init.cpp | 8 +++---
torch/csrc/profiler/python/init.cpp | 4 ---
torch/csrc/profiler/util.h | 2 --
torch/csrc/utils/pybind.h | 14 ++++++++++
torch/library.h | 1 +
17 files changed, 38 insertions(+), 69 deletions(-)
diff --git a/aten/src/ATen/functorch/Interpreter.h b/aten/src/ATen/functorch/Interpreter.h
index 11cb41ee79..c4ccbee17c 100644
--- a/aten/src/ATen/functorch/Interpreter.h
+++ b/aten/src/ATen/functorch/Interpreter.h
@@ -9,8 +9,8 @@
#include <c10/util/variant.h>
namespace std {
using ::c10::variant;