a29f1c8c
fix(cmake/midout): fix cmake midout by
2021-10-09 13:12:41 +0800
369c2ccc
style(all): reformat c++ code by
2021-10-12 10:38:41 +0800
bfb30dcb
chore(format): fix compile bugs after code format by
2021-08-20 11:05:34 +0800
eeccf2bc
ci(check): add clang-format in check stage by
2021-08-19 18:53:08 +0800
21086965
feat(mge/imperative): add local response normalization by
2021-09-24 18:14:29 +0800
d2184af3
feat(dnn/src/x86/matmul): add matmul_6x16 for x86 by
2021-09-05 09:11:52 +0800
ca4c93de
fix(mge): fix layer norm amp bug by
2021-09-28 19:24:07 +0800
c674bf0e
feat(debug): add gdb-scripts by
2021-09-29 15:01:30 +0800
dc42417f
fix(ci/build): opt macos ci and move bazel to 4.2.1 by
2021-09-30 15:02:31 +0800
dbb3232c
fix(sdk): fix error caused by incorrect parsing u16/s16 inputs in lar by
2021-09-24 19:09:15 +0800
177dec94
feat(mgb/opr): add bgr2gray mode for cvtcolor opr by
2021-09-20 19:43:59 +0800
000517c6
fix(grad): stop using exception in grad_override by
2021-09-14 18:00:59 +0800
0bb058c1
fix(mgb/lite): fix LITE_ASSERT bug while build with no exception by
2021-09-24 16:47:13 +0800
d02b0a08
feat(lite): fix typo by
2021-09-28 18:08:42 +0800
cebda6ff
feat(mge/imperative): add ctc loss by
2021-09-24 13:33:08 +0800
f5cb21ed
fix(mgb/opr): add non finite check by
2021-09-14 10:54:25 +0800
bde5cf35
feat(dnn): add resize linear for arm by
2021-09-19 21:09:57 +0800
b6142bee
feat(imperative): support tensor with uint16 date type by
2021-09-24 19:05:57 +0800
3a7bc37f
fix(lite): fix lite type define error by
2021-09-18 12:36:00 +0800
8cb20186
fix(mgb): fix fastrun cache serialization method by
2021-09-23 14:53:49 +0800
88c1eedb
feat(mgb/gopt): enable reduce for nchw44 by
2021-09-17 17:19:04 +0800
563239d3
feat(dnn): add arm_common nchw44 cwconv3x3s1p1 and cwconv5x5s1p2 by
2021-09-02 15:42:01 +0800
3344b580
feat(dnn): add elemwise for nchw88+fp16 by
2021-08-25 22:06:02 +0800
682c74df
feat(dnn): add direct nchw88 fp16 conv by
2021-09-08 13:57:09 +0800
fca19535
feat(gopt): add nhwc fuse conv typecvt optpass by
2021-09-14 18:55:42 +0800
2fc73585
Revert "feat(dnn/apicache): add generic apicache" by
2021-09-15 14:28:33 +0800
de363c04
Revert "perf(cuda/conv): cache serval cudnn api" by
2021-09-15 14:28:26 +0800
729ee649
Revert "fix(api_cache): lock api cache for thread safety" by
2021-09-15 14:17:54 +0800
64c922c4
Revert "fix(api_cache): fix serialization for conv_desc" by
2021-09-15 14:17:35 +0800
4e95c136
feat(sgd): sgd supports nesterov momentum by
2021-09-02 19:35:58 +0800
ff431e72
fix(cmake/windows): use full install path on Windows build by
2021-09-23 18:14:07 +0800
dc28a434
fix(mgb/bn): fix empty tensor input problem and other minor problems by
2021-09-23 11:12:22 +0800
270f1aa2
feat(mgb/serialization): add Accessor for OprLoader to fix BN output compatibility by
2021-09-15 17:01:05 +0800
c0ccd0ea
feat(mge/bn): add NHWC support for bn by
2021-08-23 16:51:39 +0800
3d3666b6
test(dnn/bn): add compatible configs for NHWC BN by
2021-08-23 16:51:22 +0800
b3e54ead
feat(dnn/bn): use new cudnn BN kernel to support NHWC by
2021-08-17 17:45:50 +0800
6b863cc5
feat(imperative): add pixel_shuffle opr by
2021-09-22 13:08:35 +0800
5e345043
fix(cmake/bazel/clang): remove finite-math-only opt from Ofast, for keep same build between gcc and clang by
2021-09-18 17:13:58 +0800
cccddbc0
docs(lite): opt pylite md by
2021-09-22 16:14:39 +0800
b982be56
feat(mge/imperative): add permutation support for the tensor by
2021-09-16 11:00:15 +0800
3977b7aa
feat(mgb/shuffle): add shuffle opr by
2021-07-21 10:16:00 +0800
17371e79
fix(dnn/reduce): fix reduce_mean o16c32 is incorrect for large tensor by
2021-09-04 13:44:56 +0800
884a07ff
fix(test/random): set a random seed for random unit test by
2021-09-18 23:33:25 +0800
d540d38b
(revert-211-master)
Revert "feat(imperative): add pixel_shuffle opr" by
2021-09-24 11:35:36 +0800
d17cd60d
feat(imperative): add pixel_shuffle opr by
2021-09-22 13:08:35 +0800
d7cc4628
perf(gopt): opt concat for OpenCL by
2021-09-16 17:50:07 +0800
3f0bb47a
ci(imperative): add test requires by
2021-09-16 16:58:51 +0800
34be1d6e
feat(mgb): dump static memory info by tensorboard by
2021-09-06 14:35:20 +0800
11a383bc
feat(mgb): add tensorboard tool c++ layer interface and python parsing by
2021-09-06 14:31:46 +0800
8084e4e2
feat(mgb): add tensorboard tool python layer interface by
2021-09-06 14:24:32 +0800
97b1b777
feat(mgb): add megbrain layer norm opr with subgraph by
2021-09-16 20:51:33 +0800
eca6e1d9
(tag: v1.6.0, release-1.6)
fix(ci): fixes for ci by
2021-09-14 13:17:57 +0800
19d7412a
refactor(mgb/gopt): reorganize code of global layout transform by
2021-09-10 16:10:04 +0800
8ef12bdf
feat(mgb/gopt): add user inferface for global layout transform by
2021-09-09 16:18:53 +0800
a3cd3fc7
test(mgb/gopt): add testcase for global layout transform by
2021-09-09 16:18:53 +0800
af576e9a
fix(mgb/gopt): fix auto padding for nhwc layout by
2021-08-24 19:40:32 +0800
af828ca9
feat(mgb/gopt): fix folding conv dimshuffle pass by
2021-09-07 19:26:20 +0800
c67c4b7d
feat(mgb/gopt): add layout transform pass by
2021-09-07 19:10:46 +0800
2ec7c167
feat(mgb/gopt): profiler support opr filter and var node filter by
2021-09-07 19:25:51 +0800
50ea5ae8
feat(mgb/gopt): add dynamic programming solver by
2021-09-07 19:10:14 +0800
c14e5719
feat(mgb/gopt): add profile impl for global layout transform pass by
2021-08-02 10:43:00 +0800
9c0a17d0
feat(mgb/gopt): add auto aligned reformat impls by
2021-07-28 15:36:35 +0800
2ed76b16
feat(mgb/gopt): add graph dumper for graph partition by
2021-07-23 16:52:53 +0800
76b28408
feat(mgb/gopt): add subgraph extractor by
2021-07-22 17:16:59 +0800
8a3eb05a
refactor(mgb/gopt): refactor tensor reformat opt pass by
2021-07-21 17:10:46 +0800
c33126ab
feat(mgb/gopt): add reformat manager by
2021-07-16 18:54:14 +0800
65d554ed
refactor(dtr): improve dtr in computing graph by
2021-09-03 17:33:58 +0800
4f28e146
fix(dnn): fix compatibility broken of convolution format by
2021-09-18 09:53:48 +0800
f642b05e
test(mge): update traced_module unit test by
2021-09-16 14:17:01 +0800
fb20cb36
docs(mge/traced_module): update traced_module api doc by
2021-09-15 20:28:53 +0800
c7a8d945
fix(mge/traced_module): let graph record total_id by
2021-09-16 14:15:27 +0800
8b40f577
feat(mgb/dnn): add conv1x1 algo for matrix mul by
2021-08-12 12:14:34 +0800
4de62ad6
docs(mge/device): fix docstring of get_cuda_compute_capability by
2021-09-15 17:52:44 +0800
fb49a283
refactor(mgb/dnn): refactor enum used in serializing by
2021-09-07 17:11:01 +0800
d69b5903
feat(dnn): add an get_all_algorithms_safe interface by
2021-09-06 17:54:48 +0800
103d7f33
refactor(dnn/rocm): update hip license header by
2021-04-30 14:17:35 +0800
5aa52d38
feat(dnn/rocm): add adaptive pooling opr by
2021-08-19 15:13:05 +0800
83cf4ee6
refactor(dnn/rocm): remove some useless includes by
2021-08-25 12:33:12 +0800
323a4642
feat(dnn/rocm): add topk opr by
2021-08-16 14:43:57 +0800
f4784f4a
feat(dnn/rocm): add argsort opr by
2021-08-18 10:17:36 +0800
6082c353
feat(dnn/rocm): support bool in type_cvt and elemwise by
2021-04-30 13:25:41 +0800
0ad377c7
fix(mgb/gopt): add error message when input dtype is not equal to param dtype in BN2Elemwise pass by
2021-09-14 11:05:21 +0800
93f70a95
feat(src): add MGB_CUSTOM_OP macro by
2021-09-13 15:29:58 +0800
b3e958d0
fix(src): fix the warnings and copy.bara.sky in custom op by
2021-09-09 13:24:23 +0800
cdb692d2
refactor(imperative): add TODO tag for some functions by
2021-09-07 15:28:53 +0800
90dd0716
refactor(imperative): modify the python interface of custom op by
2021-09-03 13:46:29 +0800
cbf024bf
feat(imperative): add adaptor between custom op and imperative runtime by
2021-09-01 14:39:21 +0800
39ba3021
feat(src): add adaptor between custom op and graph runtime by
2021-08-13 18:00:39 +0800
c9134717
feat(src/custom): add tensor and tensor attrs in custom namespace by
2021-08-13 17:15:49 +0800
43fb2c4a
feat(opr): let roll support empty IO by
2021-09-14 16:14:26 +0800
b2827cb1
feat(opr): let Dot, MatrixMul and BatchedMatrixMul support empty input by
2021-09-14 14:56:11 +0800
50f73877
feat(opr): let Copy and Identity support empty IO by
2021-09-14 14:54:30 +0800
74cbc10d
feat(opr): let batchnorm support empty IO by
2021-09-09 21:42:48 +0800
8ae61752
feat(opr): let nms support empty IO by
2021-09-09 21:42:10 +0800
1a1748da
feat(opr): let Argsort support empty IO by
2021-09-09 21:41:08 +0800
7234efe1
feat(opr): let random op support empty output by
2021-09-09 21:37:11 +0800
3bc94738
fix(profiler): do not assume clock duration and time_point type by
2021-08-16 20:12:49 +0800
b7e6bd7b
feat(profiler): custom event support device by
2021-08-06 17:47:36 +0800
defbc20e
fix(profiler): fix profiler in dtr by
2021-08-06 14:48:55 +0800
d4c71f92
perf(profiler): reduce profiler host overhead by
2021-07-09 18:17:20 +0800