845f42a3
fix(midout/naive/warp_perspective): fix Template Functions instantiation by
2020-07-06 20:50:30 +0800
5fb07c99
fix(dnn/x86): fix cmake error for build x86 gtest by
2020-07-06 15:55:36 +0800
763b57ad
fix(dnn/cuda): fix INTMAX overflow in warp_perspective_cuda by
2020-07-03 11:43:49 +0800
2e6e570d
feat(dnn/fallback): add armv7 im2col mk4-dot int8 and nchw44 float 3x3 s2 fuse packb speed up about 10% by
2020-06-18 17:59:24 +0800
a601fda8
(tag: v0.5.1, release-0.5)
chore(release): bump version by
2020-07-06 20:57:27 +0800
b90c1540
fix(dnn/naive): fix midout for pooling by
2020-07-04 13:44:35 +0800
df47637d
fix(dnn/naive): fix midout for relayout_format by
2020-07-04 09:44:24 +0800
0d8b9136
fix(mge/functional): reshape bias to (1, out_features) in linear by
2020-07-03 16:09:44 +0800
f60ab501
fix(mgb/opt): nchw to nchw4 pass suppport ic less than 4 by
2020-07-03 12:15:59 +0800
8ec09922
fix(dnn): fix Image2DPack4TensorFormat check by
2020-07-03 12:37:27 +0800
28d85838
feat(dnn): add relayout_format for nchw to nchw4 and ic <=4 by
2020-07-03 01:00:10 +0800
3a53872f
fix(dnn/native): also fix native logic by
2020-07-03 11:18:35 +0800
43b42a65
fix(dnn/cuda): fix indexing logic in psroi_pooling by
2020-07-03 10:45:54 +0800
3f2770fe
feat(mge/module): add quantize dtype load support for module load_state_dict by
2020-07-02 17:55:58 +0800
fdd14e09
feat(mgb/opt): add nchw->nchw4 for tensorrt replace pass by
2020-06-30 15:46:23 +0800
3eb29f5e
feat(mgb/opt): add nchw->nchw4 in tensorcore pass by
2020-06-29 12:19:18 +0800
52cb4b39
fix(mgb/gopt): fix convert format nchw->nchw4 pass by
2020-06-24 19:29:54 +0800
90d5895e
fix(mgb/gopt): remove redundant reshape in nchw->nchw4 pass by
2020-06-29 16:18:02 +0800
0d12ae80
feat(opr/standalone): import NMSKeep from MegSkull contrib by
2020-06-16 13:44:12 +0800
129fa70c
fix(mgb/serialization): fix multiple graph load error by
2020-06-30 11:22:50 +0800
4755400e
refactor(mge/quantization): add `narrow_range` to control quant dtype's lower bound by
2020-06-22 10:46:23 +0800
c8a9094b
fix(mge/data/dataloader): add refcount in _PlasmaStoreManager by
2020-06-29 11:36:45 +0800
457a1e01
refactor(imperative): initial merge of xxx and megengine by
2020-07-03 16:14:51 +0800
a808f807
feat(externcopr/nvidaopticalflow): fix output and fix cu context issue on diff thread, nvof loader outptr share with mgb ptr, upgrade lastest nvof sdk by
2020-06-24 21:36:01 +0800
1adb262a
fix(dnn/naive): fix midout for pooling by
2020-07-04 13:44:35 +0800
32d7f25b
fix(dnn/naive): fix midout for relayout_format by
2020-07-04 09:44:24 +0800
f856b170
fix(mge/functional): reshape bias to (1, out_features) in linear by
2020-07-03 16:09:44 +0800
486cbdea
fix(mgb/opt): nchw to nchw4 pass suppport ic less than 4 by
2020-07-03 12:15:59 +0800
1c3d1f86
fix(dnn): fix Image2DPack4TensorFormat check by
2020-07-03 12:37:27 +0800
7886ff9a
feat(dnn): add relayout_format for nchw to nchw4 and ic <=4 by
2020-07-03 01:00:10 +0800
1630a635
fix(dnn/native): also fix native logic by
2020-07-03 11:18:35 +0800
6078187e
fix(dnn/cuda): fix indexing logic in psroi_pooling by
2020-07-03 10:45:54 +0800
32d91d5e
fix(dnn/arm_common): fix and optimize workspacebundle copy when algo compute by
2020-07-01 14:42:38 +0800
caf77d00
feat(mge/module): add quantize dtype load support for module load_state_dict by
2020-07-02 17:55:58 +0800
dedb7a3f
feat(dnn/cuda): add cuda remap by
2020-06-23 21:33:20 +0800
cb5c5fd5
feat(mgb/opt): add nchw->nchw4 for tensorrt replace pass by
2020-06-30 15:46:23 +0800
2e70cf1d
feat(mgb/opt): add nchw->nchw4 in tensorcore pass by
2020-06-29 12:19:18 +0800
1e8337f1
fix(mgb/gopt): remove redundant reshape in nchw->nchw4 pass by
2020-06-29 16:18:02 +0800
946a340c
feat(ci/midout): opt midout and add midout ci by
2020-06-16 17:53:05 +0800
ef437f69
fix(mge/data/dataloader): add refcount in _PlasmaStoreManager by
2020-06-29 11:36:45 +0800
90aa65d9
fix(mgb/serialization): fix multiple graph load error by
2020-06-30 11:22:50 +0800
2b3a88d0
feat(mge/distributed): add parameter replica_mode by
2020-06-23 18:54:23 +0800
44c381b6
Revert "feat(dnn/naive): workspacebundle support 2D" by
2020-06-29 10:43:24 +0800
cdbe44f8
feat(dnn): add gemv supports in conv1x1 with format NCHW by
2020-06-18 19:40:08 +0800
6972fc7d
feat(opr/standalone): import NMSKeep from MegSkull contrib by
2020-06-16 13:44:12 +0800
5fa444c3
refactor(mge/quantization): add `narrow_range` to control quant dtype's lower bound by
2020-06-22 10:46:23 +0800
3b6285d7
ci(env): add nccl env by
2020-06-23 19:13:04 +0800
80ecabe8
feat(dnn/arm_common): add nchw_nchw44 aarch64 int8 3x3s2 7x7s2 asm by
2020-06-24 16:14:44 +0800
6d367454
feat(mge/opr-mm): add param local_grad for collective_comm opr by
2020-06-18 18:03:36 +0800
0ccb965c
fix(mgb/gopt): fix convert format nchw->nchw4 pass by
2020-06-24 19:29:54 +0800
1c2a525e
feat(imperative): set up imperative ci by
2020-06-25 16:58:52 +0800
7ae05ac8
feat(imperative): merge common c++ code to megbrain by
2020-06-24 21:46:04 +0800
9e904f68
fix(dnn): fix can not inline small function with GCC compiler by
2020-06-19 19:01:10 +0800
4ace67ff
test(mge/distributed): check gpu num for multi gpu test by
2020-06-24 14:13:06 +0800
9d5c5c07
feat(dnn/naive): workspacebundle support 2D by
2020-06-19 21:51:22 +0800
f268e0f8
fix(dnn/x86): fix x86 compile in gcc 5.4 by
2020-06-22 19:55:27 +0800
5d7fdd47
fix(brain): fix winodws build by
2020-06-15 17:57:27 +0800
c36e5f8d
Update Help-wanted Issue.md by
2020-06-24 15:18:30 +0800
8fdb3ea4
Create Help-wanted Issue.md by
2020-06-22 12:23:50 +0800
b532284c
chore(version): bump version. by
2020-06-19 11:37:52 +0800
8e9fa80c
feat(dnn/fallback): add matmul description for im2col by
2020-06-16 22:08:48 +0800
af3de7e1
fix(jit): fix jit doc and add NCHW44_DOT by
2020-06-19 14:41:21 +0800
7b8a9951
feat(mge/utils): export _internal.plugin to mge.utils by
2020-06-19 13:59:27 +0800
4e634d87
fix(load_and_run): fix dump_with_testcase_mge.py with import megbrain by
2020-06-19 13:42:05 +0800
318f6d75
fix(mge): fix some warnings by
2020-06-19 00:18:05 +0800
b9b53118
fix(load_and_run): fix load_and_run with --input which ignore iters by
2020-06-18 14:15:46 +0800
69fe5ab3
feat(dnn/cuda): add conv2d-sass-kernel by
2020-05-12 11:41:14 +0800
1a70842e
fix(serialization): fix model compatibility by
2020-06-17 17:01:16 +0800
6ed791bc
feat(mge/quant): add TQT quant method by
2020-06-05 02:37:03 +0800
328fb36f
feat(mgb/opr-mm): add Scatter, Gather, AllToAll oprs by
2020-06-16 20:14:08 +0800
3f51a6a0
build(third_party): Update megray version by
2020-06-23 15:23:39 +0800
05682707
(tag: v0.5.0)
chore(version): bump version. by
2020-06-19 11:37:52 +0800
5257991e
fix(jit): fix jit doc and add NCHW44_DOT by
2020-06-19 14:41:21 +0800
cdf25c4a
feat(mge/utils): export _internal.plugin to mge.utils by
2020-06-19 13:59:27 +0800
f1dd8679
fix(load_and_run): fix dump_with_testcase_mge.py with import megbrain by
2020-06-19 13:42:05 +0800
be205727
fix(mge): fix some warnings by
2020-06-19 00:18:05 +0800
5b87e8a8
fix(load_and_run): fix load_and_run with --input which ignore iters by
2020-06-18 14:15:46 +0800
80af2f93
fix(serialization): fix model compatibility by
2020-06-17 17:01:16 +0800
fda9599a
feat(mge/quant): add TQT quant method by
2020-06-05 02:37:03 +0800
285d70cb
fix(build): find_package(Threads) if not already done by
2020-06-17 13:59:06 +0800
786afef4
feat(build): install CMake config module and pkg-config descriptor by
2020-04-27 19:17:13 +0800
4d35397b
fix(dnn/fallback): fix conv1x1/im2col usable and fuse-conv-bias get fp32xfp32-->qint8 error by
2020-06-16 15:43:44 +0800
12dc36a6
feat(mgb/gopt): add interface to reproducible by
2020-06-10 23:21:04 +0800
cc4e1dfd
feat(mgb/compnode): expose mem_status and try_coalesce_all_free_memory to python by
2020-06-04 17:08:00 +0800
7bcead75
ci(docker): add build-arg option to setup yum and pip mirrors by
2020-06-10 18:36:42 +0800
6b2760dd
feat(dnn/fallback): add float32 nchw44 fuse packb 3x3 s2 by
2020-06-11 18:46:17 +0800
7aeb4f6c
fix(mge/optimizer): use static key to avoid mem leak by
2020-06-15 14:49:47 +0800
7a0c7ef4
feat(mge/module): add module for extern-c-opr by
2020-06-15 14:12:09 +0800
09d2b7c3
fix(core): make the semantics of instance id clear and correct by
2020-06-09 19:16:57 +0800
2b4b4d66
feat(dnn/fallback): add aarch64 mk4 dot 3x3 s1 fuse packb by
2020-06-12 16:43:07 +0800
a1677d7a
feat(dnn/arm_common): add fp32 gevm by
2020-06-11 15:21:18 +0800
5d950063
feat(dnn): refactor dot gemv for both aarch64 and aarch32 by
2020-06-10 14:35:28 +0800
53c288a3
fix(dnn/cuda): fix topk grid oversize by
2020-05-21 14:06:21 +0800
124767b4
fix(dnn/fallback): fix mk4_dot test after remove mk4_dot_8x6x4 matmul by
2020-06-12 18:30:04 +0800
34659c2e
fix(mgb/dnn): remove armv7 matmul mk4dot block 8x6 by
2020-06-12 15:34:56 +0800
48ac1e1a
feat(dnn/fallback): delete nopack onlypacka noneed datatype,and add im2co and conv1x1 mk4_dot support by
2020-05-29 10:30:35 +0800
3117bfb7
fix(dnn/arm): nchw44 direct int8 support 8832 by
2020-06-08 19:33:21 +0800
4e0c9ad3
feat(mgb/external): extern-c-opr dumper and loader for MACE by
2020-06-08 06:35:41 +0000
ca52a93e
fix(mge/quant): fix init value of histogram observer by
2020-06-12 16:11:08 +0800
9f352b1c
feat(megbrain/dnn): add indexing remap int32 for naive and cuda by
2020-06-05 15:14:44 +0800