50d5421a
fix(mgb/zmq): fix unused-result when compiling with c++17 by
2020-09-02 17:13:45 +0800
16324e30
feat(dnn/cuda): add remap backward by
2020-07-07 10:26:13 +0800
46b68568
ci(docker): update cmake to 3.15.2 by
2020-09-04 22:30:23 +0800
46f03b9f
perf(mge/imperative): optimize gpu memory with standalone grad_fn by
2020-08-31 15:41:07 +0800
bd73dabb
fix(dnn/build): add CUDNN_INCLUDE_DIR to the megdnn_test target by
2020-09-02 14:50:38 +0800
14235385
fix(imperative): release gil on sync by
2020-09-02 21:21:37 +0800
1c9430e2
fix(mge/imperative): clear log message by
2020-08-31 19:43:35 +0800
a226f02e
fix(mge/imperative): fix syncbn in symbolic mode by
2020-09-02 10:33:32 +0800
34333593
fix(dnn/arm): fix read invalid data in arm kernel by
2020-09-01 15:45:15 +0800
b9df2db1
build(third_party): update megray commit id by
2020-09-02 15:25:07 +0800
236c7ec8
refactor(imperative): remove unused opdef gen by
2020-09-02 15:18:12 +0800
387867f8
feat(mge/quantization): add cambricon-quantization-example by
2020-08-28 13:54:51 +0800
0067dcf0
test(mge/imperative): add module test to imperative by
2020-08-31 22:16:12 +0800
7fadc16d
refactor(mge/functional): support tensor shape in interpolate and split by
2020-08-31 12:16:01 +0800
968f74ce
chore(mgb): add no_force_inplace option to ComputingGraph by
2020-08-31 16:47:28 +0800
e1fba6ec
test(mge/distributed): add get_device_count_by_fork to fix distributed test skip by
2020-08-31 12:18:37 +0800
60076f47
test(mge/parampack): add parampack related opr test by
2020-08-29 17:51:34 +0800
d1bbbcc6
docs(mge/functional): add doc for copy, param pack split and concat by
2020-08-28 18:06:37 +0800
7e8f7209
refactor(mge/tensor): tensor reduce supports keepdims by
2020-08-31 15:08:15 +0800
59dcd3b7
fix(mgb/build): do not install cutlass by
2020-08-28 19:09:30 +0800
c5e433b5
refactor(mge): cpp license by
2020-08-31 14:12:29 +0800
6e882c1a
feat(whl/imperative): compat for build python whl imperative and legacy runtime by
2020-08-24 12:07:07 +0800
40d18c89
fix(mge/imperative): fix tests when shape is tensor by
2020-08-26 14:21:11 +0800
ea71e5c9
feat(mge): update rocm license for ACKNOWLEDGMENTS by
2020-08-28 22:00:52 +0800
7f857bd4
feat(mgb/rocm): add cmake for rocm and fix compile errors and bn by
2020-08-27 17:18:07 +0800
199eefbd
fix(dnn): generate mode files by
2020-08-27 13:14:56 +0800
95101362
fix(mgb/rocm): remove begin-internal of rocm by
2020-08-27 12:06:30 +0800
92f7cceb
fix(mge/functional): fix cond_take when index is numpy array by
2020-08-28 10:53:39 +0800
f4927db2
feat(mge/functional): support where func by
2020-08-26 13:32:05 +0800
56cb5d6a
fix(mge/functional): int operations(div, exp, pow) by
2020-08-26 15:19:06 +0800
3840c1f4
fix(mge/imperative): use logical reprentation of compnode by
2020-08-28 17:59:03 +0800
227ec04b
fix(mge/test): fix pytest failed issue by
2020-08-28 13:20:07 +0800
b9a774dd
chore(deps): add pybind11 by
2020-09-16 07:50:54 +0800
6b380e89
feat(mge/imperative): run oss test and restore cmake list build items by
2020-08-27 16:59:41 +0800
03808112
feat(dnn/arm_common): add nchw44 8x8x16 stride1 stride2 2x2 3x3 5x5 7x7 directconv by
2020-08-13 10:51:37 +0800
2dbe8194
fix(mge/opr): fix reduction static infer value by
2020-08-27 04:31:48 +0800
c20d4cc6
feat(dnn): fix opt pass nchw44 can not dump resnet by
2020-08-27 10:57:51 +0800
3dbac4f4
feat(mge): add atlas_subgraph module by
2020-08-24 23:36:03 +0800
00ef6772
fix(mgb): remove internal for cambricon and atlas by
2020-08-24 22:43:08 +0800
aeffcd58
feat(dnn/cuda): integrate cutlass nchw32 tensorcore convolution by
2020-08-20 00:20:51 +0800
19568f67
chore(gitmodule/acknowledgments): update mkldnn link and acknowledgments by
2020-08-05 11:50:44 +0800
9e5e32de
fix(dnn): restore opr_param_defs.py by
2020-08-26 11:05:40 +0800
e13fb7b9
fix(mge/quantization): fix apply observer bug which get qparam before call observer by
2020-08-25 14:51:48 +0800
d334b229
feat(imperative): add nms opr wrapper by
2020-08-18 19:55:26 +0800
bca00f2e
fix(dnn): midout at where neccessary in megdnn by
2020-08-12 18:43:03 +0800
105e4450
chroe(.gitignore): add more ignore rules for git by
2020-08-25 15:07:49 +0800
2d81e038
fix(git): make git version check work by
2020-08-25 10:44:32 +0800
b91bfd59
fix(sdk/load-and-run): fix load-and-run error after adding testcases on gpu by
2020-08-24 18:47:04 +0800
bc2dfcbb
refactor(cmake): clear cmakelist file may caused by some commit conflict fix error by
2020-08-22 17:18:43 +0800
a1e67207
feat(dnn): enable bool comparison by
2020-08-20 17:35:20 +0800
8aa34e4a
feat(imperative): add advance indexing with bool by
2020-08-19 15:09:16 +0800
101b58d1
fix(dnn): enable bool input to cond_take by
2020-08-21 14:12:14 +0800
eb0cc59d
feat(windows/python_whl/cuda): make windows MORE HAPPY for build whl with cuda by
2020-08-20 08:33:11 +0800
4a178a8d
feat(windows/cuda/cmake): support cmake cuda build on windows by
2020-08-20 22:30:17 +0800
1915593b
fix(version_depend): add a fake version on dev by
2020-08-12 14:35:58 +0800
6aade133
fix(dnn/fallback): disable im2col/conv1x1/conv1x1_gemv Quantized8Asymm in x86 by
2020-08-21 10:15:34 +0800
42124ab7
fix(mgb/cmake): fix cmake linker flags issue by
2020-08-20 10:28:50 +0800
991286e8
feat(macos/ci): add macos ci by
2020-08-20 11:23:26 +0800
56381f80
fix(dnn/arm): use vcvtq_f32_s32 for all arm code by
2020-08-20 10:59:12 +0800
11732057
fix(gopt): nchw_nchwxx useable and opt pass use nchw_nchwxx_valid by
2020-08-20 10:59:12 +0800
eb18eba8
fix(gopt): fix nchw44 nchw44_dot gopt test by
2020-08-20 10:59:12 +0800
40e79e9d
fix(dnn/x86): fix x86 matrix usable ignore format by
2020-08-20 10:59:12 +0800
eab7ab05
fix(gopt): gen nchw_nchw44 when kernel is optimized by
2020-08-20 10:59:12 +0800
777f3ea9
refactor(gopt): format code by
2020-08-20 10:59:12 +0800
b44e0549
fix(mgb): add method to check local cuda env and refactor the way to get cuda include and lib dir by
2020-08-03 11:19:13 +0800
30ce3c60
Revert "fix(mgb/opr): change EQ opr's backward_graph to nullptr instead of InvalidGrad" by
2020-08-20 03:07:18 +0800
14e71b55
feat(imperative): add helper for dnn opr caller by
2020-08-17 16:04:27 +0800
2272abe1
fix(mgb/fallback): disable nchw44 in conv1x1 and im2col in x86 by
2020-08-11 12:30:32 +0800
230ab45a
fix(mgb/naive): fix naive convolution no dispatch kernel in handle by
2020-08-06 20:00:21 +0800
22853fa2
feat(mge/quantization): add `mapping` parameter for custom modules by
2020-08-18 14:04:47 +0800
6e70fa7a
feat(dnn/arm): add fp32 asm gemm for a53 a55 and i8i8i16 gemm for a72 a53 by
2020-08-19 00:22:30 +0800
dbaf84b0
feat(imperative): add cond_take opr by
2020-08-11 12:22:14 +0800
df356635
fix(mgb/fallback): delete im2col duplicate code and fix nchw44 usable by
2020-08-14 19:58:58 +0800
4a227083
fix(mgb/fallback): fix conv1x1 and conv1x1_gemv nchw44 usable by
2020-08-14 15:39:49 +0800
b778d225
feat(mgb/fallback): add conv1x1_gemv, conv1x1 and im2col 8x8x16/8x8x32 support bias by
2020-08-07 16:53:28 +0800
c357db01
feat(mgb/arm_common): add 8x8x16 nchw44 max pooling by
2020-08-10 15:23:55 +0800
7f5f375f
feat(dnn/arm): add armv7 nchw_nchw44 3x3s2 asm kernel by
2020-07-28 17:04:00 +0800
b7d5fa7e
fix(sdk/load_and_run): fix misuse std::string::substr by
2020-08-11 16:31:48 +0800
1bce857c
fix(mgb/opr-mm): use comp_node of config as default in CollectiveComm by
2020-08-15 17:40:03 +0800
27205461
feat(mgb/opr-mm): add register info cache for multi-machine oprs by
2020-08-15 16:36:07 +0800
a7ff580e
feat(mge/utils): add net stats to calculate parameters and flops by
2020-08-14 18:07:18 +0800
96ec586d
fix(dnn): fix bool cvt by
2020-08-17 11:44:59 +0800
f26cd398
build(third_party): Update megray version by
2020-08-25 15:16:43 +0800
f829f836
test(mgb/index): add empty index desc tests by
2020-08-10 11:52:38 +0800
e73f2799
fix(mgb/index): enable index desc empty by
2020-07-30 14:48:11 +0800
b43f6a26
fix(mge/quantization): handle empty Observer in QATModule by
2020-08-13 21:53:41 +0800
13e8f00a
feat(mge/module): add forward hook support by
2020-08-13 14:11:59 +0800
ab9fa48e
feat(mge/quantization): make `q_dict` a kwarg rather than an arg by
2020-08-12 16:57:20 +0800
f8810f73
feat(mge/imperative): prepare to make whl by
2020-08-12 14:21:26 +0800
ff60fdb8
feat(dnn): add bool type cvt on gpu by
2020-08-13 16:32:26 +0800
e8571cca
fix(mgb/cuda): fix cuda host alloc set device by
2020-08-13 11:00:59 +0800
f7b5eced
refactor(mgb/opr-mm): set False as default value of local_grad by
2020-08-13 11:18:39 +0800
7a8183f4
fix(mge/quantization): fix enable observer bug by
2020-08-12 14:38:22 +0800
555ecea9
feat(mge/quantization): add bias fakequant support by
2020-08-12 10:49:51 +0800
9440842e
fix(mge/core): fix Tensor deepcopy issue by
2020-08-12 10:49:15 +0800
d4b86b84
feat(mge/dtype): add int2 lowbit support and example by
2020-08-11 16:36:49 +0800
3931099e
fix(dnn/test): fix nchw_nchw44 i8i8i16 benchmark by
2020-08-07 20:15:31 +0800
bcf5691d
feat(dnn/arm): add nchw_nchw44 i8i8i16 2x2 3x3 5x5 7x7 s1 s2 conv by
2020-07-31 19:43:53 +0800
c7b6ef35
feat(dnn/cuda): add warp perspective backward mat idx by
2020-07-08 01:37:26 +0800
a773d076
feat(dnn/arm_common): add nchw44 8x8x16 channel wise conv stride1 2x2 3x3 5x5 stride2 2x2 3x3 5x5 by
2020-07-16 22:43:46 +0800