e9db061e
fix(mgb): fix compiling error for cuda-11.1 by
2021-02-05 10:38:47 +0800
cd02d7c8
feat(mge/opr): support deformable conv2d/psroi_pooling by
2021-01-15 14:22:56 +0800
227d5b9f
Update README and fix file permissions. by
2021-02-15 15:23:20 +0800
f3378100
feat(mgb): enable output dynamic memory alloc by
2021-01-14 11:28:14 +0800
e82fa4ec
fix(gopt): using new_inp for build_chain in DelayBroadcast pass by
2021-01-29 17:33:39 +0800
a09fc5f7
fix(mgb/serialization): disable inplace arith graph opt in graph load by
2021-01-23 16:56:06 +0800
cd7090ac
fix(opencl): enable image on mali(cl2.1) by
2021-01-27 12:56:51 +0800
fc8d13cd
fix(build): fix linkage on ubuntu1604 by
2021-02-01 19:13:37 +0800
c51a687c
chore(mge): update copyright years by
2021-02-02 23:45:41 +0800
7c09e41f
refactor(mgb): add circular dependency check by
2021-01-29 16:04:34 +0800
af42ce7e
fix(megdnn): some fixes of execution policy by
2021-01-30 17:55:57 +0800
7afa422d
refactor(megdnn): refactor sub opr setter by
2021-01-28 18:19:44 +0800
821656aa
refactor(megdnn): refactor brute force algo in batched matmul by
2021-01-28 13:12:22 +0800
08ff62de
refactor(megdnn): refactor batched matmul algo in conv bias by
2021-01-26 22:08:12 +0800
8773926e
refactor(megdnn): refactor matmul algo in conv bias by
2021-01-26 21:22:02 +0800
e4b71bdf
refactor(megdnn): remove unnessary 1x1 algo by
2021-01-26 17:00:00 +0800
44c8d2d1
refactor(megdnn): refactor matmul algo in deformable conv by
2021-01-26 16:20:33 +0800
b04ad06f
refactor(megdnn): refactor matmul algo in conv backward filter by
2021-01-26 01:13:00 +0800
25089e52
refactor(megdnn): refactor matmul algo in conv backward data by
2021-01-26 00:42:31 +0800
3620a940
fix(megdnn): fix test with workspace_limits to zero by
2021-01-25 16:54:06 +0800
0d720653
refactor(megdnn): add default algo for convolution forward by
2021-01-25 16:54:06 +0800
659217ac
refactor(megdnn): refactor bfloat16 convbias to recursive inteface by
2021-01-24 01:28:16 +0800
4a1d52c9
refactor(megdnn): refactor bfloat16 matmul to recursive inteface by
2021-01-22 22:04:12 +0800
b8febaf9
refactor(megdnn): refactor bfloat16 convolutionbackwardfilter to recursive inteface by
2021-01-21 22:45:20 +0800
f14e0c17
feat(mgb): add recursive for fastrun and megdnn test by
2020-12-15 20:28:32 +0800
85fa9883
refactor(dnn): add get_algorithm_from_desc interface by
2020-12-08 16:05:09 +0800
43b4d4a4
feat(imperative/test): add more correctness tests by
2021-01-27 18:23:51 +0800
91a3580f
refactor(mge/cgtools): remove load_and_inference and use GraphInference by
2021-02-02 10:09:56 +0800
4485e780
fix(mge/functional): fix linspace device and open other trace tests by
2021-02-01 13:48:59 +0800
334eda87
refactor(mge): test trace inside opr_test by
2021-01-25 10:12:46 +0800
2b8150ab
fix(dnn): fix bazel build issue for cambricon platform by
2021-01-28 10:25:44 +0800
3b28af95
fix(sdk): add disable-optimize-for-inference option by
2021-01-25 20:31:46 +0800
938152af
fix(mge/functional): convert input type to float32 for more elemwise op by
2021-01-13 14:32:20 +0800
19466046
feat(mge/funcitonal): add cvt_color opr python interface by
2021-01-21 17:52:35 +0800
412d1f0c
fix(mge/tensor): implement abtract method to fix lint errors by
2021-01-26 11:51:39 +0800
0e8b81c2
fix(dnn/opencl): fix elemwise negative stride support by
2021-01-18 13:55:51 +0800
dbb3dd68
refactor(profiler): integrate profiler into interpreter by
2021-01-25 11:38:17 +0800
ff05667b
test(dist): refactor distributed test with fixtures by
2021-01-25 19:20:41 +0800
9fb8444d
fix(imperative): catch python exception in c++ by
2021-01-20 18:16:54 +0800
e3a3e0cd
feat(mge/functional): add warp_affine by
2021-01-22 14:22:39 +0800
329306b0
fix(cmake/cuda): fix build at cuda `copy` env caused by b278a69e1 `copy` env vs `normal` env: case a:`normal` env install CUDNN/TRT by package tool, eg, apt/yum, which will put include file(cudnn.h/NvInfer.h) to /usr/include etc. by
2021-01-25 14:55:07 +0800
8e50a6da
feat(mge/utils): add GraphInference in cgtools by
2021-01-20 14:43:09 +0800
97207d00
fix(mgb/opr): fix bazel compile error by comma in mgb_throw_if's format string by
2021-01-25 17:23:30 +0800
5a442553
fix(mgb/plugin): fix opr footprint for conv with NCHW32_NCHW4 format by
2021-01-25 12:36:16 +0800
adc49de8
feat(mge/imperative): expose c++ tensor reference count by
2021-01-06 16:41:04 +0800
fe1680b3
fix(imperative/ops): improve infer_output_attrs for broadcast by
2021-01-22 12:09:05 +0800
b1806b84
docs(mgb/opr): improve batch_norm err msg in training mode by
2021-01-22 12:13:40 +0800
27a9326c
fix(mge/module): fix frozen batch norm by
2021-01-22 13:39:56 +0800
c3ba0280
chore(mge): correct license headers by
2021-01-19 18:21:41 +0800
364afec0
chore(mge): update copyright years by
2021-01-18 11:58:06 +0800
9a09f761
fix(whl): append buildid to wheel version by
2021-01-21 17:59:39 +0800
af5ab4d9
fix(imperative/tensor): fix Py_DECREF on uninitialized pointer by
2021-01-18 13:12:15 +0800
10d9d77e
refactor(mge/distributed): remove get_free_ports func by
2021-01-18 14:20:56 +0800
0adf49b1
fix(mge/distributed): fix deadlock by mixing thread and fork by
2021-01-14 12:36:04 +0800
ae8b38f6
fix(cmake/whl): reduce wheel size as some reason: some version of trt/cudnn need -Wl,--whole-archive attribute to fix cuda issue, target A(CMAKE OBJECT,eg megbrain/megdnn): -Wl,--whole-archive depend on trt/cudnn target B(SHARED LIBS,eg libmegengine.so) --> A with `PUBLIC` depends target C(SHARED LIBS,eg _impertive.so) --> B with `PUBLIC` depends, then ld will force link fatbin section into C, which will be undesired by
2021-01-19 18:25:56 +0800
4aaae995
feat(mge/functional): add python wrapper to resize opr by
2021-01-20 16:44:37 +0800
d04b4bc0
fix(interp): thread safety for drop and swapout by
2021-01-21 13:45:56 +0800
3eb529d6
refactor(interpreter): recognize recomp on main thread rather than worker by
2021-01-19 11:32:19 +0800
df976782
feat(mge/functional): add matinv by
2021-01-20 12:48:50 +0800
3bda3347
fix(dnn/fallback): fix segmentfault caused by im2col/conv1x1 using fallback naive matmul. by
2021-01-19 18:00:07 +0800
2eed7d83
docs(mge): restore autodiff.Function docs by
2021-01-18 17:03:19 +0800
51003176
feat(mge): make Function available at autodiff.Function by
2021-01-18 17:04:11 +0800
ca1d74ad
fix(mge/utils): fix types import error by
2021-01-19 17:21:45 +0800
87ff58f7
fix(megdnn): add algo for matmul/batchedmatrixmul of naive and opencl by
2021-01-19 11:32:47 +0800
169fa53d
fix(mgb): fix execution_policy set of matmul by
2021-01-19 11:32:39 +0800
abe3c165
feat(mge): tensorrt runtime opr by
2021-01-14 14:02:57 +0800
1b1ad56a
fix(mgb/gopt): fix warp fusion opt pass by
2021-01-13 17:58:40 +0800
99b17623
fix(mgb/opr): fix Reduce static value inference by
2021-01-16 10:08:18 +0800
a3caa5d3
fix(mgb(dnn)): fix convbias cudnnConvBiasActivation by
2021-01-10 00:02:30 +0800
c418d3cd
fix(mge/data): add timeout event by
2021-01-14 13:57:48 +0800
0f739c11
fix(mgb/gopt): fix NormalizeArithChainPass to process sub-div chains by
2021-01-13 19:39:04 +0800
c53abcdf
chore(mge): minor improvements related to grad by
2021-01-14 17:36:41 +0800
0a3ca253
fix(mge): fix backward graph optimization by
2021-01-14 17:15:37 +0800
ea8eb4cf
feat(mge/distributed): scalar support for distributed functions by
2021-01-13 17:27:28 +0800
b83c77e1
test(mge/test_save_load): remove test model file by
2021-01-12 16:36:52 +0800
c5229913
test(mge/core): add host compute test, fix async_level and swap/drop test by
2020-12-28 16:18:20 +0800
c8e04ce4
fix(imperative/interpreter): use info->h_value for host compute by
2021-01-11 17:32:11 +0800
a5a60679
feat(imperative/interpreter): add more dispatch mode in apply_op by
2020-12-22 16:36:22 +0800
45e20602
feat(mgb/tensor): add proxy_to_comp_node for HostTensorND by
2020-12-28 16:12:48 +0800
248d8bf0
feat(imperative/ops): improve infer attrs validate function by
2020-12-15 11:58:24 +0800
8ed2077b
fix(imperative/interpreter): reset m_waitee on exception by
2021-01-12 16:46:18 +0800
8fe865d8
feat(imperative/ops): add infer_output_attrs for Reshape by
2021-01-13 19:40:31 +0800
267d6127
fix(imperative/mini_graph): fix no_err value in shape infer by
2021-01-12 16:49:15 +0800
bc3561c1
fix(imperative/ops): fix inplace_add's output desc, remove value by
2021-01-12 16:28:18 +0800
6295f843
fix(mge/tensor): fix mem leak when getting value in trace by
2021-01-12 18:37:41 +0800
5db17907
chore(megbrain): fix megskull cnn/img opr compatiblity for add convolutionv2 param by
2021-01-05 18:58:11 +0800
935fe781
chore(megbrain): format sereg.h code by
2020-12-22 13:54:01 +0800
55042195
chore(winograd): add Convolutionv2 param by
2020-11-13 12:39:30 +0800
7191c4bd
fix(mge/data): support timeout for serial stream dataloader by
2021-01-12 11:06:08 +0800
fe6af7cb
fix(mgb): fix fastrun log by
2021-01-09 13:46:42 +0800
b7ed0cb8
fix(mge/quantization): replace `_reset` with "=" in observer by
2021-01-11 15:37:11 +0800
495b2003
build(mgb/jit): disable halide by default by
2021-01-08 14:54:36 +0800
5d0c9bef
chore(mge/version): change dev version to 1.3 by
2021-01-12 12:23:21 +0800
7b83b0ab
ci(gitlab-ci): open windows ci log by default, as gitlab may timeout if runner do not send a log for a long time by
2021-01-11 10:56:16 +0800
f79ed946
fix(mgb/cambricon): support uint8 input for cambricon runtime opr by
2020-09-18 22:11:29 +0800
0b4918b2
test(mge/quantization): classmethod `from_float_module` of qat module by
2020-12-18 15:34:45 +0800
bb369383
chore(mge/license): remove multipledispatch from ACKNOWLEDGMENTS by
2021-01-11 11:04:14 +0800
e9a7be46
feat(mgb): adapt to imperative runtime by
2021-01-07 18:44:25 +0800
5063a206
feat(mgb): add fastrun support for matmul/batchedmatrixmul by
2021-01-07 00:32:11 +0800
409a8772
feat(dnn): add algo interface for rocm&fallback matmul and batched matrix mul by
2021-01-06 23:47:23 +0800