Commit Graph
-
7b2a76d1refactor(mgb): make conv handle noncontiguous tensors by 2021-06-09 14:49:04 +0800 -
ca2828ddfix(dnn/x86): fix x86 int8 matmul ldc bug by 2021-06-28 20:58:57 +0800 -
aa4e8476fix(interpreter): release gil when interpreter blocking by 2021-05-28 12:49:42 +0800 -
bd62a0a6fix(interpreter): remove notice flag of produce_tensor by 2021-05-18 13:38:04 +0800 -
c78a7848test(profiler): simple tests for profiler by 2021-04-26 19:39:45 +0800 -
d6079db1feat(profiler): add memory flow format for profiler by 2021-04-26 13:29:34 +0800 -
4286bb9ffeat(profiler): add launch script for profiler by 2021-04-26 13:27:51 +0800 -
9d47c3bafeat(profiler): imperative profiler support tracing by 2021-04-26 13:26:01 +0800 -
cdcb46bafeat(profiler): add chrome timeline format for profiler by 2021-05-15 20:54:36 +0800 -
1d64792brefactor(profiler): detach profiler from interpreter by 2021-05-15 18:14:44 +0800 -
f2027b8drefactor(interpreter): recompute with do_apply_op by 2021-05-15 18:11:34 +0800 -
5cb35c1brefactor(profiler): add state structs to replay recorded events by 2021-05-15 14:23:00 +0800 -
84fc5c92refactor(profiler): remove dump logic for old profiler by 2021-05-15 13:28:50 +0800 -
5b5a8261refactor(profiler): use macro to simplify event recording/definition by 2021-05-15 12:21:41 +0800 -
1ce40b5brefactor(interpreter): wrap accesses to channel/worker state by 2021-05-14 19:21:51 +0800 -
90d39057feat(mge): add mge._exit by 2021-04-26 13:14:53 +0800 -
846642a8docs(misc): add some notes by 2021-04-26 13:12:09 +0800 -
fb8f1534fix(cmake/debug): fix asan do not work at windows env by 2021-06-25 17:10:38 +0800 -
40085acbfix(mgb): remove unnecessary cudnn8 warning by 2021-06-24 18:55:04 +0800 -
54a4d70efeat(src/serialization): add support of serializing metadata by 2021-05-11 10:54:04 +0800 -
721091fafix(core): fix thread local is not supported in ios by 2021-06-18 19:24:55 +0800 -
1d865281ci(config): change pip mirror to pypi cache proxy by 2021-06-08 18:21:36 +0800 -
82b95a3afix(lar): fix build error by 2021-06-22 19:45:29 +0800 -
8cfe3c39fix(lar): update --verbose to --model-info by 2021-06-22 11:36:33 +0800 -
2e5a473dfeat(lar): support --verbose by 2021-06-21 20:40:26 +0800 -
b7e596b4perf(autograd): copy inputs before capture in backward_graph_grad_rule by 2021-06-23 13:54:04 +0800 -
2ac3c9dcfix(trace): constants in backward graph treat as ImmutableTensor corectlly by 2021-06-15 19:18:00 +0800 -
7eea1fc6feat(lite/opencl): fix CMake OpenCL and megenginelite py import and add keywords check by 2021-06-23 21:02:50 +0800 -
62bd6c82feat(cmake/debug): misc for build * add asan build option * fix cpuinfo build opt level * fix host release build with out debug info * opt "fix lite bazel/cmake symbols MR" * other misc build opt by 2021-06-09 20:28:34 +0800 -
c7a5c21achore(bazel/cmake): fix lite bazel/cmake symbols by 2021-04-02 14:08:54 +0800 -
3e4e4c46feat(mgb/jit): add graph_opt_config and jit_config interfaces by 2021-06-04 14:45:08 +0800 -
1c7d0802fix(cuda): remove cuda driver version check and runtime minor version by 2021-05-26 17:01:18 +0800 -
b87af9f7feat(dnn/cuda): topk support fp16 by 2021-05-28 17:34:25 +0800 -
34262d90fix(imperative): fix the size of blob with offset by 2021-06-17 16:46:57 +0800 -
787f187efix(imperative/src): fix dot backward error by 2021-06-10 16:07:45 +0800 -
f35687caMerge pull request #181 from haolongzhangm/update-readme by 2021-07-08 20:18:29 +0800 -
bf2e2d29Update README.md by 2021-06-15 10:25:39 +0800 -
8d76576ffix typos by 2021-04-24 20:28:36 +0800 -
e67fefcdci(yml): enable try-import branch invoke ci on push by 2021-06-30 17:08:40 +0800 -
7038a7f5fix(quant): fix spell error by 2021-05-27 17:00:21 +0800 -
4d72e707deps: update cutlass by 2021-06-23 13:26:02 +0800 -
355153e1feat(mge/dtr): add DTR in computing graph by 2021-05-14 14:24:40 +0800 -
76f4f975refactor(sublinear): add SeqModifierBase by 2021-05-24 16:52:20 +0800 -
f584416afix(dnn/bn): revise the conditions for inplace flag by 2021-05-14 14:22:08 +0800 -
a9b60fbffix(ci/lite): reopen lite_test build by cmake as some reason, lite_test need static link lite when cuda enable on gcc7 and gcc8, if not cask_trt::AbiInfo::~AbiInfo will double call at atexit stage, which will lead double free at the end of test, gcc9 do not have this issue, for compat all CI env, we use static link!!! by 2021-06-09 14:17:38 +0800 -
2eea0009feat(mgb): add fast run batch size graph option by 2021-05-19 20:37:44 +0800 -
0ac642b5fix(imperative): persistent cache write through on put by 2021-06-04 16:47:40 +0800 -
47dcdf3efix(mgb/core): fix dtype and resize modifiers for tensor by 2021-06-09 15:09:17 +0800 -
29f7cdb8fix(mgb/opr): correct nvof out shape computation by 2021-06-08 15:43:59 +0800 -
71cc814efeat(ci): add aarch64 linux ci by 2021-06-04 16:22:09 +0800 -
31a1f538feat(whl/opencl): enable OpenCL in python whl by 2021-06-04 00:25:53 +0800 -
b07f3728feat(aarch64/whl): support aarch64 whl by 2021-06-03 15:41:42 +0800 -
d8ee0d7bfix(mge/distributed): fix the mutli dataloader test error by 2021-06-03 10:48:58 +0800 -
e275dfecfeat(imperative/python): support pooling mode "average" for avg pool2d module by 2021-06-02 15:57:52 +0800 -
03ab8136fix(core): fix asan error cause by wild thread_pool ptr by 2021-06-04 18:24:54 +0800 -
24a38781feat(dnn/cuda): add nchw conv u4xs4 support by 2021-04-28 18:57:13 +0800 -
606540befeat(dnn/cuda): add nhwc 4bit warp perspective by 2021-05-24 20:09:36 +0800 -
1e601943feat(dnn/cuda): add nhwc int4 pooling by 2021-05-25 19:08:46 +0800 -
0fb9cc41fix(gopt): fix nchw64 opt pass by 2021-05-20 14:11:41 +0800 -
e661ae90feat(dnn/cuda): add base class for cutlass uint4 and int4 algos by 2021-05-21 10:12:28 +0800 -
319436ddfeat(dnn/cuda): add cutlass impls for uint4 x int4 conv bias by 2021-05-17 19:42:43 +0800 -
d28eba4efeat(dnn/cuda): add cutlass impls for int4 conv bias by 2021-05-07 17:06:29 +0800 -
14b65e4dfeat(dnn/cuda): add reduce_filter_and_update_bias by 2021-05-10 19:13:59 +0800 -
2d4e62effeat(dnn/cuda): add cuda uint4 pooling by 2021-05-07 16:59:38 +0800 -
19919384feat(dnn/cuda): add cuda uint warp perspective by 2021-05-17 13:41:56 +0800 -
01354337fix(mge/autodiff): fix incorrect handling of tuple dy by 2021-06-03 20:05:22 +0800 -
5868d1fefix(arm_common/pooling): check mode in pooling algo to avoid wrong use AVERAGE_COUNT_EXCLUDE_PADDING by 2021-06-02 17:06:05 +0800 -
86b69cacfix(dnn): fixes for int4 by 2021-04-26 17:20:49 +0800 -
4a802d21feat(dnn/cuda): add conv u4xs4 sass kernel by 2021-04-16 10:39:52 +0800 -
adf75a29perf(dnn/cuda): add sass int4 128x128 by 2021-04-21 18:13:21 +0800 -
8da2f698feat(dnn/cuda): support warp perspective/pooling op when channel not aligned to 64 by 2021-04-12 19:48:33 +0800 -
c218d4b0feat(dnn/cuda): fallback conv qs4 support channel not aligend to 64 by 2021-04-12 17:50:47 +0800 -
4fe68ac9feat(dnn/cuda): support transforming layout between nchw and nchw64 when channel not aligned to 64 by 2021-04-12 17:20:19 +0800 -
ae6ff2c5feat(mgb/gopt): add opt pass for nchw64 layout transform by 2021-04-02 17:54:40 +0800 -
63a9bd30feat(mgb/gopt): add an opt pass for padding channels to enable fast int8/int4 support on GPU by 2021-03-31 18:46:54 +0800 -
56e863b7fix(dnn/cuda): fix int4 epilogue stg bug by 2021-04-19 19:39:16 +0800 -
cff61a53perf(dnn/cuda): optimize int4 sass conv main loop and epilogue without fuse_z by 2021-04-12 16:54:26 +0800 -
12a0e615feat(dnn/cuda): add cuda elemwise int4 by 2021-04-06 17:28:14 +0800 -
df1af59bfeat(dnn): warp perspective support int4 by 2021-03-29 18:50:32 +0800 -
2398df07feat(dnn/cuda): add cuda int4 pooling by 2021-03-31 21:04:43 +0800 -
2a2a7f45test(mgb/opr): add testcase for conv bias int4 by 2021-03-30 10:27:47 +0800 -
858261affix(python_module): fix conversion between numpy-ndarray and mgb tensor for qint4 and quint4 by 2021-03-26 17:51:21 +0800 -
e250afb0feat(dnn/cuda): support conv_bias for nchw64 and qint4 by 2021-03-23 10:41:39 +0800 -
3b9b8780refactor(dnn): refactor lowbit tensor format by 2021-03-24 14:06:27 +0800 -
c74660eafix(dnn/cuda): fix invalid local read for relayout format kernel by 2021-03-24 14:05:19 +0800 -
8fef78d0feat(dnn/cuda): add relayout format when width is an odd number by 2021-03-22 17:47:55 +0800 -
91d61607feat(dnn/common): add tensor format for low-bits tensor layout by 2021-03-19 18:01:21 +0800 -
19a554d6test(dnn/cuda): add testcase for transforming tensor layout between nchw and nchw64 by 2021-03-18 15:54:21 +0800 -
71c2f612feat(dnn/cuda): add relayout format to support layout transform between NCHW and NCHW64 by 2021-03-16 13:52:40 +0800 -
df009e89feat(dnn/cuda): add cuda conv bias impls for NCHW format tensors with qint4 data type by 2021-03-12 10:30:34 +0800 -
ed922075feat(dnn/cuda): add conv bias impl for int4 data type using sass language by 2021-03-11 14:22:06 +0800 -
52b55564refactor(dnn/cuda): refactor reorder filter and bias kernel to support conv imma with data type s4 by 2021-03-10 12:06:51 +0800 -
d2673c5afix(ci/windows): add windows cuda test by 2021-05-17 20:57:57 +0800 -
2d6827c1fix(mgb/windows): temporary workround on cuda-windows python exit code(127), as windows cuda driver unloading before atexit function may remove this after upgrade cuda runtime by 2021-06-01 17:05:55 +0800 -
517cc684ci(gitlab-ci): add inline lineno checking in copybara linter by 2021-06-01 18:25:19 +0800 -
23032f50feat(dnn/cuda): support float16 for index_incr_multi_axis_vec by 2021-05-25 16:54:32 +0800 -
93894402fix(mgb/dnn): fix cudnn8 convbias by 2021-05-19 18:56:31 +0800 -
5427a67cfix(cmake/subdirectory): fix project import by other sdk by add-subdirectory by 2021-05-27 13:53:23 +0800 -
241b35a6refactor(ops): remove BackwardGraph op by 2021-05-24 17:09:19 +0800 -
d2e33af5fix(mgb): fix wrong set of strategy in lar by 2021-05-25 19:13:47 +0800