Megvii Engine Team
|
71c2f61254
|
feat(dnn/cuda): add relayout format to support layout transform between NCHW and NCHW64
GitOrigin-RevId: 1445ecfabe
|
4 years ago |
Megvii Engine Team
|
df009e89e1
|
feat(dnn/cuda): add cuda conv bias impls for NCHW format tensors with qint4 data type
GitOrigin-RevId: a0a08cf42c
|
4 years ago |
Megvii Engine Team
|
ed92207585
|
feat(dnn/cuda): add conv bias impl for int4 data type using sass language
GitOrigin-RevId: ae3d3e1c98
|
4 years ago |
Megvii Engine Team
|
52b55564d7
|
refactor(dnn/cuda): refactor reorder filter and bias kernel to support conv imma with data type s4
GitOrigin-RevId: 6827b73770
|
4 years ago |
Megvii Engine Team
|
517cc6846a
|
ci(gitlab-ci): add inline lineno checking in copybara linter
GitOrigin-RevId: 56c5068009
|
4 years ago |
Megvii Engine Team
|
23032f50f2
|
feat(dnn/cuda): support float16 for index_incr_multi_axis_vec
GitOrigin-RevId: c2ae93d568
|
4 years ago |
Megvii Engine Team
|
938944027d
|
fix(mgb/dnn): fix cudnn8 convbias
GitOrigin-RevId: 0fdbfd258c
|
4 years ago |
Megvii Engine Team
|
3591ef1f6a
|
fix(mgb): fix conv cudnnconvbackwarddata algo witch is not shake
GitOrigin-RevId: 379bfbe376
|
4 years ago |
Megvii Engine Team
|
1525a02530
|
feat(mge/module): add python wrapper for unfold
GitOrigin-RevId: 562103186f
|
4 years ago |
Megvii Engine Team
|
13b15fb08c
|
feat(megbrain): add correlation opr
GitOrigin-RevId: 6d44598891
|
4 years ago |
Megvii Engine Team
|
1997b1a289
|
feat(dnn/cuda): add correlation kernel
GitOrigin-RevId: 25e58b61e6
|
4 years ago |
Megvii Engine Team
|
acb000d07f
|
fix(api_cache): fix serialization for conv_desc
GitOrigin-RevId: 95dbc9c685
|
4 years ago |
Megvii Engine Team
|
c4032222fa
|
fix(api_cache): lock api cache for thread safety
GitOrigin-RevId: 8a244677c3
|
4 years ago |
Megvii Engine Team
|
5419a95d1e
|
perf(cuda/conv): cache serval cudnn api
GitOrigin-RevId: 188c62cdd6
|
4 years ago |
Megvii Engine Team
|
19887942c8
|
feat(dnn/apicache): add generic apicache
GitOrigin-RevId: 40b8ac2ab6
|
4 years ago |
Megvii Engine Team
|
6bb6787d9a
|
feat(mge): add a tool which can analyze the file generated by compare_binary_iodump.py
GitOrigin-RevId: 9acab0a49f
|
4 years ago |
Megvii Engine Team
|
c3f8cf04fa
|
feat(dnn): add conv_bwd_data and conv_bwd_filter accuracy shake check
GitOrigin-RevId: 4069e083d2
|
4 years ago |
Megvii Engine Team
|
f36e99d30b
|
fix(build): fix naive build
GitOrigin-RevId: 0050ff5d9c
|
4 years ago |
Megvii Engine Team
|
0a86a07096
|
fix(mgb/dnn): fix cub potential issues
Wrap cub with CUB_NS_PREFIX and remove dependency on Thrust to avoid potential linking issues
GitOrigin-RevId: 53893b0a39
|
4 years ago |
Megvii Engine Team
|
1e6ef3771f
|
feat(mgb/dnn): add accuracy shake checker
GitOrigin-RevId: 0bb52078a1
|
4 years ago |
Megvii Engine Team
|
a5a29826dc
|
feat(mgb/dnn): add accuracy_depend_on_batch attribute
GitOrigin-RevId: 2be1db43d0
|
4 years ago |
Megvii Engine Team
|
4b141f8de4
|
fix(mgb): add usable-depend-on-shape attr
GitOrigin-RevId: 3a14fa6b6f
|
4 years ago |
Megvii Engine Team
|
69a146c8c3
|
build(rocm): support rocm-3.9
GitOrigin-RevId: 85f8911736
|
4 years ago |
Megvii Engine Team
|
a31b7c6ee3
|
build(rocm): partially support hcc compilation
GitOrigin-RevId: ca9f1f8e8e
|
4 years ago |
Megvii Engine Team
|
621ae0a1e8
|
fix(dnn): replace kernel launch syntax with macro for hcc
GitOrigin-RevId: f9e69d4825
|
4 years ago |
Megvii Engine Team
|
78fff72a95
|
feat(dnn): add param_pack for rocm
GitOrigin-RevId: 2180504c71
|
4 years ago |
Megvii Engine Team
|
8163ed157d
|
fix(dnn/cuda): fix cutlass matmul splitk limit
GitOrigin-RevId: fc9a7c638c
|
4 years ago |
Megvii Engine Team
|
ef9aa80074
|
fix(mgb/dnn): fix cuda naive matmul algo
GitOrigin-RevId: 79c9bba73b
|
4 years ago |
Megvii Engine Team
|
2d18074a70
|
fix(mgb): fix spell error
GitOrigin-RevId: acae00e0a5
|
4 years ago |
Megvii Engine Team
|
ff755451d2
|
refactor(mgb): move algo's name from info to desc and delete some algo's unnecessary param() method
GitOrigin-RevId: 144ff547d1
|
4 years ago |
Megvii Engine Team
|
756c1eb7f2
|
fix(mgb/dnn): add cuda float naive matmul algo
GitOrigin-RevId: db7f7fc057
|
4 years ago |
Megvii Engine Team
|
04b1a45af4
|
fix(dnn): fix cudnn crash when finalize called after cudnn dtor
GitOrigin-RevId: b0ad639921
|
4 years ago |
Megvii Engine Team
|
c338e876ec
|
refactor(mgb/dnn): add negative attribute for algo
GitOrigin-RevId: 88b1ce94a5
|
4 years ago |
Megvii Engine Team
|
ec1a99acc2
|
refactor(mgb/dnn): replace reproducible with attribute
GitOrigin-RevId: d49015714c
|
4 years ago |
Megvii Engine Team
|
0d165399e6
|
fix(mgb): fix fastrun for imperative
GitOrigin-RevId: db54984b92
|
4 years ago |
Megvii Engine Team
|
94401ce44a
|
chore(dotprod): dotprod is enabled by default on the android platform
GitOrigin-RevId: d412108732
|
4 years ago |
Megvii Engine Team
|
85b41a90df
|
feat(dnn): add checksum opr and test
GitOrigin-RevId: e784a76e0b
|
4 years ago |
Megvii Engine Team
|
a49f4a66b7
|
feat(dnn): add indexing_one_hot and indexing_set_one_hot opr
GitOrigin-RevId: c5406c71ff
|
4 years ago |
Megvii Engine Team
|
2fd3fa8834
|
feat(cmake): update for enflame cmake compile
GitOrigin-RevId: 3c3c6b3462
|
4 years ago |
Megvii Engine Team
|
9f2af2099c
|
feat(mgb): add enflame comp node
GitOrigin-RevId: 478c8538aa
|
4 years ago |
Megvii Engine Team
|
33da8de12b
|
build(dnn/cuda): split compilation for cutlass wrapper
GitOrigin-RevId: 6365d5fdbc
|
4 years ago |
Megvii Engine Team
|
420672beca
|
fix(mgb/dnn): fix x86 matmul midout decl
GitOrigin-RevId: fe1fc977e1
|
4 years ago |
Megvii Engine Team
|
b717606989
|
fix(dnn/cuda): add block size limit for culass gemm algo
GitOrigin-RevId: c0940e4535
|
4 years ago |
Megvii Engine Team
|
55974e8cf9
|
feat(log): opt log
* opt log at release mode
* add MGE_OVERRIDE_LOG_LEVEL for runtime debug
//! env to config LogLevel
//! DEBUG = 0, INFO = 1, WARN = 2, ERROR = 3, NO_LOG = 4
//! for example , export MGE_OVERRIDE_LOG_LEVEL=0, means set LogLevel to DEBUG
GitOrigin-RevId: 16cd674c56
|
4 years ago |
Megvii Engine Team
|
58c8746e30
|
fix(opr): fix fast-run error in cuda
GitOrigin-RevId: 28dd187df9
|
4 years ago |
Megvii Engine Team
|
ba2ad46e54
|
feat(gopt): add deconv nchw4 int8 opt pass, add deconv nchw int8
GitOrigin-RevId: c0530a949e
|
4 years ago |
Megvii Engine Team
|
5d350fc843
|
feat(dnn/cuda): add deconv int8 and fix cutlass conv wrapper base on modify cutlass 2.4
GitOrigin-RevId: 49e0565e8a
|
4 years ago |
Megvii Engine Team
|
a3ea1f153c
|
feat(mgb/opr): add fast profile and combined Execution strategy
GitOrigin-RevId: 843dc3a790
|
4 years ago |
Megvii Engine Team
|
c82d88751a
|
fix(dnn/cuda): add cuda nchw int8 conv impl with nchw4 to fix cu111 compatibility
GitOrigin-RevId: 771968f9ac
|
4 years ago |
Megvii Engine Team
|
652ec9f251
|
fix(mgb/dnn): fix backward computation of tqt
GitOrigin-RevId: 850d11a5ce
|
4 years ago |