Megvii Engine Team
25ec2530ba
feat(whl/api/lar): enable megengine dll on Windows
1: reduce python whl package size
2: unify api link logic on all OS
3: add option: MGE_WINDOWS_BUILD_WITH_STATIC_CRT
--- default OFF
--- if build CRT(vc runtime) with STATIC with megengine.dll
some CRT api will crash, for example, flush, so if you
build with static megengine, and do not want to install CRT
you can set MGE_WINDOWS_BUILD_WITH_STATIC_CRT TRUE
--- how to install CRT:
https://docs.microsoft.com/en-us/cpp/windows/latest-supported-vc-redist?view=msvc-160
install VC_redist.x64.exe
4: rename megengine_export to megengine_shared(only export needed symbols ),
caused by runtime symbols conflict with pytorch
GitOrigin-RevId: 93d8d80f29
3 years ago
Megvii Engine Team
c48d58daa8
feat(dnn/arm_common): add N1HW like elemwise broadcast mode
GitOrigin-RevId: 2895135801
3 years ago
Megvii Engine Team
26634db7a8
fix(dnn): support relayout for non-contigous layout
GitOrigin-RevId: 44a0adddba
3 years ago
Megvii Engine Team
056fd6bc59
feat(dnn/arm64): support stride_m in arm64 relayout
GitOrigin-RevId: c74193a23d
3 years ago
Megvii Engine Team
c50858ee13
fix(dnn): specialize pow to make it consistent
GitOrigin-RevId: cff3bbbadd
3 years ago
Megvii Engine Team
849f0ece9d
fix(dnn): drop batched matmul cublas algo when batch is 1
GitOrigin-RevId: 71126a27b0
3 years ago
Megvii Engine Team
b5bf56e0ee
style(dnn): add bypass of clang-format for dnn foreach_opr macro
GitOrigin-RevId: 2f4bfcff42
3 years ago
Megvii Engine Team
5af52746f7
fix(mgb): fix bug caused by conv filter size is too big
GitOrigin-RevId: 7fe3a9fcf1
3 years ago
liuke
b0ba6d3201
Merge pull request #207 from togetherwhenyouwant:feat-x86-matmul-6x16x2
GitOrigin-RevId: 148ae44ba0
3 years ago
Megvii Engine Team
10af44abba
fix(dnn/cuda): fix cudnn conv impl for nchw4_nchw hybrid layout
the conv_bias algo *_IMPLICIT_GEMM in cudnn less than 8.0.0 is disabled due to the incorrect result for int8x4->f32 configs
GitOrigin-RevId: 7cc52d0a85
3 years ago
Megvii Engine Team
5885b137fa
feat(dnn/arm): support layout like NHWC channel like broadcast on arm
GitOrigin-RevId: fb4300004c
3 years ago
Megvii Engine Team
369c2ccc5a
style(all): reformat c++ code
GitOrigin-RevId: 3ffd1b211f
3 years ago
Megvii Engine Team
bfb30dcb81
chore(format): fix compile bugs after code format
GitOrigin-RevId: 11a4b06f6f
3 years ago
Megvii Engine Team
eeccf2bc0d
ci(check): add clang-format in check stage
GitOrigin-RevId: 25c24d7856
3 years ago
zjl
d2184af3b2
feat(dnn/src/x86/matmul): add matmul_6x16 for x86
3 years ago
Megvii Engine Team
177dec94c5
feat(mgb/opr): add bgr2gray mode for cvtcolor opr
GitOrigin-RevId: d50415b236
3 years ago
Megvii Engine Team
f5cb21ed3a
fix(mgb/opr): add non finite check
GitOrigin-RevId: a9fcd0a350
3 years ago
Megvii Engine Team
bde5cf3564
feat(dnn): add resize linear for arm
GitOrigin-RevId: 14ac5bda3f
3 years ago
Megvii Engine Team
8cb201868e
fix(mgb): fix fastrun cache serialization method
GitOrigin-RevId: cd8bed3cc9
3 years ago
Megvii Engine Team
563239d38f
feat(dnn): add arm_common nchw44 cwconv3x3s1p1 and cwconv5x5s1p2
GitOrigin-RevId: 9ea411d0e1
3 years ago
Megvii Engine Team
3344b580a9
feat(dnn): add elemwise for nchw88+fp16
GitOrigin-RevId: 63587975f8
3 years ago
Megvii Engine Team
682c74df27
feat(dnn): add direct nchw88 fp16 conv
GitOrigin-RevId: 44719e8b64
3 years ago
Megvii Engine Team
fca195351c
feat(gopt): add nhwc fuse conv typecvt optpass
GitOrigin-RevId: adc2301203
3 years ago
Megvii Engine Team
2fc7358517
Revert "feat(dnn/apicache): add generic apicache"
This reverts commit 40b8ac2ab6
.
GitOrigin-RevId: 0f9b2df4c8
3 years ago
Megvii Engine Team
de363c04af
Revert "perf(cuda/conv): cache serval cudnn api"
This reverts commit 188c62cdd6
.
GitOrigin-RevId: 92a82b8cd9
3 years ago
Megvii Engine Team
729ee64988
Revert "fix(api_cache): lock api cache for thread safety"
This reverts commit 8a244677c3
.
GitOrigin-RevId: 582488adeb
3 years ago
Megvii Engine Team
64c922c4bb
Revert "fix(api_cache): fix serialization for conv_desc"
This reverts commit 95dbc9c685
.
GitOrigin-RevId: ca8c67b6b3
3 years ago
Megvii Engine Team
3d3666b6e0
test(dnn/bn): add compatible configs for NHWC BN
GitOrigin-RevId: ac757ca307
3 years ago
Megvii Engine Team
b3e54eade1
feat(dnn/bn): use new cudnn BN kernel to support NHWC
GitOrigin-RevId: 9d80f2009d
3 years ago
Megvii Engine Team
3977b7aa0b
feat(mgb/shuffle): add shuffle opr
GitOrigin-RevId: 80490a6f84
3 years ago
Megvii Engine Team
17371e79b9
fix(dnn/reduce): fix reduce_mean o16c32 is incorrect for large tensor
GitOrigin-RevId: ebf03d814a
3 years ago
Megvii Engine Team
eca6e1d931
fix(ci): fixes for ci
GitOrigin-RevId: b0a432bd2e
3 years ago
Megvii Engine Team
c14e5719f8
feat(mgb/gopt): add profile impl for global layout transform pass
GitOrigin-RevId: 8ef62baf79
3 years ago
Megvii Engine Team
8a3eb05a1b
refactor(mgb/gopt): refactor tensor reformat opt pass
GitOrigin-RevId: a1b1e89b76
3 years ago
Megvii Engine Team
c33126ab5c
feat(mgb/gopt): add reformat manager
GitOrigin-RevId: b9791b131a
3 years ago
Megvii Engine Team
4f28e14684
fix(dnn): fix compatibility broken of convolution format
GitOrigin-RevId: c17f02d03d
3 years ago
Megvii Engine Team
8b40f57738
feat(mgb/dnn): add conv1x1 algo for matrix mul
GitOrigin-RevId: 585b2c045a
3 years ago
Megvii Engine Team
fb49a2834f
refactor(mgb/dnn): refactor enum used in serializing
GitOrigin-RevId: e57af4a59c
3 years ago
Megvii Engine Team
d69b59035d
feat(dnn): add an get_all_algorithms_safe interface
GitOrigin-RevId: e3734e4531
3 years ago
Megvii Engine Team
103d7f33ba
refactor(dnn/rocm): update hip license header
GitOrigin-RevId: 79d684755d
4 years ago
Megvii Engine Team
5aa52d3863
feat(dnn/rocm): add adaptive pooling opr
GitOrigin-RevId: e844b3e770
3 years ago
Megvii Engine Team
83cf4ee64e
refactor(dnn/rocm): remove some useless includes
GitOrigin-RevId: 3d2c315a36
3 years ago
Megvii Engine Team
323a4642e6
feat(dnn/rocm): add topk opr
GitOrigin-RevId: 5ecb079854
3 years ago
Megvii Engine Team
f4784f4af1
feat(dnn/rocm): add argsort opr
GitOrigin-RevId: b4c3eb4707
3 years ago
Megvii Engine Team
6082c353e7
feat(dnn/rocm): support bool in type_cvt and elemwise
GitOrigin-RevId: ad5ec7bc1c
4 years ago
Megvii Engine Team
8b94f49328
fix(dnn/cuda): fix elemwise and relayout int4 bug when last shape is 1
GitOrigin-RevId: e7d64c4987
3 years ago
Megvii Engine Team
694aa1bd92
feat(dnn): add heuristic cache
GitOrigin-RevId: 35e942b5e3
3 years ago
Megvii Engine Team
bc9cfc277a
feat(mgb): add arm resize nchwxx and naive nearest interp
GitOrigin-RevId: d5fbd59a30
3 years ago
Megvii Engine Team
722aecd437
feat(mgb): support fp16 nhwc backward
GitOrigin-RevId: 954ac6405a
3 years ago
Megvii Engine Team
0708bc780c
fix(dnn/cuda): disallow implicit dtype conversion in cublaslt matmul algos
disable tensor op matmul kernels when input and output tensors are in f32 data type to avoid potential accuracy loss
GitOrigin-RevId: 36859cba5a
3 years ago