Megvii Engine Team
00ef677249
fix(mgb): remove internal for cambricon and atlas
GitOrigin-RevId: 861e349eb4
4 years ago
Megvii Engine Team
aeffcd5897
feat(dnn/cuda): integrate cutlass nchw32 tensorcore convolution
GitOrigin-RevId: 9d6c48ed99
4 years ago
Megvii Engine Team
6e70fa7a11
feat(dnn/arm): add fp32 asm gemm for a53 a55 and i8i8i16 gemm for a72 a53
GitOrigin-RevId: a049c33f2b
4 years ago
Megvii Engine Team
b778d22523
feat(mgb/fallback): add conv1x1_gemv, conv1x1 and im2col 8x8x16/8x8x32 support bias
GitOrigin-RevId: 3d97fedc8f
4 years ago
Megvii Engine Team
c357db0134
feat(mgb/arm_common): add 8x8x16 nchw44 max pooling
GitOrigin-RevId: ed460adb7a
4 years ago
Megvii Engine Team
7f5f375fda
feat(dnn/arm): add armv7 nchw_nchw44 3x3s2 asm kernel
GitOrigin-RevId: 50ce91e41d
4 years ago
Megvii Engine Team
3931099ea7
fix(dnn/test): fix nchw_nchw44 i8i8i16 benchmark
GitOrigin-RevId: 6a68030fbf
4 years ago
Megvii Engine Team
bcf5691ddf
feat(dnn/arm): add nchw_nchw44 i8i8i16 2x2 3x3 5x5 7x7 s1 s2 conv
GitOrigin-RevId: 8ef1541665
4 years ago
Megvii Engine Team
c7b6ef35c1
feat(dnn/cuda): add warp perspective backward mat idx
GitOrigin-RevId: b4b494bb69
5 years ago
Megvii Engine Team
a773d07678
feat(dnn/arm_common): add nchw44 8x8x16 channel wise conv
stride1 2x2 3x3 5x5 stride2 2x2 3x3 5x5
GitOrigin-RevId: 43d76311c2
4 years ago
Megvii Engine Team
e258812f12
feat(dnn): add bool dtype
GitOrigin-RevId: 98c8a092b4
4 years ago
Megvii Engine Team
7ca3d579db
feat(dnn): make mk4 and mk8 matmul for winograd both on aarch64 and armv7 supports n=1
GitOrigin-RevId: 0f64b9f70f
4 years ago
Megvii Engine Team
f6018422fd
perf(dnn/arm_common): add nchw44 winograd f73
GitOrigin-RevId: 8ed98ab85b
5 years ago
Megvii Engine Team
e1e56988cd
feat(dnn/fallback): add conv1x1 filter preprocess funciton
GitOrigin-RevId: 4bd109f2da
5 years ago
Megvii Engine Team
e05c795b45
refactor(dnn/arm): refactor direct algo in algo selection
GitOrigin-RevId: d195f44dec
4 years ago
Megvii Engine Team
324af87807
feat(dnn/arm): add cpuinfo runtime check for x86 and arm
GitOrigin-RevId: c2020a344e
4 years ago
Megvii Engine Team
8b183f2c70
test(dnn/testcase): fix a testcase bug
GitOrigin-RevId: f6b9e56318
4 years ago
Megvii Engine Team
14a32ae19b
fix(cmake/cross-build): misc fix
1: fix cmake cross-ios failed caused by df118a87
build static lib for APPLE define for XCODE third_party framework including require
2: megbrain_test/megdnn_test build when MGE_INFERENCE_ONLY=ON
now u can build megbrain_test/megdnn_test by:
EXTRA_CMAKE_ARGS="-DMGE_WITH_TEST=ON" ./scripts/cmake-build/xxx.sh
example macos-cross-ios build megdnn_test for IOS by
EXTRA_CMAKE_ARGS="-DMGE_WITH_TEST=ON" ./scripts/cmake-build/cross_build_ios_arm_inference.sh
3: reuse host flatc build when cross build mode
GitOrigin-RevId: 132f4bf893567bdb1d54de506449950513a5841f
4 years ago
Megvii Engine Team
edd7e16701
feat(dnn/fallback): add im2col filterpreprocess function
GitOrigin-RevId: 61c54ad258
5 years ago
Megvii Engine Team
ef267dacf8
fix(megdnn_test/ev300): try run megdnn_test on ev300 board
GitOrigin-RevId: 5c557f082e
4 years ago
Megvii Engine Team
eed54081ab
feat(dnn/arm): add armv7 mk4 i8i8i16 gemm, optimized for A7
GitOrigin-RevId: d2f8290a8d
4 years ago
Megvii Engine Team
4d56371e0b
refactor(dnn/arm): split arm direct kernel to cut compile time
GitOrigin-RevId: b06fba83eb
5 years ago
Megvii Engine Team
fc1ce273b7
fix(dnn/cuda): fix elemwise add cuda int8 bcast
GitOrigin-RevId: 568b60e8c9
4 years ago
Megvii Engine Team
57bc36575f
style(dnn/cuda): format cuda elemwise code
GitOrigin-RevId: 246755ce20
4 years ago
Megvii Engine Team
fff2cdc7bb
feat(dnn/fallback): add winograd weight preprocess
GitOrigin-RevId: 4741298e44
5 years ago
Megvii Engine Team
d37229fa02
feat(dnn): optimize f23 and f63 nchw44 winograd
GitOrigin-RevId: 8569c9dfc6
5 years ago
Megvii Engine Team
3bd8ef3589
feat(mgb/compnode): add atlas compnode
GitOrigin-RevId: 19f3c33003
5 years ago
Megvii Engine Team
1e576e321b
feat(dnn/aarch64-arm_common): add mat_idx warppespective for aarch64/arm_common/naive
GitOrigin-RevId: 9eb0cdda5c
5 years ago
Megvii Engine Team
714cb232bb
feat(dnn): add gemv supports in conv1x1 for NCHW44 and NCHW44_DOT(aarch64 binary size grows 2KB)
GitOrigin-RevId: f8b6d7a1b7
5 years ago
Megvii Engine Team
b8b000db3b
feat(dnn/fallback): fix fallback interface of weight preprocess
GitOrigin-RevId: ca860f487e
5 years ago
Megvii Engine Team
5fb07c9964
fix(dnn/x86): fix cmake error for build x86 gtest
GitOrigin-RevId: 7613190733
5 years ago
Megvii Engine Team
763b57add7
fix(dnn/cuda): fix INTMAX overflow in warp_perspective_cuda
GitOrigin-RevId: d7354e74e2
5 years ago
Megvii Engine Team
2e6e570dfe
feat(dnn/fallback): add armv7 im2col mk4-dot int8 and
nchw44 float 3x3 s2 fuse packb speed up about 10%
GitOrigin-RevId: 3f864cef1d
5 years ago
Megvii Engine Team
7886ff9af0
feat(dnn): add relayout_format for nchw to nchw4 and ic <=4
GitOrigin-RevId: 07f2ee6c5b
5 years ago
Megvii Engine Team
dedb7a3f14
feat(dnn/cuda): add cuda remap
GitOrigin-RevId: 40a2a2ce24
5 years ago
Megvii Engine Team
946a340c3d
feat(ci/midout): opt midout and add midout ci
GitOrigin-RevId: 1e5fe75255
5 years ago
Megvii Engine Team
44c381b6f4
Revert "feat(dnn/naive): workspacebundle support 2D"
This reverts commit 4408bb9e1d
.
GitOrigin-RevId: b5b23a8aae
5 years ago
Megvii Engine Team
cdbe44f8b4
feat(dnn): add gemv supports in conv1x1 with format NCHW
GitOrigin-RevId: 97679e8526
5 years ago
Megvii Engine Team
9e904f683b
fix(dnn): fix can not inline small function with GCC compiler
GitOrigin-RevId: a23605c9e2
5 years ago
Megvii Engine Team
9d5c5c0788
feat(dnn/naive): workspacebundle support 2D
GitOrigin-RevId: 4408bb9e1d
5 years ago
Megvii Engine Team
69fe5ab3b3
feat(dnn/cuda): add conv2d-sass-kernel
GitOrigin-RevId: f284d5a4ce
5 years ago
Megvii Engine Team
786afef461
feat(build): install CMake config module and pkg-config descriptor
Also, upgrade to CMake 3.13.
The commit also contains significant refactors, as otherwise it is not
possible to properly export target `megengine` to
MegEngine-targets.cmake:
1. Optionally use system provided Flatbuffers.
2. Optionally use system provided MKL-DNN (Tested with Debian).
3. Refactor megbrain and megdnn targets into object libraries.
4. Set different path in BUILD_INTERFACE and INSTALL_INTERFACE of
various target_include_directories.
5. Specify PUBLIC/PRIVATE on various target_link_libraries.
GitOrigin-RevId: df118a879e
5 years ago
Megvii Engine Team
4d35397bdf
fix(dnn/fallback): fix conv1x1/im2col usable and fuse-conv-bias get fp32xfp32-->qint8 error
GitOrigin-RevId: 5a3bfedd8a
5 years ago
Megvii Engine Team
6b2760dd72
feat(dnn/fallback): add float32 nchw44 fuse packb 3x3 s2
GitOrigin-RevId: 3b664bb4f5
5 years ago
Megvii Engine Team
2b4b4d66d9
feat(dnn/fallback): add aarch64 mk4 dot 3x3 s1 fuse packb
GitOrigin-RevId: 3e69878d8d
5 years ago
Megvii Engine Team
a1677d7aa9
feat(dnn/arm_common): add fp32 gevm
GitOrigin-RevId: 4d348bbb34
5 years ago
Megvii Engine Team
5d950063cf
feat(dnn): refactor dot gemv for both aarch64 and aarch32
GitOrigin-RevId: 2b98867e45
5 years ago
Megvii Engine Team
53c288a304
fix(dnn/cuda): fix topk grid oversize
GitOrigin-RevId: d3c811a034
5 years ago
Megvii Engine Team
124767b4f8
fix(dnn/fallback): fix mk4_dot test after remove mk4_dot_8x6x4 matmul
GitOrigin-RevId: e3a12cf9b3
5 years ago
Megvii Engine Team
34659c2ea4
fix(mgb/dnn): remove armv7 matmul mk4dot block 8x6
GitOrigin-RevId: 4c746ef228
5 years ago