Megvii Engine Team
1e83ab638e
feat(dnn): add channelwise conv for fp16 nchw88
GitOrigin-RevId: 1bb64f82c5
3 years ago
Megvii Engine Team
7b855dc64a
fix(dnn/cuda): fix compilation for windows bazel
GitOrigin-RevId: 2023dea19c
3 years ago
Megvii Engine Team
3abe0b2462
fix(mgb): fix rocm pooling
GitOrigin-RevId: 44876d398e
3 years ago
Megvii Engine Team
16678bb998
fix(dnn): fix_short_cutlass_name_gemm
GitOrigin-RevId: cc0a2db9da
3 years ago
Megvii Engine Team
4c13bc7e1b
feat(dnn/cuda): add nhwc int8 deconv
GitOrigin-RevId: ad361a0f81
3 years ago
Megvii Engine Team
11f022ff7c
feat(dnn/cuda): add nhwc int8 imma conv and conv fuse typecvt
GitOrigin-RevId: 229e1eb4be
3 years ago
Megvii Engine Team
a0231a7920
fix(dnn/cuda): fix algo matmul for conv bwd filter
fix fastrun workspace size not available exception and device OOM error caused by the incorrect workspace size calculation of algo matmul of conv bwd filter
GitOrigin-RevId: de96b4fe11
3 years ago
Megvii Engine Team
56c1b626bf
refactor(dnn): move arch-dependant code to arch.h
GitOrigin-RevId: 52350144b1
3 years ago
Megvii Engine Team
67575d582c
feat(mge/opr): add interpolate bilinear mode
GitOrigin-RevId: f7023a3fd3
3 years ago
Megvii Engine Team
0558b2123d
feat(mge/opr): add interpolate nearest mode
GitOrigin-RevId: d384b87f50
3 years ago
Megvii Engine Team
127870a926
feat(dnn/opencl): add heuristic rule for batched matmul
GitOrigin-RevId: bd152428e6
3 years ago
Megvii Engine Team
c25125e3d2
perf(dnn/cuda): sass int8 epilogue remove shared load
GitOrigin-RevId: 2b49f5069b
3 years ago
Megvii Engine Team
55efc8e197
feat(mgb/gopt): add reformat emitter
GitOrigin-RevId: 937b20a57c
4 years ago
Megvii Engine Team
c9d060307f
feat(dnn/common): add named tensor shape
GitOrigin-RevId: 918928b8ba
4 years ago
Megvii Engine Team
ff0e6be7b9
fix(dnn/cuda): fix cutlass tensorop kernels
do not compile cutlass tensorop kernels, when using cuda version less than 10.2
GitOrigin-RevId: d4c37d5f41
3 years ago
Megvii Engine Team
336761253d
feat(dnn/cuda): add tensorcore matmul for fp16 data type
GitOrigin-RevId: 025c591f75
3 years ago
Megvii Engine Team
2c4ee99227
fix(dnn): short cutlass filename in windows
GitOrigin-RevId: 83a43fdf87
3 years ago
Megvii Engine Team
432592374d
build(dnn/cuda): fix cmake compile dependency for cutlass kernels
GitOrigin-RevId: ebe71f5a12
3 years ago
Megvii Engine Team
cc07b96f82
perf(dnn/relayout): disable copy_last_contiguous when contiguous_size is
small
GitOrigin-RevId: 7bd5e793a4
3 years ago
Megvii Engine Team
d195fdec71
refactor(mgb): refactor has-usable-algo function for global optimizer
GitOrigin-RevId: 6610516650
3 years ago
Megvii Engine Team
604bb2a569
feat(mgb/dnn): add int atomic add for megdnn
GitOrigin-RevId: 00d5d752d3
3 years ago
Megvii Engine Team
eab6afab47
feat(mgb): add padding opr for megbrain
GitOrigin-RevId: 490e0c5d5a
4 years ago
Megvii Engine Team
66c18f6054
fix(ci): fix bazel compile error in new macos
GitOrigin-RevId: d22cc87fec
3 years ago
Megvii Engine Team
c88a4e5b32
fix(mgb): fix get env macro
GitOrigin-RevId: 79c602f79c
3 years ago
Megvii Engine Team
9b4b910dc1
feat(dnn/cuda): integrate cutlass operation table and replace all cutlass wrappers
GitOrigin-RevId: 2a70335441
3 years ago
Megvii Engine Team
b18feaab33
feat(dnn/cuda): use cutlass remove shared load imma conv kernel
GitOrigin-RevId: 0b5574f526
4 years ago
Megvii Engine Team
1af350c6d2
feat(dnn): add fill kernel
GitOrigin-RevId: d2cee3a7a0
3 years ago
Megvii Engine Team
3eb0505f9b
feat(imperative): add support for quantized conv transpose2d
GitOrigin-RevId: ffd6431299
3 years ago
Megvii Engine Team
c68e669530
feat(bazel/windows/xp/sp2/inference): implement inference on windows xp
(os vesion >= sp2) build with bazel
* bazel build support(define __DEPLOY_ON_XP_SP2__ when deploy on xp sp2):
(dbg)./bazel build //brain/megbrain:load_and_run --cpu='x86_windows_xp'
--compiler='clang_cl' -c dbg --copt "-D__DEPLOY_ON_XP_SP2__=1"
(opt)./bazel build //brain/megbrain:load_and_run --cpu='x86_windows_xp'
--compiler='clang_cl' -c opt --copt "-D__DEPLOY_ON_XP_SP2__=1"
* internal behavior:
will define MGB_HAVE_THREAD=0 when enable __DEPLOY_ON_XP_SP2__
* refer to
https://docs.microsoft.com/en-us/cpp/build/configuring-programs-for-windows-xp?view=msvc-160
xp sp2(x86) do not support vc runtime fully, casused by KERNEL32.dll do not
implement some base apis for c++ std function, for example,
std::mutex/std::thread/std::condition_variable as a workround, we will
disable some MegEngine features on xp sp2 env, for exampe, multi-thread etc!
* about DNN_MUTEX/MGB_MUTEX, if your code will build in inference
code (even CPU backends), please replace std::mutex to DNN_MUTEX/MGB_MUTEX,
* about multi-thread, if you code need multi-thread support, please
enable it when MGB_HAVE_THREAD=1
* about test build env status
1: Visual Studio 2019(MSVC version <= 14.26.28801)---- pass
2: Visual Studio 2019(MSVC version > 14.26.28801) ---- failed
caused by this 'new' version will put VCR depends on win7
KERNEL32.DLL, this may be fixed at Visual Studio 2019 later version
but we do not test at this MR merge point
3: Visual Studio 2017 ---------- pass
4: Visual Studio 2014 ---------- pass
GitOrigin-RevId: 65ac48b95e
3 years ago
Megvii Engine Team
3b452d8c16
feat(mgb): cuda conv support nhwc format and fp16 dtype
GitOrigin-RevId: b8ddcd108a
3 years ago
Megvii Engine Team
10bcf75767
feat(dnn/x86): add algo for x86 max pooling for Window size bigger than 10 and S1 under NCHW88
GitOrigin-RevId: 613a18dd91
3 years ago
Megvii Engine Team
ddba5c9674
fix(core): fix nr_threads is zero
GitOrigin-RevId: 0ccbe3c69b
3 years ago
Megvii Engine Team
67f117882b
perf(arm_common): add elemwise unary multithread support
GitOrigin-RevId: 8eac123f67
3 years ago
Megvii Engine Team
3afa3893d7
perf(arm_common): optimize arm common pooling 9x9 and 13x13
GitOrigin-RevId: 33d5a62478
3 years ago
Megvii Engine Team
2c4ff5431b
fix(mgb): fix cudnn ConvolutionBackwardData
GitOrigin-RevId: 1fffc06eaa
3 years ago
Megvii Engine Team
287cab49c2
fix(mgb/sereg): fix rng operator compatibility
GitOrigin-RevId: 66d1694035
3 years ago
Megvii Engine Team
2aba0378b9
refactor(mgb/dnn): fix group conv is_available
GitOrigin-RevId: b279909168
3 years ago
Megvii Engine Team
4a92346b7a
refactor(mgb): refactor group conv3d
GitOrigin-RevId: 15360a3a41
3 years ago
Megvii Engine Team
6ce212d2e0
refactor(mgb): refactor group conv
GitOrigin-RevId: 7afd312690
4 years ago
Megvii Engine Team
f76a2cc2c6
feat(mge/opr): add silu and gelu
GitOrigin-RevId: 75aa42947e
3 years ago
Megvii Engine Team
f8b0f2cb91
build(dnn/cutlass): fix build for cutlass
GitOrigin-RevId: 9aa095fe84
3 years ago
Megvii Engine Team
869a03271b
perf(mgb): disable FoldingConvBiasDimshufflePass in cuda10 for performance
GitOrigin-RevId: d1b95a6f01
3 years ago
Megvii Engine Team
239916a997
fix(mgb/gopt): fix testcase for enable nchw64 pass
GitOrigin-RevId: 2ae8d1608d
4 years ago
Megvii Engine Team
4eda338876
feat(dnn/cuda): generate cutlass kimpls using cmake and bazel
GitOrigin-RevId: da3bcfb85a
4 years ago
Megvii Engine Team
8d248a6a9a
fix(dnn/cuda): fix testcase for fallback nchw qs8 conv
GitOrigin-RevId: 646440db59
4 years ago
Megvii Engine Team
894a2407c2
feat(dnn/cuda): add relayout format kernel for nchw <-> nhwc
GitOrigin-RevId: e11f3e5408
4 years ago
Megvii Engine Team
43c59204df
refactor(dnn/cuda): refactor relayout format kernels
GitOrigin-RevId: ab86e66533
4 years ago
Megvii Engine Team
f41a808694
feat(dnn/cuda): add nhwc int4 conv support
GitOrigin-RevId: 5236b235d0
4 years ago
Megvii Engine Team
5a14a89224
refactor(dnn/cuda): refactor cutlass kernel generator for gemm and gemv
GitOrigin-RevId: 11d78ab227
4 years ago
Megvii Engine Team
b33217d8f0
refactor(dnn/cuda): refactor cutlass kernel generator for deconv operation
GitOrigin-RevId: 88e962a912
4 years ago