提交 · 4c13bc7e1b955f01ad1b1382622ed38e9b891ecf · MegEngine 天元 / MegEngine

01 9月, 2021 2 次提交
- M
  feat(dnn/cuda): add nhwc int8 deconv · 4c13bc7e
  由 Megvii Engine Team 提交于 8月 17, 2021
```
GitOrigin-RevId: ad361a0f812458115f9a7f809345bef6fe693962
```
  4c13bc7e
- M
  feat(dnn/cuda): add nhwc int8 imma conv and conv fuse typecvt · 11f022ff
  由 Megvii Engine Team 提交于 8月 11, 2021
```
GitOrigin-RevId: 229e1eb4be0fe29932117f226523b446c9ca665e
```
  11f022ff
30 8月, 2021 1 次提交

fix(dnn/cuda): fix algo matmul for conv bwd filter · a0231a79

由 Megvii Engine Team 提交于 8月 25, 2021

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: de96b4fe117ed9691d4a5555261e9f10fa8d2ae4

a0231a79

25 8月, 2021 2 次提交
- M
  feat(mge/opr): add interpolate bilinear mode · 67575d58
  由 Megvii Engine Team 提交于 8月 13, 2021
```
GitOrigin-RevId: f7023a3fd381f36e64702893576702d59c5be2c6
```
  67575d58
- M
  feat(mge/opr): add interpolate nearest mode · 0558b212
  由 Megvii Engine Team 提交于 8月 06, 2021
```
GitOrigin-RevId: d384b87f504c7dd2731bb3c618f35f8b70d00ed2
```
  0558b212
24 8月, 2021 2 次提交
- M
  fix(dnn/cuda): fix cutlass tensorop kernels · ff0e6be7
  由 Megvii Engine Team 提交于 8月 16, 2021
```
do not compile cutlass tensorop kernels, when using cuda version less than 10.2

GitOrigin-RevId: d4c37d5f41ff6e20cf712149067594559529e6cf
```
  ff0e6be7
- M
  feat(dnn/cuda): add tensorcore matmul for fp16 data type · 33676125
  由 Megvii Engine Team 提交于 7月 13, 2021
```
GitOrigin-RevId: 025c591f75afcef8fd58034a9cdd1ae8528bbda1
```
  33676125
18 8月, 2021 2 次提交
- M
  perf(dnn/relayout): disable copy_last_contiguous when contiguous_size is · cc07b96f
  由 Megvii Engine Team 提交于 8月 11, 2021
```
small

GitOrigin-RevId: 7bd5e793a4f252510d739ae8b39046cea969cb58
```
  cc07b96f
- M
  refactor(mgb): refactor has-usable-algo function for global optimizer · d195fdec
  由 Megvii Engine Team 提交于 8月 13, 2021
```
GitOrigin-RevId: 66105166505ba81d2ef984bbc8085af5333a1c9b
```
  d195fdec
17 8月, 2021 2 次提交
- M
  feat(mgb/dnn): add int atomic add for megdnn · 604bb2a5
  由 Megvii Engine Team 提交于 7月 22, 2021
```
GitOrigin-RevId: 00d5d752d3f3d91f3fd581e816e3b1280bad4c31
```
  604bb2a5
- M
  feat(mgb): add padding opr for megbrain · eab6afab
  由 Megvii Engine Team 提交于 6月 09, 2021
```
GitOrigin-RevId: 490e0c5d5add99d54d9c9e3095147016b3c191c7
```
  eab6afab
09 8月, 2021 2 次提交
- M
  feat(dnn/cuda): integrate cutlass operation table and replace all cutlass wrappers · 9b4b910d
  由 Megvii Engine Team 提交于 7月 12, 2021
```
GitOrigin-RevId: 2a70335441e8a844dcf3c2d00bbf6db381ad9623
```
  9b4b910d
- M
  feat(dnn/cuda): use cutlass remove shared load imma conv kernel · b18feaab
  由 Megvii Engine Team 提交于 7月 06, 2021
```
GitOrigin-RevId: 0b5574f52669ba88237967d3f00fceca0857b80f
```
  b18feaab
08 8月, 2021 1 次提交
- M
  feat(dnn): add fill kernel · 1af350c6
  由 Megvii Engine Team 提交于 7月 16, 2021
```
GitOrigin-RevId: d2cee3a7a009849c2c608918e2ec51500bafa88b
```
  1af350c6
07 8月, 2021 1 次提交
- M
  feat(imperative): add support for quantized conv transpose2d · 3eb0505f
  由 Megvii Engine Team 提交于 7月 23, 2021
```
GitOrigin-RevId: ffd6431299b2ae008fbdd1eed6458437e6b6a45f
```
  3eb0505f
30 7月, 2021 5 次提交
- M
  feat(mgb): cuda conv support nhwc format and fp16 dtype · 3b452d8c
  由 Megvii Engine Team 提交于 7月 28, 2021
```
GitOrigin-RevId: b8ddcd108a4370a0b093c51bd90ebde0e007cb24
```
  3b452d8c
- M
  fix(mgb): fix cudnn ConvolutionBackwardData · 2c4ff543
  由 Megvii Engine Team 提交于 7月 27, 2021
```
GitOrigin-RevId: 1fffc06eaa6fe66435715ec5a93c86dd37de985e
```
  2c4ff543
- M
  refactor(mgb/dnn): fix group conv is_available · 2aba0378
  由 Megvii Engine Team 提交于 7月 08, 2021
```
GitOrigin-RevId: b2799091689336cfc626315885f3d296fd13e70f
```
  2aba0378
- M
  refactor(mgb): refactor group conv3d · 4a92346b
  由 Megvii Engine Team 提交于 7月 07, 2021
```
GitOrigin-RevId: 15360a3a417b34da99967525de174def66945060
```
  4a92346b
- M
  refactor(mgb): refactor group conv · 6ce212d2
  由 Megvii Engine Team 提交于 6月 30, 2021
```
GitOrigin-RevId: 7afd3126900938661927c047aac87cb4aeaa9f13
```
  6ce212d2
27 7月, 2021 1 次提交
- M
  feat(mge/opr): add silu and gelu · f76a2cc2
  由 Megvii Engine Team 提交于 7月 16, 2021
```
GitOrigin-RevId: 75aa42947e43fc86920cba11bd768dd6ff226249
```
  f76a2cc2
19 7月, 2021 19 次提交
- M
  feat(dnn/cuda): generate cutlass kimpls using cmake and bazel · 4eda3388
  由 Megvii Engine Team 提交于 6月 21, 2021
```
GitOrigin-RevId: da3bcfb85af3f148116b8b6a51f214c6cef8684e
```
  4eda3388
- M
  fix(dnn/cuda): fix testcase for fallback nchw qs8 conv · 8d248a6a
  由 Megvii Engine Team 提交于 6月 23, 2021
```
GitOrigin-RevId: 646440db59f0157a3fdbd8061167f9ac04dbd422
```
  8d248a6a
- M
  feat(dnn/cuda): add relayout format kernel for nchw <-> nhwc · 894a2407
  由 Megvii Engine Team 提交于 6月 04, 2021
```
GitOrigin-RevId: e11f3e54085929ab9919fe2070fcc9e633755b69
```
  894a2407
- M
  refactor(dnn/cuda): refactor relayout format kernels · 43c59204
  由 Megvii Engine Team 提交于 5月 31, 2021
```
GitOrigin-RevId: ab86e6653342ae9f74dd069cfb85aabca6dc637c
```
  43c59204
- M
  feat(dnn/cuda): add nhwc int4 conv support · f41a8086
  由 Megvii Engine Team 提交于 6月 11, 2021
```
GitOrigin-RevId: 5236b235d0310feecac0fbd1dc76ff0755cf9426
```
  f41a8086
- M
  refactor(dnn/cuda): refactor cutlass kernel generator for gemm and gemv · 5a14a892
  由 Megvii Engine Team 提交于 5月 31, 2021
```
GitOrigin-RevId: 11d78ab2270f0720d7d79e186124a1254c467980
```
  5a14a892
- M
  refactor(dnn/cuda): refactor cutlass kernel generator for deconv operation · b33217d8
  由 Megvii Engine Team 提交于 5月 27, 2021
```
GitOrigin-RevId: 88e962a912dc0f31b8b31787a9ab7aa1b38d1264
```
  b33217d8
- M
  refactor(dnn/cuda): refactor kernel generator for cutlass convolution kernels · 4abf7bd3
  由 Megvii Engine Team 提交于 5月 27, 2021
```
GitOrigin-RevId: 7882f9c68c0e0f35159f12a07ccdf027f4540ff1
```
  4abf7bd3
- M
  feat(dnn/cuda): add convolution with i8 input and u4 output · b4687ce8
  由 Megvii Engine Team 提交于 5月 24, 2021
```
GitOrigin-RevId: 8be439abf1f448a6be33ea0e57c48e674c5c94c4
```
  b4687ce8
- M
  fix(dnn/cuda): fix recursive algo search for fallback_nchw_qs8 · 00083d13
  由 Megvii Engine Team 提交于 5月 20, 2021
```
GitOrigin-RevId: 6be2991224bced3a38a17b6b888fd4f324d03f9f
```
  00083d13
- M
  feat(dnn/cuda): add convolution with i8 input and i4 output · 66f70578
  由 Megvii Engine Team 提交于 5月 18, 2021
```
GitOrigin-RevId: 10512645d5d5ac3d985720788760bf8a3855c1f1
```
  66f70578
- M
  feat(gopt/inference): allow Float32 output dtype in EnableNCHW4Pass · 7d3df995
  由 Megvii Engine Team 提交于 6月 11, 2021
```
GitOrigin-RevId: 81100dbaf7d9e8f5aade6c630ead79ea5ba45e52
```
  7d3df995
- M
  fix(dnn/cuda): fix AlgoFallbackNCHWQS8 to support Float32 dst · 633016a9
  由 Megvii Engine Team 提交于 6月 10, 2021
```
GitOrigin-RevId: 06f90f5cf384bc4ddb2f97860e4f530ee9a85705
```
  633016a9
- M
  feat(mge): add SlidingWindowTranspose opr · 43098fb8
  由 Megvii Engine Team 提交于 6月 29, 2021
```
BREAKING CHANGE:

GitOrigin-RevId: 54d726d2fee21f55811a482594ab81fc9fed1e3e
```
  43098fb8
- M
  feat(mge/random): add some random op and remove random/distrbution.py · b078dda9
  由 Megvii Engine Team 提交于 6月 07, 2021
```
GitOrigin-RevId: 4c05ebc2662408c098b0d5ddd42cc39b0d9d97f3
```
  b078dda9
- M
  feat(mgb/opr): add lsq opr · f30c0e06
  由 Megvii Engine Team 提交于 6月 28, 2021
```
GitOrigin-RevId: 45494a2b57fe1dacf498040c8290fcec3c335990
```
  f30c0e06
- M
  feat(dnn): add deterministic max pooling · 1cfdbc56
  由 Megvii Engine Team 提交于 5月 19, 2021
```
GitOrigin-RevId: 9ab4c7a74839b661fe5ed052a20a66b6732c09d0
```
  1cfdbc56
- M
  feat(mgb/opr): add check_has_inf kernel and opr · a5060a2b
  由 Megvii Engine Team 提交于 6月 23, 2021
```
GitOrigin-RevId: 0d042dbfce8baa51245f4189e197bf800347c6b9
```
  a5060a2b
- M
  refactor(mgb): make convolution3D handle noncontiguous tensors · d915c5a3
  由 Megvii Engine Team 提交于 6月 24, 2021
```
GitOrigin-RevId: 3d3c31b02161532637948ba9aec42d161ec05e92
```
  d915c5a3

MegEngine 天元 / MegEngine 11 个月 前同步成功

MegEngine 天元 / MegEngine
11 个月前同步成功