提交 · 3eb0505f9b9db06160fc1f4d2d8724856da9d46b · MegEngine 天元 / MegEngine

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 4 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
  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
23 7月, 2021 1 次提交
- M
  perf(mgb): disable FoldingConvBiasDimshufflePass in cuda10 for performance · 869a0327
  由 Megvii Engine Team 提交于 7月 13, 2021
```
GitOrigin-RevId: d1b95a6f01ba73f98c0094e00fee3e61e9139628
```
  869a0327
19 7月, 2021 14 次提交
- 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
  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
  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
- M
  refactor(mgb): make conv-backward-filter handle noncontiguous tensors · d04cd67f
  由 Megvii Engine Team 提交于 6月 11, 2021
```
GitOrigin-RevId: 44c586f912a235f5610f7ef40a815181726aa153
```
  d04cd67f
- M
  refactor(mgb): make conv-backward-data handle noncontiguous tensors · 44376f70
  由 Megvii Engine Team 提交于 6月 11, 2021
```
GitOrigin-RevId: 0a8f66f9d378b6466bc383a94c57ec80bcc5cb74
```
  44376f70
- M
  refactor(mgb): make conv handle noncontiguous tensors · 7b2a76d1
  由 Megvii Engine Team 提交于 6月 09, 2021
```
GitOrigin-RevId: 86282709b3013ae44e13f74639a1c45a0dad97b0
```
  7b2a76d1
- M
  feat(dnn/cuda): topk support fp16 · b87af9f7
  由 Megvii Engine Team 提交于 5月 28, 2021
```
GitOrigin-RevId: c6610d4cf04f258fd13f420e5301c273e399d67f
```
  b87af9f7
21 6月, 2021 17 次提交
- M
  feat(dnn/cuda): add nhwc 4bit warp perspective · 606540be
  由 Megvii Engine Team 提交于 5月 24, 2021
```
GitOrigin-RevId: fbec4a4a1f7a5dd184b4e76d60b26a6cac8a24ef
```
  606540be
- M
  feat(dnn/cuda): add nhwc int4 pooling · 1e601943
  由 Megvii Engine Team 提交于 5月 25, 2021
```
GitOrigin-RevId: 9cf14cde4e78d71bd8220bb6612e6acc5734d111
```
  1e601943
- M
  feat(dnn/cuda): add cutlass impls for uint4 x int4 conv bias · 319436dd
  由 Megvii Engine Team 提交于 5月 17, 2021
```
GitOrigin-RevId: cf4536855ac3faf5a929b1077dac91092b2f008f
```
  319436dd
- M
  feat(dnn/cuda): add cutlass impls for int4 conv bias · d28eba4e
  由 Megvii Engine Team 提交于 5月 07, 2021
```
GitOrigin-RevId: 878bb8c95511de578cb5c880a412a3e1c9f0dea9
```
  d28eba4e
- M
  feat(dnn/cuda): add cuda uint4 pooling · 2d4e62ef
  由 Megvii Engine Team 提交于 5月 07, 2021
```
GitOrigin-RevId: a7289772068d08deef1021f71984cb4ecfdc6702
```
  2d4e62ef
- M
  feat(dnn/cuda): add cuda uint warp perspective · 19919384
  由 Megvii Engine Team 提交于 5月 17, 2021
```
GitOrigin-RevId: 2aec72010f81ad92b726924fcdc4b65069ec0cab
```
  19919384
- M
  feat(dnn/cuda): add conv u4xs4 sass kernel · 4a802d21
  由 Megvii Engine Team 提交于 4月 16, 2021
```
GitOrigin-RevId: 4defcf5f1f33f91c5d92df0282f337e9080dc322
```
  4a802d21
- M
  perf(dnn/cuda): add sass int4 128x128 · adf75a29
  由 Megvii Engine Team 提交于 4月 21, 2021
```
GitOrigin-RevId: 1bc54821023814c3c80b5fd22c24ab0007dd6203
```
  adf75a29
- M
  feat(dnn/cuda): support warp perspective/pooling op when channel not aligned to 64 · 8da2f698
  由 Megvii Engine Team 提交于 4月 12, 2021
```
GitOrigin-RevId: 39f29ec990ebb4c61e8d217911976b51810df34c
```
  8da2f698
- M
  feat(dnn/cuda): support transforming layout between nchw and nchw64 when channel not aligned to 64 · 4fe68ac9
  由 Megvii Engine Team 提交于 4月 12, 2021
```
GitOrigin-RevId: e9ecbcf2e25ce093e61aa6c8bab8909974e288d4
```
  4fe68ac9
- M
  fix(dnn/cuda): fix int4 epilogue stg bug · 56e863b7
  由 Megvii Engine Team 提交于 4月 19, 2021
```
GitOrigin-RevId: e86da9a8a8d3d55931d55b12673c99f940a390d0
```
  56e863b7
- M
  feat(dnn/cuda): add cuda elemwise int4 · 12a0e615
  由 Megvii Engine Team 提交于 4月 06, 2021
```
GitOrigin-RevId: 8a9aaec3281c154470b15ed79ba5be288d7a752a
```
  12a0e615
- M
  feat(dnn): warp perspective support int4 · df1af59b
  由 Megvii Engine Team 提交于 3月 29, 2021
```
GitOrigin-RevId: 826a43b349a5c0b135a0b836de7863ba55dc6d3d
```
  df1af59b
- M
  feat(dnn/cuda): add cuda int4 pooling · 2398df07
  由 Megvii Engine Team 提交于 3月 31, 2021
```
GitOrigin-RevId: 14ed4e6f0095231ca87cace41842b32df18d0818
```
  2398df07
- M
  feat(dnn/cuda): support conv_bias for nchw64 and qint4 · e250afb0
  由 Megvii Engine Team 提交于 3月 23, 2021
```
GitOrigin-RevId: 1c65ba87d7a2276171addcb8883464ada95306a4
```
  e250afb0
- M
  feat(dnn/cuda): add relayout format when width is an odd number · 8fef78d0
  由 Megvii Engine Team 提交于 3月 22, 2021
```
GitOrigin-RevId: f059f1f56dd66c33633118c893027ddd50ac8f1d
```
  8fef78d0
- M
  test(dnn/cuda): add testcase for transforming tensor layout between nchw and nchw64 · 19a554d6
  由 Megvii Engine Team 提交于 3月 18, 2021
```
GitOrigin-RevId: 75d579635ad177d9391b8da6ca45fab1086d3f6a
```
  19a554d6
03 6月, 2021 3 次提交
- M
  feat(dnn/cuda): support float16 for index_incr_multi_axis_vec · 23032f50
  由 Megvii Engine Team 提交于 5月 25, 2021
```
GitOrigin-RevId: c2ae93d568892d1af6a602aed3ed7c60f9dba1bd
```
  23032f50
- M
  fix(mgb/dnn): fix cudnn8 convbias · 93894402
  由 Megvii Engine Team 提交于 5月 19, 2021
```
GitOrigin-RevId: 0fdbfd258ce2c83e2cf846c13edff090ce94b0ef
```
  93894402
- M
  feat(dnn/cuda): add correlation kernel · 1997b1a2
  由 Megvii Engine Team 提交于 4月 26, 2021
```
GitOrigin-RevId: 25e58b61e66856e7926afaa7d4e6c8147e800ae9
```
  1997b1a2

MegEngine 天元 / MegEngine 1 年多 前同步成功

MegEngine 天元 / MegEngine
1 年多前同步成功