提交 · f76a2cc2c6d96b1d563fefd3a46bd41f4ed2bc9b · MegEngine 天元 / MegEngine

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 24 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
- 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
  fix(mgb): remove unnecessary cudnn8 warning · 40085acb
  由 Megvii Engine Team 提交于 6月 24, 2021
```
GitOrigin-RevId: 04cf1bfca93273953941a2f5904dc31b5f56cc4e
```
  40085acb
- 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 15 次提交
- M
  feat(dnn/cuda): add nchw conv u4xs4 support · 24a38781
  由 Megvii Engine Team 提交于 4月 28, 2021
```
GitOrigin-RevId: 5edba47bd96371b0afa99d85645e41af5d2f3180
```
  24a38781
- 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 base class for cutlass uint4 and int4 algos · e661ae90
  由 Megvii Engine Team 提交于 5月 21, 2021
```
GitOrigin-RevId: a4d42f032c7e53f2966016092ba52c091575be77
```
  e661ae90
- 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 reduce_filter_and_update_bias · 14b65e4d
  由 Megvii Engine Team 提交于 5月 10, 2021
```
GitOrigin-RevId: 31b6e6b0abe2790029e63c9f91c64290a1801958
```
  14b65e4d
- 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
  fix(dnn): fixes for int4 · 86b69cac
  由 Megvii Engine Team 提交于 4月 26, 2021
```
GitOrigin-RevId: 845e164fd3143d2092627374167e74f37b952aed
```
  86b69cac
- 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
  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): fallback conv qs4 support channel not aligend to 64 · c218d4b0
  由 Megvii Engine Team 提交于 4月 12, 2021
```
GitOrigin-RevId: f0d080f35cc3194f6d45dba5fc58986b4f6c06ab
```
  c218d4b0
- 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
  perf(dnn/cuda): optimize int4 sass conv main loop and epilogue without fuse_z · cff61a53
  由 Megvii Engine Team 提交于 4月 12, 2021
```
GitOrigin-RevId: 4274e58d64b8de532c03a137cc86eb6274977a2b
```
  cff61a53

MegEngine 天元 / MegEngine 1 年多 前同步成功

MegEngine 天元 / MegEngine
1 年多前同步成功