提交 · e67fefcdca6555368a3d25773d2e4bb8369ae1df · MegEngine 天元 / MegEngine

30 6月, 2021 1 次提交
- H
  
  ci(yml): enable try-import branch invoke ci on push · e67fefcd
  由 huangxinda 提交于 6月 30, 2021
  
  e67fefcd
23 6月, 2021 1 次提交
- M
  
  deps: update cutlass · 4d72e707
  由 Megvii Engine Team 提交于 6月 23, 2021
  
  4d72e707
21 6月, 2021 38 次提交
- M
  feat(mge/dtr): add DTR in computing graph · 355153e1
  由 Megvii Engine Team 提交于 5月 14, 2021
```
GitOrigin-RevId: 8941810319ecc54e6751e0f584156d26d84ab1e2
```
  355153e1
- M
  refactor(sublinear): add SeqModifierBase · 76f4f975
  由 Megvii Engine Team 提交于 5月 24, 2021
```
GitOrigin-RevId: 2d0393be6b950690c5960ac63bd47931a3afb324
```
  76f4f975
- M
  fix(dnn/bn): revise the conditions for inplace flag · f584416a
  由 Megvii Engine Team 提交于 5月 14, 2021
```
GitOrigin-RevId: 59a104bf6a0dd4989c8a24f2314b7232796be818
```
  f584416a
- M
  fix(ci/lite): reopen lite_test build by cmake · a9b60fbf
  由 Megvii Engine Team 提交于 6月 09, 2021
```
as some reason, lite_test need static link lite
when cuda enable on gcc7 and gcc8, if not
cask_trt::AbiInfo::~AbiInfo will double call at
atexit stage, which will lead double free at the
end of test, gcc9 do not have this issue, for
compat all CI env, we use static link!!!

GitOrigin-RevId: 1dc2115948b1d71124e0f7953788d9201af0caef
```
  a9b60fbf
- M
  feat(mgb): add fast run batch size graph option · 2eea0009
  由 Megvii Engine Team 提交于 5月 19, 2021
```
GitOrigin-RevId: 94e333ec805d81a279365a03a9665a69f789f522
```
  2eea0009
- M
  fix(imperative): persistent cache write through on put · 0ac642b5
  由 Megvii Engine Team 提交于 6月 04, 2021
```
GitOrigin-RevId: f9408ae5046a28b06b6de914f430a15c78ab82d9
```
  0ac642b5
- M
  fix(mgb/core): fix dtype and resize modifiers for tensor · 47dcdf3e
  由 Megvii Engine Team 提交于 6月 09, 2021
```
GitOrigin-RevId: a9d95a4cd80bdd3b2fea5f89fcf9712347930bcc
```
  47dcdf3e
- M
  fix(mgb/opr): correct nvof out shape computation · 29f7cdb8
  由 Megvii Engine Team 提交于 6月 08, 2021
```
GitOrigin-RevId: 16bf086e92125cba867d9b935bb487363602014e
```
  29f7cdb8
- M
  feat(ci): add aarch64 linux ci · 71cc814e
  由 Megvii Engine Team 提交于 6月 04, 2021
```
GitOrigin-RevId: 2c0d3a8cc29745e4bceb3bd570a7e67bd1b8b651
```
  71cc814e
- M
  feat(whl/opencl): enable OpenCL in python whl · 31a1f538
  由 Megvii Engine Team 提交于 6月 04, 2021
```
GitOrigin-RevId: a1c34ef40b80b409d387698b52711533cf058f5d
```
  31a1f538
- M
  feat(aarch64/whl): support aarch64 whl · b07f3728
  由 Megvii Engine Team 提交于 6月 03, 2021
```
GitOrigin-RevId: 656a27d62be10321250652904c544fac32504e7c
```
  b07f3728
- M
  fix(mge/distributed): fix the mutli dataloader test error · d8ee0d7b
  由 Megvii Engine Team 提交于 6月 03, 2021
```
GitOrigin-RevId: 86c892591624beb210344600438ba534bb256954
```
  d8ee0d7b
- M
  feat(imperative/python): support pooling mode "average" for avg pool2d module · e275dfec
  由 Megvii Engine Team 提交于 6月 02, 2021
```
GitOrigin-RevId: 9fe442129f1ae9bafd9f3fde58c6aef1626215fd
```
  e275dfec
- M
  fix(core): fix asan error cause by wild thread_pool ptr · 03ab8136
  由 Megvii Engine Team 提交于 6月 04, 2021
```
GitOrigin-RevId: b1c1b452cd78b3db0ca778c1b31c05593dbe9e96
```
  03ab8136
- 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
  fix(gopt): fix nchw64 opt pass · 0fb9cc41
  由 Megvii Engine Team 提交于 5月 20, 2021
```
GitOrigin-RevId: dec18d1ab1b7bd0723395e490c215356f178e44a
```
  0fb9cc41
- 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(mge/autodiff): fix incorrect handling of tuple dy · 01354337
  由 Megvii Engine Team 提交于 6月 03, 2021
```
GitOrigin-RevId: beca8e37117d1b1f24d463f39903a1a74a62d167
```
  01354337
- M
  fix(arm_common/pooling): check mode in pooling algo to avoid wrong use... · 5868d1fe
  由 Megvii Engine Team 提交于 6月 02, 2021
```
fix(arm_common/pooling): check mode in pooling algo to avoid wrong use AVERAGE_COUNT_EXCLUDE_PADDING

GitOrigin-RevId: 7a2d243db7244ef82d1b63d2bbdb92e13ad19a2b
```
  5868d1fe
- 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
  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): 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
  feat(mgb/gopt): add opt pass for nchw64 layout transform · ae6ff2c5
  由 Megvii Engine Team 提交于 4月 02, 2021
```
GitOrigin-RevId: adede7cef6218b1a4871278bf45973ca599ad594
```
  ae6ff2c5
- M
  feat(mgb/gopt): add an opt pass for padding channels to enable fast int8/int4 support on GPU · 63a9bd30
  由 Megvii Engine Team 提交于 3月 31, 2021
```
GitOrigin-RevId: 94c719bb5c5410925f57d626de088b86afed4750
```
  63a9bd30
- 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
  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
- 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

MegEngine 天元 / MegEngine 大约 1 年 前同步成功

MegEngine 天元 / MegEngine
大约 1 年前同步成功