提交 · 72a70dd6a7813e6713ed78110efd25520dd38bda · MegEngine 天元 / MegEngine

21 3月, 2022 27 次提交
- M
  perf(imperative): specialize convolution implementation · 72a70dd6
  由 Megvii Engine Team 提交于 3月 14, 2022
```
GitOrigin-RevId: 33634c550f958db7d971c0b0ebfc15b89d19781b
```
  72a70dd6
- M
  refactor(fastrun): decouple fastrun from computing graph · 12a3ef8d
  由 Megvii Engine Team 提交于 3月 09, 2022
```
GitOrigin-RevId: 27abd222950e36287784e58ad56d6819e9c4cb69
```
  12a3ef8d
- M
  fix(mge/dtr): fix dtr problem · 0a6f4a88
  由 Megvii Engine Team 提交于 3月 15, 2022
```
GitOrigin-RevId: 2a703f9ee4ebf8667ac889a73f67c688dfebd9bc
```
  0a6f4a88
- M
  fix(imperative): fix profiler problem · 529b394f
  由 Megvii Engine Team 提交于 3月 16, 2022
```
GitOrigin-RevId: ea0d8a505458cb5b617e4146c5621c0aa4b4fc74
```
  529b394f
- M
  fix(imperative): fix the dtype promote problem when amp · e64536a3
  由 Megvii Engine Team 提交于 3月 11, 2022
```
GitOrigin-RevId: 43e1035fc86bc5d9212f5de00e4cce347940a7dd
```
  e64536a3
- M
  perf(imperative/src): improve dot performance · 2b80806f
  由 Megvii Engine Team 提交于 3月 08, 2022
```
GitOrigin-RevId: 35b5bd164ffde647125f8fa9c0ebf91195dc4f1f
```
  2b80806f
- M
  perf(mge/utils): move astensor1d into C++ · 2f3bc2db
  由 Megvii Engine Team 提交于 3月 11, 2022
```
GitOrigin-RevId: e7c6659020d9db5a4b17f2a35a40ea2a99f7330d
```
  2f3bc2db
- M
  perf(mge/utils): move convert_input into C++ · fa62f6c0
  由 Megvii Engine Team 提交于 3月 10, 2022
```
GitOrigin-RevId: 0d1cd362511d2d423faaeffd9d80710747cf05f2
```
  fa62f6c0
- M
  perf(mge): move Const into C++ · d98be080
  由 Megvii Engine Team 提交于 3月 09, 2022
```
GitOrigin-RevId: 31a443cffdc1b6d1470b5e0fd5ed49ab350cb4ff
```
  d98be080
- M
  perf(mge/functional): speed up Broadcast and Reshape · 1709b394
  由 Megvii Engine Team 提交于 3月 04, 2022
```
GitOrigin-RevId: a72f5460b6966f815449fc44ad6e5ac1e0ec021c
```
  1709b394
- M
  perf(mge/functional): speed up Dimshuffle · 0f736a0a
  由 Megvii Engine Team 提交于 3月 04, 2022
```
GitOrigin-RevId: 8160c9522bc59d06aab1321b87408ecb410b4a81
```
  0f736a0a
- M
  perf(mge/functional): speed up RemoveAxis · 3e5e08b0
  由 Megvii Engine Team 提交于 3月 03, 2022
```
GitOrigin-RevId: 9c5d27fe1d19bb80848ee244f7c75f309b58df6a
```
  3e5e08b0
- M
  perf(mge/functional): speed up AddAxis · a4d473c9
  由 Megvii Engine Team 提交于 3月 03, 2022
```
GitOrigin-RevId: 92a3e1bdd3c4f0d1d68d8571cad78c1c8ea0f634
```
  a4d473c9
- M
  perf(mge/functional): speed up Split · 3e206d89
  由 Megvii Engine Team 提交于 3月 03, 2022
```
GitOrigin-RevId: 43550a0706a2794421de56067a11864c10b85c67
```
  3e206d89
- M
  perf(interpreter): improve interpreter performance · 730ddc2d
  由 Megvii Engine Team 提交于 3月 09, 2022
```
GitOrigin-RevId: 88f51d15f804bdf33e64f7591d84657ab6635571
```
  730ddc2d
- M
  refactor(imperative): move typecvt code of sereval ops to c++ · 729242f9
  由 Megvii Engine Team 提交于 3月 07, 2022
```
GitOrigin-RevId: 4ffaa376c1f8edaaff7a9c9fb14cbe3ddd186515
```
  729242f9
- M
  refactor(imperative): move python code of elemwise/reduce/conv2d/bn to c++ · 3c3fc6f3
  由 Megvii Engine Team 提交于 2月 23, 2022
```
GitOrigin-RevId: 01b532439243aa2e7d40f150fcaa26fded0e4f27
```
  3c3fc6f3
- M
  perf(imperative/src): improve elemwise · 84466261
  由 Megvii Engine Team 提交于 3月 03, 2022
```
GitOrigin-RevId: 78aa487277b20bf08698ef1e100a7d4b0cc4df15
```
  84466261
- M
  perf(imperative): enable memory forwarding for imperative · e400b7ff
  由 Megvii Engine Team 提交于 2月 09, 2022
```
GitOrigin-RevId: 7c1993979c051a1b01e168eefaa03a0386ddc7bc
```
  e400b7ff
- M
  fix(imperative): do not use output_desc in rng ops · 84d1a440
  由 Megvii Engine Team 提交于 3月 03, 2022
```
GitOrigin-RevId: e6a399be171ea93d8b1a79842a6c066b06e3843d
```
  84d1a440
- M
  fix(imperative): destruct dnn handles at last · 1ce78aa0
  由 Megvii Engine Team 提交于 2月 16, 2022
```
GitOrigin-RevId: 7a67c68c55f56ef997758319d5c047c822c89e8b
```
  1ce78aa0
- M
  feat(imperative): add output_descs for apply_on_physical_tensor · 0cb60d64
  由 Megvii Engine Team 提交于 2月 23, 2022
```
GitOrigin-RevId: 5b036c2c5aa0fc11a21ed375597d6a6a4a14bad4
```
  0cb60d64
- M
  refactor(imperative): remove unnecessary reverve in small vector · c7ded2fe
  由 Megvii Engine Team 提交于 2月 17, 2022
```
GitOrigin-RevId: 85c30bc828a65bc6e626ced97095ee3c68a50c31
```
  c7ded2fe
- M
  refactor(imperative): remove some methods in proxy graph · 8c2b916e
  由 Megvii Engine Team 提交于 2月 15, 2022
```
GitOrigin-RevId: 1fb68a1da290dc0de8976c0537acc708bab07fa3
```
  8c2b916e
- M
  refactor(imperative): apply workspace limit hook to mini graph · 2348a963
  由 Megvii Engine Team 提交于 2月 16, 2022
```
GitOrigin-RevId: 27c51f31478e0a20e01368b169edd6ca0963b5ae
```
  2348a963
- M
  perf(imperative): add opr cache for apply_on_physical_tensor · fea46ea9
  由 Megvii Engine Team 提交于 5月 11, 2021
```
GitOrigin-RevId: fc5d5fb34d2379905e1130d9b0572ba596fb9fe4
```
  fea46ea9
- M
  fix(mgb/opr): fix shape cache of NvOF · ea4e6ab9
  由 Megvii Engine Team 提交于 7月 01, 2021
```
GitOrigin-RevId: 456ba478e9f5be21027f73f302291b114cc2c2d6
```
  ea4e6ab9
19 3月, 2022 3 次提交
- M
  fix(cuda): conv algo heuristic choose · 3228fb75
  由 Megvii Engine Team 提交于 3月 17, 2022
```
GitOrigin-RevId: 95c5e7d627ec8a57a8bf9572e4254947149ea6c6
```
  3228fb75
- M
  feat(dnn): cuda nhwc nearest resize support not 1 or 3 channel · 8c415f4e
  由 Megvii Engine Team 提交于 3月 15, 2022
```
GitOrigin-RevId: 764504c34162221af49b34a3eac7d20caa58f215
```
  8c415f4e
- M
  feat(opencl): add OpenCL cache compat level api · 04475744
  由 Megvii Engine Team 提交于 3月 08, 2022
```
GitOrigin-RevId: e7561e6879bcdbfee52d48387c8ec2a7f21a91d6
```
  04475744
18 3月, 2022 1 次提交
- M
  build(flatbuffer/cx2): fix cx2 build and fix uclibc build flatbuffer · 6fb5a343
  由 Megvii Engine Team 提交于 3月 16, 2022
```
GitOrigin-RevId: af851e155f131e88b400add1879077cf82259d16
```
  6fb5a343
16 3月, 2022 1 次提交
- M
  feat(gopt): fuse conv h_swish · 87de704a
  由 Megvii Engine Team 提交于 3月 02, 2022
```
GitOrigin-RevId: a3d12991fbb2e16a5e91b41fe3f3143257515c2a
```
  87de704a
15 3月, 2022 2 次提交
- M
  feat(lite): add example script and some small change for lar · 4adba378
  由 Megvii Engine Team 提交于 3月 03, 2022
```
GitOrigin-RevId: a28ed2f27a8d9fe8fb1ed2960477aeffd35f1857
```
  4adba378
- M
  fix(mge/gm): fix missing dtype checking while attach tensors · 87f00232
  由 Megvii Engine Team 提交于 3月 07, 2022
```
GitOrigin-RevId: f0aaea99b93472b893eeb3ba35c6293c5f15b122
```
  87f00232
14 3月, 2022 5 次提交
- M
  feat(gopt): merger consecutive relayout and dimshuffle to one relayout to optimize CD4 performarce · 3726f5cc
  由 Megvii Engine Team 提交于 2月 25, 2022
```
GitOrigin-RevId: a058776be35b21dfe8be14ac20247a29d6bfb78b
```
  3726f5cc
- M
  feat(gopt): merge consecutive dimshuffle and relayout to one relayout to optimize CD4 performace · 1fead9b6
  由 Megvii Engine Team 提交于 2月 22, 2022
```
GitOrigin-RevId: 16f22baa80ddf228ebc3fc8b33c340ce4635e46f
```
  1fead9b6
- M
  feat(gopt): optimize cd4 pass rule for elemwise and typecvt to let cd4 start as soon as possible · 26d1e4f7
  由 Megvii Engine Team 提交于 2月 22, 2022
```
GitOrigin-RevId: 6580dedca7d9854142fd3cf824604f52e6826093
```
  26d1e4f7
- M
  fix(cuda): fix direct conv speed and memory problem · ac26bdce
  由 Megvii Engine Team 提交于 3月 07, 2022
```
GitOrigin-RevId: 6faeeff3b80b9cd2245268bfaa9c017b1d3bac58
```
  ac26bdce
- M
  feat(cuda): add large kernel direct conv to heuristic algo chooser · f7994683
  由 Megvii Engine Team 提交于 3月 07, 2022
```
GitOrigin-RevId: bc927b6df736ee807b816d1652d602742c591f8e
```
  f7994683
11 3月, 2022 1 次提交
- M
  fix(dnn): fix the sync problem in some kernels · 6dc0c0b9
  由 Megvii Engine Team 提交于 3月 10, 2022
```
GitOrigin-RevId: df3f7dc51bdc1d8ad8073c491b522c769291c5f6
```
  6dc0c0b9

MegEngine 天元 / MegEngine 1 年多 前同步成功

MegEngine 天元 / MegEngine
1 年多前同步成功