提交 · 1ce78aa09b73987fdfd6e7872507591ad97f78b5 · MegEngine 天元 / MegEngine

21 3月, 2022 1 次提交
- M
  fix(imperative): destruct dnn handles at last · 1ce78aa0
  由 Megvii Engine Team 提交于 2月 16, 2022
```
GitOrigin-RevId: 7a67c68c55f56ef997758319d5c047c822c89e8b
```
  1ce78aa0
19 3月, 2022 2 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
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
14 3月, 2022 3 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
  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
09 3月, 2022 1 次提交
- M
  feat(dnn): add nearest mode for remap and resize · 04193e3b
  由 Megvii Engine Team 提交于 3月 01, 2022
```
GitOrigin-RevId: 31e7b72a7850d60e1d2992268c798117d8096173
```
  04193e3b
07 3月, 2022 3 次提交
- M
  feat(arm): delete the reduant implement · 93c7e451
  由 Megvii Engine Team 提交于 2月 23, 2022
```
GitOrigin-RevId: ff32a3dc8b33c264956c64ccf730d363057ac501
```
  93c7e451
- M
  feat(fallback): reduce support general intrinsic · e34a642b
  由 Megvii Engine Team 提交于 2月 17, 2022
```
GitOrigin-RevId: f250aa7b2a145a66699636c11e5602f02693ed2a
```
  e34a642b
- M
  feat(fallback): add simd general intrinsic · 10f23778
  由 Megvii Engine Team 提交于 2月 17, 2022
```
GitOrigin-RevId: ad78ba689fe5a1a142964bc508a4da625705cdb8
```
  10f23778
04 3月, 2022 3 次提交
- M
  feat(dnn): differentiate sass kernel with cuda version · 286051ed
  由 Megvii Engine Team 提交于 3月 01, 2022
```
GitOrigin-RevId: 40bb4423b86ec5d0016cffe405ba51c66d502f16
```
  286051ed
- M
  feat(bazel): make bazel gensass depend on cuda toolchain version automatically · f78b60ec
  由 Megvii Engine Team 提交于 11月 24, 2021
```
GitOrigin-RevId: 9433f21a918fea46c8c409f7b173fdac480fc752
```
  f78b60ec
- M
  feat(mgb): show more details for cuda driver api call · f48227c0
  由 Megvii Engine Team 提交于 11月 17, 2021
```
GitOrigin-RevId: 40e63d9dac062615ab533463b46c7dfc7a8c920f
```
  f48227c0
03 3月, 2022 1 次提交
- M
  fix(cuda): fix fp16 tensorcore gemm split k workspace · d8bb3ff5
  由 Megvii Engine Team 提交于 3月 02, 2022
```
GitOrigin-RevId: d04a0e098541e7e1b519bb392966f598cde3639a
```
  d8bb3ff5
27 2月, 2022 6 次提交
- M
  feat(cuda): add fp16 compute 16 kernel · d7b0994a
  由 Megvii Engine Team 提交于 2月 23, 2022
```
GitOrigin-RevId: e03435be021ccf3d8eff357a80d5203e903aca96
```
  d7b0994a
- M
  refactor(cuda): depthwish large kernel · 8a2e92bd
  由 Megvii Engine Team 提交于 2月 21, 2022
```
GitOrigin-RevId: dade8710b4122dce27dcebfd74491e810288a160
```
  8a2e92bd
- M
  feat(cuda): float16 depthwise large kernel conv compute fp32 · 6b8a69d5
  由 Megvii Engine Team 提交于 2月 15, 2022
```
GitOrigin-RevId: 3050d48f2691faeeda4fb054134041cc620b5a35
```
  6b8a69d5
- M
  feat(cuda): support float16 depthwise large kernel conv · bc385b53
  由 Megvii Engine Team 提交于 2月 07, 2022
```
GitOrigin-RevId: fdc1b15fbcb3968e695601bff6b6a953bf66f115
```
  bc385b53
- M
  perf(cuda): speedup conv backward data with small feature map and large filter size · 7d2063e3
  由 Megvii Engine Team 提交于 1月 26, 2022
```
GitOrigin-RevId: 85592bca6b06239a3aa30621a386738553597023
```
  7d2063e3
- M
  perf(cuda): speedup chanwise conv with small feature map and large filter size · 72403e89
  由 Megvii Engine Team 提交于 1月 21, 2022
```
GitOrigin-RevId: e65b2ce85626730ed772bf49acfc45bc812ce166
```
  72403e89
25 2月, 2022 1 次提交
- M
  feat(mge): add conv padding mode · ab6d12ca
  由 Megvii Engine Team 提交于 1月 26, 2022
```
GitOrigin-RevId: 147ced856e196437cfc5b371c91094ebcbba6ea8
```
  ab6d12ca
23 2月, 2022 5 次提交
- M
  feat(dnn/cuda): add implicit bmm kernels for large kernel depthwise convolution backward filter opr · 47fe7663
  由 Megvii Engine Team 提交于 2月 14, 2022
```
GitOrigin-RevId: 932e7689e89f2864884546935ca13f656842f41c
```
  47fe7663
- M
  feat(dnn/cuda): add heuristic rule for implicit batched gemm large kernel dwconv2d kernels · dcc96935
  由 Megvii Engine Team 提交于 2月 11, 2022
```
GitOrigin-RevId: 2d2c213bfdf91e85b2513cafb1dda0f6940199e5
```
  dcc96935
- M
  fix(dnn/cuda): fix ci · 6cefabe7
  由 Megvii Engine Team 提交于 2月 11, 2022
```
GitOrigin-RevId: 8267e5f9ddd5c6813fcfccf8df197f3c8112fa98
```
  6cefabe7
- M
  feat(dnn/cuda): add implicit bmm large kernel dwconv2d dgrad kernels · 888f4e46
  由 Megvii Engine Team 提交于 2月 10, 2022
```
GitOrigin-RevId: fcb7974d62f4cc032544d507144e121bcf01c565
```
  888f4e46
- M
  feat(dnn/cuda): add implicit bmm large kernel dwconv2d fprop impl · 08d8635f
  由 Megvii Engine Team 提交于 2月 07, 2022
```
GitOrigin-RevId: feb09ebb5836d26433c4a82940bb5f22795da381
```
  08d8635f
25 1月, 2022 1 次提交
- M
  perf(aarch64): optimize aarch64 uint16 relayout with block_w==3 · 260923e1
  由 Megvii Engine Team 提交于 1月 24, 2022
```
GitOrigin-RevId: fe6aaaac0cbd594ad80dcf2e9763fce9c99f5a4e
```
  260923e1
19 1月, 2022 3 次提交
- M
  feat(dnn,mgb,imperative): add diag opr implement · 95ac0555
  由 Megvii Engine Team 提交于 11月 30, 2021
```
GitOrigin-RevId: 43016ffa2b99b3b459eea63e43ccb11deaf9eb36
```
  95ac0555
- M
  feat(arm): add arm rnn_cell/lstm_cell/lstm optimized kernel · 39d77fb5
  由 Megvii Engine Team 提交于 1月 04, 2022
```
GitOrigin-RevId: b9bb7352bc1921c7bbc4168fb3cdea6c6bb955c5
```
  39d77fb5
- M
  fix(build): split elemwise_multi_type cpp · f509b1be
  由 Megvii Engine Team 提交于 1月 11, 2022
```
GitOrigin-RevId: 13267e9db6fa3194291965f50fe08eb892815e8a
```
  f509b1be
18 1月, 2022 3 次提交
- M
  fix(mgb/cuda-stub): add libcuda-wrap_11.4.h to fit the CUDA11.4 toolchain · 3251f501
  由 Megvii Engine Team 提交于 1月 13, 2022
```
GitOrigin-RevId: efa38f00d15518ed3622eab085a210c5018d782d
```
  3251f501
- M
  feat(dnn/elemwise/arm_common): support part of arm ternary elemwise multithread · ee0b95e9
  由 Megvii Engine Team 提交于 1月 14, 2022
```
BCAST111C_VEC_BCAST111C and BCAST101_VEC_BCAST101

GitOrigin-RevId: 0e26553c90563b6d5f752882749ef049cf9d1d31
```
  ee0b95e9
- M
  feat(mge): add softmax op use cudnn api · cbbca5fb
  由 Megvii Engine Team 提交于 12月 17, 2021
```
GitOrigin-RevId: 7734ebf8c47362ab670e90b2692d7074e64d1631
```
  cbbca5fb
16 1月, 2022 1 次提交
- M
  fix(dnn): add naive lstm kernel · 20b42a8c
  由 Megvii Engine Team 提交于 1月 04, 2022
```
GitOrigin-RevId: f08ef810cf936768a022c10f226d80e499355659
```
  20b42a8c
13 1月, 2022 1 次提交
- M
  test(dnn): fix compute capability requirement for NCHWX test · 82be0aac
  由 Megvii Engine Team 提交于 1月 12, 2022
```
GitOrigin-RevId: d2f8022be112d164cb8ff1f90dbe15a4ec1bda04
```
  82be0aac
12 1月, 2022 2 次提交
- M
  fix(mgb): change caffepooling log level · 3b41840b
  由 Megvii Engine Team 提交于 12月 31, 2021
```
GitOrigin-RevId: 290d243ef52fee14d2b241344a255ed9e0a99360
```
  3b41840b
- M
  feat(mgb/opr): add dropout kernel · 19993070
  由 Megvii Engine Team 提交于 1月 01, 2022
```
GitOrigin-RevId: d248bd20050c030196cf9febeefed8c6e7a5b06b
```
  19993070

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