提交 · 5f44203d7bbaf34f2fc4a3e82b056b32431e1de7 · MegEngine 天元 / MegEngine

11 1月, 2021 2 次提交
- M
  feat(dnn/cuda): add a cutlass impl for fusing convolution and dimshuffle · 5f44203d
  由 Megvii Engine Team 提交于 11月 17, 2020
```
GitOrigin-RevId: 3fc6faef01202867f54206367d32ab01659326d0
```
  5f44203d
- M
  feat(dnn/cuda): add impl for fusing warp perspective and dimshuffle · 61f917fb
  由 Megvii Engine Team 提交于 11月 21, 2020
```
GitOrigin-RevId: 51e025973f58ba75ea96765bcb8507581c6c1c25
```
  61f917fb
05 1月, 2021 1 次提交
- M
  fix(dnn): forbid pooling window size smaller than padding · eb826422
  由 Megvii Engine Team 提交于 12月 15, 2020
```
GitOrigin-RevId: 9ad61c409d741ded70c88473374e1de426b27008
```
  eb826422
18 12月, 2020 5 次提交
- M
  chore(winograd): remove winograd transform code · fc0fcd2f
  由 Megvii Engine Team 提交于 11月 13, 2020
```
GitOrigin-RevId: 78c3cfceae019cc75d74b21df4a845d685296fc6
```
  fc0fcd2f
- M
  fix(dnn): fix opencl algo search · d1adc9a2
  由 Megvii Engine Team 提交于 12月 16, 2020
```
GitOrigin-RevId: 25997d0ef168769b18f06434b32a99739f857bb0
```
  d1adc9a2
- M
  fix(dnn/test): delete large size in ARM_COMMON.FP32_GEVM · 7e2b2dbf
  由 Megvii Engine Team 提交于 12月 14, 2020
```
GitOrigin-RevId: 581ef438165d2c879af4583e6a17105da15570bb
```
  7e2b2dbf
- M
  feat(imperative): auto generated opdef header and python binding · 69e3e322
  由 Megvii Engine Team 提交于 12月 18, 2020
```
GitOrigin-RevId: d2f22ad5fe0b15f45afa1ea31af8874e8b18fef9
```
  69e3e322
- M
  fix(build/windows/cuda/llvm): fix windows bazel build with cuda · 0398a786
  由 Megvii Engine Team 提交于 12月 11, 2020
```
* Adapt to the new version llvm/clang-11
* fix windows bazel build with cuda
* add windows bazel build cuda ci
* opt windows bazel ci scripts

GitOrigin-RevId: 6ea7c66585348590d8f8a34150853c776c16352c
```
  0398a786
11 12月, 2020 6 次提交
- M
  feat(dnn): add cuda preprocess fusion · 3bf73ff1
  由 Megvii Engine Team 提交于 12月 01, 2020
```
GitOrigin-RevId: d789c99e59ce713a075061aacf6acdba78af43d3
```
  3bf73ff1
- M
  feat(dnn/aarch64): add quantizeds4 matmul int4x4x16_k8x8x8 · 86cf7490
  由 Megvii Engine Team 提交于 9月 16, 2020
```
GitOrigin-RevId: 781290024466459c292cb46d7c97c5d39454525f
```
  86cf7490
- M
  perf(dnn/cuda): change conv_bias heu, prefer dnn chanwise impl, dislike dnn batch gemm conv1x1 · 142f31a8
  由 Megvii Engine Team 提交于 12月 07, 2020
```
GitOrigin-RevId: 323bf6073a7b6e97ce9dfedc8caa49d351e5f96f
```
  142f31a8
- M
  refactor(mgb/cuda): use single implementation of get_device_prop from utils · f214e146
  由 Megvii Engine Team 提交于 12月 07, 2020
```
GitOrigin-RevId: 5cc95472b9f27339380f74a4a2828368af56c038
```
  f214e146
- M
  perf(mgb/cuda): do not call cudaGetDeviceProperties to avoid io traffic · 54e79dd1
  由 Megvii Engine Team 提交于 12月 06, 2020
```
GitOrigin-RevId: 6aa35928c8ec737d244fdb3ca9639ae49b03b284
```
  54e79dd1
- M
  refactor(dnn): refactor opr proxy in test · 98a74e4a
  由 Megvii Engine Team 提交于 12月 02, 2020
```
GitOrigin-RevId: a1d8682e6f6957a212cc8793c2f0cc9b58f2192b
```
  98a74e4a
04 12月, 2020 4 次提交
- M
  feat(dnn): add uint16 support · 7066ad5b
  由 Megvii Engine Team 提交于 11月 23, 2020
```
GitOrigin-RevId: f4c4b1c7b9076dd8c4c13b23753d6f1aa5570ae5
```
  7066ad5b
- M
  refactor(dnn): refactor algo interface, use algoinfo instead of global algorithm · a1877ee0
  由 Megvii Engine Team 提交于 11月 16, 2020
```
GitOrigin-RevId: 479718ac7557af090996757063cbed58d1c78f2e
```
  a1877ee0
- M
  perf(dnn/cuda): enhance performance for pooling forward · 6f5d0feb
  由 Megvii Engine Team 提交于 11月 25, 2020
```
GitOrigin-RevId: 55fb2a9b25f2059daf91bf597a9192ca3094db7c
```
  6f5d0feb
- M
  chore(dnn/test): refactor megdnn arm_common test · 0560a218
  由 Megvii Engine Team 提交于 11月 21, 2020
```
GitOrigin-RevId: 41689103017705eb1c0c933b0406d48c5d669931
```
  0560a218
27 11月, 2020 1 次提交
- M
  feat(dnn): support conv bias activation for nchw4 input tensor format and nchw output tensor format · 6856ce9c
  由 Megvii Engine Team 提交于 11月 12, 2020
```
GitOrigin-RevId: 29cd73f87b57b0ee6bfa608d629b75aed1491df0
```
  6856ce9c
21 11月, 2020 9 次提交
- M
  feat(mbg/core): support bias preprocess in conv_bias · 60c6d59f
  由 Megvii Engine Team 提交于 11月 13, 2020
```
GitOrigin-RevId: d2e1e14d414a9dcb9a6816515100b854eedb8a0e
```
  60c6d59f
- M
  docs(dnn): add comments of weight prerpocess interface · ff8ef9ed
  由 Megvii Engine Team 提交于 11月 11, 2020
```
GitOrigin-RevId: bb496ed219dcb3bbb400617036273d5798cb053e
```
  ff8ef9ed
- M
  ci(midout): fix midout and reopen midout test · 1f75c7ad
  由 Megvii Engine Team 提交于 10月 28, 2020
```
we just test function, do not check size

GitOrigin-RevId: dce5387e836621399ef5c36af7c8f6709ada844a
```
  1f75c7ad
- M
  refactor(dnn): refactor deconv algo · 1e71e0af
  由 Megvii Engine Team 提交于 11月 17, 2020
```
GitOrigin-RevId: 422be792ebc0de98b0ba8ada823e720a1c5a86d8
```
  1e71e0af
- M
  feat(dnn/cuda): support weight preprocessing for cutlass algorithms · 89ad33ae
  由 Megvii Engine Team 提交于 11月 04, 2020
```
GitOrigin-RevId: 7b77579acd8b61864dee06b3bba00d5c1b88cc8e
```
  89ad33ae
- M
  feat(dnn/opr): add megdnn fake quant opr · c03249c0
  由 Megvii Engine Team 提交于 11月 10, 2020
```
GitOrigin-RevId: 5a04b6da2ffefe1f7f76da0bca1fcc66b7389bf1
```
  c03249c0
- M
  feat(dnn/cuda): opt dp4a conv for small channel base on cutlass · 739f927c
  由 Megvii Engine Team 提交于 11月 09, 2020
```
GitOrigin-RevId: 2a74c35f27ea8186d79313ad607bb8fa89e846fb
```
  739f927c
- M
  fix(mkl): fix windows mkl LOG compute exception · 1f8e4075
  由 Megvii Engine Team 提交于 11月 13, 2020
```
GitOrigin-RevId: cd2ebaaec17e143cf496e1888a5d33a2b936210e
```
  1f8e4075
- M
  refactor(dnn/cuda): misc · 4aa277a2
  由 Megvii Engine Team 提交于 11月 06, 2020
```
GitOrigin-RevId: 1f8f91a0cc7778493f5ed11310c19e2a5e080739
```
  4aa277a2
20 11月, 2020 1 次提交
- M
  refactor(dnn): refactor algorithm type interface · f7b2bdae
  由 Megvii Engine Team 提交于 11月 07, 2020
```
GitOrigin-RevId: 843d885f82a42456c8b0f1018290a3b5c04a3f00
```
  f7b2bdae
13 11月, 2020 4 次提交
- M
  refactor(dnn): remove unused costmodel in cuda · 18ec5341
  由 Megvii Engine Team 提交于 11月 10, 2020
```
GitOrigin-RevId: b15f0607b91b138c593d55b847bcc71b3ffee613
```
  18ec5341
- M
  refactor(dnn): remove ProfileCache and matmul algo in x86 · e39f9386
  由 Megvii Engine Team 提交于 11月 10, 2020
```
GitOrigin-RevId: 55a700d7476a8cff404ef990eb39be393eaee5b5
```
  e39f9386
- M
  feat(megdnn/rocm): add bn for rocm backend · 89303cd8
  由 Megvii Engine Team 提交于 11月 04, 2020
```
GitOrigin-RevId: 8bd49599b28847e51735ff57293bc16ff0033924
```
  89303cd8
- M
  feat(megdnn/rocm): add average inclusive mode for pooling · aea829c9
  由 Megvii Engine Team 提交于 11月 05, 2020
```
GitOrigin-RevId: 08162bd1fb42d5dae04c0733d843a9f3595ef7f2
```
  aea829c9
05 11月, 2020 4 次提交
- M
  perf(mge): add opdef for broadcast · 12178011
  由 Megvii Engine Team 提交于 10月 29, 2020
```
GitOrigin-RevId: 92f0af29eb000b3e37f059e83fda52d26f21b383
```
  12178011
- M
  refactor(dnn/arm): refactor CPU heuristic algo selection · 2a3f4d09
  由 Megvii Engine Team 提交于 8月 03, 2020
```
GitOrigin-RevId: 60d2646bb33316411caa18686eec724dc1f6c430
```
  2a3f4d09
- M
  feat(dnn): add nchw_fp32 nchw44_qint8 cuda dct · ba66e1d0
  由 Megvii Engine Team 提交于 10月 23, 2020
```
GitOrigin-RevId: 581e31fc209008418f9821f32f7c71db76f84ddf
```
  ba66e1d0
- M
  refactor(meg/internal): move interal codes back to megbrain · a9f98e9c
  由 Megvii Engine Team 提交于 10月 22, 2020
```
GitOrigin-RevId: b2dbda96be73e27733bada2858011d2c08b221ff
```
  a9f98e9c
30 10月, 2020 1 次提交
- M
  build(3516): fix some cpu flags build failed and fix 3516 ycm · 44b27f0d
  由 Megvii Engine Team 提交于 10月 26, 2020
```
GitOrigin-RevId: a0c73fa18a98e924b45265af736349d76f8dbc75
```
  44b27f0d
26 10月, 2020 2 次提交
- M
  feat(dnn/cuda): add volta dp4a int8 sass kernel · 8764a6c8
  由 Megvii Engine Team 提交于 9月 22, 2020
```
GitOrigin-RevId: 9fefd39678729ec185c1b09c5b4abd88ebbde3a0
```
  8764a6c8
- M
  style(atlas): add comment for async d2d · 3635af62
  由 Megvii Engine Team 提交于 10月 21, 2020
```
GitOrigin-RevId: 606a56ac4ecc7a3b76f2c10a93eef10fe330564a
```
  3635af62

MegEngine 天元 / MegEngine 10 个月 前同步成功

MegEngine 天元 / MegEngine
10 个月前同步成功