提交 · 09dab38748edcf52688a883ea3447c9d360fd587 · MegEngine 天元 / MegEngine

01 4月, 2022 1 次提交
- M
  feat(cuda): support int1 simplewq conv · 09dab387
  由 Megvii Engine Team 提交于 3月 11, 2022
```
GitOrigin-RevId: 9c37c41bc7e450f3df81e6059603101de3f14416
```
  09dab387
26 11月, 2021 1 次提交
- M
  fix(dnn/cuda): fix cudnn_conv algo of conv_bias opr for fp16 add z cases · ba2f0c2e
  由 Megvii Engine Team 提交于 11月 16, 2021
```
GitOrigin-RevId: b29b009de0803d0d8c4c6f1995ed10f2920652a8
```
  ba2f0c2e
25 11月, 2021 1 次提交
- M
  feat(dnn): use ref ptr interface for all backends · c85631aa
  由 Megvii Engine Team 提交于 11月 24, 2021
```
GitOrigin-RevId: f65feae5ccd9d1fa97eb8831ec0685d98f24e9a7
```
  c85631aa
15 11月, 2021 1 次提交
- M
  fix(mgb/dnn): fix cudnnConvBiasActivation crash on nchw32 int8 with oc > 256 · 9b4cd92b
  由 Megvii Engine Team 提交于 9月 30, 2021
```
GitOrigin-RevId: 20c0b90575ece88da0d1aafd2ec1751b575137ea
```
  9b4cd92b
15 10月, 2021 1 次提交

fix(dnn/cuda): fix cudnn conv impl for nchw4_nchw hybrid layout · 10af44ab

由 Megvii Engine Team 提交于 10月 14, 2021

the conv_bias algo *_IMPLICIT_GEMM in cudnn less than 8.0.0 is disabled due to the incorrect result for int8x4->f32 configs

GitOrigin-RevId: 7cc52d0a85c5ba345af52c61534e7e82f42cc088

10af44ab

12 10月, 2021 1 次提交
- M
  style(all): reformat c++ code · 369c2ccc
  由 Megvii Engine Team 提交于 10月 12, 2021
```
GitOrigin-RevId: 3ffd1b211f140e8ca05661a3c71801083ae4a951
```
  369c2ccc
26 9月, 2021 1 次提交

Revert "perf(cuda/conv): cache serval cudnn api" · de363c04

由 Megvii Engine Team 提交于 9月 15, 2021

This reverts commit 188c62cdd65ba27793b0af89164c6eee0122b21f.

GitOrigin-RevId: 92a82b8cd9f1bed053c7be89fa61ffb05acd1279

de363c04

30 7月, 2021 1 次提交
- M
  feat(mgb): cuda conv support nhwc format and fp16 dtype · 3b452d8c
  由 Megvii Engine Team 提交于 7月 28, 2021
```
GitOrigin-RevId: b8ddcd108a4370a0b093c51bd90ebde0e007cb24
```
  3b452d8c
19 7月, 2021 5 次提交
- 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
  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
21 6月, 2021 3 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): support conv_bias for nchw64 and qint4 · e250afb0
  由 Megvii Engine Team 提交于 3月 23, 2021
```
GitOrigin-RevId: 1c65ba87d7a2276171addcb8883464ada95306a4
```
  e250afb0
- M
  refactor(dnn): refactor lowbit tensor format · 3b9b8780
  由 Megvii Engine Team 提交于 3月 24, 2021
```
GitOrigin-RevId: b646dc085b171589d5dec24650ccf7148f8a03af
```
  3b9b8780
03 6月, 2021 2 次提交
- M
  fix(mgb/dnn): fix cudnn8 convbias · 93894402
  由 Megvii Engine Team 提交于 5月 19, 2021
```
GitOrigin-RevId: 0fdbfd258ce2c83e2cf846c13edff090ce94b0ef
```
  93894402
- M
  perf(cuda/conv): cache serval cudnn api · 5419a95d
  由 Megvii Engine Team 提交于 3月 16, 2021
```
GitOrigin-RevId: 188c62cdd65ba27793b0af89164c6eee0122b21f
```
  5419a95d
25 5月, 2021 1 次提交
- M
  feat(mgb/dnn): add accuracy shake checker · 1e6ef377
  由 Megvii Engine Team 提交于 4月 16, 2021
```
GitOrigin-RevId: 0bb52078a1e338f2ff3a0c1887121dc3a0a7c698
```
  1e6ef377
29 3月, 2021 1 次提交

feat(log): opt log · 55974e8c

由 Megvii Engine Team 提交于 3月 19, 2021

* opt log at release mode
* add MGE_OVERRIDE_LOG_LEVEL for runtime debug
//! env to config LogLevel
//!  DEBUG = 0, INFO = 1, WARN = 2, ERROR = 3, NO_LOG = 4
//! for example , export MGE_OVERRIDE_LOG_LEVEL=0, means set LogLevel to DEBUG

GitOrigin-RevId: 16cd674c56445b1f034359a2ac48beb7f86183a4

55974e8c

22 3月, 2021 1 次提交
- M
  fix(dnn/cuda): disable cudnn conv_bias kernels for NCHW4_NCHW tensor format · c69359d0
  由 Megvii Engine Team 提交于 2月 08, 2021
```
due to illegal memory access errors caused by the kernels

GitOrigin-RevId: 584cedeeab86034ca527ba69162df538e01a9460
```
  c69359d0
02 2月, 2021 1 次提交
- M
  chore(mge): update copyright years · 364afec0
  由 Megvii Engine Team 提交于 1月 18, 2021
```
GitOrigin-RevId: 3c0690bcc130e60568ee0bbfafdd1c19dc073273
```
  364afec0
21 1月, 2021 1 次提交
- M
  fix(mgb(dnn)): fix convbias cudnnConvBiasActivation · a3caa5d3
  由 Megvii Engine Team 提交于 1月 10, 2021
```
GitOrigin-RevId: c0e44feffbe91f299227175e25b1e5c88b9b4724
```
  a3caa5d3
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
19 6月, 2020 1 次提交
- M
  fix(dnn/cuda): enable cuda algos for nchw quantized · 32c86211
  由 Megvii Engine Team 提交于 6月 04, 2020
```
GitOrigin-RevId: 4d1e167b86764ea18a0ea45e58491428b778aa74
```
  32c86211
12 5月, 2020 1 次提交
- M
  feat(mge): add bfloat16 support · 0293d58a
  由 Megvii Engine Team 提交于 12月 24, 2019
```
GitOrigin-RevId: a942ce67915dac6b1203c974daee7ad8d0902e0a
```
  0293d58a
25 3月, 2020 1 次提交
- M
  MegEngine: Initial commit of MegEngine. · f91881ff
  由 Megvii Engine Team 提交于 2月 14, 2020
```
GitOrigin-RevId: f0c8338beb9cac953bd2d8b76710790940dc9300
```
  f91881ff

MegEngine 天元 / MegEngine 1 年多 前同步成功

MegEngine 天元 / MegEngine
1 年多前同步成功