提交 · 8e5410e41faa37de65bde5651f414b472146148f · MegEngine 天元 / MegEngine

27 2月, 2022 12 次提交
- M
  feat(cuda): add fp16 compute 16 kernel · 8e5410e4
  由 Megvii Engine Team 提交于 2月 23, 2022
```
GitOrigin-RevId: e03435be021ccf3d8eff357a80d5203e903aca96
```
  8e5410e4
- M
  refactor(cuda): depthwish large kernel · 472e2f96
  由 Megvii Engine Team 提交于 2月 21, 2022
```
GitOrigin-RevId: dade8710b4122dce27dcebfd74491e810288a160
```
  472e2f96
- M
  feat(cuda): float16 depthwise large kernel conv compute fp32 · e698ec20
  由 Megvii Engine Team 提交于 2月 15, 2022
```
GitOrigin-RevId: 3050d48f2691faeeda4fb054134041cc620b5a35
```
  e698ec20
- M
  feat(cuda): support float16 depthwise large kernel conv · 48406382
  由 Megvii Engine Team 提交于 2月 07, 2022
```
GitOrigin-RevId: fdc1b15fbcb3968e695601bff6b6a953bf66f115
```
  48406382
- M
  perf(cuda): speedup conv backward data with small feature map and large filter size · 7042f76b
  由 Megvii Engine Team 提交于 1月 26, 2022
```
GitOrigin-RevId: 85592bca6b06239a3aa30621a386738553597023
```
  7042f76b
- M
  perf(cuda): speedup chanwise conv with small feature map and large filter size · 87a2aeeb
  由 Megvii Engine Team 提交于 1月 21, 2022
```
GitOrigin-RevId: e65b2ce85626730ed772bf49acfc45bc812ce166
```
  87a2aeeb
- M
  feat(mge): add conv padding mode · 2293385e
  由 Megvii Engine Team 提交于 1月 26, 2022
```
GitOrigin-RevId: 147ced856e196437cfc5b371c91094ebcbba6ea8
```
  2293385e
- M
  feat(dnn/cuda): add implicit bmm kernels for large kernel depthwise convolution backward filter opr · afe9c4b5
  由 Megvii Engine Team 提交于 2月 14, 2022
```
GitOrigin-RevId: 932e7689e89f2864884546935ca13f656842f41c
```
  afe9c4b5
- M
  feat(dnn/cuda): add heuristic rule for implicit batched gemm large kernel dwconv2d kernels · e8a16929
  由 Megvii Engine Team 提交于 2月 11, 2022
```
GitOrigin-RevId: 2d2c213bfdf91e85b2513cafb1dda0f6940199e5
```
  e8a16929
- M
  fix(dnn/cuda): fix ci · 38067472
  由 Megvii Engine Team 提交于 2月 11, 2022
```
GitOrigin-RevId: 8267e5f9ddd5c6813fcfccf8df197f3c8112fa98
```
  38067472
- M
  feat(dnn/cuda): add implicit bmm large kernel dwconv2d dgrad kernels · 1da58ae1
  由 Megvii Engine Team 提交于 2月 10, 2022
```
GitOrigin-RevId: fcb7974d62f4cc032544d507144e121bcf01c565
```
  1da58ae1
- M
  feat(dnn/cuda): add implicit bmm large kernel dwconv2d fprop impl · 96050073
  由 Megvii Engine Team 提交于 2月 07, 2022
```
GitOrigin-RevId: feb09ebb5836d26433c4a82940bb5f22795da381
```
  96050073
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
11 1月, 2022 2 次提交
- M
  fix(build): split some cpp, which consume two many mem when build · 32717b0c
  由 Megvii Engine Team 提交于 1月 07, 2022
```
make build possible at 8G ddr env, when -j8

GitOrigin-RevId: d0c442b41d6633ef9c6304ecad0b3bddae9908e3
```
  32717b0c
- M
  feat(mgb/opr): add layernorm forward and backward kernel · a9374181
  由 Megvii Engine Team 提交于 12月 08, 2021
```
GitOrigin-RevId: 0cd484e753a4fbfb88cf81ddbde6ad80b844e69d
```
  a9374181
10 1月, 2022 1 次提交
- M
  fix(mgb/src): add tensorRT version check · a404cd7d
  由 Megvii Engine Team 提交于 1月 07, 2022
```
GitOrigin-RevId: 7abfd30caba14fedff28f89978a6a1f88b4b09f3
```
  a404cd7d
07 1月, 2022 2 次提交
- M
  feat(cmake): format all cmake file · c53cad20
  由 Megvii Engine Team 提交于 1月 06, 2022
```
GitOrigin-RevId: 0a4ecab99b0251bd4c4ee5c49d155343c7759c84
```
  c53cad20
- M
  fix(dnn/x86): opt algo order · a5803058
  由 Megvii Engine Team 提交于 1月 06, 2022
```
GitOrigin-RevId: 6dd14f9a9628f5276399dff2b8328020468541b5
```
  a5803058
05 1月, 2022 1 次提交
- M
  fix(mgb/gopt): fix cpu global layout transform fastrun error · 93310c0e
  由 Megvii Engine Team 提交于 12月 27, 2021
```
GitOrigin-RevId: ea254297e5e46134490c9aa9f1e98861493ba015
```
  93310c0e
29 12月, 2021 2 次提交
- M
  perf(arm): optimize arm uint16 relayout with n=4 · c90e0b54
  由 Megvii Engine Team 提交于 12月 27, 2021
```
GitOrigin-RevId: 5779c6b9c15aa52447e32f8d95d1b845c6d21e18
```
  c90e0b54
- M
  feat(dnn): add elemwise multi type support i16xf32 and u8xf32 · f6d99094
  由 Megvii Engine Team 提交于 12月 17, 2021
```
GitOrigin-RevId: 2fe469bb4ec9a0b7d20f88a54d2a87e7ad42385b
```
  f6d99094
28 12月, 2021 1 次提交
- M
  fix(dnn): correct behaviour of floor div for int tensor · d9a46ea4
  由 Megvii Engine Team 提交于 11月 18, 2021
```
GitOrigin-RevId: 1444f69cce7fea7c3bf34206bce239caad6cc1d0
```
  d9a46ea4
24 12月, 2021 1 次提交
- M
  feat(mgb/gopt): global layout transform support opencl · 0ad5eeae
  由 Megvii Engine Team 提交于 12月 23, 2021
```
GitOrigin-RevId: 132605c7d946d403dc2164a71cd3769b29ccfb31
```
  0ad5eeae
22 12月, 2021 1 次提交
- K
  
  feat(mgb/opr): add cell level rnn/lstm and sequence level rnn/lstm · 8f48da7f
  由 kxz@thumt102-1 提交于 10月 09, 2021
  
  8f48da7f
21 12月, 2021 2 次提交
- M
  feat(dnn/check_non_finite): addmul scale to check_non_finite opr · 2881934c
  由 Megvii Engine Team 提交于 12月 07, 2021
```
GitOrigin-RevId: c35a219e52f852b37a311c6de505c6eb08e2ab8a
```
  2881934c
- M
  feat(dnn/src): add images2neibs kernel of opencl and related test · 6bb54099
  由 Megvii Engine Team 提交于 11月 15, 2021
```
GitOrigin-RevId: 82242b7437c0efa920277bd4909354137497c42c
```
  6bb54099
15 12月, 2021 1 次提交
- M
  feat(dnn): add fallback postprocess · 6ce4a344
  由 Megvii Engine Team 提交于 12月 14, 2021
```
GitOrigin-RevId: 4201a0f158b1439cfa9d900c498aff1e242992ed
```
  6ce4a344
10 12月, 2021 1 次提交
- M
  fix(dnn/arm_common): support more monotonous case in arm typecvt for performance · c96dbd29
  由 Megvii Engine Team 提交于 12月 03, 2021
```
GitOrigin-RevId: 9e28a64d93799a04501910f5a05a988715ff4b04
```
  c96dbd29
02 12月, 2021 2 次提交
- M
  perf(dnn): slightly improve arm neon transcendental function performance · ead611e1
  由 Megvii Engine Team 提交于 11月 30, 2021
```
GitOrigin-RevId: 210d88f81e23efd104ff32ddb57c06b39d0e3e03
```
  ead611e1
- M
  fix(mgb/cuda): fix conv error when the input tensor is too large · 0d169524
  由 Megvii Engine Team 提交于 11月 25, 2021
```
GitOrigin-RevId: 1b1d693795e665630a65a28997078694a78cb214
```
  0d169524
01 12月, 2021 1 次提交
- M
  fix(mgb/x86): fix convbias crash on X86 · 02d5f46d
  由 Megvii Engine Team 提交于 11月 30, 2021
```
GitOrigin-RevId: cc7283c6a26cd35445594d2ff2d8e5567f61142b
```
  02d5f46d

MegEngine 天元 / MegEngine 1 年多 前同步成功

MegEngine 天元 / MegEngine
1 年多前同步成功