提交 · 3977b7aa0bccb09b1b29780990d7c774ca0e1ae0 · MegEngine 天元 / MegEngine

24 9月, 2021 2 次提交
- M
  feat(mgb/shuffle): add shuffle opr · 3977b7aa
  由 Megvii Engine Team 提交于 7月 21, 2021
```
GitOrigin-RevId: 80490a6f848d524111bee097f11b591b5a3956c8
```
  3977b7aa
- M
  fix(dnn/reduce): fix reduce_mean o16c32 is incorrect for large tensor · 17371e79
  由 Megvii Engine Team 提交于 9月 04, 2021
```
GitOrigin-RevId: ebf03d814a893efca9dd9e09bb58001c22093fd4
```
  17371e79
18 9月, 2021 2 次提交
- M
  feat(mgb/gopt): add reformat manager · c33126ab
  由 Megvii Engine Team 提交于 7月 16, 2021
```
GitOrigin-RevId: b9791b131a996f4f7c55173294abf0715fac97fc
```
  c33126ab
- M
  feat(mgb/dnn): add conv1x1 algo for matrix mul · 8b40f577
  由 Megvii Engine Team 提交于 8月 12, 2021
```
GitOrigin-RevId: 585b2c045a894abfbca5606ce620df6418d97787
```
  8b40f577
17 9月, 2021 5 次提交
- M
  feat(dnn): add an get_all_algorithms_safe interface · d69b5903
  由 Megvii Engine Team 提交于 9月 06, 2021
```
GitOrigin-RevId: e3734e4531c2f91205814ec6b490f5d6392f3fa3
```
  d69b5903
- M
  refactor(dnn/rocm): update hip license header · 103d7f33
  由 Megvii Engine Team 提交于 4月 30, 2021
```
GitOrigin-RevId: 79d684755d31a982446a2c136ce800d70afae91f
```
  103d7f33
- M
  feat(dnn/rocm): add adaptive pooling opr · 5aa52d38
  由 Megvii Engine Team 提交于 8月 19, 2021
```
GitOrigin-RevId: e844b3e7706af483bbcf9d5cd20f4c009a7f7e27
```
  5aa52d38
- M
  feat(dnn/rocm): add topk opr · 323a4642
  由 Megvii Engine Team 提交于 8月 16, 2021
```
GitOrigin-RevId: 5ecb07985491359bb8063427cc142fbcec3da943
```
  323a4642
- M
  feat(dnn/rocm): add argsort opr · f4784f4a
  由 Megvii Engine Team 提交于 8月 18, 2021
```
GitOrigin-RevId: b4c3eb4707bb4def8739d8ece8c26d4aa5af4147
```
  f4784f4a
16 9月, 2021 1 次提交
- M
  fix(dnn/cuda): fix elemwise and relayout int4 bug when last shape is 1 · 8b94f493
  由 Megvii Engine Team 提交于 9月 10, 2021
```
GitOrigin-RevId: e7d64c49871032deeda4176289f0457d4b9d85b8
```
  8b94f493
15 9月, 2021 1 次提交
- M
  feat(mgb): add arm resize nchwxx and naive nearest interp · bc9cfc27
  由 Megvii Engine Team 提交于 8月 29, 2021
```
GitOrigin-RevId: d5fbd59a3021ac82527b1f093d97a828b3d725a4
```
  bc9cfc27
11 9月, 2021 1 次提交
- M
  feat(mgb): support fp16 nhwc backward · 722aecd4
  由 Megvii Engine Team 提交于 8月 18, 2021
```
GitOrigin-RevId: 954ac6405a2e7b8b6719916c57e31a80f623b0c1
```
  722aecd4
10 9月, 2021 1 次提交

fix(dnn/cuda): disallow implicit dtype conversion in cublaslt matmul algos · 0708bc78

由 Megvii Engine Team 提交于 8月 31, 2021

disable tensor op matmul kernels when input and output tensors are in f32 data type to avoid potential accuracy loss

GitOrigin-RevId: 36859cba5a526a7778e12c03ac32815144fe0505

0708bc78

06 9月, 2021 1 次提交
- M
  feat(dnn): add channelwise conv for fp16 nchw88 · 1e83ab63
  由 Megvii Engine Team 提交于 8月 15, 2021
```
GitOrigin-RevId: 1bb64f82c5cc4e512e9141a2bebd35ee8033fc5d
```
  1e83ab63
01 9月, 2021 2 次提交
- M
  feat(dnn/cuda): add nhwc int8 deconv · 4c13bc7e
  由 Megvii Engine Team 提交于 8月 17, 2021
```
GitOrigin-RevId: ad361a0f812458115f9a7f809345bef6fe693962
```
  4c13bc7e
- M
  feat(dnn/cuda): add nhwc int8 imma conv and conv fuse typecvt · 11f022ff
  由 Megvii Engine Team 提交于 8月 11, 2021
```
GitOrigin-RevId: 229e1eb4be0fe29932117f226523b446c9ca665e
```
  11f022ff
25 8月, 2021 2 次提交
- M
  feat(mge/opr): add interpolate bilinear mode · 67575d58
  由 Megvii Engine Team 提交于 8月 13, 2021
```
GitOrigin-RevId: f7023a3fd381f36e64702893576702d59c5be2c6
```
  67575d58
- M
  feat(mge/opr): add interpolate nearest mode · 0558b212
  由 Megvii Engine Team 提交于 8月 06, 2021
```
GitOrigin-RevId: d384b87f504c7dd2731bb3c618f35f8b70d00ed2
```
  0558b212
24 8月, 2021 4 次提交
- M
  perf(dnn/cuda): sass int8 epilogue remove shared load · c25125e3
  由 Megvii Engine Team 提交于 8月 04, 2021
```
GitOrigin-RevId: 2b49f5069b3183e3dd1dc5c9f0bb0a64095022e1
```
  c25125e3
- M
  feat(dnn/common): add named tensor shape · c9d06030
  由 Megvii Engine Team 提交于 6月 30, 2021
```
GitOrigin-RevId: 918928b8ba673cc1c0fa97aa3bb3d4459203fd8f
```
  c9d06030
- M
  fix(dnn/cuda): fix cutlass tensorop kernels · ff0e6be7
  由 Megvii Engine Team 提交于 8月 16, 2021
```
do not compile cutlass tensorop kernels, when using cuda version less than 10.2

GitOrigin-RevId: d4c37d5f41ff6e20cf712149067594559529e6cf
```
  ff0e6be7
- M
  feat(dnn/cuda): add tensorcore matmul for fp16 data type · 33676125
  由 Megvii Engine Team 提交于 7月 13, 2021
```
GitOrigin-RevId: 025c591f75afcef8fd58034a9cdd1ae8528bbda1
```
  33676125
17 8月, 2021 1 次提交
- M
  feat(mgb): add padding opr for megbrain · eab6afab
  由 Megvii Engine Team 提交于 6月 09, 2021
```
GitOrigin-RevId: 490e0c5d5add99d54d9c9e3095147016b3c191c7
```
  eab6afab
09 8月, 2021 1 次提交
- M
  feat(dnn/cuda): use cutlass remove shared load imma conv kernel · b18feaab
  由 Megvii Engine Team 提交于 7月 06, 2021
```
GitOrigin-RevId: 0b5574f52669ba88237967d3f00fceca0857b80f
```
  b18feaab
08 8月, 2021 1 次提交
- M
  feat(dnn): add fill kernel · 1af350c6
  由 Megvii Engine Team 提交于 7月 16, 2021
```
GitOrigin-RevId: d2cee3a7a009849c2c608918e2ec51500bafa88b
```
  1af350c6
07 8月, 2021 1 次提交
- M
  feat(imperative): add support for quantized conv transpose2d · 3eb0505f
  由 Megvii Engine Team 提交于 7月 23, 2021
```
GitOrigin-RevId: ffd6431299b2ae008fbdd1eed6458437e6b6a45f
```
  3eb0505f
30 7月, 2021 8 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
- M
  feat(dnn/x86): add algo for x86 max pooling for Window size bigger than 10 and S1 under NCHW88 · 10bcf757
  由 Megvii Engine Team 提交于 7月 27, 2021
```
GitOrigin-RevId: 613a18dd916f575fcd29448c2048dde9dae70d1e
```
  10bcf757
- M
  fix(core): fix nr_threads is zero · ddba5c96
  由 Megvii Engine Team 提交于 7月 29, 2021
```
GitOrigin-RevId: 0ccbe3c69b5e94afe2068700252c23fca57ac910
```
  ddba5c96
- M
  perf(arm_common): add elemwise unary multithread support · 67f11788
  由 Megvii Engine Team 提交于 7月 27, 2021
```
GitOrigin-RevId: 8eac123f67224e283b368c515bf0b8e7ef565158
```
  67f11788
- M
  perf(arm_common): optimize arm common pooling 9x9 and 13x13 · 3afa3893
  由 Megvii Engine Team 提交于 7月 27, 2021
```
GitOrigin-RevId: 33d5a624784a5dde61b6c9cfe461297a0f2950fe
```
  3afa3893
- M
  refactor(mgb/dnn): fix group conv is_available · 2aba0378
  由 Megvii Engine Team 提交于 7月 08, 2021
```
GitOrigin-RevId: b2799091689336cfc626315885f3d296fd13e70f
```
  2aba0378
- M
  refactor(mgb): refactor group conv3d · 4a92346b
  由 Megvii Engine Team 提交于 7月 07, 2021
```
GitOrigin-RevId: 15360a3a417b34da99967525de174def66945060
```
  4a92346b
- M
  refactor(mgb): refactor group conv · 6ce212d2
  由 Megvii Engine Team 提交于 6月 30, 2021
```
GitOrigin-RevId: 7afd3126900938661927c047aac87cb4aeaa9f13
```
  6ce212d2
23 7月, 2021 1 次提交
- M
  perf(mgb): disable FoldingConvBiasDimshufflePass in cuda10 for performance · 869a0327
  由 Megvii Engine Team 提交于 7月 13, 2021
```
GitOrigin-RevId: d1b95a6f01ba73f98c0094e00fee3e61e9139628
```
  869a0327
19 7月, 2021 5 次提交
- M
  fix(dnn/cuda): fix testcase for fallback nchw qs8 conv · 8d248a6a
  由 Megvii Engine Team 提交于 6月 23, 2021
```
GitOrigin-RevId: 646440db59f0157a3fdbd8061167f9ac04dbd422
```
  8d248a6a
- M
  feat(dnn/cuda): add relayout format kernel for nchw <-> nhwc · 894a2407
  由 Megvii Engine Team 提交于 6月 04, 2021
```
GitOrigin-RevId: e11f3e54085929ab9919fe2070fcc9e633755b69
```
  894a2407
- M
  feat(dnn/cuda): add nhwc int4 conv support · f41a8086
  由 Megvii Engine Team 提交于 6月 11, 2021
```
GitOrigin-RevId: 5236b235d0310feecac0fbd1dc76ff0755cf9426
```
  f41a8086
- 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
  feat(mge): add SlidingWindowTranspose opr · 43098fb8
  由 Megvii Engine Team 提交于 6月 29, 2021
```
BREAKING CHANGE:

GitOrigin-RevId: 54d726d2fee21f55811a482594ab81fc9fed1e3e
```
  43098fb8

MegEngine 天元 / MegEngine 1 年多 前同步成功

MegEngine 天元 / MegEngine
1 年多前同步成功