Code

Add ResNetUnit Python API#35426

Code

Merged

Xreki merged 60 commits intoPaddlePaddle:developPaddlePaddle/Paddle:developfrom ZzSean:cudnn_v8_fusionZzSean/Paddle:cudnn_v8_fusionCopy head branch name to clipboardOct 15, 2021

+289-14

Commits

Commits on Sep 24, 2021

• first commit

  ZzSean

  committed
  57a6edc

  Copy full SHA for 57a6edc

• Support full compile

  ZzSean

  committed
  ae796a4

  Copy full SHA for ae796a4

• add mean and var

  ZzSean

  committed
  889d430

  Copy full SHA for 889d430

• modify based on 'resnet_unit_op' branch

  ZzSean

  committed
  01841a6

  Copy full SHA for 01841a6

• modify bn test

  ZzSean

  committed
  c65b5df

  Copy full SHA for c65b5df

• sync cudnn_bn_add_relu_test

  ZzSean

  committed
  af42837

  Copy full SHA for af42837

• support bn+add+relu backward and compile success

  ZzSean

  committed
  a59f3f3

  Copy full SHA for a59f3f3

• add OpMaker and InferShape for backward

  ZzSean

  committed
  8b5ca97

  Copy full SHA for 8b5ca97

Commits on Sep 26, 2021

• add conv backward

  ZzSean

  committed
  b519f8c

  Copy full SHA for b519f8c

• fix run error

  ZzSean

  committed
  b0165ee

  Copy full SHA for b0165ee

• add param dtype check

  ZzSean

  committed
  da04c42

  Copy full SHA for da04c42

• fix run error

  ZzSean

  committed
  3aaeb44

  Copy full SHA for 3aaeb44

• fix

  ZzSean

  committed
  450c80c

  Copy full SHA for 450c80c

• fix bwd error

  ZzSean

  committed
  236a1f6

  Copy full SHA for 236a1f6

Commits on Sep 27, 2021

• decrease intermediate space and fix bad malloc

  ZzSean

  committed
  adba900

  Copy full SHA for adba900

• Enhance the cache of make fused cudnn ops.

  Xreki

  committed
  ccb13a5

  Copy full SHA for ccb13a5

• Merge branch 'develop' into cudnn_v8_fusion

  Xreki

  committed
  ed67cf0

  Copy full SHA for ed67cf0

• Fix a bug of std::vector.

  Xreki

  committed
  c82017b

  Copy full SHA for c82017b

• Use independent forward and backward cache.

  Xreki

  committed
  5339c4e

  Copy full SHA for 5339c4e

Commits on Sep 28, 2021

• Enable inplace addto for resnet_unit_grad op.

  Xreki

  committed
  1a1c3bf

  Copy full SHA for 1a1c3bf

• python test

  ZzSean

  committed
  688a779

  Copy full SHA for 688a779

• Merge branch 'develop' into cudnn_v8_fusion

  Xreki

  committed
  686452f

  Copy full SHA for 686452f

Commits on Sep 29, 2021

• Merge branch 'develop' into cudnn_v8_fusion

  Xreki

  committed
  97e56eb

  Copy full SHA for 97e56eb

• Merge branch 'cudnn_v8_fusion' of https://github.com/ZzSean/Paddle into cudnn_v8_fusion

  Xreki

  committed
  34864b1

  Copy full SHA for 34864b1

Commits on Oct 8, 2021

• change bn_add_relu backward implementation

  ZzSean

  committed
  097187b

  Copy full SHA for 097187b

• Merge branch 'develop' into cudnn_v8_fusion

  Xreki

  committed
  ac0da0e

  Copy full SHA for ac0da0e

• Merge branch 'cudnn_v8_fusion' of https://github.com/ZzSean/Paddle into cudnn_v8_fusion

  Xreki

  committed
  d19d0a5

  Copy full SHA for d19d0a5

• refactor the cudnn fusion op

  ZzSean

  committed
  959b429

  Copy full SHA for 959b429

• Polish the codes of cudnn_bn_add_relu_test.

  Xreki

  committed
  a267dbb

  Copy full SHA for a267dbb

• Merge branch 'cudnn_v8_fusion' of https://github.com/ZzSean/Paddle into cudnn_v8_fusion

  Xreki

  committed
  dfc3c40

  Copy full SHA for dfc3c40

• add backward test for bn_add_relu

  ZzSean

  committed
  2e77e80

  Copy full SHA for 2e77e80

• Add more test case.

  Xreki

  committed
  1c06aa9

  Copy full SHA for 1c06aa9

Commits on Oct 9, 2021

• Merge branch 'cudnn_v8_fusion' of https://github.com/ZzSean/Paddle into cudnn_v8_fusion

  Xreki

  committed
  4a52f07

  Copy full SHA for 4a52f07

• delete debug info

  ZzSean

  committed
  f50f291

  Copy full SHA for f50f291

• Polish the codes of resnet_unit_op and fix two bugs.

  Xreki

  committed
  41a4a84

  Copy full SHA for 41a4a84

• Merge branch 'cudnn_v8_fusion' of https://github.com/ZzSean/Paddle into cudnn_v8_fusion

  Xreki

  committed
  c7bf91b

  Copy full SHA for c7bf91b

• add more test

  ZzSean

  committed
  70b3f2d

  Copy full SHA for 70b3f2d

• Polish and remove some unused codes.

  Xreki

  committed
  b3dd740

  Copy full SHA for b3dd740

• Merge branch 'cudnn_v8_fusion' of https://github.com/ZzSean/Paddle into cudnn_v8_fusion

  Xreki

  committed
  e2c0172

  Copy full SHA for e2c0172

• Merge branch 'develop' into cudnn_v8_fusion

  Xreki

  committed
  0d8eb92

  Copy full SHA for 0d8eb92

Commits on Oct 11, 2021

• add attr is_test and delete unused code

  ZzSean

  committed
  d0dee01

  Copy full SHA for d0dee01

• change ptr to tensor in forward

  ZzSean

  committed
  7bd83d4

  Copy full SHA for 7bd83d4

• change ptr to tensor in backward

  ZzSean

  committed
  794565d

  Copy full SHA for 794565d

Commits on Oct 12, 2021

• update develop

  ZzSean

  committed
  f8a4847

  Copy full SHA for f8a4847

• Merge branch 'develop' into cudnn_v8_fusion

  ZzSean

  committed
  b35afee

  Copy full SHA for b35afee

• unify the format of the params of fusion inferences

  ZzSean

  committed
  13efb64

  Copy full SHA for 13efb64

• add more assert

  ZzSean

  committed
  ab369a1

  Copy full SHA for ab369a1

• fix

  ZzSean

  committed
  fce8a19

  Copy full SHA for fce8a19

• delete unused code

  ZzSean

  committed
  2ec78a4

  Copy full SHA for 2ec78a4

• add notes

  ZzSean

  committed
  aa4cc57

  Copy full SHA for aa4cc57

• fix some potential bug

  ZzSean

  committed
  212ab3f

  Copy full SHA for 212ab3f

Commits on Oct 14, 2021

• Merge branch 'develop' into cudnn_v8_fusion

  ZzSean

  committed
  0d5bdae

  Copy full SHA for 0d5bdae

• modify

  ZzSean

  committed
  fd4e10b

  Copy full SHA for fd4e10b

• Revert "modify"

  Show description for 33e261b
  ZzSean

  committed
  33e261b

  Copy full SHA for 33e261b

• Revert "Revert "modify""

  Show description for a2b1044
  ZzSean

  committed
  a2b1044

  Copy full SHA for a2b1044

• fix bugs

  ZzSean

  committed
  57289d7

  Copy full SHA for 57289d7

• Merge branch 'develop' into cudnn_v8_fusion

  ZzSean

  committed
  64968bf

  Copy full SHA for 64968bf

• support inplace addto and add python API

  ZzSean

  committed
  c66c4b2

  Copy full SHA for c66c4b2

• delete assert

  ZzSean

  committed
  e755d16

  Copy full SHA for e755d16

Commits on Oct 15, 2021

• fix

  ZzSean

  committed
  d43a950

  Copy full SHA for d43a950