【Hackathon 5th No.38】为 Paddle 新增 FractionalMaxPool2d / FractionalMaxPool3d API by megemini · Pull Request #59130 · PaddlePaddle/Paddle

megemini · 2023-11-19T10:58:15Z

PR types

New features

PR changes

APIs

Description

【Hackathon 5th No.38】为 Paddle 新增 FractionalMaxPool2d / FractionalMaxPool3d API

RFC：PaddlePaddle/community#698

涉及文件：

paddle/phi/api/yaml/backward.yaml 反向算子描述
paddle/phi/api/yaml/ops.yaml 前向算子描述
paddle/phi/infermeta/backward.cc 反向算子
paddle/phi/infermeta/backward.h 反向算子
paddle/phi/infermeta/unary.cc 算子 InferMeta
paddle/phi/infermeta/unary.h 算子 InferMeta
paddle/phi/kernels/funcs/pooling.cc cpu 算子实现
paddle/phi/kernels/funcs/pooling.cu gpu 算子实现
paddle/phi/kernels/funcs/pooling.h 算子头文件
paddle/phi/kernels/impl/pool_grad_kernel_impl.h
paddle/phi/kernels/impl/pool_kernel_impl.h
paddle/phi/kernels/pool_grad_kernel.h
paddle/phi/kernels/pool_kernel.h
python/paddle/nn/__init__.py 暴露 API
python/paddle/nn/functional/__init__.py 暴露 API
python/paddle/nn/functional/pooling.py API 实现
python/paddle/nn/layer/__init__.py 暴露 layer API
python/paddle/nn/layer/pooling.py layer API 实现
test/legacy_test/test_fractional_max_pool2d.py 2d api/layer 单测
test/legacy_test/test_fractional_max_pool3d.py 3d api/layer 单测
test/legacy_test/test_pool_max_op.py 算子单测
paddle/phi/kernels/xpu/pool_kernel.cc xpu 算子
paddle/phi/kernels/xpu/pool_grad_kernel.cc xpu 反向

请评审～

… hack5_38

paddle-bot · 2023-11-19T10:58:20Z

你的PR提交成功，感谢你对开源项目的贡献!
请关注后续CI自动化测试结果，详情请参考Paddle-CI手册。
Your PR has been submitted. Thanks for your contribution!
Please wait for the result of CI firstly. See Paddle CI Manual for details.

… hack5_38

megemini · 2023-11-28T04:23:49Z

Update 20231128

@Charles-hit

API 应该已经写好了，CI 已经通过～相较 RFC 有几点改变：

增加了 random_u 参数
这个参数就是公式 ceiling(α(i + u)) 里面的那个 u，也就是 (0, 1) 范围内的随机数。
此次实现，这个随机数可以通过 paddle.seed 来固定，虽然可以固定住随机数，但实际上却无法得知 u 是多少～
考虑到能够更好的复现与测试，因此在具体实现的时候增加了这么个 random_u 参数～
增加了 xpu 算子的参数
目前只是增加了参数，从而能够编译通过，不过没有具体内部实现～
xpu 好像没有这部分算子？
参考 adaptive max pool 增加了 test_fractional_max_pool2d.py/test_fractional_max_pool3d.py 的单测文件，由于复用的算子 MaxPoolXdWithIndex 还没有实现 data_format 参数，因此内部的测试用例也暂时注释掉了～

请评审～谢谢！；）

Charles-hit · 2023-12-04T03:20:53Z

-  forward : max_pool2d_with_index(Tensor x, int[] kernel_size, int[] strides = {1, 1}, int[] paddings = {0, 0}, bool global_pooling = false, bool adaptive = false) -> Tensor(out), Tensor(mask)
-  args : (Tensor x, Tensor mask, Tensor out_grad, int[] kernel_size, int[] strides, int[] paddings, bool global_pooling, bool adaptive)
+  forward : max_pool2d_with_index(Tensor x, int[] kernel_size, int[] strides = {1, 1}, int[] paddings = {0, 0}, bool global_pooling = false, bool adaptive = false, bool fractional = false, float random_u = 0.0) -> Tensor(out), Tensor(mask)
+  args : (Tensor x, Tensor mask, Tensor out_grad, int[] kernel_size, int[] strides, int[] paddings, bool global_pooling, bool adaptive, bool fractional, float random_u)


这儿直接改变原来phi算子定义会有不兼容升级的问题，新增api的话新增一个phi接口，可以复用已有的max_pool2d_with_index接口即可

Charles-hit · 2023-12-04T03:21:00Z

-  forward : max_pool3d_with_index(Tensor x, int[] kernel_size, int[] strides = {1, 1, 1}, int[] paddings = {0, 0, 0}, bool global_pooling = false, bool adaptive = false) -> Tensor(out), Tensor(mask)
-  args : (Tensor x, Tensor mask, Tensor out_grad, int[] kernel_size, int[] strides, int[] paddings, bool global_pooling, bool adaptive)
+  forward : max_pool3d_with_index(Tensor x, int[] kernel_size, int[] strides = {1, 1, 1}, int[] paddings = {0, 0, 0}, bool global_pooling = false, bool adaptive = false, bool fractional = false, float random_u = 0.0) -> Tensor(out), Tensor(mask)
+  args : (Tensor x, Tensor mask, Tensor out_grad, int[] kernel_size, int[] strides, int[] paddings, bool global_pooling, bool adaptive, bool fractional, float random_u)


Charles-hit · 2023-12-04T03:21:32Z


 - op : max_pool2d_with_index
-  args : (Tensor x, int[] kernel_size, int[] strides= {1, 1}, int[] paddings = {0, 0}, bool global_pooling = false, bool adaptive = false)
+  args : (Tensor x, int[] kernel_size, int[] strides= {1, 1}, int[] paddings = {0, 0}, bool global_pooling = false, bool adaptive = false, bool fractional = false, float random_u = 0.0)


Charles-hit · 2023-12-04T03:21:38Z


 - op : max_pool3d_with_index
-  args : (Tensor x, int[] kernel_size, int[] strides = {1, 1, 1}, int[] paddings = {0, 0, 0}, bool global_pooling = false, bool adaptive = false)
+  args : (Tensor x, int[] kernel_size, int[] strides = {1, 1, 1}, int[] paddings = {0, 0, 0}, bool global_pooling = false, bool adaptive = false, bool fractional = false, float random_u = 0.0)


Charles-hit · 2023-12-04T03:21:59Z

                                   const std::vector<int>& paddings,
                                   bool global_pooling,
                                   bool adaptive,
+                                   bool fractional,


infermeta也需要新增一个复用之前的即可

Charles-hit · 2023-12-04T03:22:18Z

                               const std::vector<int>& paddings,
                               bool global_pooling,
                               bool adaptive,
+                               bool fractional,


Charles-hit · 2023-12-04T03:22:26Z

                               const std::vector<int>& paddings,
                               bool global_pooling,
                               bool adaptive,
+                               bool fractional,


Charles-hit · 2023-12-04T03:23:03Z

                  const std::vector<int>& strides,
                  const std::vector<int>& paddings,
                  bool adaptive,
+                  bool fractional,


直接修改原kernel定义也会有不兼容升级的问题，可以新增一个kernel来复用当前kernel

Charles-hit · 2023-12-04T06:40:43Z

                                   const std::vector<int>& paddings,
                                   bool global_pooling,
                                   bool adaptive,
+                                   bool fractional,


这儿不可以直接修改原有kernel签名，会有不兼容升级的问题，新增的算子可以考虑复用这个算子。

Charles-hit · 2023-12-04T06:43:29Z

+from paddle import base
+from paddle.base import core
+
+


为了方便管理，测试最好继承自optest，需要支持低精度fp16， bfp16等，尽可能覆盖全算子类别。

测试有两部分：

test/legacy_test/test_pool_max_op.py 这里是测试底层算子的单测文件，继承了 OpTest，覆盖了全算子类别～

test/legacy_test/test_fractional_max_pool2d.py 和 test/legacy_test/test_fractional_max_pool3d.py 测试的是 python api ，参考的是 test/legacy_test/test_adaptive_max_pool2d.py 和 test/legacy_test/test_adaptive_max_pool3d.py ，继承 unittest.TestCase ，这里也要继承 OpTest ?

测试有两部分：

test/legacy_test/test_pool_max_op.py 这里是测试底层算子的单测文件，继承了 OpTest，覆盖了全算子类别～

test/legacy_test/test_fractional_max_pool2d.py 和 test/legacy_test/test_fractional_max_pool3d.py 测试的是 python api ，参考的是 test/legacy_test/test_adaptive_max_pool2d.py 和 test/legacy_test/test_adaptive_max_pool3d.py ，继承 unittest.TestCase ，这里也要继承 OpTest ?

这儿需要在yaml中新增两个算子定义，也就是新增两个op，我们针对单个op的测试都是用OpTest去统一管理的，所以继承OpTest是比较合理的测试方式，也便于后续维护。针对API的话，OpTest是可以覆盖到动态图API测试的，check_dygraph开关就是测试动态图API，静态图API测试可以继承自 unittest.TestCase。

megemini · 2023-12-04T07:17:39Z

@Charles-hit

关于复用 max_pool2d_with_index 以及相关 kernel，能否帮忙指导一下具体怎么做？

fractional max pool 的 strides 和 ksize 都是跟随 index 改变的，与 adaptive max pool 的流程基本一样，因此，adaptive max pool 的实现也是修改了算子签名、增加了 adaptive 参数后修改 max_pool2d_with_index 相关 kernel 代码从而实现的～

如果不修改目前的 kernel 算子签名，如何复用？

还请帮忙看一下～

非常感谢！

是否可以在 ops.yaml/backward.yaml 等处新增算子描述，然后在 MaxPool3dWithIndexGradFunctor 等处增加 operator() 的重载？

也就是说，是保留原来的算子，但是内部复用现在增加了 fractional max pool 参数的函数？原来的算子复用现在的，而不是现在复用原来的～

… hack5_38

Charles-hit · 2023-12-05T07:13:40Z

@Charles-hit

关于复用 max_pool2d_with_index 以及相关 kernel，能否帮忙指导一下具体怎么做？

fractional max pool 的 strides 和 ksize 都是跟随 index 改变的，与 adaptive max pool 的流程基本一样，因此，adaptive max pool 的实现也是修改了算子签名、增加了 adaptive 参数后修改 max_pool2d_with_index 相关 kernel 代码从而实现的～

如果不修改目前的 kernel 算子签名，如何复用？

还请帮忙看一下～

非常感谢！

是否可以在 ops.yaml/backward.yaml 等处新增算子描述，然后在 MaxPool3dWithIndexGradFunctor 等处增加 operator() 的重载？

也就是说，是保留原来的算子，但是内部复用现在增加了 fractional max pool 参数的函数？原来的算子复用现在的，而不是现在复用原来的～

yaml处需要新增算子描述，新增两个算子，如果想复用原有kernel的话确实可以考虑你说的那种情况，增加一个重载，不要改变原有kernel的参数以及行为就好。

… hack5_38

megemini added 11 commits October 27, 2023 17:57

[Init] init commit

34c6d6f

Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into…

71bc3c3

… hack5_38

[Add] fractional max pool python api

8efe56a

[Add] FractionalMaxPool layer

88362e8

[Add] fractional index generation

f265203

[Add] add fractional max pooling to __init__

d7457e0

[Fix] add default value False to max_poolNd_with_index

f79b8d2

Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into…

19cbb42

… hack5_38

[Update] __init__ add fractional funcs

594e130

[Add] test file

e667f09

[Add] test_pool_max_op for fractional

ab256ac

paddle-bot Bot added the contributor External developers label Nov 19, 2023

luotao1 mentioned this pull request Nov 20, 2023

【PaddlePaddle Hackathon 5th】开源贡献个人挑战赛 #57262

Closed

luotao1 added the PaddlePaddle Hackathon label Nov 20, 2023

luotao1 assigned luotao1 and Charles-hit Nov 20, 2023

megemini added 12 commits November 20, 2023 13:48

[Fix] test index cast to int

bf7a640

[Change] pooling cu rand

44f44c0

Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into…

a88aeec

… hack5_38

[Change] skip fractional op test

e0679f0

[Update] cu seed

99196b2

[Add] add param random_u for solid fractional

89297c8

[Fix] fractional pool 3d random_u

ba9d583

[Add] add xpu support for fractional max pool 2d with index

6d6f93b

[Add] fractional api unittest

8ea0e52

Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into…

0c7254c

… hack5_38

Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into…

a302293

… hack5_38

[Add] docstring for fractional max pool

62b1c13

megemini changed the title ~~【暂用于 RFC】【Hackathon 5th No.38】为 Paddle 新增 FractionalMaxPool2d / FractionalMaxPool3d API~~ 【Hackathon 5th No.38】为 Paddle 新增 FractionalMaxPool2d / FractionalMaxPool3d API Nov 27, 2023

Charles-hit reviewed Dec 4, 2023

View reviewed changes

megemini added 3 commits December 4, 2023 15:42

Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into…

4fe49f7

… hack5_38

Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into…

ae13ac2

… hack5_38

Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into…

71ededc

… hack5_38

Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into…

1782dda

… hack5_38

megemini mentioned this pull request Dec 8, 2023

【Hackathon 5th No.38】为 Paddle 新增 FractionalMaxPool2d / FractionalMaxPool3d API -kernel #59847

Merged

megemini closed this Dec 21, 2023

Conversation

megemini commented Nov 19, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

PR types

PR changes

Description

Uh oh!

paddle-bot Bot commented Nov 19, 2023

Uh oh!

megemini commented Nov 28, 2023

Update 20231128

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Charles-hit Dec 4, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Charles-hit Dec 5, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

megemini commented Dec 4, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Charles-hit commented Dec 5, 2023

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

megemini commented Nov 19, 2023 •

edited

Loading

Charles-hit Dec 4, 2023 •

edited

Loading

Charles-hit Dec 5, 2023 •

edited

Loading

megemini commented Dec 4, 2023 •

edited

Loading