[0-size Tensor Job2 No.60] Add 0-size Tensor support for paddle.nn.functional.instance_norm [fluid_ops] by co63oc · Pull Request #73854 · PaddlePaddle/Paddle

co63oc · 2025-07-06T12:31:10Z

PR Category

Operator Mechanism

PR Types

Bug fixes

Description

[0-size Tensor Job2 No.60] Add 0-size Tensor support for paddle.nn.functional.instance_norm [fluid_ops]

InstanceNormInferMeta 取消维度为0判断
符号推导没有对应代码

Paddle/paddle/fluid/pir/dialect/operator/interface/infer_symbolic_shape/multiary_infer_sym.cc

Line 2729 in 1b1ad5a

bool InstanceNormOpInferSymbolicShape(

InstanceNormGradInferMeta 修改scale、bias中维度设置为0部分，设置为scale、bias对应维度，增加bias参数

修改 backward.yaml配置文件，kernel增加bias参数, 使用到 composite 配置，修改paddle/fluid/prim/api/composite_backward/composite_backward_api.h文件增加bias参数
修改kernel前向，反向，填充为0
增加单测

torch 测试反向填充

import torch

_args = []
input = torch.randn((8, 32, 32, 0))
input.requires_grad = True
weight = torch.randn([32])
weight.requires_grad = True
bias = torch.randn([32])
bias.requires_grad = True
_args.append(x)
_args.append(None)
_args.append(None)
_args.append(weight)
_args.append(bias)
result = torch.nn.functional.instance_norm(*_args)
result.sum().backward()
print(weight.grad)

tensor([0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.,
        0., 0., 0., 0., 0., 0., 0., 0.])

PaddleAPITest测试通过

codecov-commenter · 2025-07-07T05:58:47Z

Codecov Report

Attention: Patch coverage is 44.82759% with 16 lines in your changes missing coverage. Please review.

Please upload report for BASE (develop@8341678). Learn more about missing BASE report.

Files with missing lines	Patch %	Lines
paddle/phi/kernels/cpu/instance_norm_kernel.cc	10.00%	9 Missing ⚠️
...addle/phi/kernels/cpu/instance_norm_grad_kernel.cc	22.22%	7 Missing ⚠️

❌ Your patch status has failed because the patch coverage (44.82%) is below the target coverage (90.00%). You can increase the patch coverage or adjust the target coverage.

Additional details and impacted files

@@            Coverage Diff             @@
##             develop   #73854   +/-   ##
==========================================
  Coverage           ?   44.82%           
==========================================
  Files              ?        4           
  Lines              ?       29           
  Branches           ?        0           
==========================================
  Hits               ?       13           
  Misses             ?       16           
  Partials           ?        0

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

co63oc · 2025-07-18T04:33:16Z

@DanielSun11 CI已完成需要review

co63oc · 2025-07-19T10:30:09Z

/re-run all-failed

DanielSun11

LGTM

DanielSun11 · 2025-07-21T05:16:09Z

代码改动较大，请 @wanghuancoder review一下

wanghuancoder

请 @luotao1 协调云飞帮忙看看

wanghuancoder · 2025-07-21T06:25:37Z

paddle/phi/ops/yaml/backward.yaml

+  args : (Tensor x, Tensor scale, Tensor bias, Tensor saved_mean, Tensor saved_variance, Tensor y_grad, float epsilon=1e-5)
  output : Tensor(x_grad), Tensor(scale_grad), Tensor(bias_grad)
  infer_meta :
    func : InstanceNormGradInferMeta
    spmd_rule : InstanceNormGradInferSpmd
  kernel :
    func : instance_norm_grad
    data_type : x
-  optional : scale
+  optional : scale, bias
  backward : instance_norm_double_grad
-  composite: instance_norm_grad(x, scale, saved_mean, saved_variance, y_grad, epsilon, x_grad, scale_grad, bias_grad)
+  composite: instance_norm_grad(x, scale, bias, saved_mean, saved_variance, y_grad, epsilon, x_grad, scale_grad, bias_grad)


这里有几个问题需要云飞评估：

修改反向签名是否涉及不兼容升级？

如果只是InferMeta用，好像有办法不修改函数签名，在kernel->func中标记需要传给Kernel的参数有哪些，可以参考lp_pool2d_grad的param。也需要云飞确认，这种用法用的比较少

这个bias如果只是InferMeta用，又是optional的是否可以标记为no need buffer？也需要云飞确认

2. lp_pool2d_grad

和lp_pool2d_grad不同的是，这里同时使用了composite，使用func传参数，composite中会报错

改反向一般不影响推理兼容性
配置no need buffer这里不太确定有没有问题，可能需要实际测试下

已修改增加 no_need_buffer : bias

wanghuancoder · 2025-07-21T06:26:11Z

paddle/fluid/prim/api/composite_backward/composite_backward_api.h

 template <typename T>
 void instance_norm_grad(const Tensor& x,
                        const paddle::optional<Tensor>& scale,
+                        const paddle::optional<Tensor>& bias UNUSED,


这里需要改InferMeta吗？

不增加bias参数，composite这部分报错

zyfncg · 2025-07-21T07:10:31Z

paddle/phi/ops/yaml/backward.yaml

+  args : (Tensor x, Tensor scale, Tensor bias, Tensor saved_mean, Tensor saved_variance, Tensor y_grad, float epsilon=1e-5)
  output : Tensor(x_grad), Tensor(scale_grad), Tensor(bias_grad)
  infer_meta :
    func : InstanceNormGradInferMeta
    spmd_rule : InstanceNormGradInferSpmd
  kernel :
    func : instance_norm_grad
    data_type : x
-  optional : scale
+  optional : scale, bias
  backward : instance_norm_double_grad
-  composite: instance_norm_grad(x, scale, saved_mean, saved_variance, y_grad, epsilon, x_grad, scale_grad, bias_grad)
+  composite: instance_norm_grad(x, scale, bias, saved_mean, saved_variance, y_grad, epsilon, x_grad, scale_grad, bias_grad)


改反向一般不影响推理兼容性
配置no need buffer这里不太确定有没有问题，可能需要实际测试下

co63oc · 2025-07-21T10:40:02Z

/re-run all-failed

co63oc · 2025-07-21T12:40:58Z

/re-run all-failed

co63oc · 2025-07-21T22:55:57Z

@zyfncg 已增加no_need_buffer，CI已完成

co63oc requested review from JiabinYang and xiaoguoguo626807 as code owners July 6, 2025 12:31

paddle-bot bot added the contributor External developers label Jul 6, 2025

co63oc changed the title ~~[0-size Tensor Job2 No.60] Add 0-size Tensor support for paddle.nn.functional.instance_norm [fluid_ops]~~ CI测试不review paddle.nn.functional.instance_norm [fluid_ops] Jul 6, 2025

co63oc requested review from ForFishes, LiYuRio and zhiqiu as code owners July 6, 2025 23:10

co63oc force-pushed the g41 branch 2 times, most recently from bf3b13d to 35db413 Compare July 11, 2025 06:14

co63oc changed the title ~~CI测试不review paddle.nn.functional.instance_norm [fluid_ops]~~ [0-size Tensor Job2 No.60] Add 0-size Tensor support for paddle.nn.functional.instance_norm [fluid_ops] Jul 11, 2025

Fix

802fc24

co63oc force-pushed the g41 branch from fe81d9e to 802fc24 Compare July 11, 2025 11:56

luotao1 mentioned this pull request Jul 11, 2025

【开源任务】 Paddle API 0-size 机制建设-第2批 #73182

Closed

luotao1 added the HappyOpenSource Pro 进阶版快乐开源活动，更具挑战性的任务 label Jul 14, 2025

luotao1 assigned luotao1 and DanielSun11 Jul 14, 2025

Merge branch 'develop' into g41

4f8167f

Merge branch 'develop' into g41

770955c

DanielSun11 previously approved these changes Jul 21, 2025

View reviewed changes

wanghuancoder reviewed Jul 21, 2025

View reviewed changes

zyfncg previously approved these changes Jul 21, 2025

View reviewed changes

Fix

013ec81

co63oc dismissed stale reviews from zyfncg and DanielSun11 via 013ec81 July 21, 2025 08:06

zyfncg approved these changes Jul 22, 2025

View reviewed changes

luotao1 added the skip-ci: approval label Jul 22, 2025

luotao1 merged commit 81a6159 into PaddlePaddle:develop Jul 22, 2025
81 of 87 checks passed

co63oc deleted the g41 branch July 23, 2025 04:06

Conversation

co63oc commented Jul 6, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

PR Category

PR Types

Description

Uh oh!

codecov-commenter commented Jul 7, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

co63oc commented Jul 18, 2025

Uh oh!

co63oc commented Jul 19, 2025

Uh oh!

DanielSun11 left a comment

Choose a reason for hiding this comment

Uh oh!

DanielSun11 commented Jul 21, 2025

Uh oh!

wanghuancoder left a comment

Choose a reason for hiding this comment

Uh oh!

wanghuancoder Jul 21, 2025

Choose a reason for hiding this comment

Uh oh!

co63oc Jul 21, 2025

Choose a reason for hiding this comment

Uh oh!

zyfncg Jul 21, 2025

Choose a reason for hiding this comment

Uh oh!

co63oc Jul 21, 2025

Choose a reason for hiding this comment

Uh oh!

wanghuancoder Jul 21, 2025

Choose a reason for hiding this comment

Uh oh!

co63oc Jul 21, 2025

Choose a reason for hiding this comment

Uh oh!

zyfncg Jul 21, 2025

Choose a reason for hiding this comment

Uh oh!

co63oc commented Jul 21, 2025

Uh oh!

co63oc commented Jul 21, 2025

Uh oh!

co63oc commented Jul 21, 2025

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants

co63oc commented Jul 6, 2025 •

edited

Loading

codecov-commenter commented Jul 7, 2025 •

edited

Loading