Support backward vlog guard

[DanielSun11]

PR Category

Execute Infrastructure

PR Types

New features

Description

支持在前向执行时添加一个backward_vlog_guard,实现在反向执行时打印guard内捕获的反向GradNode的日志。

使用方法

import paddle
from paddle.base.framework import capture_backward_subgraph_guard,backward_vlog_guard
x = paddle.randn([2,3],dtype="float32")
y = paddle.randn([2,3],dtype="float64")
x.stop_gradient = False
y.stop_gradient = False
z1 = x - y
z2 = x + y
with backward_vlog_guard(4):
    z3 = z1 * z2
z4 = z3.sum()
loss = z4
loss.sum().backward()

输出的日志：

只有multiply_grad 的日志信息，执行其他GradNode的时候不打印任何日志

W1128 18:11:17.496543 330082 gpu_resources.cc:116] Please NOTE: device: 0, GPU Compute Capability: 9.0, Driver API Version: 12.9, Runtime API Version: 12.9
W1128 18:11:17.506551 330082 dygraph_functions.cc:127938] Got different data type, run type promotion automatically, this may cause data type been changed.
W1128 18:11:17.639621 330082 dygraph_functions.cc:119953] Got different data type, run type promotion automatically, this may cause data type been changed.
4
I1128 18:11:17.656737 330082 backward.cc:501] multiply1GradNode(0x15d65c70) begin run 
I1128 18:11:17.656770 330082 multiply_node.cc:49] 
==========================Running_AD_API_GRAD: multiply_grad==========================
I1128 18:11:17.656828 330082 multiply_node.cc:136] { Input: [ 
( grad_out , [{
        Name: multiply1_out_float64_2x3@Grad,
        Initialized: 1,
        Tensor_Ptr:0x15d651f0,
        Tensor_Impl_Ptr: 0x172a0100 }]),  
( x , [{
        Name: None,
        Initialized: 1,
        Tensor_Ptr:0x7ffcdb762880,
        Tensor_Impl_Ptr: 0xd1a9410 }]),  
( y , [{
        Name: None,
        Initialized: 1,
        Tensor_Ptr:0x7ffcdb7628c0,
        Tensor_Impl_Ptr: 0x15d63360 }]), ]} 
I1128 18:11:17.656849 330082 multiply_node.cc:146] 
==========================Running_C++_API: multiply_grad1==========================
I1128 18:11:17.656862 330082 backward_api_base.cc:80976] multiply_grad API kernel key: [GPUDNN, NCHW, float64]
I1128 18:11:17.656877 330082 backward_api_base.cc:80983] multiply_grad kernel: {"input":[
        backend: GPU,  layout: STRIDED,  dtype: float64,
        backend: GPU,  layout: STRIDED,  dtype: float64,
        backend: GPU,  layout: STRIDED,  dtype: float64
],
"output":[
        backend: GPU,  layout: STRIDED,  dtype: float64,
        backend: GPU,  layout: STRIDED,  dtype: float64
],
"attribute":[
        "int"
]}
I1128 18:11:17.656929 330082 multiply_node.cc:166] Fused api multiply_grad is called 
I1128 18:11:17.656931 330082 multiply_node.cc:168] 
==========================Finish_C++_API: multiply_grad1==========================
I1128 18:11:17.656947 330082 multiply_node.cc:279] { Input: [ 
( grad_out , [{
        Name: multiply1_out_float64_2x3@Grad,
        Initialized: 1,
        Tensor_Ptr:0x15d651f0,
        Tensor_Impl_Ptr: 0x172a0100 }]),  
( x , [{
        Name: None,
        Initialized: 1,
        Tensor_Ptr:0x7ffcdb762880,
        Tensor_Impl_Ptr: 0xd1a9410 }]),  
( y , [{
        Name: None,
        Initialized: 1,
        Tensor_Ptr:0x7ffcdb7628c0,
        Tensor_Impl_Ptr: 0x15d63360 }]), ],  
 Output: [ 
 ( grad_x , [{
        Name: subtract1_out_float64_2x3@Grad,
        Initialized: 1,
        Tensor_Ptr:0x17498790,
        Tensor_Impl_Ptr: 0x174987f0 }]),  
 ( grad_y , [{
        Name: add1_out_float64_2x3@Grad,
        Initialized: 1,
        Tensor_Ptr:0x172a0050,
        Tensor_Impl_Ptr: 0x15d67350 }]), ] } 
I1128 18:11:17.656982 330082 multiply_node.cc:293] 
==========================Finish_AD_API_GRAD: multiply_grad==========================
I1128 18:11:17.656986 330082 backward.cc:548] multiply1GradNode(0x15d65c70) Found pending node: subtract1GradNode(0xd1a8ae0)
I1128 18:11:17.656993 330082 backward.cc:548] multiply1GradNode(0x15d65c70) Found pending node: add1GradNode(0x15d62af0)

[paddle-bot]

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.

[codecov-commenter]

Codecov Report

❌ Patch coverage is 91.93548% with 5 lines in your changes missing coverage. Please review.
⚠️ Please upload report for BASE (develop@5f1358c). Learn more about missing BASE report.

Files with missing lines	Patch %	Lines
paddle/fluid/eager/api/utils/global_utils.h	78.26%	5 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             develop   #76685   +/-   ##
==========================================
  Coverage           ?   91.93%           
==========================================
  Files              ?        4           
  Lines              ?       62           
  Branches           ?        0           
==========================================
  Hits               ?       57           
  Misses             ?        5           
  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.

[wanghuancoder]

LGTM