[PHI] Skip check in DeformableConvInferMeta on dynamic dim

[DrRyanHuang]

PR Category

Execute Infrastructure

PR Types

Bug fixes

Description

目前运行包含 Deformable Conv2d 的模型，如： ttfnet^pafnet_10x_coco_单卡动转静训练 和 ppyolo^ppyolov2_r101vd_dcn_365e_coco_单卡动转静训练，会出现如下问题：

  File "/workspace/Paddle/build/python/paddle/vision/ops.py", line 907, in deform_conv2d
    pre_bias = _C_ops.deformable_conv(

ValueError: (InvalidArgument) The number of input channels should be equal to filter channels * groups. The difference is [256]: [-1]
  [Hint: Expected in_dims[1] == filter_dims[1] * groups, but received in_dims[1]:256 != filter_dims[1] * groups:-1.] (at /workspace/Paddle/paddle/phi/infermeta/multiary.cc:1859)

如下代码的 in_dims[1] 和 filter_dims[1] 都会存在 -1 的情况：

Paddle/paddle/phi/infermeta/multiary.cc

Lines 1856 to 1863 in 2bec694

	PADDLE_ENFORCE_EQ(
	in_dims[1],
	filter_dims[1] * groups,
	common::errors::InvalidArgument(
	"The number of input channels should be equal to filter "
	"channels * groups. The difference is [%d]: [%d]",
	in_dims[1],
	filter_dims[1] * groups));

所以在编译期且存在 -1 的情况直接跳过参数检查

pcard-91591

[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.

[SigureMo]

LGTMeow

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
⚠️ Please upload report for BASE (develop@409515d). Learn more about missing BASE report.

Additional details and impacted files

@@             Coverage Diff             @@
##             develop    #74650   +/-   ##
===========================================
  Coverage           ?   100.00%           
===========================================
  Files              ?         1           
  Lines              ?         5           
  Branches           ?         0           
===========================================
  Hits               ?         5           
  Misses             ?         0           
  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.