memory optimize pass crashes on ppyoloe

[sfraczek]

bug描述 Describe the Bug

Error description

I built paddle develop (SHA a0f4388) and run ppyoloe test and have a crash in memory_optimize_pass.
I tried this SHA and it worked: 93099bb

Command

python3.9 test_ppyoloe_infer.py --model_path=models/ppyoloe_crn_l_300e_coco_quant --reader_config=configs/ppyoloe_reader.yml --device=CPU --use_mkldnn=True --cpu_threads=10 --precision=fp32

Error message

其他补充信息 Additional Supplementary Information

No response

[paddle-bot]

您好，我们已经收到了您的问题，会安排技术人员尽快解答您的问题，请耐心等待。请您再次检查是否提供了清晰的问题描述、复现代码、环境&版本、报错信息等。同时，您也可以通过查看官网API文档、常见问题、历史Issue、AI社区来寻求解答。祝您生活愉快～

Hi! We've received your issue and please be patient to get responded. We will arrange technicians to answer your questions as soon as possible. Please make sure that you have posted enough message to demo your request. You may also check out the API，FAQ，Github Issue and AI community to get the answer.Have a nice day!

[Shixiaowei02]

We are following up on this issue, thanks.

[jiweibo]

I found that memory_optimize_pass does not take effect when mkldnn and memory_optimize_pass are enabled at the same time, but I don't know why.

[jiweibo]

@sfraczek If I use config.enable_mkldnn(), can I enable memory_optimize? Is there a conflict between these two ways?

[jiweibo]

@sfraczek If I use config.enable_mkldnn(), can I enable memory_optimize? Is there a conflict between these two ways?

I found the code, which cause the error. Do you know the background of this code? @sfraczek

https://github.com/PaddlePaddle/Paddle/blob/develop/paddle/fluid/inference/api/analysis_config.cc#L943

[sfraczek]

I only read the issue and PR threads but I don't know the background of this problem. I ran ppyoloe with your fix - reenabling this pass for mkldnn and it seems to work. It fixed ppyoloe problem. I'm waiting for some more int8 and bf16 tests.

[sfraczek]

I run into some problems with our validation. Sorry for the delay. I don't have the results yet.