Fix clone function of Program to avoid memory leak. #10358
Conversation
kexinzhao
left a comment
kexinzhao
left a comment
There was a problem hiding this comment.
LGTM! Thank you for fixing this!
| def inference_optimize(self): | ||
| res = Program() | ||
| res.desc = core.inference_optimize(self.desc) | ||
| res.desc = core.ProgramDesc(self.desc) |
There was a problem hiding this comment.
我感觉为什么core.inference_optimize的实现有memory leak，原因还不是很确定。这里加个comment说明一下吧，说明这里的功能和core.inference_optimize是一样的以及为什么不用core.inference_optimize的原因吧。
There was a problem hiding this comment.
暂时还不确定core.inference_optimize导致memory leak的原因，我加了个comment说明这里的功能和core.inference_optimize一样
python/paddle/fluid/framework.py
Outdated
| p.desc = core.inference_optimize(self.desc) | ||
| else: | ||
| p.desc = core.ProgramDesc(self.desc) | ||
| for i in xrange(p.desc.num_blocks()): |
There was a problem hiding this comment.
这里可以调Python的inference_optimize吗？
There was a problem hiding this comment.
@wanghaoshuang 觉得这里可以直接调用Python里的: def inference_optimize(self)，另外，直接在Python里修正attr is_test=True后，有打印看下inference_program检查过吗？is_test是否变过来了？
There was a problem hiding this comment.
这里可以调Python的inference_optimize吗？
Fixed. Thx.
直接在Python里修正attr is_test=True后，有打印看下inference_program检查过吗？is_test是否变过来了？
嗯，已经验证了。
4 participants
fix #10254
for_test=Trueat herewill lead to memory leak while inference is active.
Reference assignment would be more suitablehere.than the pointer assignment