[spine-cpp] Reduce the number of calls to malloc

[zhiyangyou]

• ‘setAnimation ’causes a lot of memory request calls AnimationState::computeHold will result in a call to _propertyIDs.addAll(), which will result in a large number of Entry objects being created in HashMap, is it possible to selectively pool Entry objects to reduce malloc caused by setAnimation https://github.com/EsotericSoftware/spine-runtimes/blob/b1d5e7727d7cb61e1328a40218dab21f59dfbb87/spine-cpp/spine-cpp/src/spine/AnimationState.cpp#L1027

• ‘Animation’ Ctor, it would be better to change the left-value to a reference

  https://github.com/EsotericSoftware/spine-runtimes/blob/b1d5e7727d7cb61e1328a40218dab21f59dfbb87/spine-cpp/spine-cpp/src/spine/Animation.cpp#L47

• Implicitly calls the constructor of spine::String https://github.com/EsotericSoftware/spine-runtimes/blob/b1d5e7727d7cb61e1328a40218dab21f59dfbb87/spine-cpp/spine-cpp/src/spine/SkeletonBinary.cpp#L353

• use spine::String ( xxx, true) ctor and spine::String::unown() avoid temp string data copy and memory alloc such as : https://github.com/EsotericSoftware/spine-runtimes/blob/b1d5e7727d7cb61e1328a40218dab21f59dfbb87/spine-cpp/spine-cpp/src/spine/SkeletonBinary.cpp#L527 https://github.com/EsotericSoftware/spine-runtimes/blob/b1d5e7727d7cb61e1328a40218dab21f59dfbb87/spine-cpp/spine-cpp/src/spine/SkeletonBinary.cpp#L828

String strPathForAtt(path.buffer(), true); RegionAttachment region = _attachmentLoader->newRegionAttachment(skin, String(name), strPathForAtt); strPathForAtt.unown();

[badlogic]

Thanks, those are some great suggestions. Ideally, the allocator could be self-defined, so one could e.g. use a bump allocator for loading skeletons, which would significantly cut the number of mallocs.

[zhiyangyou]

Thanks, those are some great suggestions. Ideally, the allocator could be self-defined, so one could e.g. use a bump allocator for loading skeletons, which would significantly cut the number of mallocs.

bump allocator , this is an excellent suggestion for me! ^_^，thanks a lot ！

I am a newbie in c++ :( , can you please recommend 1 such allocator to me.

[badlogic]

I don't know of any 3rd party libraries for that I'm afraid.

[zhiyangyou]

I have noticed the SpineExtension class, but I don't know how to introduce some high-performance allocators to avoid some memory fragmentation. I don't know how other applications integrate the spine-cpp runtime library.

:( I checked a lot of blogs and they told me that malloc calls can only be minimized, not avoided. Ask if there is a better deal for me..

[zhiyangyou]

Do you think it's a good idea to encapsulate the deallocate and allocate methods of std::allocator, and add some cookie information to the header of the requested memory block?

[badlogic]

I don't see how that'd help. Implementing a bump allocator that can beat malloc as we use it now is not hard. The issue is with object life times. A first step would be for anything allocated during loading of SkeletonData via SkeletonJson or SkeletonBinary to be allocated by a bump allocator, while any other object type like Skeleton uses malloc. This requires some refactoring of the runtime. There are a lot of reallocations happening during parsing of data, which doesn't lend itself well to bump allocation. I'm afraid there's no simple, immediate fix for this.

[zhiyangyou]

Ok, my doubt is gone, thanks for your reply. :）

Thanks to the spine-cpp runtime, it solved a performance hotspot in my game.