cocos / cocos-engine

Cocos simplifies game creation and distribution with Cocos Creator, a free, open-source, cross-platform game engine. Empowering millions of developers to create high-performance, engaging 2D/3D games and instant web entertainment.
https://www.cocos.com/en/creator
Other
7.84k stars 1.89k forks source link

Custom Pipeline #16466

Open whaqzhzd opened 11 months ago

whaqzhzd commented 11 months ago

Cocos Creator version

3.8.1

System information

ALL

Issue description

Now when starting the 3.8.1 custom pipelines one after another, the data of each item has improved too significantly compared to the built-in pipelines.

When using the built-in pipelines, the profile data is as follows: image

When we enable the Custom pipelines, the data of each item is as follows: image

The pipeline settings are as follows, no post-processing has been enabled yet. image

I have reported this issue on the 3.8.0 forum before. At that time, after optimization, the situation was alleviated to some extent. But when I updated to 3.8.1, this problem occurred again.

The related 3.8.0 forum post is as follows: https://forum.cocos.org/t/topic/149932/386?u=544430497

Relevant error log output

No response

Steps to reproduce

-

Minimal reproduction project

No response

whaqzhzd commented 1 month ago

I have learned that version 3.8.4 will designate the new pipeline as the default, and it appears that the issue has been resolved. Is the performance of the new default pipeline superior to that of the classic pipeline, particularly on devices with limited capabilities such as iOS and WeChat games? This is a crucial consideration for game engines.

star-e commented 1 month ago

We will continue to improve the performance of the builtin-pipeline, currently it should be on a par with the legacy-pipeline.

We have done some optimizations compared with the legacy-pipeline.

  1. The new builtin-pipeline will discard depth-stencil buffer by default, which might imply better performance on mobile platform.
  2. We have supported MSAA on all native platforms, and utilized on-chip memory on mobile platforms. So it will have better image-quality with relatively low performance cost (it might be free on some hardware).

The new builtin-pipeline also inherited several post-process features from the previous version. These features are optimized and redundant draw calls are totally removed.

whaqzhzd commented 1 month ago

I can't wait!