WebGL matrix performance fix #5072
Conversation
|
It would be nice to also have some hard perf numbers for this from a benchmark. I will see if there is a benchmark that can catch this and try to post some numbers tomorrow. |
|
Nice work and quick fix! |
|
Added a small question around object creation... |
|
Firefox benchmark:
Chrome:
It seems that the float32 fix works, but there is still a performance regression somewhere. |
| private _posMatrixCache: {[_: string]: mat4}; | ||
| private _fogMatrixCache: {[_: string]: mat4}; | ||
| private _alignedPosMatrixCache: {[_: string]: mat4}; | ||
| private _posMatrixCacheF32: Map<string, mat4> = new Map(); |
There was a problem hiding this comment.
I would advise to create an object/type that has both 32 and 64 matrices to avoid cache misalignment.
There was a problem hiding this comment.
I'm not sure what you mean exactly. As in not allowing the maps to have different sets of key-value pairs by accident?
There was a problem hiding this comment.
Instead of two maps holding the same keys, have one map holding an object with two fields.
|
Codecov is not running on PRs lately for somereason, I'll see if I can make it work again. |
Fixes #5070 by passing float32 matrices to WebGL instead of float64. Note that the performance degradation only seemed to affect Chrome, not Firefox. Updates relevant unit & render tests.