I noticed when enabling "accelerated readback" with userptr enabling
writes directly into client memory was that we often missed the
tiled-memcpy fastpaths; in particular it did not handle subimages. Since
I was relying on the tiled-memcpy fastpath to avoid using the GPU for
readback where unnecessary, this is unfortunate! ;)
Quickly checking with x11perf on glamor (bdw) shows a slight benefit:
1 2 Operation
------------ ------------------------- -------------------------
587000.0 628000.0 ( 1.070) ShmPutImage 10x10 square
12100.0 13300.0 ( 1.099) ShmPutImage 500x500 square
114000.0 117000.0 ( 1.026) ShmGetImage 10x10 square
10700.0 9930.0 ( 0.928) ShmGetImage 500x500 square
_______________________________________________
mesa-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/mesa-dev
