Quoting Chris Wilson (2020-11-09 11:28:04) > Quoting Matthew Auld (2020-11-09 11:12:49) > > When performing an allocation we try split it down into the largest > > possible power-of-two blocks/pages-sizes, and for the common case we > > expect to allocate the blocks in descending order. This also naturally > > fits with our GTT alignment tricks(including the hugepages selftest), > > where we sometimes try to align to the largest possible GTT page-size > > for the allocation, in the hope that translates to bigger GTT > > page-sizes. Currently, we seem to incorrectly add the blocks in the > > opposite order, which is definitely not the intended behaviour. > > Right, so currently we end up with small -> large blocks, and you want > to change it so that we use large -> small blocks, because we want to > keep things aligned for as long as possible (if we start small, we > immediately spoil the alignment provided for us). Reviewed-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx> > How can we test this? Create a 2MiB + 4KiB object and ask if > page_sizes.gtt has a 2MiB? I don't think it can be directly observed > from userspace, right? Is it possible to generalise the characteristic > we expect and prevent future errors? But we should try and capture some testing :) -Chris _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/intel-gfx
