On 2018-08-24 03:09 PM, Christian König wrote: > Am 24.08.2018 um 20:59 schrieb Felix Kuehling: >> On 2018-08-24 03:32 AM, Christian König wrote: >>> The roundup to a power of two here is not 100% correct, but I don't >>> think it will hurt us much. >>> >>> For now the patch is Reviewed-by: Christian König >>> <christian.koenig at amd.com>. >> Thanks. Do you mind elaborating on what's not 100% correct here? > > Well it's not a power of two what give optimal root PD sizes. > > IIRC the optimal configuration for different vm sizes are: > 1GB:   4k/4k > 2GB:   8k/4k or 4k/8k > 4GB:   8k/8k > 6GB:   12K/8k > 8GB:   16K/8k or 8k/16k > 12GB:   12k/16k > .... > > As you can see 6GB and 12GB are still using all the allocated PD, but > are not power of two. Oh, I see. I guess my mistake was to assume the PT/PD sizes were powers of two. With my change they will be. > >> Maybe do the min(..., max_size) after roundup_pow_of_two? > > We would need to increase the vm_size based on the PD size after > figuring out all the parameters. > > But the win is probably not worth the effort. Right. Then I'll go ahead and submit it as is. Thanks,  Felix > > Regards, > Christian. > >> >> Regards, >>   Felix > > _______________________________________________ > amd-gfx mailing list > amd-gfx at lists.freedesktop.org > https://lists.freedesktop.org/mailman/listinfo/amd-gfx
