On Thu, Apr 7, 2016 at 12:11 PM, Marek Olšák <mar...@gmail.com> wrote: > On Thu, Apr 7, 2016 at 2:18 AM, Bas Nieuwenhuizen > <b...@basnieuwenhuizen.nl> wrote: >> Hi Nicolai, >> >> Patches 1-2 and 5-6 are >> >> Reviewed-by: Bas Nieuwenhuizen <b...@basnieuwenhuizen.nl> >> >> However, for increasing the limits there are several cases which still >> use signed shifts (i.e. 1 << ...) which is undefined behavior shifting >> into bit 31. mesa/st contains several of those, not sure which need to >> be updated due to having lower limits, but I would presume at least >> st_glsl_to_tgsi.cpp needs updating. >> >> I will probably revisit using the constant engine once my compute >> shader and robust_buffer_access_behavior work is mostly done. Tracking >> some kind of highest used descriptor (probably better via shader >> usage, as we can apply the correct bound checking during compilation), >> is probably a good idea independently of CE usage. Uploading 2 KiB for >> every texture change is indeed a bit much. > > The CE code should be able to do uploads at per-slot granularity. > There is no reason to upload everything if you can do it fine-grained.
The upload to CE RAM can be per slot, but I think the CE RAM to L2 upload has to upload a complete new set of descriptors though, as the previous draw still needs the old descriptors and we don't want to flush the SMEM/constant L1 cache. I still need to test the bandwidth of that upload though. - Bas _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev
