On Fri, Feb 3, 2017 at 9:45 PM, Brian Paul <bri...@vmware.com> wrote: > On 02/01/2017 02:23 PM, Brian Paul wrote: >> >> On 01/27/2017 04:00 AM, Marek Olšák wrote: >>> >>> On Fri, Jan 27, 2017 at 10:05 AM, Nicolai Hähnle <nhaeh...@gmail.com> >>> wrote: >>>> >>>> On 27.01.2017 00:51, Marek Olšák wrote: >>>>> >>>>> >>>>> From: Marek Olšák <marek.ol...@amd.com> >>>>> >>>>> For lower memory usage and more efficient updates of the buffer >>>>> residency >>>>> list. (e.g. if drivers keep seeing the same buffer for many consecutive >>>>> "add" calls, the calls can be turned into no-ops trivially) >>>> >>>> >>>> >>>> This makes sense to me, but how are you planning to deal with the bind >>>> flags? They are currently set differently for different upload mgrs. We >>>> should probably do away with them entirely anyway. >>> >>> >>> Drivers can set the bind flags they need. Some drivers will set all 3 >>> bind flags. Other drivers don't have to set any. >> >> >> I need to look into this part more closely. I think we may have trouble >> mixing constants with index/vertex data in our VMware driver... > > > Marek, > > Your patch series, as-is, did indeed cause trouble with our VMware driver. > We need to keep constants in a separate buffer. > > The good news is I don't think this is a huge problem and I've updated (a > subset of) your patches to accommodate both your needs and ours. > > The basic idea is to add a pipe_context::get_stream_uploader() hook that > allows drivers to use just one or separate uploaders for > vertex/index/constant data. Plus, I added a > pipe_context::unmap_stream_uploaders() helper, but this isn't strictly > necessary. > > WIP patch attached (only lightly tested). Let me know what you think.
Can we simply add these 2 fields into pipe_context instead of the callback? pipe_context::stream_uploader // vertex + index pipe_context::const_uploader Marek _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org https://lists.freedesktop.org/mailman/listinfo/mesa-dev
