An update on the freedreno graphics driver

Posted Apr 8, 2015 23:35 UTC (Wed) by liam (guest, #84133)
Parent article: An update on the freedreno graphics driver

Clark said that the extension could be called the "we hate tilers" extension, since handling it will perform badly on a tile-based GPU.

Given that adreno3xx+ support immediate mode rendering (to some degree) is this really that bad?

An update on the freedreno graphics driver

Posted Apr 9, 2015 18:57 UTC (Thu) by robclark (subscriber, #74945) [Link] (2 responses)

>> Clark said that the extension could be called the "we hate tilers" extension, since handling it will perform badly on a tile-based GPU.
> Given that adreno3xx+ support immediate mode rendering (to some degree) is this really that bad?

Well, at least for a3xx the immediate mode (gmem bypass) is only for when blending/depth/stencil are not required. So probably little overlap between that case and use of NV_conditional_render.

An update on the freedreno graphics driver

Posted Apr 9, 2015 20:57 UTC (Thu) by liam (guest, #84133) [Link] (1 responses)

Well... that's a rather limited number of scenarios. I'm sure it's useful but now I know why Qualcomm didn't specify that the immediate mode wasn't functionally complete.
Perhaps it's mostly used for chrome elements.

An update on the freedreno graphics driver

Posted Apr 9, 2015 23:33 UTC (Thu) by robclark (subscriber, #74945) [Link]

yeah, mostly for UI type stuff (and texture upload, mipmap level generation, etc)

I guess it would have cost too many transistors, relative to the usefulness, to make it functionally complete.