Kenneth Graunke <kenn...@whitecape.org> writes: > On 10/08/2013 10:36 AM, Eric Anholt wrote: >> We had a fixup for gen4's 3d-layout cubemaps (which, iirc, we'd >> experimentally found to be necessary!), but while the spec still requires >> it on gen5, we'd been missing it in the array-layout cubemaps. > > Ah, I see it now: > > From the Sandybridge PRM, Volume 1, Part 1, Section 7.19...OR...the > Ivybridge PRM, Volume 1, Part 1, Section 6.19 (Surface Padding > Requirements): > > "For cube surfaces, an additional two rows of padding are required at > the bottom of the surface. This must be ensured regardless of whether > the surface is stored tiled or linear. This is due to the potential > rotation of cache line orientation from memory to cache." > > So it sounds like this is indeed necessary on all platforms. Unless > they just forgot to remove this text from the docs. Have you observed > it to fix anything?
No, but it's hard to trigger -- you need your cube map to be large, unaligned in the y direction, and to land at the very end of the address space.
pgpUsgn7q0JHt.pgp
Description: PGP signature
_______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev