video: tegra: nvmap: Add new NVMAP_PAGES_ZAP_AND_CLEAN option
CUDA needs this zap but not reserve pages in prefetch operations.
That way, if no cpu accesses in between the prefetch operation and
till cuda kernel launch, none of the pages will be marked dirty and
there won't be any zap or cache writeback operation. So, a lot of
overhead can be avoided.
bug
200167711
Change-Id: I3b5d8d12b82ca0b2ebdd94c0f3701f71dd7a7667
Signed-off-by: Sri Krishna chowdary <schowdary@nvidia.com>
Reviewed-on: http://git-master/r/935713
(cherry picked from commit
6368766defcbfeb71f649713e89713295664f638)
Reviewed-on: http://git-master/r/
1013284
(cherry picked from commit
2812b20d575513fd438ee5a3d147575659c0bf4d)
Reviewed-on: http://git-master/r/
1022503
Reviewed-by: Sachin Nikam <snikam@nvidia.com>
Tested-by: Sachin Nikam <snikam@nvidia.com>