gpu_group_references: distinguish between shared_schedule and copy_schedule
commitcf470867baedce340eff08a80afb6e9b3018d7ea
authorSven Verdoolaege <skimo@kotnet.org>
Wed, 2 Oct 2013 19:54:23 +0000 (2 21:54 +0200)
committerSven Verdoolaege <skimo@kotnet.org>
Fri, 3 Jun 2016 14:22:58 +0000 (3 16:22 +0200)
tree6b52b33347373d4e54d9e096d8b943284e271cdc
parente26f43a08a61335358bdb6a04c9b762d17db21dd
gpu_group_references: distinguish between shared_schedule and copy_schedule

The initial mapping to shared memory is currently computed right outside
the band that is mapped to threads.  In a subsequent commit, it will
be made possible to introduce this mapping at a higher level.
In preparation, adjust gpu_group_references to take into account
this distinction by introducing a copy_schedule that is used for
determining the initial overlap between accesses.
In particular, when a reference group can only be mapped to private
memory, overlap will still be determined at the deepest level.

Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
gpu_group.c
gpu_group.h