gpu: move force_private from gpu_array_info to gpu_local_array_info
commit2150a0aa460c2aaf8b48d7a5e5794b5a2000ac12
authorSven Verdoolaege <skimo@kotnet.org>
Fri, 22 Aug 2014 20:30:59 +0000 (22 22:30 +0200)
committerSven Verdoolaege <skimo@kotnet.org>
Mon, 16 Mar 2015 13:35:00 +0000 (16 14:35 +0100)
tree64f4caec17e0a2e758b33305c2987fa6884d6f48
parenta7f9ec2156b4a76dbda013b33aa30ad0a05ce9cc
gpu: move force_private from gpu_array_info to gpu_local_array_info

The gpu_array_info elements are shared by the entire gpu_prog,
while the decision of whether to forcibly put arrays into registers
is kernel specific.
We currently use this flag before we consider the next kernel,
overwriting the old flag.
In the future, however, we will be constructing a single schedule tree
for the entire program, containing information about all kernels
together.  We therefore need to put this information in kernel
specific data structures.

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