| commit | 3f56bd1e0586085555234dbb9acf8f46555b3815 | |
| author | Sven Verdoolaege <skimo@kotnet.org> | |
| Mon, 21 Apr 2014 15:37:39 +0000 (21 17:37 +0200) | ||
| committer | Sven Verdoolaege <skimo@kotnet.org> | |
| Wed, 15 Apr 2015 20:50:21 +0000 (15 22:50 +0200) | ||
| tree | 0518d3ee60fea2b55c1913016bb238742a5f4d1b | treesnapshot (tar.gz zip) |
| parent | fa4e683fb467eb5d89733b3dd1e638a9d050968a | commitdiff |
gpu: only add synchronization after writes from core computation that require it
If a write to global memory from within the core computation
is not needed within the same kernel invocation and is also not overwritten,
or if it is only needed inside the same iteration of the bands where
synchronization is inserted, then there is no need to insert synchronization
after that write.
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
If a write to global memory from within the core computation
is not needed within the same kernel invocation and is also not overwritten,
or if it is only needed inside the same iteration of the bands where
synchronization is inserted, then there is no need to insert synchronization
after that write.
Signed-off-by: Sven Verdoolaege <skimo@kotnet.org>
| gpu.c | diffblobblamehistory | |
| gpu.h | diffblobblamehistory |