 Samuel Thibault
|
69e9aa32cb
simplify unpartitioning data, and make it actually wait for completion of all transfers
|
11 anos atrás |
|
Samuel Thibault
|
436af1714c
Fix flushing all memchunk at termination
|
11 anos atrás |
|
Samuel Thibault
|
8790d2b00e
fix warning
|
11 anos atrás |
 Nathalie Furmento
|
4f6b40a1da
src/ examples/ tests/ tools/: fix coding style, curly braces go on a newline
|
11 anos atrás |
|
Samuel Thibault
|
1316e8c48e
memchunks may be used as normal or redux data alternatively, e.g. cg example
|
11 anos atrás |
|
Nathalie Furmento
|
efea5f1a8f
src/ tests/ tools/: fix coding style, curly braces go on a newline
|
11 anos atrás |
 Olivier Aumage
|
b37f386a18
- modify the OpenMP tasks locking scheme to avoid multi-locks ordering issues
|
11 anos atrás |
|
Samuel Thibault
|
bb817246ac
Factorize task submission code
|
11 anos atrás |
|
Samuel Thibault
|
3914f3f061
revert r14967 and r14972: sched_ctx should actually call _starpu_push_task instead of _starpu_task_submit_nodeps (which does many things sched_ctx does not need)
|
11 anos atrás |
|
Samuel Thibault
|
5a0fcbf328
Add anticipative writeback, to flush dirty data asynchronously before the
|
11 anos atrás |
|
Olivier Aumage
|
3832fd72cf
- address a Helgrind complaint about some early cpu workers reading global_size[0] while some late cpu workers are still writing its value
|
11 anos atrás |
|
Olivier Aumage
|
02a993ca5f
- STARPU_HG_DISABLE_CHECKING has a too broad effect, use another strategy based on a dedicated Helgrind path
|
11 anos atrás |
|
Samuel Thibault
|
271b6a3f7a
use STARPU_MAIN_RAM instead of hardcoded 0
|
11 anos atrás |
|
Samuel Thibault
|
3762ec1d0f
trace memory reclaiming done by tidy
|
11 anos atrás |
|
Samuel Thibault
|
ffa7e695ab
Only disable checking on the terminated field. enable/disable is unfortunately global, not per thread.
|
11 anos atrás |
|
Olivier Aumage
|
76b420bcb3
- address a Helgrind comment
|
11 anos atrás |
|
Nathalie Furmento
|
9abcc3b7d0
src/core/dependencies/data_concurrency.c: check a task defines a codelet before accessing its buffers
|
11 anos atrás |
|
Olivier Aumage
|
bbc050832f
- address some Helgrind comments
|
11 anos atrás |
|
Olivier Aumage
|
5feb89677f
- add Valgrind support for OpenMP stacks
|
11 anos atrás |
 Andra Hugo
|
09de7cee21
when moving a task from one context to another we create a new job for the corresponding task, take care to sort its data before resubmiting it because starpu wants a sorted list of buffers at the end
|
11 anos atrás |
|
Samuel Thibault
|
f3370dd0b9
Make valgrind ignore the uninitialized values
|
11 anos atrás |
|
Samuel Thibault
|
219ae045bb
fix build
|
11 anos atrás |
|
Samuel Thibault
|
b96308f41e
Fix GPU deallocation on OpenCL too
|
11 anos atrás |
|
Samuel Thibault
|
0adb694f41
Fix freeing GPU memory
|
11 anos atrás |
|
Samuel Thibault
|
f4b34d59ad
Factorize code, no behavior change
|
11 anos atrás |
|
Nathalie Furmento
|
b8a32f9b4c
minor fixes
|
11 anos atrás |
|
Nathalie Furmento
|
992aac301e
minor fixes
|
11 anos atrás |
|
Samuel Thibault
|
f68ea334d8
fix mingw build
|
11 anos atrás |
|
Samuel Thibault
|
e13b6b5ff7
fix windows build
|
11 anos atrás |
|
Samuel Thibault
|
5af6d13d14
Fix lock ordering: when one has to allocate destination room for reclaiming, we have to unlock the data handle lock and acquire mc_lock, so we need to release the mc_lock before calling try_to_free_mem_chunk or try_to_reuse_mem_chunk. Also add a stressing disk reclaiming testcase.
|
11 anos atrás |
