 Samuel Thibault
|
61599d233e
Regenerated task structures can not be deinitialized, as waiting for the callback is not enough to wait for starpu to be done with it, and some of these regenerated tasks are actually re-scheduled so starpu can't know how to properly wait for them
|
13 years ago |
 Nathalie Furmento
|
a51401af10
tests/helper.h: inform user application retuern value is ignored when running on valgrind (as computation may have been skipped)
|
13 years ago |
|
Nathalie Furmento
|
6b3e7c0780
tests/: further reduce test values when running on valgrind
|
13 years ago |
 Cyril Roelandt
|
795aab35ea
function_call_termination_condition.cocci: add a patch mode.
|
13 years ago |
|
Cyril Roelandt
|
4cdf0a494c
Properly free memory at the end of the test.
|
13 years ago |
|
Cyril Roelandt
|
230c8f72bd
Prevent increment_redux_v2 from failing in Valgrind.
|
13 years ago |
|
Cyril Roelandt
|
57b544ba53
_starpu_data_unregister: let _starpu_request_mem_chunk_removal remove all chunks allocated on the given node for the given handle.
|
13 years ago |
|
Samuel Thibault
|
2f31d773a6
make clear that register should be NULL for scratch, else it would be ignored.
|
13 years ago |
|
Samuel Thibault
|
381d4b3bdc
make clear that register should be NULL for scatch
|
13 years ago |
|
Samuel Thibault
|
66f7704ad4
clearer
|
13 years ago |
|
Nathalie Furmento
|
cde1f9b0c9
tests/core/static_restartable.c: temporarily skip the test when running on valgrind
|
13 years ago |
|
Samuel Thibault
|
01cc5ce2c7
Tell the user that he should ask the administrator to use EXCLUSIVE-PROCESS, not EXCLUSIVE-THREAD
|
13 years ago |
|
Nathalie Furmento
|
467a125447
tests/core/starpu_task_wait.c: temporarily skip the test when running on valgrind
|
13 years ago |
|
Nathalie Furmento
|
d89d465314
configure.ac: fix detection of CUDA workability and typo
|
13 years ago |
|
Nathalie Furmento
|
1f89f72a89
configure.ac: check if CUDA is working properly before enabling it
|
13 years ago |
|
Samuel Thibault
|
fba139b857
|
13 years ago |
|
Samuel Thibault
|
749b4d31ef
comment
|
13 years ago |
|
Cyril Roelandt
|
a0b69c2602
Fix a division by zero.
|
13 years ago |
|
Nathalie Furmento
|
1ff1fed8c5
mpi: fix equality test
|
13 years ago |
|
Nathalie Furmento
|
b141c60fd3
mpi: for now, mpi can only exchange basic data interfaces, and not the ones declared at application level
|
13 years ago |
|
Nathalie Furmento
|
de38f6c394
doc: document starpu_data_interface_get_next_id
|
13 years ago |
|
Nathalie Furmento
|
203765dbd8
Data interface: add new function int starpu_data_interface_get_next_id() which returns the next id available for a data interface. That ensures different data interface registered by application all have different ids.
|
13 years ago |
|
Cyril Roelandt
|
ab1f11fd5e
Coccinelle: add a script that checks for mutex that are locked and unlocked in a function, but not unlocked when treating error cases.
|
13 years ago |
|
Cyril Roelandt
|
5731ba7783
Exit nicely when running on valgrind.
|
13 years ago |
 Nicolas Collin
|
19c0b51a3a
Bug fixed : pre_exec_hook was not initialized in 'policy'.
|
13 years ago |
|
Nathalie Furmento
|
5b25d475bb
src/core/dependencies/cg.c: include core/task.h for _starpu_task_destroy
|
13 years ago |
|
Nathalie Furmento
|
70707c3fcf
src/datawizard/interfaces/variable_interface.c: cast to the correct type
|
13 years ago |
|
Samuel Thibault
|
50c1628116
Even when setting task dependencies due to implicit data dependencies, check that the job is not already terminated.
|
13 years ago |
|
Samuel Thibault
|
29c58ebf73
Document that we actually create several workers, but use only one at a time
|
13 years ago |
|
Samuel Thibault
|
37845e1678
When requested for only a single combined worker, also create workers of smaller sizes, but always using the same cpus, so that we keep the property of only one parallel task at a time
|
13 years ago |
