 Nathalie Furmento
|
ef3823b148
mpi/src/starpu_mpi_insert_task.c: change the algorithm for starpu_mpi_redux_data (see comments inside function implementation), to make sure all the steps are executed after data_handle being reduced finished its last read access.
|
12 years ago |
|
Nathalie Furmento
|
1bd474839b
mpi/src/starpu_mpi.c: fix memory cleaning
|
12 years ago |
|
Nathalie Furmento
|
83b95ee7cb
mpi/tests: new testcase for receiving early several datas
|
12 years ago |
|
Nathalie Furmento
|
e2675531f5
mpi: fix for receiving early several datas (and not only one)
|
12 years ago |
|
Nathalie Furmento
|
31753c0497
mpi/src/starpu_mpi.c: use the tag which is defined for the communication, and not the tag of the data (as they may be different)
|
12 years ago |
 Samuel Thibault
|
5c113f4053
While unpartitioning, do not acquire the child data when they are not in redux mode. Otherwise we would not be able to unpartition a data in a callback
|
12 years ago |
|
Samuel Thibault
|
ce1dc108c3
make sure the buffer we are dropping does not contain data
|
12 years ago |
|
Samuel Thibault
|
2c4d2a6c7e
Fix releasing buffers of partitioned data used with reduction
|
12 years ago |
|
Nathalie Furmento
|
b78f57263a
mpi/src/starpu_mpi.c: improve debug and error messages
|
12 years ago |
|
Nathalie Furmento
|
d889f9df2d
mpi/tests/gather.c: use same tag for all the nodes, and skip test if
|
12 years ago |
|
Nathalie Furmento
|
f3288d4dda
Changelog: mention MPI fixes
|
12 years ago |
|
Nathalie Furmento
|
4e5b001319
mpi: fix for being able to receive a data with the same tag from
|
12 years ago |
|
Nathalie Furmento
|
90f8e97cde
mpi/tests/gather.c: skip test when less than 3 nodes
|
12 years ago |
|
Samuel Thibault
|
77b58f8dd8
Fix memory leak
|
12 years ago |
|
Nathalie Furmento
|
1071eb99e2
mpi/tests/mpi_reduction.c: initialize memory
|
12 years ago |
|
Nathalie Furmento
|
59353eccf8
src/core/perfmodel/perfmodel_bus.c: free hwloc bitmap
|
12 years ago |
|
Nathalie Furmento
|
133ee36825
src/datawizard/interfaces/data_interface.c: improve error message
|
12 years ago |
|
Samuel Thibault
|
e2b16b3138
Avoid getting build failures just because of automake warnings
|
12 years ago |
|
Samuel Thibault
|
e3625c944d
make starpu_data_unpartition acquire the child pieces, otherwise we may not have collapsed the reduction
|
12 years ago |
|
Samuel Thibault
|
b5e6986273
fix protecting access to worker_exp_end with the sched mutex
|
12 years ago |
|
Samuel Thibault
|
41dee18d0c
fix protecting post_sync_tasks_cnt with sequential_consistency_mutex. But no need to take the mutex when it is known to be 0, just tell hellgrind to be fine with it
|
12 years ago |
|
Samuel Thibault
|
242066fe76
Tell hellgrind about header spinlock vs busy_mutex in the same order as in _starpu_data_check_not_busy
|
12 years ago |
|
Nathalie Furmento
|
2c71c896d0
mpi: test functionality provided in previous commit
|
12 years ago |
|
Nathalie Furmento
|
29bb94205b
mpi/src: fix when receiving too early data which have not yet been registered, the data is received as as a raw memory, and will be given it to the application when it post a equivalent receive
|
12 years ago |
|
Nathalie Furmento
|
c8be1347be
mpi/src: properly set the size of the data in the envelope
|
12 years ago |
|
Nathalie Furmento
|
73d80820ab
mpi/src: rename variables to have a more appropriate name
|
12 years ago |
 Andra Hugo
|
9502f53f25
introduce min max as a constraint for the linear program
|
12 years ago |
|
Nathalie Furmento
|
2d5990eadd
tests/disk/disk_copy.c: decrease vector size on 32 bits machine
|
12 years ago |
 Pierre-André Wacrenier
|
9c421fc50e
unlock worker's sched_mutex before busy waiting
|
12 years ago |
|
Nathalie Furmento
|
8b523baf9c
examples/Makefile.am: only check sched_ctx/parallel_code when long-check is enabled
|
12 years ago |
