 Samuel Thibault
|
d1bbfcfdc8
improve data fetch efficiency, by not releasing data after the asynchronous transfers have completed
|
8 years ago |
|
Samuel Thibault
|
bfc1b8ff4e
Fix paje trace when pipelining tasks and transfers
|
8 years ago |
|
Samuel Thibault
|
d6e4271db8
factorize fetch_task_input changes from starpu-on-mpi, use it for CPU driver too
|
8 years ago |
 Corentin Salingue
|
82d0b3921a
merge from trunk
|
8 years ago |
 Nathalie Furmento
|
d917eaa107
src: use _STARPU_MSG instead of fprintf and avoid direct calls to fprintf(stderr, ...)
|
8 years ago |
|
Corentin Salingue
|
18812205bf
make transfers asynchronous for MPI and MIC workers
|
8 years ago |
|
Samuel Thibault
|
de0bb767b0
port r19310 from 1.2: Fix locking memory replicates when e.g. invalidating data: we need to lock all nodes to prevent e.g. an eviction to disk.
|
9 years ago |
|
Samuel Thibault
|
26f550995f
Add starpu_data_set_user_data and starpu_data_get_user_data.
|
9 years ago |
|
Samuel Thibault
|
b643914328
fix warnings
|
9 years ago |
|
Samuel Thibault
|
1496e10d1f
finish fixing asynchronous partitioning of temporary data: data_partition_submit should not do coherency when the data is not initialized yet
|
9 years ago |
|
Samuel Thibault
|
759c851da1
drop unused field
|
9 years ago |
|
Samuel Thibault
|
9aa325e691
Fix multiple partitioning planning of the same data: they share the same switching codelet, but may have varying nparts
|
9 years ago |
|
Samuel Thibault
|
1f841a3f0d
Add origin of data requests for debugging
|
9 years ago |
|
Samuel Thibault
|
02508a3ed7
Add magic number to data, to check against multiple unregistrations
|
9 years ago |
|
Samuel Thibault
|
cc2c83286c
Allocate per_worker array lazily, to divide data handle weight by something by an order of 10x
|
9 years ago |
|
Samuel Thibault
|
245509cfa6
reduce memory footprint of handles a bit
|
9 years ago |
|
Samuel Thibault
|
654ad27766
Add STARPU_LOCALITY flag
|
9 years ago |
|
Samuel Thibault
|
a5f7d96fea
Add rbtree and rbtree+list implementation. Use it to introduce priorities in data requests, which default to the task priority or workerorder
|
10 years ago |
|
Samuel Thibault
|
d650d6dcd1
Avoid recursing when unlocking data requests, otherwise we can end up stacking calls
|
10 years ago |
|
Samuel Thibault
|
91fd2cfc15
Fix mem_reclaim: on write request, if some requests are pending (notably reclaiming), create an additional write request which will wait for them before invalidating all copies
|
10 years ago |
|
Samuel Thibault
|
4c91b21b7c
Fix nowhere support without fetching data on a node: we still need to call __starpu_push_task_output
|
10 years ago |
|
Samuel Thibault
|
c8077021d0
Support asynchronous partitioning through the home node of data, not only MAIN_RAM
|
10 years ago |
|
Samuel Thibault
|
68605a2f15
Add asynchronous partition planning. It only supports coherency through the main RAM for now.
|
10 years ago |
|
Samuel Thibault
|
75d0d38234
Add STARPU_NOWHERE to create synchronization tasks with data.
|
10 years ago |
|
Samuel Thibault
|
c8efbb83e8
Inline list heads instead of keeping allocating them
|
10 years ago |
|
Samuel Thibault
|
3434b52b4c
Add arbiter support
|
10 years ago |
|
Samuel Thibault
|
8fbc9b2c5a
Complete arbiter implementation
|
10 years ago |
|
Samuel Thibault
|
381eb29a05
Remove some WIP tags where it should be fine on the long run
|
10 years ago |
|
Samuel Thibault
|
67c7976ba1
s/commute_/arbitered_/g
|
10 years ago |
|
Samuel Thibault
|
a5b6105706
Use arbiters instead of a global state. Does not support several arbiters yet
|
10 years ago |
