 Samuel Thibault
|
8f26d26ad7
Introduce the STARPU_CUDA_ASYNC and STARPU_OPENCL_ASYNC flags. No optimization is done yet for now.
|
11 years ago |
 Nathalie Furmento
|
fa66bfbbf9
tests/perfmodels/user_base.c: make sure function always return a value
|
11 years ago |
|
Samuel Thibault
|
f7033edbc6
Drop spurious test
|
11 years ago |
|
Samuel Thibault
|
384ffc0f62
Add missing ghost dependencies in DAG
|
11 years ago |
|
Samuel Thibault
|
1bb1440053
Make DAG readable
|
11 years ago |
|
Samuel Thibault
|
8e3d4676f4
Make _starpu_data_wait_until_available take a synchronization name, so that the node in the DAG becomes understandable
|
11 years ago |
|
Samuel Thibault
|
1a62c13496
reformat comment
|
11 years ago |
|
Samuel Thibault
|
b59aff6ad5
Fix starpu_data_unregister_submit and starpu_data_unregister_no_coherency, to make them properly wait for pending tasks
|
11 years ago |
|
Nathalie Furmento
|
d023eb38ae
fix function name to starpu_data_unregister_submit() and starpu_data_invalidate_submit()
|
11 years ago |
|
Samuel Thibault
|
8cc1f6c0db
Use separate OpenCL command queues for CPU->GPU, GPU->CPU, and GPU->GPU transfers
|
11 years ago |
|
Samuel Thibault
|
dc80d3edf4
Add changelog for previous commit
|
11 years ago |
|
Samuel Thibault
|
09974968a8
Let the application provide the task footprint itself
|
11 years ago |
|
Samuel Thibault
|
4899b89184
Fix GPU thread binding: document logical cores/PUs, simplify _starpu_initialize_workers_deviceid initialization loop, fix confusion between PUs and cores in _starpu_get_next_bindid
|
11 years ago |
|
Samuel Thibault
|
0cd8850c0f
Also print the number of errors. We also actually already do some aging
|
11 years ago |
|
Samuel Thibault
|
9fe394f47d
Allow performance models to have bigger variation: 50% instead of just 10%
|
11 years ago |
|
Samuel Thibault
|
7c03c945f7
Add support for letting a schedule choose its own set of workers without having to care about combined workers. This only works for SPMD however, not FORKJOIN
|
11 years ago |
|
Samuel Thibault
|
4095f15601
Re-enable parallel worker execution to actually test for explicit combined worker
|
11 years ago |
|
Samuel Thibault
|
c1a4dd4e9b
Fix pushing tasks to specific combined workers
|
11 years ago |
|
Samuel Thibault
|
d35a5da619
Use STARPU_COMMUTE in cholesky examples. Ignore the COMMUTE flag in MPI code.
|
11 years ago |
|
Samuel Thibault
|
e4c9c8774a
Fix GPU memory size in the simgrid case
|
11 years ago |
|
Samuel Thibault
|
8d795023f2
Drop documentation about stack size, we already set it to 8MiB
|
11 years ago |
|
Samuel Thibault
|
8606711d0f
Use 8MiB stack size by default in simgrid, instead of its default tiny size.
|
11 years ago |
|
Samuel Thibault
|
883669a501
Fix warning
|
11 years ago |
|
Samuel Thibault
|
823cdcc9e0
Fix warning
|
11 years ago |
|
Samuel Thibault
|
5ccb4bf6fc
Silent warning
|
11 years ago |
|
Samuel Thibault
|
2c3729244e
Silent warning
|
11 years ago |
|
Samuel Thibault
|
941eaebeb7
silent warning
|
11 years ago |
|
Samuel Thibault
|
e36b471bb8
Silent warning
|
11 years ago |
|
Samuel Thibault
|
c5ed3dff55
Silent warning
|
11 years ago |
|
Samuel Thibault
|
9df7f32dba
Fix allocation size
|
11 years ago |
