 Cédric Augonnet
|
cf48d2f6c6
In spite of the MPI norm, some MPI implementation seems to require that the
|
16 years ago |
|
Cédric Augonnet
|
9e9015d718
implement starpu_get_block_elemsize
|
16 years ago |
|
Cédric Augonnet
|
480ee416d5
- Cleanup the code managing performance models and per model "debug" files. We
|
16 years ago |
|
Cédric Augonnet
|
9711586171
fix previous commit
|
16 years ago |
|
Cédric Augonnet
|
89fe8a7605
Support 2D matrices ("blas" interface)
|
16 years ago |
|
Cédric Augonnet
|
fda086c7ae
Implement starpu_get_blas_elemsize
|
16 years ago |
|
Cédric Augonnet
|
1470ba3508
- Provide the status returned by the MPI_Irecv function.
|
16 years ago |
|
Cédric Augonnet
|
bf8bb00b8a
Start to implement starpu_mpi_handle_to_datatype.
|
16 years ago |
|
Cédric Augonnet
|
22e39e050a
starpu_get_handle_interface_id returns the identifier of the interface of a
|
16 years ago |
|
Cédric Augonnet
|
e51aadfda3
- The different data interfaces are now indexed between 0 and
|
16 years ago |
|
Cédric Augonnet
|
869320b7a2
Small hack to avoid side effects from third party libs that also try to bind
|
16 years ago |
|
Cédric Augonnet
|
5763bd2d0c
Implement starpu_get_vector_elemsize that was somehow missing.
|
16 years ago |
|
Cédric Augonnet
|
2a2758ac3e
Make the "ring" example more challenging as we now have both MPI and CUDA
|
16 years ago |
|
Cédric Augonnet
|
c3c1b0de02
Introduce the MPI helper lib (enabled with --with-mpi at configure time).
|
16 years ago |
|
Cédric Augonnet
|
c94a373e97
Fix the starpu_sync_data_with_mem API, it is now required to call
|
16 years ago |
|
Cédric Augonnet
|
fff603e54c
Make sure that the directory that should contain the performance models is
|
16 years ago |
|
Cédric Augonnet
|
0363a81ed4
fix compilation in case CUDA is not used
|
16 years ago |
|
Cédric Augonnet
|
0934f728e3
We don't hardcode the performance of the bus anymore, so that tool is useless
|
16 years ago |
|
Cédric Augonnet
|
c054754a75
Implement bus performance sampling: this is no longer hardcoded.
|
16 years ago |
|
Cédric Augonnet
|
a3be10b313
The performande models associated to the codelets are now put in
|
16 years ago |
|
Cédric Augonnet
|
3cad2c6bbb
The src and dst nodes were inverted in the matrices
|
16 years ago |
|
Cédric Augonnet
|
e9b8da269e
add some dirty program to measure the bandwith/latency used by the performance models
|
16 years ago |
|
Cédric Augonnet
|
0b80205b82
It is possible to generate a file with the distribution of the duration of the
|
16 years ago |
|
Cédric Augonnet
|
bcd9c67fc4
Fix debug assertions, it is possible that no request is associated to a data
|
16 years ago |
|
Cédric Augonnet
|
e561dd511a
We now assume that the header lock is already taken when calling
|
16 years ago |
|
Cédric Augonnet
|
eb5c03ad79
BUG FIX: when a data dependency is unlocked, we have to update the current mode
|
16 years ago |
|
Cédric Augonnet
|
485ce7b82f
cudaMemset only fills one byte at a time, so we use a char array instead of an
|
16 years ago |
|
Cédric Augonnet
|
20420637e6
starpu_wait_all_tasks and starpu_wait_task are blocking too.
|
16 years ago |
|
Cédric Augonnet
|
6f21718f48
The helper functions to allocate/liberate pinned memory use tasks internally,
|
16 years ago |
|
Cédric Augonnet
|
5284a61274
When a data request needs to actually perform a data transfer from a source, we
|
16 years ago |
