 Nathalie Furmento
|
59640039cb
tests/datawizard/data_lookup.c: detect when no PU can execute the codelet and skip the test
|
hace 14 años |
 Cédric Augonnet
|
d8e4ed1724
Only compile CUDA code if STARPU_USE_CUDA is defined
|
hace 14 años |
|
Nathalie Furmento
|
0d914cc730
tests/datawizard/lazy_allocation: properly implement the CUDA codelet to check buffer content
|
hace 14 años |
|
Nathalie Furmento
|
ce21c179fc
MPI: properly set id when FxT is enabled
|
hace 14 años |
 Ludovic Courtès
|
1e153b7a83
gcc: In task submission, pass raw pointers to `starpu_insert_task'.
|
hace 14 años |
|
Ludovic Courtès
|
007a5cf66c
gcc: Initialize the `nbuffers' field of codelets.
|
hace 14 años |
|
Ludovic Courtès
|
8177b305f4
gcc: Factorize the test stubs into <lib.h>.
|
hace 14 años |
|
Ludovic Courtès
|
91e09797ef
gcc: Pass field initializers to the codelet constructor in the right order.
|
hace 14 años |
|
Nathalie Furmento
|
fa4f1b9a81
cuda: fix type for cudaError
|
hace 14 años |
 Samuel Thibault
|
37772c92ec
Use cl_mem for OpenCL buffer pointers, as that's what they actually are, and to avoid confusing beginners with CPU buffers.
|
hace 14 años |
|
Samuel Thibault
|
ca5dd22771
move C99-ism from cuda/opencl helper headers to libstarpu itself.
|
hace 14 años |
|
Nathalie Furmento
|
1f33d0ed4a
examples/reductions/dot_product: add a cuda codelet for the reduction
|
hace 14 años |
|
Nathalie Furmento
|
69da56dbba
examples/: on enodev errors, tests are marked as skipped
|
hace 14 años |
|
Nathalie Furmento
|
1e62c8c5c8
tests/: on enodev errors, tests are marked as skipped
|
hace 14 años |
|
Nathalie Furmento
|
0b2f6a8c1c
tests/: on enodev errors, tests are marked as skipped
|
hace 14 años |
|
Nathalie Furmento
|
ccde4d7412
starpu_insert_task returns the return code of the task submission
|
hace 14 años |
|
Ludovic Courtès
|
01d162d2ae
Add `gdbinit' to the distribution.
|
hace 14 años |
|
Ludovic Courtès
|
c5900310ca
Rewrite `starpu_handle_get_local_ptr' in terms of `starpu_handle_to_pointer'.
|
hace 14 años |
|
Nathalie Furmento
|
3410893c43
MPI/examples/cholesky: allocate matrix per blocks and not as a whole. This should improve MPI communications as they will now deal with contiguous data.
|
hace 14 años |
|
Cédric Augonnet
|
bb0dc7df61
In case we have CPU parallel tasks, the archtype is not that of the CPU, but a
|
hace 14 años |
|
Cédric Augonnet
|
e0ef0c401b
- Make the generated gnuplot script executable
|
hace 14 años |
|
Cédric Augonnet
|
f732a91778
Fix the CUDA kernel of the handle_to_pointer test which cannot directly assign
|
hace 14 años |
|
Cédric Augonnet
|
b57821046d
The starpu_regression_display tool needs to parse FxT traces.
|
hace 14 años |
|
Cédric Augonnet
|
3b6eb44eba
Remove the perfmodel_display_gnuplot which is not needed anymore
|
hace 14 años |
|
Cédric Augonnet
|
0e31e57c71
Rewrite the tool to display performance models based on regression.
|
hace 14 años |
|
Cédric Augonnet
|
65c62f7649
Minor cleanup/simplification of the examples: do not forget to unregister data,
|
hace 14 años |
|
Cédric Augonnet
|
853e462da8
- Make sure there is no overflow when we write strings in FxT events.
|
hace 14 años |
|
Cédric Augonnet
|
2a2eaf9b5b
Use the proper arguments in the FxT even to find the name of the codelet
|
hace 14 años |
|
Samuel Thibault
|
febecc5fa2
Fix confusion between local pointer, ram pointer, and home pointer:
|
hace 14 años |
|
Samuel Thibault
|
ea265fa757
fix missing declarations
|
hace 14 años |
