 Cédric Augonnet
|
0828950dd7
Add a wrapper for DDOT
|
15 years ago |
|
Cédric Augonnet
|
b910d275eb
The async task overhead now indicates precisely how time is spent
|
15 years ago |
|
Cédric Augonnet
|
e2c5593252
Add new fields in the per-task profiling info structure
|
15 years ago |
|
Cédric Augonnet
|
b42f38c8b2
Avoid overflow when computing pi
|
15 years ago |
|
Cédric Augonnet
|
e4ed360572
When a data handle is accessed in a R/W mode for the first time after it was
|
15 years ago |
|
Cédric Augonnet
|
193905f07a
Use starpu_task_wait_for_all instead of condition variables to simplify the
|
15 years ago |
|
Cédric Augonnet
|
8d2b6c2760
Rename the STARPU_TASK_BLOCKED_ON_JOB status into STARPU_TASK_BLOCKED_ON_DATA
|
15 years ago |
|
Cédric Augonnet
|
23a21d47db
Fix compilation without CUDA
|
15 years ago |
|
Cédric Augonnet
|
3e6aadaca8
The pi_redux example is now one of the tests used by make check
|
15 years ago |
|
Cédric Augonnet
|
1b0a67230d
No need to lock the header again
|
15 years ago |
|
Cédric Augonnet
|
af9ea8975f
StarPU automatically calls starpu_data_start_reduction_mode when the data is
|
15 years ago |
|
Cédric Augonnet
|
e4c85b9e07
Implement the Monte Carlo algorithm used to compute Pi by the means of
|
15 years ago |
|
Cédric Augonnet
|
9ae657bdb4
Fix compilation for machines without CUDA
|
15 years ago |
|
Cédric Augonnet
|
3b5af731f6
Don't use CUDA/CUBLAS if CUDA is not available
|
15 years ago |
|
Cédric Augonnet
|
11561287cf
The dot_product example computes the scalar product of two vectors by the means
|
15 years ago |
|
Cédric Augonnet
|
8e094a1fec
First draft of implementation of the STARPU_REDUX mode.
|
15 years ago |
|
Cédric Augonnet
|
58215b1860
- Add sanity checks
|
15 years ago |
|
Cédric Augonnet
|
d3dd8e5bd6
The "manual_reduction" test performs a reduction operation by the means of
|
15 years ago |
|
Cédric Augonnet
|
78faf9849a
The starpu_data_unregister_no_coherency does the same as starpu_data_unregister
|
15 years ago |
|
Cédric Augonnet
|
461d4d99b7
Provide an implementation of the spinlocks that actually rely on mutexes (with
|
15 years ago |
|
Cédric Augonnet
|
b566d09898
Use a rw-lock instead of a mutex to protect the list of progression hooks.
|
15 years ago |
|
Cédric Augonnet
|
280aa1e569
Document STARPU_LIMIT_GPU_MEM
|
15 years ago |
|
Cédric Augonnet
|
e3df8c20c8
Implement STARPU_LIMIT_GPU_MEM on both CUDA and OpenCL devices.
|
15 years ago |
 Nathalie Furmento
|
a5f704ef6b
mpi_lu example: modifications towards the new API
|
15 years ago |
|
Nathalie Furmento
|
6b8dcf733f
StarPU MPI: Add extra functionalities for MPI Implementations which are not thread safe
|
15 years ago |
|
Cédric Augonnet
|
9557513f74
Do not forget to take back the spinlock if the transfer was successful, this
|
15 years ago |
|
Cédric Augonnet
|
dabb837619
Memory reclaiming bug fix: while reclaiming memory, we should not hold the lock of the header describing the entire state of the handle while other handles are inspected to find available memory.
|
15 years ago |
|
Cédric Augonnet
|
0eb2226d5d
Check return values for the pthread_spin_* methods
|
15 years ago |
|
Cédric Augonnet
|
1607a0cb41
Document --with-mkl-cflags and --with-mkl-ldflags
|
15 years ago |
|
Cédric Augonnet
|
9817b32691
Add support for the MKL. Since it is really hard to guess the flags for MKL, we
|
15 years ago |
