 Samuel Thibault
|
fad8cc39f7
simplify yet more storing codelet start/stop: put it in the worker instead of jobs, so various parts of StarPÜ can access it directly
|
лет назад: 8 |
|
Samuel Thibault
|
454637e88d
Always use j->cl_start instead of a local variable, to make debugging easier
|
лет назад: 8 |
 Nathalie Furmento
|
1d90cf39b7
src: avoid calling fprint directly, use instead starpu message functions
|
лет назад: 8 |
 Olivier Aumage
|
cf5be99691
update copyrights for files modified in 2017
|
лет назад: 8 |
|
Olivier Aumage
|
408f4d3d98
factorize code
|
лет назад: 8 |
|
Olivier Aumage
|
98508d981f
merge from trunk
|
лет назад: 8 |
|
Olivier Aumage
|
b71f91a262
drop the right task from the current_tasks pipeline on CUDA
|
лет назад: 8 |
|
Olivier Aumage
|
bc6e66ef68
reset current_task field when the task is refused by the cuda driver
|
лет назад: 8 |
|
Samuel Thibault
|
25c5266794
Fix accessing the CUDA properties before it gets initialized
|
лет назад: 8 |
 Luka Stanisic
|
6971a2c68c
simgrid: adding new codelet flag
|
лет назад: 8 |
|
Samuel Thibault
|
b6110e1eff
Fix tracing GPU thread progress: stop and restart progress close to the main loop, which avoids intermixing with other trace events
|
лет назад: 8 |
|
Samuel Thibault
|
d1bbfcfdc8
improve data fetch efficiency, by not releasing data after the asynchronous transfers have completed
|
лет назад: 8 |
|
Samuel Thibault
|
74ecd02370
Fix boolean vs binary test
|
лет назад: 8 |
|
Samuel Thibault
|
6b3556b030
Fix execution with CUDA_PIPELINE=0
|
лет назад: 8 |
|
Samuel Thibault
|
8ba41f91be
Fix simgrid build without cuda
|
лет назад: 8 |
|
Nathalie Furmento
|
a017fcbdeb
src: driver operations are stored in a struct _starpu_driver_ops which are used to implement the public driver API
|
лет назад: 8 |
|
Samuel Thibault
|
0452a05f97
Warn about missing concurrent kernel execution support even when running multistream workers in separate threads
|
лет назад: 8 |
|
Samuel Thibault
|
bfcac01ad1
Avoid emitting progress probes repeatedly, allowing to re-enable them in the trace
|
лет назад: 8 |
|
Samuel Thibault
|
7cdeadd447
Add STARPU_CUDA_THREAD_PER_DEV environment variable to support driving all
|
лет назад: 8 |
|
Samuel Thibault
|
43a3c904ed
directly lookup destination node instead of relying on the node of the current worker
|
лет назад: 8 |
|
Samuel Thibault
|
c71f05d00b
fix warning
|
лет назад: 8 |
|
Samuel Thibault
|
5a5eb593d1
Always reset the local worker key to worker0 for datawizard to get the proper wait queue when using multiple workers in the same driver thread
|
лет назад: 8 |
|
Samuel Thibault
|
6a9541027d
comments
|
лет назад: 8 |
|
Samuel Thibault
|
6734d1fd95
Fix initial value for number of cuda devices
|
лет назад: 8 |
|
Samuel Thibault
|
03c9afe025
Do not set ncudagpus each time _starpu_init_cuda is called, it is useless
|
лет назад: 8 |
|
Samuel Thibault
|
f3f9acf198
Remove mutexes and conds from simgrid tasks and transfers since we are using queues everywhere now
|
лет назад: 8 |
|
Samuel Thibault
|
bfc1b8ff4e
Fix paje trace when pipelining tasks and transfers
|
лет назад: 8 |
|
Samuel Thibault
|
3b6ce806ec
Fix warnings
|
лет назад: 8 |
|
Samuel Thibault
|
64115beec5
Use asynchronous task input fetching for cuda and opencl too
|
лет назад: 8 |
|
Samuel Thibault
|
d6e4271db8
factorize fetch_task_input changes from starpu-on-mpi, use it for CPU driver too
|
лет назад: 8 |
