 Samuel Thibault
|
38fbca49e9
Fix _starpu_datawizard_progress prototype: it does not actually return anything
|
11 rokov pred |
|
Samuel Thibault
|
8bfbe1ee39
Drop mentioning Gordon
|
11 rokov pred |
|
Samuel Thibault
|
8bcae573b8
Drop duplicate wait
|
11 rokov pred |
|
Samuel Thibault
|
e8a62c8fce
move spurious variable initialization
|
11 rokov pred |
|
Samuel Thibault
|
c60d91cb3f
Enable asynchronous OpenCL wherever possible
|
11 rokov pred |
|
Samuel Thibault
|
840af37490
push pending requests at the back of the list of pending requests, not the front
|
11 rokov pred |
|
Samuel Thibault
|
058e113825
port r12533 from 1.1: Do not let non-CUDA workers use non-0 streams, CUDA seems not very threadsafe with that. Make the coherency engine avoid selecting non-CUDA workers to issue transfers, to avoid letting them use the 0 stream.
|
11 rokov pred |
|
Samuel Thibault
|
9e7c9936ce
Use STARPU_MAIN_RAM where appropriate
|
11 rokov pred |
|
Samuel Thibault
|
9aa1c5ba6c
Use STARPU_MAIN_RAM where appropriate
|
11 rokov pred |
|
Samuel Thibault
|
2bafbc890a
fix build
|
11 rokov pred |
|
Samuel Thibault
|
92d443d920
Make documentation clearer
|
11 rokov pred |
|
Samuel Thibault
|
ea5d8f596b
Make a copy of the interface to the memchunk only when the latter gets detached from the data, and thus the interface code will not work on it. Drop the copy when the memchunk gets reattached. This allows interfaces to modify pointers in the interface in unpack, notably
|
11 rokov pred |
|
Samuel Thibault
|
fb03301fdf
port r12523 from 1.1: Fix detaching a memchunk from a handle: the memchunk is not supposed to still have a pointer to the replicate
|
11 rokov pred |
|
Samuel Thibault
|
1aad553cc2
Fix crash with perfmodels having type common
|
11 rokov pred |
 Andra Hugo
|
aed609a645
rollback r12518: don't allow unregister in callback or task codelets
|
11 rokov pred |
|
Andra Hugo
|
84e77b5ad7
add prolog callbacks to mpi
|
11 rokov pred |
|
Andra Hugo
|
a78e7bdc6a
allow unregisters in callbacks and codelets when we have only CPUs
|
11 rokov pred |
|
Andra Hugo
|
2938dd4805
patch Terry: add a new prologue callback (one executed at pop time)
|
11 rokov pred |
|
Andra Hugo
|
e991168dec
rollback book_workers
|
11 rokov pred |
|
Samuel Thibault
|
5c4a168e72
Do not destroy containers before using them
|
11 rokov pred |
|
Samuel Thibault
|
5a9ffce947
backport r12506: fix disabling out-of-order for data transfers
|
11 rokov pred |
|
Samuel Thibault
|
bf8aa6338d
Let the OpenCL driver progress while the GPU is computing
|
11 rokov pred |
|
Samuel Thibault
|
7d2bf0c630
Enable asynchronous CUDA kernels wherever possible
|
11 rokov pred |
|
Samuel Thibault
|
7ce4ced85d
Let the CUDA driver progress while the GPU is computing
|
11 rokov pred |
|
Samuel Thibault
|
42925b7f8e
Do not define _XOPEN_SOURCE, as it disables GNU extensions, which users don't expect
|
11 rokov pred |
 Olivier Aumage
|
40432cdc2c
- first word after \section is the section name
|
11 rokov pred |
|
Samuel Thibault
|
163ef6b095
Make statistics only on the part of the trace between start_profiling and stop_profiling
|
11 rokov pred |
|
Andra Hugo
|
aa59ce38c7
rollback
|
11 rokov pred |
|
Andra Hugo
|
1c2f50cec5
fix exec_parallel code when called by the app thread
|
11 rokov pred |
|
Andra Hugo
|
0c40ac6f8a
make the hypervisor point to the good includes
|
11 rokov pred |
