 Samuel Thibault
|
69040bc437
port and complete r16119 from 1.1: put ANNOTATE_HAPPENS_BEFORE after atomic_adds, so that the reading part of the atomic add is considered before the update which signals potential termination
|
10 years ago |
|
Samuel Thibault
|
133291affe
Fix reused_requests size, we get overflows otherwise
|
10 years ago |
|
Samuel Thibault
|
6ab978fd5d
revert 16092 and part of 16066 to be on the safe side: really wait for all existing requests, to avoid letting any through which would break coherency
|
10 years ago |
|
Samuel Thibault
|
fdbdfd23f7
Drop spurious change
|
10 years ago |
|
Samuel Thibault
|
832ef25b5e
fix r16066: we still need to account for the self-hop requests, which we will wait for
|
10 years ago |
|
Samuel Thibault
|
99dc8831f2
Optimize write invalidation a bit: we don't need a self-hop request when data is already there and we will add a write invalidation request. Also, there is no need for an write invalidation request for waiting for a request whose target is the target of the write request
|
10 years ago |
|
Samuel Thibault
|
3e3823c78e
optimize requests a bit: only one trivial request is needed when data is already there, and no request is needed if it's also a prefetch request. Also make sure that we don't post the invalidation request
|
10 years ago |
|
Samuel Thibault
|
91fd2cfc15
Fix mem_reclaim: on write request, if some requests are pending (notably reclaiming), create an additional write request which will wait for them before invalidating all copies
|
10 years ago |
|
Samuel Thibault
|
4c91b21b7c
Fix nowhere support without fetching data on a node: we still need to call __starpu_push_task_output
|
10 years ago |
|
Samuel Thibault
|
c8077021d0
Support asynchronous partitioning through the home node of data, not only MAIN_RAM
|
10 years ago |
|
Samuel Thibault
|
75d0d38234
Add STARPU_NOWHERE to create synchronization tasks with data.
|
10 years ago |
|
Samuel Thibault
|
1683d35a0d
Fix msg.h inclusion according to new location
|
10 years ago |
|
Samuel Thibault
|
c8efbb83e8
Inline list heads instead of keeping allocating them
|
10 years ago |
 Nathalie Furmento
|
e3b44209ed
update copyright
|
10 years ago |
|
Nathalie Furmento
|
bb61f3808d
update cnrs copyright
|
10 years ago |
|
Nathalie Furmento
|
efea5f1a8f
src/ tests/ tools/: fix coding style, curly braces go on a newline
|
10 years ago |
|
Samuel Thibault
|
5a0fcbf328
Add anticipative writeback, to flush dirty data asynchronously before the
|
10 years ago |
|
Samuel Thibault
|
1da595f4fa
marking as requested would falsely make schedulers believe the data will soon be transferred while that's not true
|
10 years ago |
|
Samuel Thibault
|
3b8830c844
Add missing requested flag
|
10 years ago |
|
Samuel Thibault
|
074d020c3d
Add starpu_idle_prefetch_task_input_on_node, and call it at task submission for tasks which execute on a specific node
|
10 years ago |
|
Samuel Thibault
|
4c6de74837
Add starpu_data_idle_prefetch_on_node
|
10 years ago |
|
Samuel Thibault
|
479ab4ad04
port r14410 from 1.2: Also check against zeroed mode
|
10 years ago |
|
Samuel Thibault
|
9de161f711
backport r14407 from 1.2: Check that the data access mode is not bogus
|
10 years ago |
|
Nathalie Furmento
|
5e2bf3755b
src/datawizard/coherency.c: clarify error message
|
10 years ago |
|
Samuel Thibault
|
88a89370aa
Add a memcpy_peer property to CUDA gpus, to record whether it is able to run gpu-gpu transfers, so that simgrid can enable them when appropriate
|
11 years ago |
|
Samuel Thibault
|
c10e9e185f
Fix fix
|
11 years ago |
|
Samuel Thibault
|
551444f1f4
Fix crash when running with cuda peer support in simgrid mode
|
11 years ago |
|
Samuel Thibault
|
476b96feea
merge trunk
|
11 years ago |
|
Samuel Thibault
|
8f6f4fa49a
Add hierarchical schedulers
|
11 years ago |
|
Samuel Thibault
|
26a6543b6f
merge trunk
|
11 years ago |
