Samuel Thibault
|
05e36c83b3
Follow 14268: Fix transfers not starting immediately when a handle or request lock is busy
|
11 years ago |
Samuel Thibault
|
7f5a5c5a21
Fix longer transfer times in simgrid when a handle lock is busy
|
11 years ago |
Samuel Thibault
|
35d6a3a824
fix uninitialized field
|
11 years ago |
Samuel Thibault
|
ae1892fa7f
Improve the disk documentation, and clean the interface a bit
|
11 years ago |
Samuel Thibault
|
4ceefa0477
Fix logic: we want to enable the racy optimization when we use non-blocking drivers
|
11 years ago |
Samuel Thibault
|
58c1502e08
backport r13712 from trunk: Also prevent the worker from sleeping when there are prefetch requests to be pushed
|
11 years ago |
Samuel Thibault
|
7c3fe024e9
port r13705 from 1.1: Add missing wakeups when pushing requests on nodes
|
11 years ago |
Samuel Thibault
|
fcd3349e03
backport r13702 from 1.1: Disable yet more racy optimization when using blocking drivers
|
11 years ago |
Samuel Thibault
|
5ae68359cf
Fix memory leak on busy requests lists
|
11 years ago |
Samuel Thibault
|
8766e66f5d
backport r13698 from 1.1: Do not use racy optimization when using blocking drivers
|
11 years ago |
Samuel Thibault
|
4b1a16edde
Fix University name after the fusion..
|
11 years ago |
Nathalie Furmento
|
353c98cbf2
src/datawizard/data_request.c: fix call to STARPU_HG_DISABLE_CHECKING and STARPU_HG_ENABLE_CHECKING
|
11 years ago |
Samuel Thibault
|
fc76d5ee50
Lighten pressure on the spinlock of the request we are waiting for. This should fix the livelock in simgrid.
|
11 years ago |
Samuel Thibault
|
f2f024fd29
Fix erroneously not taking the mutex if it wasn't available at first when terminating
|
11 years ago |
Samuel Thibault
|
9218801246
Do not expose a known-to-be-racy-but-we-re-fine optimization to helgrind
|
11 years ago |
Nathalie Furmento
|
4ce96baf81
src/datawizard/data_request.c: add parenthesis to enhance clarity
|
12 years ago |
Samuel Thibault
|
840af37490
push pending requests at the back of the list of pending requests, not the front
|
12 years ago |
Samuel Thibault
|
7ce4ced85d
Let the CUDA driver progress while the GPU is computing
|
12 years ago |
Samuel Thibault
|
5ccb4bf6fc
Silent warning
|
12 years ago |
Samuel Thibault
|
0423864961
port r11516 from 1.1: considerably reduce the amount of requests submitted at the same time. Pushing more does not seem to really improve performance, and on the contrary increases the latency of possibly urging requests
|
12 years ago |
Samuel Thibault
|
f4c7eb0c8e
port r11378 from 1.1: Do not allocate data on the critical path
|
12 years ago |
Samuel Thibault
|
e9c70aac17
port r11376 from 1.1: Introduce rwlock_tryrdlock and trywrlock variants. Use them in the progression hooks to avoid getting stuck due to idling workers.
|
12 years ago |
Samuel Thibault
|
5f1af2ec79
port r11140 from 1.1: When prefetching, do not try too hard to lock a data handle, better do something else instead
|
12 years ago |
Samuel Thibault
|
b217a5399a
data_requests[i] may change, it is still fine for _starpu_handle_node_data_requests to read it unprotected
|
12 years ago |
Samuel Thibault
|
e5bea6f293
port r11205 from trunk: Do not push too many data requests at the same time, CUDA would actually enter synchronous mode...
|
12 years ago |
Samuel Thibault
|
edce30c7f0
port r11194 from 1.1: Also disable checking _head of renewed request lists
|
12 years ago |
Samuel Thibault
|
aea35461df
Document each racy spot normally reported by helgrind
|
12 years ago |
Samuel Thibault
|
0d1b0375b3
backport r11155 from 1.1: Disabling access to a variable locally does not actually seem to prevent helgrind from raising the race issues, on the contrary it prevents it from recording where the race comes from... Simply just disable checking these variables completely.
|
12 years ago |
Samuel Thibault
|
54460ab919
port r11138 from 1.1: Do not lie to helgrind about having taken a mutex, it produces erroneous races inside mutexes. Rather make helgrind ignore the precise memory accesses we know we do not need to care about
|
12 years ago |
Thibaud Lambert
|
3b7abcc2cf
mic: Fix bugs
|
12 years ago |