 Samuel Thibault
|
b195529cd4
Add missing calls to can_execute hook
|
12 anos atrás |
|
Samuel Thibault
|
14016e793f
port r11537 from 1.1: cudaFree takes much more than 125µs on average. 750µs is an average on the cholesky example. This still needs to be somehow tuned
|
12 anos atrás |
|
Samuel Thibault
|
c77d312268
port r 11535 from 1.1: Add missing termination of data transfers for reclaiming write-back
|
12 anos atrás |
|
Samuel Thibault
|
098414fc11
port r11533 from 1.1: Also explicitly show in traces the writing back time during reclaiming
|
12 anos atrás |
|
Samuel Thibault
|
9a05d0c3b8
port r11531 from 1.1: Also show free periods in the trace, to distinguish with data transfers and algorithmic complexity in reclaiming
|
12 anos atrás |
|
Samuel Thibault
|
5f6f2bf55b
port r11529 from 1.1: Ignore data races on statistics
|
12 anos atrás |
|
Samuel Thibault
|
0001dad2f6
port r11527 from 1.1: Do not care about races on number of tasks, this is used as a statistic only
|
12 anos atrás |
 Nathalie Furmento
|
f3a3538ac7
examples/stencil/stencil-kernels.c: check return value of function clEnqueueCopyBuffer
|
12 anos atrás |
 Marc Sergent
|
b2f8911c89
Adding configure option --enable-calibration-heuristic which allows the user to set the maximum authorized deviation of the history-based calibrator
|
12 anos atrás |
|
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 anos atrás |
|
Samuel Thibault
|
631d6e3185
port r 11514 from 1.1: Add unpartition state in trace
|
12 anos atrás |
|
Nathalie Furmento
|
3221332ec5
mpi: minor fixes for function declarations
|
12 anos atrás |
|
Nathalie Furmento
|
5fdde967b0
tests: partly revert #11510 and turn all cpu implementations codelets not static so that MIC code is able to successfully call dlsym on the function
|
12 anos atrás |
|
Nathalie Furmento
|
8ec3cd5612
tests: forgot to add file in previous commit
|
12 anos atrás |
|
Nathalie Furmento
|
0c6c658cb1
tests: turn function static when it makes sense to do so
|
12 anos atrás |
|
Samuel Thibault
|
d80744f490
port r11508 from 1.1: Fix passing key to hasthable: the passed pointer has to be pointing to the key at all time
|
12 anos atrás |
|
Samuel Thibault
|
806d128705
port r11506 from 1.1: Do not remove reused memchunk from the cache, it was already removed in _starpu_memchunk_cache_lookup_locked. Thanks Cyril for managing to find a case where it crashes :)
|
12 anos atrás |
|
Samuel Thibault
|
dc2959f5f2
port r11504 from 1.1: Make run_driver cover the case when the task is not finished when we call starpu_drivers_request_termination. This is actually not working any more in the trunk, and needs fixing
|
12 anos atrás |
|
Nathalie Furmento
|
402e568b27
src/datawizard/memalloc: store interface size of the memchunk as the handle may no longer be valid when reusing the memchunk (thanks to Cyril Bordage for reporting the bug)
|
12 anos atrás |
|
Nathalie Furmento
|
9bb5a09263
src: minor fixes
|
12 anos atrás |
|
Nathalie Furmento
|
e1166e9eb0
include/starpu_bitmap.h: add missing starpu_bitmap_has_next() function prototype
|
12 anos atrás |
|
Nathalie Furmento
|
5f35f4c737
src/datawizard/interfaces/data_interface.c: rename internal function
|
12 anos atrás |
|
Nathalie Furmento
|
59dc8c395f
src/datawizard: add missing include starpu_scheduler.h
|
12 anos atrás |
|
Nathalie Furmento
|
ddf6a00144
src/datawizard/malloc: define prototypes
|
12 anos atrás |
|
Samuel Thibault
|
01764c5ddc
ignore race about the watchdog state, we are ok with it
|
12 anos atrás |
|
Nathalie Furmento
|
15acb4be0f
doc/Makefile.am: add showcheck target
|
12 anos atrás |
|
Nathalie Furmento
|
78f4820844
mpi/src/starpu_mpi.c: initialisation can only be done once mpi itself has been initialised
|
12 anos atrás |
|
Nathalie Furmento
|
a31d46c22b
mpi/src/starpu_mpi.c: make sure all internal structures are initialised before starting mpi communication thread
|
12 anos atrás |
|
Nathalie Furmento
|
7667e0b20a
tests/perfmodels/value_nan.c: delete temporary file
|
12 anos atrás |
|
Samuel Thibault
|
b801a66994
port r 11483 from 1.1: Default to using only 90% of the available GPU memory, to avoid seeing cudaMemset run out of memory..
|
12 anos atrás |
