Samuel Thibault
|
7c6d72edfc
expose memory manager functions to applications
|
il y a 11 ans |
Samuel Thibault
|
a60fedf197
Add STARPU_MALLOC_NORECLAIM flag to allocate without running a reclaim if the node is out of memory
|
il y a 11 ans |
Olivier Aumage
|
793eba8411
- modify the way to unregister OpenMP tasks' data
|
il y a 11 ans |
Samuel Thibault
|
501de79951
Fix counting allocated memory
|
il y a 11 ans |
Samuel Thibault
|
6757bca1b5
Remove implementation array terminator in tests, examples, and documentation, since it is not actually used
|
il y a 11 ans |
Samuel Thibault
|
4b1a16edde
Fix University name after the fusion..
|
il y a 11 ans |
Samuel Thibault
|
33226eb62f
Disable starpu_malloc_on_node when CUDA does not support peer access
|
il y a 11 ans |
Samuel Thibault
|
feda650297
Fix starpu_malloc_on_node for OpenCL when not called from the GPU worker
|
il y a 11 ans |
Samuel Thibault
|
52cfd05a46
fix starpu_malloc_on_node when called outside of the GPU worker
|
il y a 11 ans |
Samuel Thibault
|
9327d30712
revert spurious changes
|
il y a 11 ans |
Samuel Thibault
|
c209bb90fd
port r13570 from 1.1: Fix mc memory leak
|
il y a 11 ans |
Nathalie Furmento
|
c18275f016
src/datawizard/malloc.c: improve error message
|
il y a 11 ans |
Samuel Thibault
|
a9b47f084e
Automatically disable memory pinning when running in valgrind
|
il y a 12 ans |
Samuel Thibault
|
143aa5f194
Add STARPU_DISABLE_PINNING environment variable, to be able to disable memory pinning for Valgrind
|
il y a 12 ans |
Samuel Thibault
|
2f62b5cd10
Use big fat abortions when one tries to make a task or callback sleep, instead of just returning EDEADLCK which few people will test
|
il y a 12 ans |
Nathalie Furmento
|
3989f504bd
src: give name to internal tasks
|
il y a 12 ans |
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
|
il y a 12 ans |
Nathalie Furmento
|
ddf6a00144
src/datawizard/malloc: define prototypes
|
il y a 12 ans |
Samuel Thibault
|
a9cd1dc130
port r11478 from 1.1: Fix comment
|
il y a 12 ans |
Samuel Thibault
|
bb88fe0e68
port r11466 from 1.1: Even without allocation cache, data release is done lazily, we can not count on precision of value returned by starpu_memory_get_available
|
il y a 12 ans |
Nathalie Furmento
|
0cf9cf4000
src/datawizard/malloc.c: fix misplaced }
|
il y a 12 ans |
Samuel Thibault
|
3830a76d5b
port r11435 from 1.1: make sure we do not overflow
|
il y a 12 ans |
Samuel Thibault
|
0fec9f8acd
port r11433 from 1.1: Fix simgrid simulation with batched allocation: we really need to return different allocated values
|
il y a 12 ans |
Samuel Thibault
|
cd3cfcd51d
port r11372 from 1.1: Fix simgrid cuda allocation times
|
il y a 12 ans |
Samuel Thibault
|
4ce5fe9027
port r11357 from 1.1: Fix memleak
|
il y a 12 ans |
Samuel Thibault
|
336612abd1
port r11320 from 1.1: Do not error out if the user didn't release memory allocated with starpu_malloc_on_node
|
il y a 12 ans |
Samuel Thibault
|
0e385974a9
port r11312 from 1.1: Small CUDA allocations (<= 4MiB) are now batched to avoid the huge cudaMalloc overhead.
|
il y a 12 ans |
Samuel Thibault
|
49b7fe4fd2
Fix automatic RAM allocation on CUDA without memcpy_peer: we can not allocated pinned memory, since that requires waiting for a task, and starpu_malloc_on_node is called with a spinlock locked.
|
il y a 12 ans |
Corentin Salingue
|
613bc5f403
Out-of-core support, through registration of disk areas as additional memory nodes.
|
il y a 12 ans |
Corentin Salingue
|
792e96b06f
fix all bug
|
il y a 12 ans |