 Nathalie Furmento
|
452a0e24dc
src/datawizard/ rename field interface in struct starpu_data_replicate_t into data_interface (on windows, when including <windows.h>, one gets #define interface struct, ...)
|
%!s(int64=14) %!d(string=hai) anos |
|
Nathalie Furmento
|
345116afd3
src/datawizard/memalloc: rename field interface into chunk_interface (on windows, when including <windows.h>, one gets #define interface struct, ...)
|
%!s(int64=14) %!d(string=hai) anos |
 Samuel Thibault
|
0d9f08ae8d
Fix copyrights, thanks Ludo!
|
%!s(int64=14) %!d(string=hai) anos |
 Cédric Augonnet
|
20e50d126a
Remove dead code
|
%!s(int64=14) %!d(string=hai) anos |
|
Cédric Augonnet
|
8e094a1fec
First draft of implementation of the STARPU_REDUX mode.
|
%!s(int64=14) %!d(string=hai) anos |
|
Cédric Augonnet
|
dabb837619
Memory reclaiming bug fix: while reclaiming memory, we should not hold the lock of the header describing the entire state of the handle while other handles are inspected to find available memory.
|
%!s(int64=14) %!d(string=hai) anos |
|
Cédric Augonnet
|
d20dcb2005
Revert commit r2772. We actually cannot really merge replicates and mem_chunks
|
%!s(int64=14) %!d(string=hai) anos |
|
Cédric Augonnet
|
07c5a7654b
Properly check the return value of the pthread_rwlock_* methods
|
%!s(int64=14) %!d(string=hai) anos |
|
Cédric Augonnet
|
2c5f2890b2
It is not possible to perform superflous memory allocation (for other memory
|
%!s(int64=14) %!d(string=hai) anos |
|
Cédric Augonnet
|
e3620a36d3
- Implement the SCRATCH access mode using per-worker data replicates in the
|
%!s(int64=14) %!d(string=hai) anos |
|
Cédric Augonnet
|
40ed79b80f
Some more minor cleanups
|
%!s(int64=14) %!d(string=hai) anos |
|
Cédric Augonnet
|
25b2878d1f
Code cleanup
|
%!s(int64=14) %!d(string=hai) anos |
|
Cédric Augonnet
|
630c27c6fb
Bug fix: in case of cache hit, we don't return -ENOMEM but the size of the buffer
|
%!s(int64=14) %!d(string=hai) anos |
|
Cédric Augonnet
|
e344b2b6cc
This is a first step toward the merge of the data replicatate and mem_chunk
|
%!s(int64=14) %!d(string=hai) anos |
|
Cédric Augonnet
|
c21c9078b4
Fix typo
|
%!s(int64=14) %!d(string=hai) anos |
|
Cédric Augonnet
|
ddb5759a50
Minor code cleanups
|
%!s(int64=14) %!d(string=hai) anos |
|
Cédric Augonnet
|
1c3235beac
Bug fix: we put a copy of the saved interface into the replicate, there should
|
%!s(int64=14) %!d(string=hai) anos |
|
Cédric Augonnet
|
fc5da2fa67
_starpu_allocate_memory_on_node now manipulates a data replicate rather than a
|
%!s(int64=14) %!d(string=hai) anos |
|
Cédric Augonnet
|
f57828e9d1
Don't try to allocate a piece of data that is already allocated.
|
%!s(int64=14) %!d(string=hai) anos |
|
Cédric Augonnet
|
bfe3303c4b
The generic method to copy a data handle from a node to another node now
|
%!s(int64=14) %!d(string=hai) anos |
|
Cédric Augonnet
|
0b718bb11a
- Rename the starpu_local_data_state_t structure into starpu_data_replicate_s.
|
%!s(int64=14) %!d(string=hai) anos |
|
Cédric Augonnet
|
e84341dbfa
In order to be a little more coherent, we only use -ENOMEM and not a mix of
|
%!s(int64=14) %!d(string=hai) anos |
|
Cédric Augonnet
|
57ab0708b3
The allocate_data_on_node method of the starpu_data_interface_ops_t structure
|
%!s(int64=14) %!d(string=hai) anos |
|
Cédric Augonnet
|
d59bd9cfe2
Remove duplicate test
|
%!s(int64=14) %!d(string=hai) anos |
|
Nathalie Furmento
|
4a983beb04
Update copyright
|
%!s(int64=15) %!d(string=hai) anos |
|
Cédric Augonnet
|
75aa25c52d
Handle hash collisions in the memory allocation cache
|
%!s(int64=15) %!d(string=hai) anos |
|
Cédric Augonnet
|
ac81ad2867
- Start to rewrite the memory allocation cache mechanisms
|
%!s(int64=15) %!d(string=hai) anos |
|
Cédric Augonnet
|
d901a076c7
Minor cleanups
|
%!s(int64=15) %!d(string=hai) anos |
|
Cédric Augonnet
|
4e8827c5d3
Cache the size and the footprint of the data into the handle structure rather
|
%!s(int64=15) %!d(string=hai) anos |
|
Cédric Augonnet
|
7f2a7c8108
Properly handle the case of memory chunks associated to a handle that was
|
%!s(int64=15) %!d(string=hai) anos |
