 Samuel Thibault
|
f8d3a939b0
Make memory tidying less agressive, matching the numbers with memory reclaiming
|
10 years ago |
|
Samuel Thibault
|
105bef2aaa
revert 16092: the test is already performed above
|
10 years ago |
|
Samuel Thibault
|
8f29983c73
Check that a replicate is not busy before writing it back
|
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
|
2510a99377
Revert 16000, we will fix this another, better way
|
10 years ago |
|
Samuel Thibault
|
04ead5605d
Fix concurrency between reclaiming and tasks writing to data. Reclaiming has to acquire the rwlock on data to prevent tasks from writing to data. Also fix the mem_reclaim test to actually test the values, and not submit too many writing tasks at the same time
|
10 years ago |
|
Samuel Thibault
|
d5b8385488
When a handle got unregistered while evicting it, we still have to properly handle the remove notification of the mc we were targetting
|
10 years ago |
|
Samuel Thibault
|
681aa0a73e
Cope with the case when a data handle is unregistered just while being evicted
|
10 years ago |
|
Samuel Thibault
|
e5bad70a23
replace reclaiming's explicit transfer with a proper transfer request, to fix interaction with concurrent requests
|
10 years ago |
|
Samuel Thibault
|
d06eb3bbd4
Make reclaiming prevent more fetches on the node being cleaned up, to avoid permanent battle between them
|
10 years ago |
|
Samuel Thibault
|
5dd0687128
fix assertion: a request may have slipped through while unlocking the handle when allocating on the destination
|
10 years ago |
|
Samuel Thibault
|
03f5b6f62a
Clean ram pointer of handles when freeing an interface, notably useful with NULL data registration
|
10 years ago |
|
Samuel Thibault
|
e09ed840c2
Always free RAM data immediately when STARPU_USE_ALLOCATION_CACHE is not defined
|
10 years ago |
|
Samuel Thibault
|
bf8e9c78cb
Cache values gotten from environment variables, to fix concurrency with putenv() calls, and also reduce cost
|
10 years ago |
|
Samuel Thibault
|
d72bcabf98
Do not use alloca, it is not standard. Use ugly workaround instead
|
10 years ago |
|
Samuel Thibault
|
dfa363d81f
make sure to get alloca
|
10 years ago |
|
Samuel Thibault
|
b39dab678e
Fix undefined nul-size variable length array when using void data
|
10 years ago |
|
Samuel Thibault
|
db8e8261a4
update TSAN suppressions
|
11 years ago |
|
Samuel Thibault
|
c8efbb83e8
Inline list heads instead of keeping allocating them
|
11 years ago |
 Nathalie Furmento
|
bb61f3808d
update cnrs copyright
|
11 years ago |
|
Samuel Thibault
|
d7133eab35
Put just-cleaned up mc at the end of the clean list, so it will actually be clean by the time it gets at the front of the clean list
|
11 years ago |
|
Samuel Thibault
|
7b2d1621c4
* Add starpu_data_wont_use to advise that a piece of data will not be used
|
11 years ago |
|
Samuel Thibault
|
436af1714c
Fix flushing all memchunk at termination
|
11 years ago |
|
Samuel Thibault
|
8790d2b00e
fix warning
|
11 years ago |
|
Samuel Thibault
|
1316e8c48e
memchunks may be used as normal or redux data alternatively, e.g. cg example
|
11 years ago |
|
Nathalie Furmento
|
efea5f1a8f
src/ tests/ tools/: fix coding style, curly braces go on a newline
|
11 years ago |
|
Samuel Thibault
|
5a0fcbf328
Add anticipative writeback, to flush dirty data asynchronously before the
|
11 years ago |
|
Samuel Thibault
|
3762ec1d0f
trace memory reclaiming done by tidy
|
11 years ago |
|
Samuel Thibault
|
f4b34d59ad
Factorize code, no behavior change
|
11 years ago |
|
Samuel Thibault
|
5af6d13d14
Fix lock ordering: when one has to allocate destination room for reclaiming, we have to unlock the data handle lock and acquire mc_lock, so we need to release the mc_lock before calling try_to_free_mem_chunk or try_to_reuse_mem_chunk. Also add a stressing disk reclaiming testcase.
|
11 years ago |
