 Samuel Thibault
|
ad0f7bbef7
Set reduction codelet modes in codelet, not at execution
|
%!s(int64=13) %!d(string=hai) anos |
|
Samuel Thibault
|
40b9725a65
fix build
|
%!s(int64=13) %!d(string=hai) anos |
|
Samuel Thibault
|
af6fbb8766
Tell helgrind that it's fine to check for empty fifo without actual mutex (it's just a pointer)
|
%!s(int64=13) %!d(string=hai) anos |
|
Samuel Thibault
|
9dc7323531
Note TODO about MPI reduction
|
%!s(int64=13) %!d(string=hai) anos |
 Andra Hugo
|
a0074841dc
fix ispeed_lp (consider the bw only if the gpu does not belong to the ctx, in this case a forced data transfer could be needed)
|
%!s(int64=13) %!d(string=hai) anos |
|
Andra Hugo
|
4c3d206a67
wake up workers when there's a task pushed
|
%!s(int64=13) %!d(string=hai) anos |
 Sylvain Henry
|
a56bf9a099
SOCL: fix race
|
%!s(int64=13) %!d(string=hai) anos |
|
Samuel Thibault
|
6f424b2cdf
Make setting mc->data atomic with removing it from the mc_list, instead of dealing with data being NULL
|
%!s(int64=13) %!d(string=hai) anos |
|
Samuel Thibault
|
7f41445f74
add -lm for sqrt
|
%!s(int64=13) %!d(string=hai) anos |
|
Samuel Thibault
|
8162e8cfed
explain profiling safety to helgrind through annotation rather than suppression
|
%!s(int64=13) %!d(string=hai) anos |
|
Samuel Thibault
|
38e722860a
Initialize cond and mutex dynamically, so they can be destroyed dynamically by helgrind
|
%!s(int64=13) %!d(string=hai) anos |
|
Samuel Thibault
|
d59644d146
Add more synchronization hints about config->running
|
%!s(int64=13) %!d(string=hai) anos |
|
Samuel Thibault
|
5f35a0321d
Add more synchronization to explain helgrind that config.running is safe
|
%!s(int64=13) %!d(string=hai) anos |
 Nathalie Furmento
|
a646d6a75a
src/common/utils.h: old versions of valgrind do not provide the macros
|
%!s(int64=13) %!d(string=hai) anos |
|
Samuel Thibault
|
2a81ad3400
Add -lm for sqrtf and sqrt
|
%!s(int64=13) %!d(string=hai) anos |
|
Samuel Thibault
|
65505adaba
Also fix freeing opencl data of vector and matrix interfaces
|
%!s(int64=13) %!d(string=hai) anos |
|
Samuel Thibault
|
73b8f5a1c3
fix build
|
%!s(int64=13) %!d(string=hai) anos |
|
Samuel Thibault
|
0346982735
fix freeing opencl data from the block interface
|
%!s(int64=13) %!d(string=hai) anos |
|
Nathalie Furmento
|
4828cc67c5
mpi/src/starpu_mpi.c: test return values of MPI functions
|
%!s(int64=13) %!d(string=hai) anos |
|
Nathalie Furmento
|
6cacff4980
mpi/tests/user_defined_datatype.c: make sure tags are different for each iteration
|
%!s(int64=13) %!d(string=hai) anos |
|
Nathalie Furmento
|
8ef6b950f3
tests/loader.c: send SIGQUIT on timeout to generate core file
|
%!s(int64=13) %!d(string=hai) anos |
|
Andra Hugo
|
5c1a97e5c7
fix resize for lp2 (compute correctly the bw && the latency)
|
%!s(int64=13) %!d(string=hai) anos |
|
Samuel Thibault
|
d95e1c8398
explain helgrind about some races we are fine with
|
%!s(int64=13) %!d(string=hai) anos |
|
Samuel Thibault
|
95e0b19b3e
Add some missing mutex/cond/rwlock_destroy
|
%!s(int64=13) %!d(string=hai) anos |
|
Samuel Thibault
|
4b2844fb36
Also automatically release opencl memory on shutdown
|
%!s(int64=13) %!d(string=hai) anos |
|
Andra Hugo
|
a603573008
fix valgrind issue
|
%!s(int64=13) %!d(string=hai) anos |
|
Samuel Thibault
|
32e124f6b6
revert spurious change in r9000, keep the suppression for now
|
%!s(int64=13) %!d(string=hai) anos |
|
Samuel Thibault
|
c44a90ff25
after r8987, mem chunks can now not have a data pointer
|
%!s(int64=13) %!d(string=hai) anos |
|
Samuel Thibault
|
add756f1af
Note about racyness. We can't really add a suppression for this case, it would hide potential other races
|
%!s(int64=13) %!d(string=hai) anos |
|
Samuel Thibault
|
0cd466b90a
revert r8991: the busy_mutex must be taken after the spinlock, so we can not take the spinlock after the mutex. Reading an int really is atomic on all our targets anyway.
|
%!s(int64=13) %!d(string=hai) anos |
