Patrik

Osgnach

Age: 29

Nov
4
comment Designing task dependencies with openmp
Well, this is nor really an option. First because with this kind of program, every saved minute is golden. Second because I am the only developer, so complex refactoring is acceptable. I just need to figure out how I can enforce a dependency like f<-g1,..g5<-h using openmp tasks
Nov
3
revised Designing task dependencies with openmp
added 245 characters in body
Nov
3
comment Designing task dependencies with openmp
Hi, thank you for your input. I'll take a look at these clauses. In the meantime, there is something more I have to add. I don't think it is a scheduling issue. The issue is caused by the fact that parallelism is implemented as a for loop over i. If I use all 32 hardware threads that I have available, when processing the last 5 elements only 5 threads are active. I need a deeper parallelism
Oct
31
asked Designing task dependencies with openmp
Oct
16
awarded Popular Question
Jul
31
comment MATLAB parfor how to slice a matrix
Ok, I'll accept this answer. Thank you
Jul
31
accepted MATLAB parfor how to slice a matrix
Jul
30
comment MATLAB parfor how to slice a matrix
Edric's answer makes sense, but my question is: why isn't MATLAB smart enough to just send exactly one column of in to exactly one worker? Slicing should be trivial. Is it just an issue of the analyser or of the implementation of parfor?
Jul
30
asked MATLAB parfor how to slice a matrix
Jul
2
awarded Curious
Mar
31
awarded Yearling
Oct
21
awarded Tumbleweed
Oct
14
awarded Student
Oct
14
asked 1D PDE decoupling
Apr
23
accepted OpenMP: synchronization inside parallel for
Apr
19
comment OpenMP: synchronization inside parallel for
the thread id is passed to f. the id is used to break the array p into chunks of equal size. What f does is irrelevant, but I'll add a sample
Apr
19
revised OpenMP: synchronization inside parallel for
added 148 characters in body
Apr
19
asked OpenMP: synchronization inside parallel for
Nov
26
comment Segfault on free()
Valgrind found the issue. A buffer overflow was overwriting the 16 magic bytes. I still don't understand why watch didn't catch this, though
Nov
26
accepted Segfault on free()
1 2 3 4 5