When parallelizing my optimization code using parfor, there are notable differences from the serial "for" loop. Digging into the internet, I found two possible cause causes.
Intel MKL
My optimization involves matrix manipulations (inverse, multiplicaiton, etc), which relies on the Intel MKL. It seems to be an know issue that mkl produces different results depending on the number of threads involved.
Matlab quirks
For me, the speed/precision tradeoff is not worth it atm. I'll see what can be done.
2061

被折叠的 条评论
为什么被折叠?



