|
|
|
|
|
|
by boulos
1723 days ago
|
|
|
I don't have MKL to try this out, but I'd check that the MKL threading choice actually didn't break the initialization to 1.0 loop. That is, instead of checking after doing the x[i] *= SCALE bit with cblas, I would check both before and after the scaling. |
|
|