4 ms·
Except that, when you are writing scientific code to solve a large linear problem, like a weather model or asupernova simulation, the difference between some na
by ziotom78 8y ago
Except that, when you are writing scientific code to solve a large linear problem, like a weather model or asupernova simulation, the difference between some naïve code and a heavily optimized one can be ten hours against ten days...
- carlmr 8y agoIf it's within 20% of BLAS performance it shouldn't be the difference between ten hours and ten days. That must be something else then.
- ziotom78 8y agoThe figure "20%" alone means nothing. You are probably meaning, "20% in this specific benchmark with "non-huge matrices" (as said by the OP). Algorithms can show wildly different performance when run on small-scale and large-scale inputs.
- carlmr 8y agoWell then you have to still test it if it's still 20% or not. I haven't written the code, I just think you shouldn't have to resort to hyperbole (ten hours, ten days) if the comment above doesn't warrant it.