Sur un quadcore
Intel(R) Xeon(R) CPU X3350 @ 2.66GHz
Les résultats sont assez aléatoires, même si rien d'autre de gourmand ne tourne.
inplace sequential prediction : 0.402000 ms
inplace sequential prediction += : 0.354000 ms
outplace sequential prediction : 1.446000 ms
inplace parallel prediction += using 1 threads: 2.075000 ms
inplace parallel prediction += using 2 threads: 0.966000 ms
inplace parallel prediction += using 4 threads: 0.837000 ms
inplace parallel prediction += using 8 threads: 1.835000 ms
inplace parallel prediction += using 16 threads: 0.755000 ms
outplace parallel prediction using 1 threads: 2.175000 ms
outplace parallel prediction using 2 threads: 1.123000 ms
outplace parallel prediction using 4 threads: 0.774000 ms
outplace parallel prediction using 8 threads: 0.853000 ms
outplace parallel prediction using 16 threads: 0.771000 ms
ou alors
inplace sequential prediction : 0.310000 ms
inplace sequential prediction += : 0.276000 ms
outplace sequential prediction : 1.365000 ms
inplace parallel prediction += using 1 threads: 2.280000 ms
inplace parallel prediction += using 2 threads: 1.083000 ms
inplace parallel prediction += using 4 threads: 0.632000 ms
inplace parallel prediction += using 8 threads: 0.507000 ms
inplace parallel prediction += using 16 threads: 0.610000 ms
outplace parallel prediction using 1 threads: 1.108000 ms
outplace parallel prediction using 2 threads: 1.177000 ms
outplace parallel prediction using 4 threads: 1.319000 ms
outplace parallel prediction using 8 threads: 0.628000 ms
outplace parallel prediction using 16 threads: 0.724000 ms
Je pense que les temps sont beaucoup trop court et la moindre préemption/interruption fausse le résultat.
Du coup je me demande si c'est pas simplement un patch RT la solution
[^] # Re: creation des threads + cache
Posté par _Sil_ . En réponse au message Parallelisation d'une boucle (théoriquement) trivialement parallélisable. Évalué à 1.
Intel(R) Xeon(R) CPU X3350 @ 2.66GHz
Les résultats sont assez aléatoires, même si rien d'autre de gourmand ne tourne.
inplace sequential prediction : 0.402000 ms
inplace sequential prediction += : 0.354000 ms
outplace sequential prediction : 1.446000 ms
inplace parallel prediction += using 1 threads: 2.075000 ms
inplace parallel prediction += using 2 threads: 0.966000 ms
inplace parallel prediction += using 4 threads: 0.837000 ms
inplace parallel prediction += using 8 threads: 1.835000 ms
inplace parallel prediction += using 16 threads: 0.755000 ms
outplace parallel prediction using 1 threads: 2.175000 ms
outplace parallel prediction using 2 threads: 1.123000 ms
outplace parallel prediction using 4 threads: 0.774000 ms
outplace parallel prediction using 8 threads: 0.853000 ms
outplace parallel prediction using 16 threads: 0.771000 ms
ou alors
inplace sequential prediction : 0.310000 ms
inplace sequential prediction += : 0.276000 ms
outplace sequential prediction : 1.365000 ms
inplace parallel prediction += using 1 threads: 2.280000 ms
inplace parallel prediction += using 2 threads: 1.083000 ms
inplace parallel prediction += using 4 threads: 0.632000 ms
inplace parallel prediction += using 8 threads: 0.507000 ms
inplace parallel prediction += using 16 threads: 0.610000 ms
outplace parallel prediction using 1 threads: 1.108000 ms
outplace parallel prediction using 2 threads: 1.177000 ms
outplace parallel prediction using 4 threads: 1.319000 ms
outplace parallel prediction using 8 threads: 0.628000 ms
outplace parallel prediction using 16 threads: 0.724000 ms
Je pense que les temps sont beaucoup trop court et la moindre préemption/interruption fausse le résultat.
Du coup je me demande si c'est pas simplement un patch RT la solution