multicore toolkit

[altenbach]

There is always a parallelization overhead (splitting up the problem and reassembling the result). There is also the loop overhead. That's why you need to benchmark it.

I still don't understand what problem you are trying to solve. A factor of 2 or 3 in speed difference is not really significant and the toolkit does give you better performance.

[ssara]

hi altenbach

@altenbach wrote:

I still don't understand what problem you are trying to solve. A factor of 2 or 3 in speed difference is not really significant and the toolkit does give you better performance.

---

in my work i used GPU and i got high speeds, but i also need to test muticore toolkit, compare and discuss the benchmark on it.

so, why the plain AxB slower than 1 thread MASM AxB?? i expected to get the same time delay, the function node in plain call c, in MASM is the function node call C or openMP ??

when multiplying square matrices, the situation is reversed ! the plain became faster than 1thread MASM  

any explanation plz

[altenbach]

---

@ssara wrote:

when multiplying square matrices, the situation is reversed ! the plain became faster than 1thread MASM  

any explanation plz

---

The MASM toolkit is significantly newer and uses code libraries that are different from the regular tools. It is possible that they are e.g. more aware of the exact cache level sizes, etc. so the performance will also depend on the exact CPU architecture and such. It could even depend on how you benchmark.

In summary, a factor of two in either direction is nothing serious and not worth losing sleep over. It would need significant research to find the explanation. There are many subtly different ways to do this calculation.

[ssara]

thanks altenbach

---