How make algorithm processsor cache friendly?

  • Thread starter Thread starter Borneqa
  • Start date Start date
B

Borneqa

Guest
I tested methods of Suffix Array creating, naive method, although using quick sort was more cache friendly than SA-IS.

SA-IS for 150 kB block is even 4x faster than naive, but for 10 MB block is only 1.1 x faster, although should be linear compared to n*log(n) naive. Probably it is processor cache problem. How change it to "external"? For small blocks would be a bit slower, but for large blocks faster (?). How determine processor cache size? Is any tool for cache analyzing?

Continue reading...
 
Back
Top