[Gllug] Patterns of locality of reference in xterm
james.dutton at gmail.com
Mon Jun 20 08:31:53 UTC 2011
On 20 June 2011 00:40, Nix <nix at esperi.org.uk> wrote:
> On 19 Jun 2011, Adrian McMenamin verbalised:
>> A few people expressed interest in this (honestly, they did).
>> My first, rather crude, results can be seen here:
> Doing it for GCC might *very* well be useful, because locality of
> reference (particularly within single compilation passes) translates
> directly into speed: GCC is distinctly cache-unfriendly and fixing that
> is a long, long slog.
Why not use a cache-oblivious algorithm.
I used one for some crypto implementation recently and it sped it up
by about 20x.
But for my optimization I only had to arrange the data to be cache
oblivious. The code itself was so small it would all fit in the Level
1 cache of any modern CPU.
I suppose you are wishing the code to be cache oblivious with the code
size being larger than the cache?
Gllug mailing list - Gllug at gllug.org.uk
More information about the GLLUG