Source-Changes archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: CVS commit: src/sys/netipsec



In message: <87r83oflh9.fsf%snark.piermont.com@localhost>
            "Perry E. Metzger" <perry%piermont.com@localhost> writes:
: 
: David Laight <david%l8s.co.uk@localhost> writes:
: > Indeed, on a modern x86 you do not want (ever) to execute movs{b,w,l}
: > unless it is repeated AND the repeat count is considerable
: > (unless you are optimising for space).
: > 
: > ISTR that something like:
: > 1:  mov     (%esi,%ecx,4),%eax
: >     mov     %eax,(%edi,%ecx,4)
: >     dec     %ecx
: >     jnc     1b
: > is faster than 'rep movsw' for %ecx < (about) 16.
: 
: Are you including the overhead of calling the function?
: 
: In any case, it is possible to teach GCC to use different sequences
: than the ones it uses right now.

how does one teach gcc to generate optimal code for all sizes from 16
to 64k?

Warner



Home | Main Index | Thread Index | Old Index