[fpc-devel] Patch, speed improvement for ARM-rtl procedure 'move'

[Bernd Mueller]

Hello,

the attached patch improves the performance of the rtl Move_blended 
procedure by about 16 % on ARMv4, when both destination and source 
addresses are aligned. I rearranged the code and could remove one 
instruction of the inner loop 'Ldwordloop'. I think the patch should 
work for the Move_pld procedure too (ARMv5), but I could not test it.

Regards, Bernd.

-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: armrtlmove
URL: <http://lists.freepascal.org/pipermail/fpc-devel/attachments/20080310/64998d53/attachment.ksh>