Re: Fastcode AnsiStringReplace B&V 1.7.1
- From: "Dennis" <marianndkc@xxxxxxxxxxxxxxx>
- Date: Thu, 21 Dec 2006 16:17:36 +0100
Hi John
A question to John: StringReplace_JOH_IA32_15 has IFDEF Small.
Would it be better to have two functions with different names?
That is necessary. Otherwise we would only benchmark and validate one of the
two functions. Old discussion.
I was experimenting with reducing the size of the function. The {$IFDEFof
SMALL} reduces the size of the function by a few hundred bytes at a cost
a 5% reduction in speed. I left this in (but undefined) just in casethe
CodeGear think that the function size is too excessive for inclusion in
Highlander RTL.
Bad idea. Then you are selling unvalidated and unbenchmarked code to them
that migth not be the winner of the IA32 size penalty target.
Please make two versions and make us able to benchmark validate and elect
winners as we always do.
Best regards
Dennis Kjaer Christensen
.
- References:
- Fastcode AnsiStringReplace B&V 1.7.1
- From: Jouni Turunen
- Re: Fastcode AnsiStringReplace B&V 1.7.1
- From: Dennis
- Re: Fastcode AnsiStringReplace B&V 1.7.1
- From: Jouni Turunen
- Re: Fastcode AnsiStringReplace B&V 1.7.1
- From: John O'Harrow
- Fastcode AnsiStringReplace B&V 1.7.1
- Prev by Date: Re: Fastcode AnsiStringReplace B&V 1.7.1
- Next by Date: Re: Fastcode AnsiStringReplace B&V 1.7.1
- Previous by thread: Re: Fastcode AnsiStringReplace B&V 1.7.1
- Next by thread: Data alignment code... What goes on?
- Index(es):
Relevant Pages
|