Re: Fastcode MM Rules

From: Fredrik Lysholm (fredrik.lysholm_at_mogul.com)
Date: 02/10/05 

Date: Thu, 10 Feb 2005 17:26:43 +0100

"Dennis" <marianndkc@home3.gvdnet.dk> wrote in message
news:420b7a2b$1@newsgroups.borland.com...
> Hi Fredrik
>
>> 2. CPUID code
>> This project has two categories with very specific rules, why debate
>> that?
>> The RTL doesn't allow CPUID code and the blend does.. If you (as me) want
>> the fastest version just use Blend..
>
> We are not! We are debating the use of patching not the use of CPUID.
>
You're not what?
Having both a Blend or RTL target?
Talking about CPUID? Well I might be wrong but I thought I read you talking
about why Borland didn't allow CPUID etc.. all I wanted to say is live with
Borlands rules and use CPUID code in the Blend category instead..

>> For the heroes actually coding this challanging "IFDEF:ing" the two cases
>> shouldn't be impossible.. And if you feel you don't just make a Blend and
>> I'll make a RTL-entry out of that one if you wish ;)
>
> If you use IFDEF then you make two functions in one and only one is
> benchmarked and validated.
>

Hmm.. well I'd assume most people doing the benchmarks would actually set
the IFDEF switches as I did with my results to show differnces to the devs
with different setup but I guess that some wouldn't do that :)

>> Regards,
>> Fredrik
>
> Regards
> Dennis
>
>