| View previous topic :: View next topic |
| Author |
Message |
Dennis Guest
|
Posted: Tue Nov 01, 2005 1:02 pm Post subject: CPUID Based Move Library with Fastcode winners |
|
|
Hi
I added CPUID based selection of Move functions to the FastcodeMoveDC
library made by Lars G. and uploaded it to attachments.
It would be nice to have it tested in the Move B&V.
It is going to be used in the Fastcode MM B&V.
Best regards
Dennis Kjaer Christensen
New Fastcode site
http://www.fastcodeproject.org/
|
|
| Back to top |
|
 |
John O'Harrow Guest
|
Posted: Tue Nov 01, 2005 7:33 pm Post subject: Re: CPUID Based Move Library with Fastcode winners |
|
|
Hi Dennis,
This unit does not follow the new Fastcode library format recently agreed.
It also contains a bug. If you run it on an old processor that does not
support MMX, FastcodeTarget in FastcodeCPUID.PAS would get set
fctRTLReplacement, the variable FastMove in FastcodeMoveCPUID would wrongly
get set to MoveFastcodeBlended, causing MMX based moves to be called on
non-MMX processors.
--
regards,
John
The Fastcode Project:
http://www.fastcodeproject.org/
"Dennis" <marianndkc (AT) home3 (DOT) gvdnet.dk> wrote
| Quote: | Hi
I added CPUID based selection of Move functions to the FastcodeMoveDC
library made by Lars G. and uploaded it to attachments.
It would be nice to have it tested in the Move B&V.
It is going to be used in the Fastcode MM B&V.
Best regards
Dennis Kjaer Christensen
New Fastcode site
http://www.fastcodeproject.org/
|
|
|
| Back to top |
|
 |
Dennis Guest
|
Posted: Tue Nov 01, 2005 9:09 pm Post subject: Re: CPUID Based Move Library with Fastcode winners |
|
|
Hi John
I agree.
Best regards
Dennis Kjaer Christensen
|
|
| Back to top |
|
 |
Dennis Guest
|
Posted: Thu Nov 03, 2005 8:47 pm Post subject: Re: CPUID Based Move Library with Fastcode winners |
|
|
Hi
Uploaded new version.
Best regards
Dennis Kjaer Christensen
|
|
| Back to top |
|
 |
Roman Ganz Guest
|
Posted: Fri Nov 04, 2005 9:43 am Post subject: Re: CPUID Based Move Library with Fastcode winners |
|
|
Hello Dennis,
I noticed, that your new FastcodeMoveCPUID uses not the latest Functions
from John O'Harrow. He posted his latest FastMove recently under
..attachments "Reposted FastMove Ver 3.02"
You use for example this:
//Author: John O'Harrow
//Date:
//Optimized for: Intel P4 Prescott
//Instructionset(s): SSE3
//Original name: MoveJOH_SSE3_9
procedure MoveFastcodeP4P(const Source; var Dest; Count : Integer);
but latest Version is "MoveJOH_SSE3_10"
Thanks for your great work!
Greets Roman
|
|
| Back to top |
|
 |
John O'Harrow Guest
|
Posted: Fri Nov 04, 2005 9:47 am Post subject: Re: CPUID Based Move Library with Fastcode winners |
|
|
The MoveJOH_XXX_10 functions have not yet been benchmarked on all CPU's, so
I cannot claim that they are faster yet. I can conform that they are faster
on my P4.
--
regards,
John
The Fastcode Project:
http://www.fastcodeproject.org/
"Roman Ganz" <knightkiller (AT) NOS (DOT) PAMshadowgames.ch> wrote
| Quote: | Hello Dennis,
I noticed, that your new FastcodeMoveCPUID uses not the latest Functions
from John O'Harrow. He posted his latest FastMove recently under
.attachments "Reposted FastMove Ver 3.02"
You use for example this:
//Author: John O'Harrow
//Date:
//Optimized for: Intel P4 Prescott
//Instructionset(s): SSE3
//Original name: MoveJOH_SSE3_9
procedure MoveFastcodeP4P(const Source; var Dest; Count : Integer);
but latest Version is "MoveJOH_SSE3_10"
Thanks for your great work!
Greets Roman
|
|
|
| Back to top |
|
 |
Dennis Guest
|
Posted: Fri Nov 04, 2005 11:34 am Post subject: Re: CPUID Based Move Library with Fastcode winners |
|
|
Hi John and Roman
_10 version functions have not been released in the Move B&V, have not been
validated on all targets and as John say - not been benchmarked. Therefore
they are not winners in any targets and not included in the unofficial
library and should not be in any official Fastcode library either.
There is also no reason to always asume that latest is greatest. Test and
you will know ;-)
eg. Someone needs to do the above steps.
I am sure that some of the functions will turn up to be winners :-)
Regards
Dennis
|
|
| Back to top |
|
 |
|