| View previous topic :: View next topic |
| Author |
Message |
Dennis
Guest
|
|
| Back to top |
|
 |
Dennis
Guest
|
 Posted: Fri Apr 20, 2007 1:05 am Post subject: Re: Fastcode Trim B&V 0.6.1 |
 |
|
Hi
Uploaded AMD 64 and Yonah results.
Now we need the remaning benchmark results and some bench stability runs.
Best regards
Dennis Kjaer Christensen |
|
| Back to top |
|
|
Dennis
Guest
|
| Posted: Fri Apr 20, 2007 8:53 pm Post subject: Re: Fastcode Trim B&V 0.6.1 |
|
|
Hi
The sorting of results is turned of in this release. This was a mistake.
Best regards
Dennis Kjaer Christensen |
|
| Back to top |
|
|
Dennis
Guest
|
| Posted: Fri Apr 20, 2007 9:17 pm Post subject: Re: Fastcode Trim B&V 0.6.1 |
|
|
Hi
In the Excel file BenchStability.xls you see a bunch of bench stability test
results. The conclusion of these results is that some of my functions where
the main reason that the bench stability numbers where so bad. I removed
those functions.
Together with some other finetrimmings of the benchmark stability is now <
2%. I am working on a table of exact numbers. This is for a 3 times rerun of
the benchmark all and considering the huge amount of functions we have it is
not exceptionally bad. I hope we can get below the 1% limit by removing some
functions and further trimming the benchmark.
If you have some functions that can be removed then tell me soon.
I will need all the functions sizes for the next release. The a full run of
the benchmark on all targets will enable the SpreadsheetCreator to make a
list of functions that are candidates for removal.
Best regards
Dennis Kjaer Christensen |
|
| Back to top |
|
|
Dennis
Guest
|
| Posted: Fri Apr 20, 2007 9:22 pm Post subject: Re: Fastcode Trim B&V 0.6.1 |
|
|
Hi
FastMM4 is included in the release and it revealed a 1 byte TStringList leak
which is now fixed with
procedure TMainForm.FormClose(Sender: TObject; var Action: TCloseAction);
begin
.......
FResultList.Free;
end;
FastMM4 is enabled/disabled in the dpr file
program TrimBV;
uses
// Fastmm4,
// Fastmm4messages,
.................
Best regards
Dennis Kjaer Christensen |
|
| Back to top |
|
|
Dennis
Guest
|
| Posted: Fri Apr 20, 2007 9:29 pm Post subject: Re: Fastcode Trim B&V 0.6.1 |
|
|
Hi
| Quote: | FastMM4 is enabled/disabled in the dpr file
program TrimBV;
uses
// Fastmm4,
// Fastmm4messages,
|
but be aware of this bug
IFDEF's in DPR uses clause
http://qc.codegear.com/wc/qcmain.aspx?d=8718
That will cause the two FastMM lines to be removed if the IDE modifies the
dpr.
Best regards
Dennis Kjaer Christensen |
|
| Back to top |
|
|
Dennis
Guest
|
| Posted: Sat Apr 21, 2007 12:32 am Post subject: Re: Fastcode Trim B&V 0.6.1 |
|
|
Hi
The Yonah has now run for approx. 6 hours and is half way through the 3
bench all run. Max benchmark spread is 1.4%
The Dothan has run approx 3 hours and it is at 1.3%
The Prescott is always a little worse and it is at 1.9% after approx. 3
hours.
I will give the final numbers some time tomorrow when the runs have finished
;-)
Best regards
Dennis Kjaer Christensen |
|
| Back to top |
|
|
|
FAQ
Search
Memberlist
Usergroups
Register
Profile
Log in to check your private messages
Log in
