Board index » delphi » FastCode MM B&V 0.56

FastCode MM B&V 0.56


2005-12-05 10:30:22 AM
delphi219
This is another incremental update to the B&V. The major rewrite is
still planned and will be discussed in a separate thread.
ToDo:
- Replace quality label validations with supported/not supported.
- Update validation and benchmark descriptions to be more descriptive -
need this information from the original authors.
- Add author names to benchmark descriptions (where available).
- Include the latest FastMM - currently 4.48.
- include the latest DCK_IA32 MM - currently 0.24.
- Add new benchmark from Dennis - SingleFPBenchmark2.
Please add your wish list items.
--
Regards,
Bruce McGee
Glooscap Software
 
 

Re:FastCode MM B&V 0.56

Hi Bruce
Very nice to see some action. I think we still have time to reach 1.0
quality before 12/12. My opinion is that the validation needs a little
improvement and that is it.
ToDo:
- Replace quality label validations with supported/not supported.
- Update validation and benchmark descriptions to be more descriptive -
need this information from the original authors.
- Add author names to benchmark descriptions (where available).
- Include the latest FastMM - currently 4.48.
- include the latest DCK_IA32 MM - currently 0.24.
- Add new benchmark from Dennis - SingleFPBenchmark2
-Megamultithreaded validation with interthread communication
-Integration of Dll test
-Latest Nexus MM (maybe we have the latest)
-DeXter support
-3GB validation ?
Benchmark descriptions are quite OK.
Validation descriptions are in the Source.
Best regards
Dennis Kjaer Christensen
 

Re:FastCode MM B&V 0.56

Hi Bruce,
I really think the MM B&V is good enough to be called 1.0. If you're
planning a rewrite then you can always call it 2.0.
I use the B&V to test FastMM before I release a new version. I have found
that if it passes all the validations and finishes the benchmarks then I
haven't broken anything.
Regards,
Pierre
 

Re:FastCode MM B&V 0.56

Hi Dennis,
Quote
quality before 12/12. My opinion is that the validation needs a little
improvement and that is it.
Remember that the benchmarks are in effect also part of the validation. It
is not like other challenges where an incorrect result in the benchmarks may
go unnoticed - if the MM is buggy then the B&V will crash in the benchmarks.
Regards,
Pierre
 

Re:FastCode MM B&V 0.56

Hi Pierre
Only exceptions are found.
This is true in other benchmarks as well.
Best regards
Dennis Kjaer Christensen
 

Re:FastCode MM B&V 0.56

Hi Pierre
I also think it is quite good. I'd like to have seen a few improvements
before accepting it as 1.0.0, but time is running out.
We simply need to push it to 1.0.0 before 12/12.
Not having the MM challenge in this years competition would not make much
sense.
Best regards
Dennis Kjaer Christensen
 

Re:FastCode MM B&V 0.56

Pierre le Riche writes:
Quote
Hi Bruce,

I really think the MM B&V is good enough to be called 1.0. If you're
planning a rewrite then you can always call it 2.0.

I use the B&V to test FastMM before I release a new version. I have
found that if it passes all the validations and finishes the
benchmarks then I haven't broken anything.

Regards,
Pierre
Think of this one as a clean up submission based on the previous B&V.
The rewrite won't happen until after.
--
Regards,
Bruce McGee
Glooscap Software
 

Re:FastCode MM B&V 0.56

Hi Pierre
Quote
Remember that the benchmarks are in effect also part of the validation. It
Correct. They also are listed in the spreadsheet under validation. An MM has
to run through all the benchmarks to be considered valid. Actually some of
the MM's only fail in the benchmark and not in the validation. I'd like
to improve the validation to cover these holes as well.
Regards
Dennis
 

Re:FastCode MM B&V 0.56

Dennis writes:
Quote
-Megamultithreaded validation with interthread communication
-3GB validation ?
I can add any new validations or benchmarks that anyone cares to send.
Quote
-Integration of Dll test
The DLL test causes some stability problems, so can not be integrated in
to the B&V. I expect this is why it was broken out in the first place.
this will be taken care of after 1.0.
Quote
-Latest Nexus MM (maybe we have the latest)
The latest NexusMM files are included. I have asked them about Delphi
2006 support but haven't heard back.
Quote
-DeXter support
Validations are run on Delphi 2006. Delphi 2005 is still the base
compiler, so it is still the one that benchmarks are run against.
Should the base compiler be changed to Delphi 2006 now that it's
released?
--
Regards,
Bruce McGee
Glooscap Software
 

Re:FastCode MM B&V 0.56

Release 0.56 is on the FastCode site. This is a clean up of the last
release. unless something serious is found, it will be promoted to 1.0.
www.fastcodeproject.org/
FastCode Memory Manager B&V 0.56 - 12-Dec-2005
Added the latest DKC_IA32_MM - 0.24.
Added the latest FastMM - 4.52.
Added new FastcodeCPUID and FastcodeMoveCPUID units from Dennis.
Added new benchmark SingleFPBenchmark2 from Dennis.
Updated the descriptions of Pierre's benchmarks to include his name.
Changed the description of the ArrayUpsizeSingleThread benchmark with
Dennis' new text.
Added Delphi 2006 support. Delphi 2005 is still the base compiler for
benchmarks.
Quote
- Replace quality label validations with supported/not supported.
This wasn't finished in time for 0.56.
--
Regards,
Bruce McGee
Glooscap Software
 

Re:FastCode MM B&V 0.56

"Bruce McGee" <XXXX@XXXXX.COM>writes
Quote

Validations are run on Delphi 2006. Delphi 2005 is still the base
compiler, so it is still the one that benchmarks are run against.
Should the base compiler be changed to Delphi 2006 now that it's
released?
I would say yes (for next year). I find it much better than D2005.
--
regards,
John
The Fastcode Project:
www.fastcodeproject.org/
 

Re:FastCode MM B&V 0.56

Hi Bruce
Very good.
Best regards
Dennis Kjaer Christensen
 

Re:FastCode MM B&V 0.56

Hi
Quote
Release 0.56 is on the FastCode site.
It is not. 0.54 is.
Regards
Dennis
 

Re:FastCode MM B&V 0.56

Dennis writes:
Quote
Hi

>Release 0.56 is on the FastCode site.

It is not. 0.54 is.

Regards
Dennis
Premature announciation. :)
I put it on the Yahoo site. Dennis (the other one) will move it from
there to the FastCodeProject.org site shortly. I will e-mail you the zip
file for testing.
--
Regards,
Bruce McGee
Glooscap Software
 

Re:FastCode MM B&V 0.56

Hi Bruce
Thanks.
Best regards
Dennis Kjaer Christensen