From: Kenneth J.Davis (jeremyd@computer.org)
Date: Tue Aug 13 2002 - 12:40:10 EDT
Win32 builds I have compiled:
[Any differences between AbiWord cvs and any builds I compile are
documented and patches provided; I do not change the optimizer
setting, except for my VC7 builds where I explicitly set OPTIMIZER
to empty on the command line so the tinderbox log is not cluttered
with messages of standard compiler does not support optimizations.
Nightly builds are only done with unmodified AbiWord cvs source.]
Official builds by me (1.0.1/1.0.2) and any periodic builds I make
are compiled using Microsoft Visual C/C++ version 6 Enterprise Ed
with whatever the standard optimization options are.
Which should be (if I didn't overlook any): OPTIMIZER = -O2 -Ob1 ==
/Og use global optimizations
/Oi generate intrinsic functions
/Ot favors fast code
/Oy omits frame pointer
/Ob1 limits inlining to explicit inlined and C++ member function in clas def
/Gs (I think) control stack probe (controls stack size used by functions)
/Gf enable string pooling
/Gy enable function level linking
I believe the builds are then done with the default architecture
optimization /GB, which according to my VC6 docs optimizes in
favor of the Pentium but is still a blend of 386/486/Pentium/ and
Pentium Pro optimizations. (_M_IX86 == 500) With the note
future versions may change to optimize for dominant processor.
Stable nightly builds are done with Microsoft Visual C/C++
version 5 Enterprise Ed (with required updated platform SDK) and
also with whatever the standard optimizations are, so it should be the
same as above VC6 builds (including blended architecture optimizations,
_M_IX86 == 500).
Head nightly builds (when it builds :-) are done with Microsoft
Visual C/C++ 7 Standard Ed, so it does not support any optimizations,
hence OPTIMIZER = ; thus limiting its options to -GF -Gy (string
pooling and function level linking) and any other default options.
Any other questions or I wasn't clear enough, let me know.
Jeremy Davis
jeremyd@computer.org
Jordi Mas <jmas@softcatala.org> wrote on 8/13/2002 5:39:00 AM:
>
>En/na Tomas Frydrych ha escrit:
>> I have been in touch with the reporter of bug 3556 and it looks that
>> our official win32 binaries are over-optimized for a particular
>> processor, crashing on his AMD; he pointed out that the only
>> binaries that work for him are those from Jeremy's site.
>>
>> We really need to avoid any optimatization that relies on processor
>> specific instructions in the official binaries.
>>
>> Tomas
>>
>
>Do we know which optimisation parameters is Jeremy not using? Is he trying the
>same exact version of the software? Is he the only user that has reported this
>behaviour?
>
>Thanks,
>--
>
>Jordi Mas
>http://www.softcatala.org
This archive was generated by hypermail 2.1.4 : Tue Aug 13 2002 - 13:51:40 EDT