michael cuthbertson uttered:
Brannon:
Thank you for your thoughts.
To be clear, the 'optimize for speed' setting in MY release is actually
slower than MY debug version - I know nothing about Ralf's settings.
That issue is separate from SQLiteSpy - I didn't mean to conflate them.
And the issue is not which version of VS I'm using.
If I simply put a 'return 0' in my callback, the time is within 2%.
Thus, I am timing the dll only and I am using the pre-compiled version,
not a VS 6 compiled version.
Ralf is compiling his own version, in Borland, and gets 2.5 times my speed.
Therefore, I believe the speed difference lies in the differences between
the two compiled versions.
The precompiled version from sqlite.org is compiled using Mingw32, which
is a Win32 targeted version of gcc. Being cross platform, gcc can not be
as aggressive in optimisations as VC or Borland, which are mostly targeted
at Intel based processors and have large, paid development teams wringing
every last drop of performance out of them.
At a guess, as SQLite code is heavy in branching, the Borland compiler is
making better use of your processor pipeline by reordering instructions,
especially if you have a P4.
Michael
Christian
--
/"\
\ / ASCII RIBBON CAMPAIGN - AGAINST HTML MAIL
X - AGAINST MS ATTACHMENTS
/ \
