Hi Thomas, hi everybody else,
it would be great, if you all could send comments to this
compiler-flag-discussion. I believe it's really important, that
the ./configure - make - make install - way of compiling won't create
slower code then the old fashioned one does.
What about a 'compiler-flag-database': let the configure check, which
compiler it detects, and add all known well working compiler flags to the
CFLAGS.
Which one these are exactly, we have to discuss here:
gcc 2.7.2 and egcs 1.1.1 produce correct and (on pentium class cpu's)
faster code on the computers I tested with
$ gcc -malign-loops=2 -malign-jumps=2 -malign-functions=2 -O3 -fPIC
-ffast-math -funroll-loops -fomit-frame-pointer.
Dieter Nuetzel reported some other flags especially for pgcc; (could
you post them again, Dieter ??)
Some commercial compilers have optimizer flags up to -O6 or -O9, but I
don't know which ones this are and what the flags are especially for.
I'm not shure, if it would be a good idea to force the compiler to produce
processor-specific optimisations (I know, pgcc knows various pentium
class cpu's and k6 - ), but it could be - anybody knows, how much faster
the generated code is ?? Since 486/pentium optimized code didn't worked
on 386 cpu's it could be time again to accept that optimisations are
mostly specific to a particular target architecture.
A way out could be a
--disable-processor-specific-optimisations-at-compile-time
switch for distributors and people, which don't know the target processor
exactly.
Should the final release generate a stripped library, which is smaller ?
Thomas, could you create a README or INSTALL file, which describes all
switches and a 'how to install Mesa, if ./configure fails' ??
- Holger
The autoconf stuff works now -- I'd like to see it in the next
release. Test it all and let's make it better !!
_______________________________________________
Mesa-dev maillist - [EMAIL PROTECTED]
http://lists.mesa3d.org/mailman/listinfo/mesa-dev
