Re: GCC for Mac OS X

[Brad Lucier <lucier at math dot purdue dot edu>]

Re:

> 4. Compiler turnaround.  Mac developers are used to the fast
> compile times offered by Metrowerks, and are dismayed by GCC's
> slowness.  Do we need precompiled headers?  Should file system
> probes be cached?  There are a lot of theories going around, but
> facts are in short supply.  But however it's accomplished, the
> compiler needs to go faster.

Every once in a while I profile the compiler when it seems to be going
too slowly on my code.  Right now, it seems that compute_flow_dominators,
which uses an O(N^2) algorithm instead of an available O(N) algorithm is
a problem for some code, see

http://gcc.gnu.org/ml/gcc/2000-02/msg00359.html

and

http://gcc.gnu.org/ml/gcc-patches/2000-02/msg00315.html

Re:

> 3. PowerPC code quality.  Apple has more than a few bit-twiddling
> graphics engineers who pore over inner loops and such.  They need
> the compiler to generate faster code.  Exception handling and
> AltiVec support are two specific areas; for instance, Apple is
> using a big hairy patch from Motorola for AltiVec code gen, but
> from what I've seen of the patch so far, it seems problematic
> for mainstream GCC.

And, as has been said several times already, the FSF needs a copyright
assignment to include the Altivec code; perhaps you can talk to Kalpesh
Gala, pesh@ncct.sps.mot.com, Motorola's Altivec evangelist, to get
Motorola to move faster on this.

Brad