gcc 3.2.1

[Chris Croswhite]

Thanks for the reply and sorry for being so vague.


> You mean you removed the -g from the GCC build flags and the resulting
> GCC's bad? That shouldn't happen. Which platform are you using? What
> compiler did you start from? Did you 'make bootstrap' or just 'make'?
> If you did bootstrap, the compiler has proven that it can compile
> itself. Did you run the GCC testsuite? If so, do you get similar
> results to other reports for your platform? (follow links from the
> GCC-3.2.1 build status page or look in the gcc-testsuite mailing list
> archives.)

I need to correct my statement, as I have done more testing since I wrote
the previous email.  By removing the dubug option, it had no affect (other
than get ride of some of the debug stuff we didn't need).  However, we think
now we have hit a bug (looks like we are getting some bad pointers from a
linked tcl library [compiled with 3.2.1 as well].  I will follow up with a
test case.

>
> Is this limited to one application? With or without -g should not
> affect the generated code but you may have found a bug where it does;
> if you can reproduce this, please consider submitting a bug report.

We are using this on our internal code.


>
>> Also, I ahve found that gcc 3.2.1 has increased the size of my
>> executable by 35% (10.1M -> 15.2M).  Is there a way to reduce this?
>

We are using g++ (for mixed c and c++).  The comparison is to gcc2.95.3,
although, with version 3.0.2, we also experienced a code bloat (about 30%).
   We are using 02 and have found 03 to produce slightly smaller code (~20k
smaller). As far as dead code elimination, any way to truely eliminate it?

On a positive note, if we can nail down this bug (which appears also in head
and 3.1.1), we can gain a 6-8% run time performance...this is great!

> Versus which compiler? Is your application C++? g++ 3.x have a
> different, more conformant implentation of the STL, etc., which could
> account for extra code size - I understand this is mostly tuning
> problems with GCC's code inliner and this is discussed periodically on
> the main GCC mailing list. Alternatively, you might not have enabled
> dead code elimination; are you compiling with -O2 or above?
>
> Rup.