This is the mail archive of the
gcc-bugs@gcc.gnu.org
mailing list for the GCC project.
Re: Bug in code generation of gcc-2.95.2
- To: ak at muc dot de (Andi Kleen)
- Subject: Re: Bug in code generation of gcc-2.95.2
- From: root at jacob dot remcomp dot fr (root)
- Date: Tue, 4 Jan 2000 22:39:16 +0100 (CET)
- Cc: gcc-bugs at gcc dot gnu dot org
Hi!
Well, today I corrected a bug in the code generation of lcc-win32, my compiler
system... I can understand you guys. I thank you for taking care of this, but
the job seems an order of magnitude bigger than in lcc-win32: gcc code size is
approx 10MB, lcc-win32 code size is approx 1MB...
I wonder though, if all this complex optimizations are really worth... My
system generates for the same loop:
xorl %edi,%edi
_$10:
movl _PickList+4(,%edi,4),%ebx
movl %ebx,_PickList(,%edi,4)
incl %edi
cmpl $7,%edi
jl _$10
Straightforward, easy, no complex transformations and...
>>> This code is exactly as fast as the one generated by gcc!! <<<
One memory read, one memory write per pass. ANY compiler can't do better than
that.
If possible, keep me in carbon copy of this followup. I find it very interesting
and lets me learn about gcc. I would like to help, but the last time I tried
it looked like an impossible job without some years training...
I spent a whole week trying to correct a bug to no avail. To find the exact
reason of this bug by the way it took me 1 day of work.
--
Jacob Navia Logiciels/Informatique
41 rue Maurice Ravel Tel 01 48.23.51.44
93430 Villetaneuse Fax 01 48.23.95.39
France