linux-kernels and -foptimize-sibling-calls

[Franz Sirl]

On Wednesday 01 November 2000 22:43, Richard Henderson wrote:
> On Wed, Nov 01, 2000 at 09:09:50PM +0100, Franz Sirl wrote:
> > Well, with gcc-2.96 20000928 (I can't currently build with yesterdays CVS
> > due to cpp macro problems) out of the ~500 .o files in my typical
> > 2.4.0-t10 kernel build ~80 differ in code. Now what?
>
> Pick one, preprocess it, and I'll try and figure out what went wrong.

Done. I used net/ipv4/tcp.c since the resulting assembler really looks quite 
different in some areas and TCP was the area I had problems with. 
Unfortunately I'm not sure if there's really a bug in compiling this file as 
a lot of the assembler differences look like scheduling/register allocation 
differences on a quick browse.

On Wednesday 01 November 2000 22:45, Geoff Keating wrote:
> Pick one of them, get two .s files, run 'diff', and show us the first
> 50 lines of differences.  (Also provide a .i file so we can debug it
> for ourselves).

Diffing is quite difficult, cause the labels change their numbering :-(.

The kernel uses these options for compile:

cc1 tcp.i -quiet -dumpbase tcp.c -msoft-float -mmultiple -mstring -O2 -Wall 
-Wstrict-prototypes -Wno-uninitialized -version -fomit-frame-pointer 
-fno-strict-aliasing -fsigned-char -ffixed-r2 -o tcp.s

Franz.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: tcp.i.bz2
Type: application/x-bzip2
Size: 67263 bytes
Desc: not available
URL: <https://gcc.gnu.org/pipermail/gcc/attachments/20001102/bbb34979/attachment.bz2>