GCC 3.3 Branch

Jan Hubicka jh@suse.cz
Sat Dec 14 10:36:00 GMT 2002


> > 
> > BIB still has the -march=pentium4 failures that prevents
> > compiling most SPEC2000 benchmarks.  This was true as of
> > 2002-12-12.  I think this is a significant problem that should be
> > fixed before BIB is merged into the trunk.
> The GCC misscompilation, is, unrforunately present at mainline too.
> We are just discussing the fix with Richard, but it is getting involved.
> THe other bug is typo.  I fixed it last week, but apparently forgot to
> commit the patch.  I will do that shortly in case BIB is currently open
> for obvious bugfixes.
> 
> Honza
OK, the patch is the following.  I am installing it as obivous to BIB
branch.  It passes bootstrap+regtest and I am now rerunning it with
-march=athlon -mcpu=pentium4 that should excercise the code in question
same way as -march=pentium4 does.

I apologize for the breakage.

Honza

Sat Dec 14 16:57:54 CET 2002  Jan Hubicka  <jh@suse.cz>
	* i386.md (movqicc splitter): Fix template.
Index: i386.md
===================================================================
RCS file: /cvs/gcc/gcc/gcc/config/i386/i386.md,v
retrieving revision 1.380.4.37
diff -c -3 -p -r1.380.4.37 i386.md
*** i386.md	11 Dec 2002 14:24:03 -0000	1.380.4.37
--- i386.md	14 Dec 2002 15:57:08 -0000
***************
*** 16270,16277 ****
    "#"
    "&& reload_completed"
    [(set (match_dup 0)
! 	(if_then_else:SI (match_operator 1 "ix86_comparison_operator" 
! 				[(reg 17) (const_int 0)])
  		      (match_dup 2)
  		      (match_dup 3)))]
    "operands[0] = gen_lowpart (SImode, operands[0]);
--- 16270,16276 ----
    "#"
    "&& reload_completed"
    [(set (match_dup 0)
! 	(if_then_else:SI (match_op_dup 1 [(reg 17) (const_int 0)])
  		      (match_dup 2)
  		      (match_dup 3)))]
    "operands[0] = gen_lowpart (SImode, operands[0]);



More information about the Gcc mailing list