[Bug target/11156] new bootstrap failure on mips-sgi-irix6.5 stage2 libgcc2 fp-bit.c

["ghazi at caip dot rutgers dot edu" <gcc-bugzilla at gcc dot gnu dot org>]

PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=11156



------- Additional Comments From ghazi@caip.rutgers.edu  2003-06-13 03:21 -------
Subject: Re:  New: new bootstrap failure on mips-sgi-irix6.5 stage2 libgcc2 fp-bit.c

 > From: Richard Sandiford <rsandifo@redhat.com>
 > 
 > "ghazi@gcc.gnu.org" <gcc-bugzilla@gcc.gnu.org> writes:
 > > PLEASE REPLY TO gcc-bugzilla@gcc.gnu.org ONLY, *NOT* gcc-bugs@gcc.gnu.org.
 > > Sometime between June 7 and June 8, I started getting a new bootstrap failure 
 > > on mainline for mips-sgi-irix6.5 in stage2 building libgcc2.  It looks like 
 > > this:
 > > 
 > > fp-bit.c: In function `__pack_f':
 > > fp-bit.c:205: internal compiler error: Segmentation fault
 > > Please submit a full bug report,
 > > with preprocessed source if appropriate.
 > > See <URL:http://gcc.gnu.org/bugs.html> for instructions.
 > > make[3]: *** [libgcc/./_pack_sf.o] Error 1
 > 
 > Ah, so it _isn't_ just me ;)
 > 
 > I believe it's a bug in the native irix assembler.  In fold(),
 > we have the code:
 > 
 >         .set    noreorder
 >         .set    nomacro
 >         bne     $2,$0,.L3414
 >         move    $6,$2
 >         .set    macro
 >         .set    reorder
 > 
 > which gets assembled as:
 > 
 >    2caf4:       10400004        beqz    v0,2cb08 <.LCFI407+0x184fc>
 >    2caf8:       8f810001        lw      at,1(gp)
 >                         2caf8: R_MIPS_GOT16     .text+0x14fd0
 >    2cafc:       24214fd0        addiu   at,at,20432
 >                         2cafc: R_MIPS_LO16      .text
 >    2cb00:       00200008        jr      at
 >    2cb04:       00000000        nop
 >    2cb08:       00403025        move    a2,v0
 > 
 > In other words, the bne has been changed into a long branch sequence,
 > 
 > I'm suprised the assembler does this for branches inside a set
 > noreorer/set nomacro block, especially given that the original
 > branch _was_ within range.
 > 
 > Anyway, the problem is that delay slot insn is now only executed
 > when the branch isn't taken.
 > 
 > I'm not sure what's triggered it.  It might be the recent growth in
 > fold(), or it might be related to the 'label in delay slot' problem
 > that was seen on sparc (the irix assembler warns about the same thing).
 > But whatever caused it, it's still a bug in the assembler.
 > Richard

Thanks for the analysis.  Do you have any ideas on a solution?
(Perhaps to somehow avoid triggering the assembler bug?)

		Thanks,
		--Kaveh
--
Kaveh R. Ghazi			ghazi@caip.rutgers.edu