LO_SUM still breaking rs6000, revert patch?

Geoff Keating geoffk@geoffk.org
Tue Jul 10 11:52:00 GMT 2001

Hi Richard,

this patch:

+       * emit-rtl.c (adjust_address, adjust_address_nv): Handle an
+       address that is a LO_SUM specially.
+       * explow.c (plus_constant_wide, case LO_SUM): Deleted.

is the cause of the current build failure on powerpc:

/maat/heart/tbox/powerpc-eabisim/build/gcc/xgcc -B/maat/heart/tbox/powerpc-eabisim/build/gcc/ -B/maat/heart/tbox/objs/powerpc-eabisim/bin/ -B/maat/heart/tbox/objs/powerpc-eabisim/lib/ -isystem /maat/heart/tbox/objs/powerpc-eabisim/include -c -DSkip_f2c_Undefs -DAllow_TYQUAD -I. -I/maat/heart/tbox/cvs-gcc/gcc/libf2c/libI77 -I.. -I/maat/heart/tbox/cvs-gcc/gcc/libf2c/libI77/..  -DHAVE_CONFIG_H -g -O2 /maat/heart/tbox/cvs-gcc/gcc/libf2c/libI77/lread.c
cc1: warning: changing search order for system directory "/maat/heart/tbox/objs/powerpc-eabisim/include"
cc1: warning:   as it has already been specified as a system directory
/maat/heart/tbox/cvs-gcc/gcc/libf2c/libI77/lread.c: In function `l_R':
/maat/heart/tbox/cvs-gcc/gcc/libf2c/libI77/lread.c:261: Internal error: Segmentation fault

I think your subsequent patches have not succeeded in restoring the
behaviour of GCC to the state before this patch:

        * explow.c (plus_constant_wide, case LO_SUM): New case.
        (plus_constant_for_output_wide): Delete.
        * rtl.h (plus_constant_for_output): Delete.
        * alias.c (canon_rtx, init_alias_analysis): Call plus_constant
        instead of plus_constant_for_output.
        * recog.c (offsettable_address_p, adj_offsettable_operand): Likewise.
        * config/darwin.c, config/arm/arm.c, config/m68k/m68k.c: Likewise.
        * config/m88k/m88k.c, config/mips/mips.c, config/pa/pa.c: Likewise.
        * config/rs6000/rs6000.c, config/sparc/sparc.c: Likewise.
        * config/sparc/sparc.md: Likewise.
        Convert some change_address calls to adjust_address.

Should this patch (plus the subsequent patches) be reverted pending a
better study of the situation?

- Geoffrey Keating <geoffk@geoffk.org>

More information about the Gcc-regression mailing list