[Bug target/79170] [7 regression] memcmp builtin expansion sequence can overflow

[acsawdey at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=79170

--- Comment #2 from acsawdey at gcc dot gnu.org ---
Author: acsawdey
Date: Mon Jan 30 23:24:24 2017
New Revision: 245041

URL: https://gcc.gnu.org/viewcvs?rev=245041&root=gcc&view=rev
Log:
2017-01-27  Aaron Sawdey  <acsaw...@linux.vnet.ibm.com>

        PR target/79170
        * gcc.dg/memcmp-1.c: Improved to catch failures seen in PR 79170.

2017-01-27  Aaron Sawdey  <acsaw...@linux.vnet.ibm.com>

        PR target/79170
        * config/rs6000/altivec.md (*setb_internal): Rename to setb_signed.
        (setb_unsigned) New pattern for setb with CCUNS.
        * config/rs6000/rs6000.c (expand_block_compare): Use a different
        subfc./subfe sequence to avoid overflow problems.  Generate a
        shorter sequence with cmpld/setb for power9.
        * config/rs6000/rs6000.md (subf<mode>3_carry_dot2): Add a new pattern
        for generating subfc. instruction.
        (cmpstrsi): Add TARGET_POPCNTD predicate as the generate sequence
        now uses this instruction.



Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/config/rs6000/altivec.md
    trunk/gcc/config/rs6000/rs6000.c
    trunk/gcc/config/rs6000/rs6000.md
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/testsuite/gcc.dg/memcmp-1.c