Fix middle-end/17767 (take 2)

[Uros Bizjak]

Roger Sayle wrote:

>I agree with RTH that a better solution to PR middle-end/17767 would
>be to correctly implement vector comparisons in both fold_rtx and
>simplify_relational_operation_1 (perhaps in addition to a change like
>yours above).
>  
>
I think that proper vector handling in these functions is 4.1 material. 
A minimum fix for PR17767 and its dupes is atached to this message. 
Patch is bootstrapped on pentium4-linux-gnu, regtested c,c++. A testcase 
is also attached.

2005-01-04  Uros Bizjak  <uros@kss-loka.si>

    PR middle-end/17767
    * cse.c (fold_rtx) [RTX_COMPARE, RTX_COMM_COMPARE]: Don't attempt
    any simplifications of vector mode comparison operators.
    * simplify-rtx.c (simplify_relational_operation): Fix variable name.

testsuite ChangeLog:
  
2005-01-04  Uros Bizjak  <uros@kss-loka.si>

    PR middle-end/17767
    * gcc.dg/i386-mmx-6.c: New test case.

Uros.

-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: pr17767.diff
URL: <http://gcc.gnu.org/pipermail/gcc-patches/attachments/20050104/0d2a3687/attachment.ksh>
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
Name: i386-mmx-6.c
URL: <http://gcc.gnu.org/pipermail/gcc-patches/attachments/20050104/0d2a3687/attachment.c>