Re: [PATCH] PR 40049, prevent garbage collect errors on machines with vector/vector shift

[Richard Guenther <richard dot guenther at gmail dot com>]

On Fri, May 8, 2009 at 7:23 PM, Michael Meissner
<meissner@linux.vnet.ibm.com> wrote:
> The patch was a quick patch from the existing code. ?The following patch
> eliminates the unused variable.

This is ok with a testcase.  It would be nice to have a testcase with variable
shift that tests for correctness though.

Thanks,
Richard.

> 2009-05-07 ?Michael Meissner ?<meissner@linux.vnet.ibm.com>
>
> ? ? ? ?PR tree-optimization/40049
> ? ? ? ?* tree-vect-stmts.c (vectorizable_operation): If the machine has
> ? ? ? ?only vector/vector shifts, convert the type of the constant to the
> ? ? ? ?appropriate type to avoid building incorrect trees, which
> ? ? ? ?eventually have problems with garbage collection.
>
> Index: gcc/tree-vect-stmts.c
> ===================================================================
> --- gcc/tree-vect-stmts.c ? ? ? (revision 147246)
> +++ gcc/tree-vect-stmts.c ? ? ? (working copy)
> @@ -1974,11 +1974,20 @@ vectorizable_operation (gimple stmt, gim
> ? ? ? ? ?else
> ? ? ? ? ? ?{
> ? ? ? ? ? ? ?optab = optab_for_tree_code (code, vectype, optab_vector);
> - ? ? ? ? ? ? if (vect_print_dump_info (REPORT_DETAILS)
> - ? ? ? ? ? ? ? ? && optab
> + ? ? ? ? ? ? if (optab
> ? ? ? ? ? ? ? ? ?&& (optab_handler (optab, TYPE_MODE (vectype))->insn_code
> ? ? ? ? ? ? ? ? ? ? ?!= CODE_FOR_nothing))
> - ? ? ? ? ? ? ? fprintf (vect_dump, "vector/vector shift/rotate found.");
> + ? ? ? ? ? ? ? {
> + ? ? ? ? ? ? ? ? if (vect_print_dump_info (REPORT_DETAILS))
> + ? ? ? ? ? ? ? ? ? fprintf (vect_dump, "vector/vector shift/rotate found.");
> +
> + ? ? ? ? ? ? ? ? /* Unlike the other binary operators, shifts/rotates have
> + ? ? ? ? ? ? ? ? ? ?the rhs being int, instead of the same type as the lhs,
> + ? ? ? ? ? ? ? ? ? ?so make sure the scalar is the right type if we are
> + ? ? ? ? ? ? ? ? ? ?dealing with vectors of short/char. ?*/
> + ? ? ? ? ? ? ? ? if (dt[1] == vect_constant_def)
> + ? ? ? ? ? ? ? ? ? op1 = fold_convert (TREE_TYPE (vectype), op1);
> + ? ? ? ? ? ? ? }
> ? ? ? ? ? ?}
> ? ? ? ?}
>
>
>
> --
> Michael Meissner, IBM
> 4 Technology Place Drive, MS 2203A, Westford, MA, 01886, USA
> meissner@linux.vnet.ibm.com
>