[PATCH] Fix tree-ssa-coalesce ICE on ASM_EXPR (PR tree-optimization/31866)

[Richard Guenther]

On 6/20/07, Jakub Jelinek <jakub@redhat.com> wrote:
> Hi!
>
> create_outofssa_var_map ICEs on the attached testcase, because
> DECL_REGISTER VAR_DECL input isn't SSA_NAME (that's correct).
> create_outofssa_var_map needs SSA_NAME_VERSION of both the input
> and output to coalesce them, therefore trying to do anything to
> coalesce a VAR_DECL with SSA_NAME wouldn't fit well into the
> current coalescing framework.
>
> Ok for trunk?

This is ok.

Thanks,
Richard.

> BTW, the constraint analysis seems to be too simplistic to me,
> although info gcc requests that if a number is together with some
> other letters in one variant, it comes last (so "0rm" is invalid),
> e.g. "0,1,2" or "0,rm" are valid constraints and I'm not sure if it
> is a good idea to coalesce in that case.  My guess would be we want
> to coalesce only if the constraint contains just a number (say "5"),
> or if each variant contains the same number (say "1,1" or "5,5,5").
>
> 2007-06-20  Jakub Jelinek  <jakub@redhat.com>
>
>         PR tree-optimization/31866
>         * tree-ssa-coalesce.c (create_outofssa_var_map): Do nothing
>         if ASM_EXPR's input is not a SSA_NAME.
>
>         * gcc.dg/pr31866.c: New test.
>