Re: tree-ssa bootstrap failure on Alpha

[Diego Novillo <dnovillo at redhat dot com>]

On Tue, 2003-06-03 at 17:01, Falk Hueffner wrote:

> the bootstrap fails as follows:
> 
Thanks.  Fixed with the attached patch.  Jeff's patch to avoid
gimplifying an already gimplified expression triggered a bug in the
gimplification of CALL_EXPRs.

The problem is that there are some machine-dependent builtins that we
refuse to gimplify because the back ends typically want those builtins
to have specific arguments (this used to trigger compilation problems on
IA64).  So, we refuse to gimplify those CALL_EXPRs.

Before, the is_simple_call_expr predicate would just return 'true' on MD
builtins and then the gimplifier would proceed to mark the builtin as
non-gimple in simplify_call_expr.  But now, we have the short circuit
that will refuse to gimplify expressions that are already in gimple
form.

The patch moves the code to mark the CALL_EXPR non-gimple into the
is_simple_call_expr predicate.  I'm not sure I like having a predicate
with such a side-effect, but I see no other obvious place to move this
code to.  Jason, any suggestions?


Diego.

Attachment: 20030603-fix-alpha-bootstrap.diff
Description: Text document