[PATCH] PR30391: GIMPLE_MODIFY_STMT vs. middle-end

Ian Lance Taylor iant@google.com
Mon Feb 5 20:03:00 GMT 2007


"Roger Sayle" <roger@eyesopen.com> writes:

> ! /* Build a GIMPLE_MODIFY_STMT node.  This tree code doesn't have a
> !    type, so we can't use build2 (a.k.a. build2_stat).  */
>   
>   tree
> ! build_gimple_stmt_stat (tree arg0, tree arg1 MEM_STAT_DECL)
>   {
>     tree t;
>   
> !   t = make_node_stat (GIMPLE_MODIFY_STMT PASS_MEM_STAT);
>     /* ?? We don't care about setting flags for tuples...  */
>     GIMPLE_STMT_OPERAND (t, 0) = arg0;
>     GIMPLE_STMT_OPERAND (t, 1) = arg1;
>     return t;
>   }

This kind of change doesn't make sense to me.  We want to add many
more tuples, in order to reduce memory size.  GIMPLE_MODIFY_STMT was
just the first tupelization exercise.  We shouldn't write code which
assumes that it is the only one.

Ian



More information about the Gcc-patches mailing list