This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

Re: [new-regalloc] tidy rest_of_compilation




On Fri, 2 Feb 2001, Geert Bosch wrote:

> Hi Daniel,
>
> Unfortunately the latest changes to the new-regalloc.c cause my compiler
> to blowup with a signal 6. I'm using a gcc-2.8.1 based compiler for
> bootstrapping and never had any trouble with compiling the current GCC.
>
> The patch below fixes this and it makes the code a tiny bit more
> readable at the same time I think.
>
> Index: new-regalloc.c
> ===================================================================
> RCS file: /cvs/gcc/gcc/gcc/Attic/new-regalloc.c,v
> retrieving revision 1.1.2.11
> diff -c -r1.1.2.11 new-regalloc.c
> *** new-regalloc.c	2001/02/02 06:01:14	1.1.2.11
> --- new-regalloc.c	2001/02/02 16:01:37
> ***************
> *** 881,897 ****
>        rtx move;
>   {
>     rtx body = PATTERN (move);
>     enum reg_class lhs;
>     enum reg_class rhs;
>     if (HARD_REGISTER_NUM_P (REGNO (SET_DEST (body))))
>       lhs = REGNO_REG_CLASS (REGNO (SET_DEST (body)));
>     else
>       lhs = reg_preferred_class (REGNO (SET_DEST (body)));
>     if (HARD_REGISTER_NUM_P (REGNO (SET_SRC (body))))
>       rhs = REGNO_REG_CLASS (REGNO (SET_SRC (body)));
>     else
>       rhs = reg_preferred_class (REGNO (SET_SRC (body)));
> !   return (BLOCK_FOR_INSN (move)->loop_depth + 1)
>       * REGISTER_MOVE_COST (GET_MODE (SET_SRC (body)), rhs, lhs);
>   }
>
> --- 881,901 ----
>        rtx move;
>   {
>     rtx body = PATTERN (move);
> +   struct basic_block_def * bb = BLOCK_FOR_INSN (move);
>     enum reg_class lhs;
>     enum reg_class rhs;
> +
>     if (HARD_REGISTER_NUM_P (REGNO (SET_DEST (body))))
>       lhs = REGNO_REG_CLASS (REGNO (SET_DEST (body)));
>     else
>       lhs = reg_preferred_class (REGNO (SET_DEST (body)));
> +
>     if (HARD_REGISTER_NUM_P (REGNO (SET_SRC (body))))
>       rhs = REGNO_REG_CLASS (REGNO (SET_SRC (body)));
>     else
>       rhs = reg_preferred_class (REGNO (SET_SRC (body)));
> !
> !   return (bb->loop_depth + 1)
>       * REGISTER_MOVE_COST (GET_MODE (SET_SRC (body)), rhs, lhs);
>   }
>
>

These two are equivalent code, if one works, and the other doesn't,
something else is wrong.


Most likely compute_bb_for_insn info isn't up to date.
--Dan


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]