This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [CFG] unrolling heuristic tweek
- From: Daniel Berlin <dberlin at dberlin dot org>
- To: Jan Hubicka <jh at suse dot cz>
- Cc: gcc-patches at gcc dot gnu dot org, <gcc-pdo at atrey dot karlin dot mff dot cuni dot cz>
- Date: Mon, 6 May 2002 12:20:41 -0400 (EDT)
- Subject: Re: [CFG] unrolling heuristic tweek
On Mon, 6 May 2002, Jan Hubicka wrote:
>
> Mon May 6 18:09:58 CEST 2002 Jan Hubicka <jh@suse.cz>
> * params.def (PARAM_MAX_UNROLL_TIMES): set to 8
^^^^^^^^^^
> * unroll-new.c (unroll_or_peel_loop): Disable peeling when number
> of iterations is unknown.
>
> Index: params.def
> ===================================================================
> RCS file: /cvs/gcc/egcs/gcc/params.def,v
> retrieving revision 1.11.2.9
> diff -c -3 -p -r1.11.2.9 params.def
> *** params.def 1 May 2002 20:45:34 -0000 1.11.2.9
> --- params.def 6 May 2002 15:58:29 -0000
> *************** DEFPARAM(PARAM_MAX_UNROLLED_INSNS,
> *** 152,158 ****
> DEFPARAM(PARAM_MAX_UNROLL_TIMES,
> "max-unroll-times",
> "The maximum number of unrollings of a single loop",
> ! 4)
> /* The maximum number of insns of a peeled loop. */
> DEFPARAM(PARAM_MAX_PEELED_INSNS,
> "max-peeled-insns",
> --- 152,158 ----
> DEFPARAM(PARAM_MAX_UNROLL_TIMES,
> "max-unroll-times",
> "The maximum number of unrollings of a single loop",
> ! 16)
^^^^^^^^^^^^^^^
> /* The maximum number of insns of a peeled loop. */
> DEFPARAM(PARAM_MAX_PEELED_INSNS,
> "max-peeled-insns",
> *************** DEFPARAM(PARAM_MAX_PEELED_INSNS,
> *** 162,168 ****
> DEFPARAM(PARAM_MAX_PEEL_TIMES,
> "max-peel-times",
> "The maximum number of peelings of a single loop",
> ! 15)
> /* The maximum number of insns of an unswitched loop. */
> DEFPARAM(PARAM_MAX_UNSWITCH_INSNS,
> "max-unswitch-insns",
> --- 162,168 ----
> DEFPARAM(PARAM_MAX_PEEL_TIMES,
> "max-peel-times",
> "The maximum number of peelings of a single loop",
> ! 16)
^^^^^^^^^^^^^^^^^
> /* The maximum number of insns of an unswitched loop. */
> DEFPARAM(PARAM_MAX_UNSWITCH_INSNS,
> "max-unswitch-insns",
> Index: unroll-new.c
> ===================================================================
> RCS file: /cvs/gcc/egcs/gcc/Attic/unroll-new.c,v
> retrieving revision 1.1.2.25
> diff -c -3 -p -r1.1.2.25 unroll-new.c
> *** unroll-new.c 6 May 2002 13:42:12 -0000 1.1.2.25
> --- unroll-new.c 6 May 2002 15:58:31 -0000
> *************** unroll_or_peel_loop (loops, loop, flags)
> *** 1142,1147 ****
> --- 1142,1157 ----
> flags &= ~(UAP_UNROLL | UAP_UNROLL_ALL);
> }
> }
> + else
> + {
> + /* For now we have no good heuristics to decide whether loop peeling
> + will be effective, so disable it. */
> + if ((flags & UAP_PEEL) && rtl_dump_file)
> + fprintf (rtl_dump_file,
> + ";; Not peeling loop, no evidence it will be profitable\n",
> + niter, npeel);
> + flags &= ~UAP_PEEL;
> + }
>
> /* We might have lost simpleness when counting loop iterations. */
> if (!simple && !(flags & UAP_UNROLL_ALL))
>