public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed
From: Richard Guenther <richard.guenther@gmail.com>
To: gcc-patches@gcc.gnu.org
Subject: Re: [PATCH] fix arm neon ICE by widening tree_type's precision field
Date: Mon, 08 Jun 2009 17:11:00 -0000	[thread overview]
Message-ID: <84fc9c000906081010y1ae87a77y73132f8953186fef@mail.gmail.com> (raw)
In-Reply-To: <20090608153204.GW21107@codesourcery.com>

On Mon, Jun 8, 2009 at 11:32 AM, Nathan Froyd<froydnj@codesourcery.com> wrote:
> ARM's NEON support in GCC includes some rather wide types: XImode, the
> biggest, is a 64-byte integer type.  Compiling the simple testcase in
> the patch below, which uses XImode, results in an ICE.
>
> The reason behind this is because XImode is defined as a
> FRACTIONAL_INT_MODE of 511 bits, rather than an INT_MODE of 64 bytes.
> This hack was done because the 'precision' field of tree_type is only 9
> bits wide...and therefore can't store XImode's actual precision of 512
> bits.  So when we ask mode_for_size for a MODE_INT of 512 bits, it
> thinks that the only reasonable choice is BLKmode--since XImode is not
> a MODE_INT mode.  We eventually wind up calling simplify_subreg with the
> input mode of BLKmode, which triggers an ICE.
>
> The fix is to widen the 'precision' field of tree_type to 10 bits.
> Unfortunately, there's not much space in tree_type to widen that field;
> the structure is nicely packed for 32-bit and 64-bit hosts.  The
> approach I've taken in the patch below is to move packed_flag into
> tree_base and rearrange the bitfields in tree_type.  I realize this is
> slightly gross, but I don't see a better way--suggestions welcome.

Ugh.

> Tested with cross to arm-none-eabi.  OK to commit?  (Need middle-end
> approval for the tree.h changes and ARM approval for arm-modes.def.)

Ok.

Thanks,
Richard.

> -Nathan
>
> 2009-06-08  Nathan Froyd  <froydnj@codesourcery.com>
>
>        * tree.h (tree_base): Add packed_flag, moved from...
>        (tree_type): ...here.  Widen precision field to 10 bits and
>        rearrange fields to pack nicely.
>        (TYPE_PACKED): Update for new location of packed_flag.
>        * config/arm/arm-modes.def (XImode): Define as a INT_MODE.
>
> Index: config/arm/arm-modes.def
> ===================================================================
> --- config/arm/arm-modes.def    (revision 148220)
> +++ config/arm/arm-modes.def    (working copy)
> @@ -62,6 +62,4 @@ VECTOR_MODES (FLOAT, 16);     /*       V
>  INT_MODE (EI, 24);
>  INT_MODE (OI, 32);
>  INT_MODE (CI, 48);
> -/* ??? This should actually have 512 bits but the precision only has 9
> -   bits.  */
> -FRACTIONAL_INT_MODE (XI, 511, 64);
> +INT_MODE (XI, 64);
> Index: testsuite/ChangeLog
> ===================================================================
> --- testsuite/ChangeLog (revision 148220)
> +++ testsuite/ChangeLog (working copy)
> @@ -1,3 +1,7 @@
> +2009-06-08  Nathan Froyd  <froydnj@codesourcery.com>
> +
> +       * gcc.target/arm/neon-dse-1.c: New test.
> +
>  2009-06-05  Jakub Jelinek  <jakub@redhat.com>
>
>        PR middle-end/40340
> Index: testsuite/gcc.target/arm/neon-dse-1.c
> ===================================================================
> --- testsuite/gcc.target/arm/neon-dse-1.c       (revision 0)
> +++ testsuite/gcc.target/arm/neon-dse-1.c       (revision 0)
> @@ -0,0 +1,14 @@
> +/* { dg-do compile } */
> +/* { dg-require-effective-target arm_neon_ok } */
> +/* { dg-options "-O1" } */
> +/* { dg-add-options arm_neon } */
> +
> +#include <arm_neon.h>
> +
> +void neon_internal_error(int *dst, int *src)
> +{
> +  uint16x8x4_t sval;
> +
> +  sval = vld4q_u16((void *)src);
> +  vst4q_u16((void *)dst,sval);
> +}
> Index: tree.h
> ===================================================================
> --- tree.h      (revision 148220)
> +++ tree.h      (working copy)
> @@ -373,8 +373,10 @@ struct GTY(()) tree_base {
>   unsigned lang_flag_6 : 1;
>
>   unsigned visited : 1;
> +  /* For tree_type.  */
> +  unsigned packed_flag : 1;
>
> -  unsigned spare : 23;
> +  unsigned spare : 22;
>
>   union tree_ann_d *ann;
>  };
> @@ -2219,7 +2221,7 @@ extern enum machine_mode vector_type_mod
>
>  /* Indicated that objects of this type should be laid out in as
>    compact a way as possible.  */
> -#define TYPE_PACKED(NODE) (TYPE_CHECK (NODE)->type.packed_flag)
> +#define TYPE_PACKED(NODE) (TYPE_CHECK (NODE)->common.base.packed_flag)
>
>  /* Used by type_contains_placeholder_p to avoid recomputation.
>    Values are: 0 (unknown), 1 (false), 2 (true).  Never access
> @@ -2237,17 +2239,16 @@ struct GTY(()) tree_type {
>   tree attributes;
>   unsigned int uid;
>
> -  unsigned int precision : 9;
> -  ENUM_BITFIELD(machine_mode) mode : 7;
> -
> -  unsigned string_flag : 1;
> +  unsigned int precision : 10;
>   unsigned no_force_blk_flag : 1;
>   unsigned needs_constructing_flag : 1;
>   unsigned transparent_union_flag : 1;
> -  unsigned packed_flag : 1;
>   unsigned restrict_flag : 1;
>   unsigned contains_placeholder_bits : 2;
>
> +  ENUM_BITFIELD(machine_mode) mode : 7;
> +  unsigned string_flag : 1;
> +
>   unsigned lang_flag_0 : 1;
>   unsigned lang_flag_1 : 1;
>   unsigned lang_flag_2 : 1;
>

  reply	other threads:[~2009-06-08 17:11 UTC|newest]

Thread overview: 39+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2009-06-08 15:32 Nathan Froyd
2009-06-08 17:11 ` Richard Guenther [this message]
2009-06-08 18:04 ` Paolo Bonzini
2009-06-08 20:28   ` Nathan Froyd
2009-06-08 20:32     ` Steven Bosscher
2009-06-09  6:36     ` Paolo Bonzini
2009-06-09 14:54       ` Nathan Froyd
2009-06-09 15:00         ` Richard Guenther
2009-06-09 15:07           ` Richard Guenther
2009-06-09 15:44           ` Steven Bosscher
2009-06-10  2:50           ` Eric Botcazou
2009-06-09 15:40         ` Steven Bosscher
2009-06-09 15:53           ` Richard Guenther
2009-06-09 16:31             ` Nathan Froyd
2009-06-09 16:34               ` Richard Guenther
2009-06-09 16:36               ` Diego Novillo
2009-06-09 17:30               ` Paolo Bonzini
2010-07-28 19:01               ` [4.5 regression] C++ ignores some aligned attributes (Re: [PATCH] fix arm neon ICE by widening tree_type's precision field) Ulrich Weigand
2010-07-28 19:40                 ` Richard Guenther
2010-07-29 12:29                   ` [4.5 regression] C++ ignores some aligned attributes (Re: [PATCH] fix arm neon ICE by widening tree_type's precision field Ulrich Weigand
2010-07-28 22:07                 ` [PATCH] [4.5 regression] C++ ignores some aligned attributes Ulrich Weigand
2010-07-30 16:00                   ` Ulrich Weigand
2010-07-30 16:22                     ` Richard Guenther
2010-07-30 16:22                       ` Ulrich Weigand
2010-07-31 17:45                   ` Eric Botcazou
2010-07-31 19:38                     ` Ulrich Weigand
2010-08-04 14:03                   ` Paul Brook
2010-08-04 14:19                     ` Ulrich Weigand
2010-08-04 14:27                       ` Mark Mitchell
2010-08-04 15:04                       ` Paul Brook
2010-08-04 16:42                         ` Ulrich Weigand
2010-08-04 15:33                       ` Martin Sebor
2010-08-04 15:47                         ` Mark Mitchell
2010-08-05  9:02                           ` Martin Sebor
2009-06-08 19:40 ` [PATCH] fix arm neon ICE by widening tree_type's precision field Joseph S. Myers
2009-06-08 19:52   ` Daniel Jacobowitz
2009-06-08 20:12     ` Andrew Pinski
2009-06-08 20:20     ` Jakub Jelinek
2009-06-08 20:33       ` Daniel Jacobowitz

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=84fc9c000906081010y1ae87a77y73132f8953186fef@mail.gmail.com \
    --to=richard.guenther@gmail.com \
    --cc=gcc-patches@gcc.gnu.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).