From: Richard Biener <richard.guenther@gmail.com>
To: Sujoy Saraswati <ssaraswati@gmail.com>
Cc: GCC Patches <gcc-patches@gcc.gnu.org>
Subject: Re: Fix 61441
Date: Tue, 01 Sep 2015 12:21:00 -0000 [thread overview]
Message-ID: <CAFiYyc0rwpQ_v1x4WzbrzaPfiFQetfq_hLur44otKwceqo6CDg@mail.gmail.com> (raw)
In-Reply-To: <CA+ZXfhXZWO8eOCYGPv3nomrfK7zzWHkXSJE++jw0heHxbQoqaA@mail.gmail.com>
On Tue, Sep 1, 2015 at 12:23 PM, Sujoy Saraswati <ssaraswati@gmail.com> wrote:
> The following patch fixes 61441. It converts sNaN to qNaN on folding
> when -fno-signaling-nans is used.
>
> Bootstrap and regression tests on x86_64-linux-gnu and
> aarch64-unknown-linux-gnu passed with changes done on trunk.
>
> Is this fix fine ?
Note that I'm curious what
the actual bug is - is it that (double) sNaN creates a sNaN? Then the fix
should be elsewhere, in constant folding itself
(fold_convert_const_real_from_real
or real_convert).
If that isn't the bug you have very many other passes to fix for the
same problem.
So - can you please explain?
Thanks,
Richard.
> Regards,
> Sujoy
>
> 2015-09-01 Sujoy Saraswati <ssaraswati@gmail.com>
>
> PR tree-optimization/61441
> * tree-ssa-ccp.c (convert_snan_to_qnan): Convert sNaN to qNaN when
> flag_signaling_nans is off.
> (ccp_fold_stmt, visit_assignment, visit_cond_stmt): call
> convert_snan_to_qnan to convert sNaN to qNaN on constant folding.
>
> PR tree-optimization/61441
> * gcc.dg/pr61441.c: New testcase.
>
> Index: gcc/tree-ssa-ccp.c
> ===================================================================
> --- gcc/tree-ssa-ccp.c (revision 226965)
> +++ gcc/tree-ssa-ccp.c (working copy)
> @@ -560,6 +560,24 @@ value_to_wide_int (ccp_prop_value_t val)
> return 0;
> }
>
> +/* Convert sNaN to qNaN when flag_signaling_nans is off */
> +
> +static void
> +convert_snan_to_qnan (tree expr)
> +{
> + if (expr
> + && (TREE_CODE (expr) == REAL_CST)
> + && !flag_signaling_nans)
> + {
> + REAL_VALUE_TYPE *d = TREE_REAL_CST_PTR (expr);
> +
> + if (HONOR_NANS (TYPE_MODE (TREE_TYPE (expr)))
> + && REAL_VALUE_ISNAN (*d)
> + && d->signalling)
> + d->signalling = 0;
> + }
> +}
> +
> /* Return the value for the address expression EXPR based on alignment
> information. */
>
> @@ -2156,6 +2174,7 @@ ccp_fold_stmt (gimple_stmt_iterator *gsi)
> if (val.lattice_val != CONSTANT
> || val.mask != 0)
> return false;
> + convert_snan_to_qnan (val.value);
>
> if (dump_file)
> {
> @@ -2197,7 +2216,10 @@ ccp_fold_stmt (gimple_stmt_iterator *gsi)
> bool res;
> if (!useless_type_conversion_p (TREE_TYPE (lhs),
> TREE_TYPE (new_rhs)))
> + {
> new_rhs = fold_convert (TREE_TYPE (lhs), new_rhs);
> + convert_snan_to_qnan (new_rhs);
> + }
> res = update_call_from_tree (gsi, new_rhs);
> gcc_assert (res);
> return true;
> @@ -2216,6 +2238,7 @@ ccp_fold_stmt (gimple_stmt_iterator *gsi)
> tree new_rhs = fold_builtin_alloca_with_align (stmt);
> if (new_rhs)
> {
> + convert_snan_to_qnan (new_rhs);
> bool res = update_call_from_tree (gsi, new_rhs);
> tree var = TREE_OPERAND (TREE_OPERAND (new_rhs, 0),0);
> gcc_assert (res);
> @@ -2260,7 +2283,10 @@ ccp_fold_stmt (gimple_stmt_iterator *gsi)
> {
> tree rhs = unshare_expr (val);
> if (!useless_type_conversion_p (TREE_TYPE (lhs), TREE_TYPE (rhs)))
> + {
> rhs = fold_build1 (VIEW_CONVERT_EXPR, TREE_TYPE (lhs), rhs);
> + convert_snan_to_qnan (rhs);
> + }
> gimple_assign_set_rhs_from_tree (gsi, rhs);
> return true;
> }
> @@ -2292,6 +2318,7 @@ visit_assignment (gimple stmt, tree *output_p)
> /* Evaluate the statement, which could be
> either a GIMPLE_ASSIGN or a GIMPLE_CALL. */
> val = evaluate_stmt (stmt);
> + convert_snan_to_qnan (val.value);
>
> /* If STMT is an assignment to an SSA_NAME, we only have one
> value to set. */
> @@ -2324,6 +2351,7 @@ visit_cond_stmt (gimple stmt, edge *taken_edge_p)
> if (val.lattice_val != CONSTANT
> || val.mask != 0)
> return SSA_PROP_VARYING;
> + convert_snan_to_qnan (val.value);
>
> /* Find which edge out of the conditional block will be taken and add it
> to the worklist. If no single edge can be determined statically,
>
> Index: gcc/testsuite/gcc.dg/pr61441.c
> ===================================================================
> --- gcc/testsuite/gcc.dg/pr61441.c (revision 0)
> +++ gcc/testsuite/gcc.dg/pr61441.c (working copy)
> @@ -0,0 +1,17 @@
> +/* { dg-do run } */
> +/* { dg-options "-O1 -lm" } */
> +
> +#define _GNU_SOURCE
> +#include <stdio.h>
> +#include <math.h>
> +
> +int main (void)
> +{
> + float sNaN = __builtin_nansf ("");
> + double x = (double) sNaN;
> + if (issignaling(x))
> + {
> + __builtin_abort();
> + }
> + return 0;
> +}
next prev parent reply other threads:[~2015-09-01 12:21 UTC|newest]
Thread overview: 17+ messages / expand[flat|nested] mbox.gz Atom feed top
2015-09-01 10:23 Sujoy Saraswati
2015-09-01 12:21 ` Richard Biener [this message]
2015-09-02 11:36 ` Sujoy Saraswati
2015-09-02 11:56 ` Richard Biener
2015-09-02 12:16 ` Sujoy Saraswati
2015-09-10 7:33 ` Sujoy Saraswati
2015-09-14 13:50 ` Richard Biener
2015-09-14 20:39 ` Joseph Myers
2015-09-16 13:01 ` Sujoy Saraswati
2015-09-16 17:03 ` Joseph Myers
2015-10-13 10:46 ` Sujoy Saraswati
2015-10-28 17:19 ` Joseph Myers
2015-11-05 11:29 ` Sujoy Saraswati
2015-11-05 16:58 ` Joseph Myers
2015-11-06 4:56 ` Sujoy Saraswati
2015-11-06 13:09 ` Joseph Myers
2015-11-26 8:28 ` Saraswati, Sujoy (OSTL)
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=CAFiYyc0rwpQ_v1x4WzbrzaPfiFQetfq_hLur44otKwceqo6CDg@mail.gmail.com \
--to=richard.guenther@gmail.com \
--cc=gcc-patches@gcc.gnu.org \
--cc=ssaraswati@gmail.com \
/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).