From libc-alpha-return-50517-siddhesh=redhat.com@sourceware.org Thu Jun  5 10:36:33 2014
Received: (qmail 11948 invoked by alias); 4 Jun 2014 17:51:01 -0000
Mailing-List: contact libc-alpha-help@sourceware.org; run by ezmlm
List-Id: <libc-alpha.sourceware.org>
Sender: libc-alpha-owner@sourceware.org
Date: Wed, 4 Jun 2014 17:50:46 +0000
From: "Joseph S. Myers" <joseph@codesourcery.com>
To: <libc-alpha@sourceware.org>
Subject: Fix pow overflow in non-default rounding modes (bug 16315)
Message-ID: <Pine.LNX.4.64.1406041749420.3719@digraph.polyomino.org.uk>
MIME-Version: 1.0
Content-Type: multipart/mixed;
	boundary="-1152306461-1522705971-1401904246=:3719"
Content-Length: 24171

---1152306461-1522705971-1401904246=:3719
Content-Type: text/plain; charset="none"
Content-Transfer-Encoding: QUOTED-PRINTABLE

This patch, relative to a tree with
<https://sourceware.org/ml/libc-alpha/2014-06/msg00076.html> applied,
fixes bug 16315, bad pow handling of overflow/underflow in non-default
rounding modes.  Tests of pow are duly converted to ALL_RM_TEST to run
all tests in all rounding modes.

There are two main issues here.  First, various implementations
compute a negative result by negating a positive result, but this
yields inappropriate overflow / underflow values for directed
rounding, so either overflow / underflow results need recomputing in
the correct sign, or the relevant overflowing / underflowing operation
needs to be made to have a result of the correct sign.  Second, the
dbl-64 implementation sets FE_TONEAREST internally; in the overflow /
underflow case, the result needs recomputing in the original rounding
mode.

Tested x86_64 and x86 and ulps updated accordingly.

(auto-libm-test-out diffs omitted below.)

2014-06-04  Joseph Myers  <joseph@codesourcery.com>

=09[BZ #16315]
=09* sysdeps/i386/fpu/e_pow.S (__ieee754_pow): Ensure possibly
=09overflowing or underflowing operations take place with sign of
=09result.
=09* sysdeps/i386/fpu/e_powf.S (__ieee754_powf): Likewise.
=09* sysdeps/i386/fpu/e_powl.S (__ieee754_powl): Likewise.
=09* sysdeps/ieee754/dbl-64/e_pow.c: Include <math.h>.
=09(__ieee754_pow): Recompute overflowing and underflowing results in
=09original rounding mode.
=09* sysdeps/x86/fpu/powl_helper.c: Include <stdbool.h>.
=09(__powl_helper): Allow negative argument X and scale negated value
=09as needed.  Avoid passing value outside [-1, 1] to f2xm1.
=09* sysdeps/x86_64/fpu/e_powl.S (__ieee754_powl): Ensure possibly
=09overflowing or underflowing operations take place with sign of
=09result.
=09* sysdeps/x86_64/fpu/multiarch/e_pow.c [HAVE_FMA4_SUPPORT]:
=09Include <math.h>.
=09* math/auto-libm-test-in: Add more tests of pow.
=09* math/auto-libm-test-out: Regenerated.
=09* math/libm-test.inc (pow_test): Use ALL_RM_TEST.
=09(pow_tonearest_test_data): Remove.
=09(pow_test_tonearest): Likewise.
=09(pow_towardzero_test_data): Likewise.
=09(pow_test_towardzero): Likewise.
=09(pow_downward_test_data): Likewise.
=09(pow_test_downward): Likewise.
=09(pow_upward_test_data): Likewise.
=09(pow_test_upward): Likewise.
=09(main): Don't call removed functions.
=09* sysdeps/i386/fpu/libm-test-ulps: Update.
=09* sysdeps/x86_64/fpu/libm-test-ulps: Likewise.

diff --git a/sysdeps/x86_64/fpu/multiarch/e_pow.c b/sysdeps/x86_64/fpu/mult=
iarch/e_pow.c
index a740b6c..433cce0 100644
--- a/sysdeps/x86_64/fpu/multiarch/e_pow.c
+++ b/sysdeps/x86_64/fpu/multiarch/e_pow.c
@@ -1,5 +1,6 @@
 #ifdef HAVE_FMA4_SUPPORT
 # include <init-arch.h>
+# include <math.h>
 # include <math_private.h>
=20
 extern double __ieee754_pow_sse2 (double, double);

--=20
Joseph S. Myers
joseph@codesourcery.com
---1152306461-1522705971-1401904246=:3719--