Avoid excess range in results from i386 scalb functions (bug 18981).

i386 scalb / scalbn / scalbln (and thus ldexp) functions for float and
double can return results with excess range (and consequently excess
precision for subnormal results).  As the results of these functions
are fully determined by reference to IEEE 754 operations, this is
unambiguously a bug, apart from the testsuite failures it causes.

This patch makes those functions store their results on the stack and
load them back to eliminate the excess range.  Double rounding is not
a problem, as the only cases where it could occur are when the result
overflows or underflows for extended precision, and then the
double-rounded results are the same as the single-rounded results.

The new macros will be used for more functions, more such macros
added, and existing code refactored to use such macros, in subsequent
patches.

Tested for x86.  Committed.

	[BZ #18981]
	* sysdeps/i386/fpu/i386-math-asm.h: New file.
	* sysdeps/i386/fpu/e_scalb.S: Include <i386-math-asm.h>.
	(__ieee754_scalb): Use DBL_NARROW_EVAL.
	* sysdeps/i386/fpu/e_scalbf.S: Include <i386-math-asm.h>.
	(__ieee754_scalbf): Use FLT_NARROW_EVAL.
	* sysdeps/i386/fpu/s_scalbn.S: Include <i386-math-asm.h>.
	(__scalbn): Use DBL_NARROW_EVAL.
	* sysdeps/i386/fpu/s_scalbnf.S: Include <i386-math-asm.h>.
	(__scalbnf): Use FLT_NARROW_EVAL.

7 files changed, 61 insertions, 1 deletions

diff --git a/ChangeLog b/ChangeLog
index 67c837da5a..cafe59e3fb 100644
--- a/ChangeLog
+++ b/ChangeLog
@@ -1,3 +1,16 @@
+2015-09-18  Joseph Myers  <joseph@codesourcery.com>
+
+	[BZ #18981]
+	* sysdeps/i386/fpu/i386-math-asm.h: New file.
+	* sysdeps/i386/fpu/e_scalb.S: Include <i386-math-asm.h>.
+	(__ieee754_scalb): Use DBL_NARROW_EVAL.
+	* sysdeps/i386/fpu/e_scalbf.S: Include <i386-math-asm.h>.
+	(__ieee754_scalbf): Use FLT_NARROW_EVAL.
+	* sysdeps/i386/fpu/s_scalbn.S: Include <i386-math-asm.h>.
+	(__scalbn): Use DBL_NARROW_EVAL.
+	* sysdeps/i386/fpu/s_scalbnf.S: Include <i386-math-asm.h>.
+	(__scalbnf): Use FLT_NARROW_EVAL.
+
 2015-09-18  Wilco Dijkstra  <wdijkstr@arm.com>
 
 	* math/w_tgamma.c (__ieee754_gamma_r): Use explicit sign check.
diff --git a/NEWS b/NEWS
index 8f487fac8d..4d65df21b8 100644
--- a/NEWS
+++ b/NEWS
@@ -16,7 +16,7 @@ Version 2.23
   18647, 18661, 18674, 18675, 18681, 18757, 18778, 18781, 18787, 18789,
   18790, 18795, 18796, 18820, 18823, 18824, 18857, 18863, 18870, 18872,
   18873, 18875, 18887, 18921, 18951, 18952, 18961, 18966, 18967, 18970,
-  18977.
+  18977, 18981.
 
 * The obsolete header <regexp.h> has been removed.  Programs that require
   this header must be updated to use <regex.h> instead.
diff --git a/sysdeps/i386/fpu/e_scalb.S b/sysdeps/i386/fpu/e_scalb.S
index 6a01aa64d6..370924c29f 100644
--- a/sysdeps/i386/fpu/e_scalb.S
+++ b/sysdeps/i386/fpu/e_scalb.S
@@ -6,6 +6,7 @@
  */
 
 #include <machine/asm.h>
+#include <i386-math-asm.h>
 
 	.section .rodata
 
@@ -52,6 +53,7 @@ ENTRY(__ieee754_scalb)
 	jne	4f
 	fscale
 	fstp	%st(1)
+	DBL_NARROW_EVAL
 	ret
 
 	/* y is -inf */
diff --git a/sysdeps/i386/fpu/e_scalbf.S b/sysdeps/i386/fpu/e_scalbf.S
index 12b25159ac..4f2dfa3acf 100644
--- a/sysdeps/i386/fpu/e_scalbf.S
+++ b/sysdeps/i386/fpu/e_scalbf.S
@@ -7,6 +7,7 @@
  */
 
 #include <machine/asm.h>
+#include <i386-math-asm.h>
 
 	.section .rodata
 
@@ -54,6 +55,7 @@ ENTRY(__ieee754_scalbf)
 	jne	4f
 	fscale
 	fstp	%st(1)
+	FLT_NARROW_EVAL
 	ret
 
 	/* y is -inf */
diff --git a/sysdeps/i386/fpu/i386-math-asm.h b/sysdeps/i386/fpu/i386-math-asm.h
new file mode 100644
index 0000000000..fd4313016c
--- /dev/null
+++ b/sysdeps/i386/fpu/i386-math-asm.h
@@ -0,0 +1,39 @@
+/* Helper macros for x86 libm functions.
+   Copyright (C) 2015 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <http://www.gnu.org/licenses/>.  */
+
+#ifndef _I386_MATH_ASM_H
+#define _I386_MATH_ASM_H 1
+
+/* Remove excess range and precision by storing a value on the stack
+   and loading it back.  */
+#define FLT_NARROW_EVAL				\
+	subl	$4, %esp;			\
+	cfi_adjust_cfa_offset (4);		\
+	fstps	(%esp);				\
+	flds	(%esp);				\
+	addl	$4, %esp;			\
+	cfi_adjust_cfa_offset (-4);
+#define DBL_NARROW_EVAL				\
+	subl	$8, %esp;			\
+	cfi_adjust_cfa_offset (8);		\
+	fstpl	(%esp);				\
+	fldl	(%esp);				\
+	addl	$8, %esp;			\
+	cfi_adjust_cfa_offset (-8);
+
+#endif /* i386-math-asm.h.  */
diff --git a/sysdeps/i386/fpu/s_scalbn.S b/sysdeps/i386/fpu/s_scalbn.S
index dab75801c5..4e90903115 100644
--- a/sysdeps/i386/fpu/s_scalbn.S
+++ b/sysdeps/i386/fpu/s_scalbn.S
@@ -4,6 +4,7 @@
  */
 
 #include <machine/asm.h>
+#include <i386-math-asm.h>
 
 RCSID("$NetBSD: s_scalbn.S,v 1.4 1995/05/09 00:19:06 jtc Exp $")
 
@@ -12,6 +13,7 @@ ENTRY(__scalbn)
 	fldl	4(%esp)
 	fscale
 	fstp	%st(1)
+	DBL_NARROW_EVAL
 	ret
 END (__scalbn)
 strong_alias (__scalbn, __scalbln)
diff --git a/sysdeps/i386/fpu/s_scalbnf.S b/sysdeps/i386/fpu/s_scalbnf.S
index ad97be86f4..f8353c4c75 100644
--- a/sysdeps/i386/fpu/s_scalbnf.S
+++ b/sysdeps/i386/fpu/s_scalbnf.S
@@ -4,6 +4,7 @@
  */
 
 #include <machine/asm.h>
+#include <i386-math-asm.h>
 
 RCSID("$NetBSD: s_scalbnf.S,v 1.3 1995/05/09 00:19:59 jtc Exp $")
 
@@ -12,6 +13,7 @@ ENTRY(__scalbnf)
 	flds	4(%esp)
 	fscale
 	fstp	%st(1)
+	FLT_NARROW_EVAL
 	ret
 END (__scalbnf)
 strong_alias (__scalbnf, __scalblnf)