Fix asin missing underflows (bug 16351).

Similar to various other bugs in this area, some asin implementations
do not raise the underflow exception for subnormal arguments, when the
result is tiny and inexact.  This patch forces the exception in a
similar way to previous fixes.

Tested for x86_64, x86, powerpc and mips64.

	[BZ #16351]
	* sysdeps/i386/fpu/e_asin.S (dbl_min): New object.
	(MO): New macro.
	(__ieee754_asin): Force underflow exception for results with small
	absolute value.
	* sysdeps/i386/fpu/e_asinf.S (flt_min): New object.
	(MO): New macro.
	(__ieee754_asinf): Force underflow exception for results with
	small absolute value.
	* sysdeps/ieee754/dbl-64/e_asin.c: Include <float.h> and <math.h>.
	(__ieee754_asin): Force underflow exception for results with small
	absolute value.
	* sysdeps/ieee754/flt-32/e_asinf.c: Include <float.h>.
	(__ieee754_asinf): Force underflow exception for results with
	small absolute value.
	* sysdeps/ieee754/ldbl-128/e_asinl.c: Include <float.h>.
	(__ieee754_asinl): Force underflow exception for results with
	small absolute value.
	* sysdeps/ieee754/ldbl-128ibm/e_asinl.c: Include <float.h>.
	(__ieee754_asinl): Force underflow exception for results with
	small absolute value.
	* sysdeps/ieee754/ldbl-96/e_asinl.c: Include <float.h>.
	(__ieee754_asinl): Force underflow exception for results with
	small absolute value.
	* sysdeps/x86_64/fpu/multiarch/e_asin.c [HAVE_FMA4_SUPPORT]:
	Include <math.h>.
	* math/auto-libm-test-in: Do not mark underflow exceptions as
	possibly missing for bug 16351.
	* math/auto-libm-test-out: Regenerated.

8 files changed, 102 insertions, 3 deletions

diff --git a/sysdeps/i386/fpu/e_asin.S b/sysdeps/i386/fpu/e_asin.S
index 9a44cb62ac..ea8cb6f1b2 100644
--- a/sysdeps/i386/fpu/e_asin.S
+++ b/sysdeps/i386/fpu/e_asin.S
@@ -7,8 +7,26 @@
 
 RCSID("$NetBSD: e_asin.S,v 1.4 1995/05/08 23:45:40 jtc Exp $")
 
+	.section .rodata.cst8,"aM",@progbits,8
+
+	.p2align 3
+	.type dbl_min,@object
+dbl_min:	.byte 0, 0, 0, 0, 0, 0, 0x10, 0
+	ASM_SIZE_DIRECTIVE(dbl_min)
+
+#ifdef PIC
+# define MO(op) op##@GOTOFF(%ecx)
+#else
+# define MO(op) op
+#endif
+
+	.text
+
 /* asin = atan (x / sqrt((1-x) (1+x))) */
 ENTRY(__ieee754_asin)
+#ifdef  PIC
+	LOAD_PIC_REG (cx)
+#endif
 	fldl	4(%esp)			/* x */
 	fld	%st
 	fld1				/* 1 : x : x */
@@ -18,6 +36,20 @@ ENTRY(__ieee754_asin)
 	fmulp				/* 1 - x^2 */
 	fsqrt				/* sqrt (1 - x^2) */
 	fpatan
-	ret
+	fldl	MO(dbl_min)
+	fld	%st(1)
+	fabs
+	fucompp
+	fnstsw
+	sahf
+	jnc	1f
+	subl	$8, %esp
+	cfi_adjust_cfa_offset (8)
+	fld	%st(0)
+	fmul	%st(0)
+	fstpl	(%esp)
+	addl	$8, %esp
+	cfi_adjust_cfa_offset (-8)
+1:	ret
 END (__ieee754_asin)
 strong_alias (__ieee754_asin, __asin_finite)
diff --git a/sysdeps/i386/fpu/e_asinf.S b/sysdeps/i386/fpu/e_asinf.S
index 5c1065dd4d..f7bda77fa6 100644
--- a/sysdeps/i386/fpu/e_asinf.S
+++ b/sysdeps/i386/fpu/e_asinf.S
@@ -8,8 +8,26 @@
 
 RCSID("$NetBSD: $")
 
+	.section .rodata.cst4,"aM",@progbits,4
+
+	.p2align 2
+	.type flt_min,@object
+flt_min:	.byte 0, 0, 0x80, 0
+	ASM_SIZE_DIRECTIVE(flt_min)
+
+#ifdef PIC
+# define MO(op) op##@GOTOFF(%ecx)
+#else
+# define MO(op) op
+#endif
+
+	.text
+
 /* asin = atan (x / sqrt(1 - x^2)) */
 ENTRY(__ieee754_asinf)
+#ifdef  PIC
+	LOAD_PIC_REG (cx)
+#endif
 	flds	4(%esp)			/* x */
 	fld	%st
 	fmul	%st(0)			/* x^2 */
@@ -17,6 +35,20 @@ ENTRY(__ieee754_asinf)
 	fsubp				/* 1 - x^2 */
 	fsqrt				/* sqrt (1 - x^2) */
 	fpatan
-	ret
+	flds	MO(flt_min)
+	fld	%st(1)
+	fabs
+	fucompp
+	fnstsw
+	sahf
+	jnc	1f
+	subl	$4, %esp
+	cfi_adjust_cfa_offset (4)
+	fld	%st(0)
+	fmul	%st(0)
+	fstps	(%esp)
+	addl	$4, %esp
+	cfi_adjust_cfa_offset (-4)
+1:	ret
 END (__ieee754_asinf)
 strong_alias (__ieee754_asinf, __asinf_finite)
diff --git a/sysdeps/ieee754/dbl-64/e_asin.c b/sysdeps/ieee754/dbl-64/e_asin.c
index e90f47ccc2..b87e557988 100644
--- a/sysdeps/ieee754/dbl-64/e_asin.c
+++ b/sysdeps/ieee754/dbl-64/e_asin.c
@@ -39,6 +39,8 @@
 #include "powtwo.tbl"
 #include "MathLib.h"
 #include "uasncs.h"
+#include <float.h>
+#include <math.h>
 #include <math_private.h>
 
 #ifndef SECTION
@@ -67,7 +69,15 @@ __ieee754_asin(double x){
   m = u.i[HIGH_HALF];
   k = 0x7fffffff&m;              /* no sign */
 
-  if (k < 0x3e500000) return x;  /* for x->0 => sin(x)=x */
+  if (k < 0x3e500000)
+    {
+      if (fabs (x) < DBL_MIN)
+	{
+	  double force_underflow = x * x;
+	  math_force_eval (force_underflow);
+	}
+      return x;  /* for x->0 => sin(x)=x */
+    }
   /*----------------------2^-26 <= |x| < 2^ -3    -----------------*/
   else
   if (k < 0x3fc00000) {
diff --git a/sysdeps/ieee754/flt-32/e_asinf.c b/sysdeps/ieee754/flt-32/e_asinf.c
index 91ca9afd00..00bad4239b 100644
--- a/sysdeps/ieee754/flt-32/e_asinf.c
+++ b/sysdeps/ieee754/flt-32/e_asinf.c
@@ -39,6 +39,7 @@
 static char rcsid[] = "$NetBSD: e_asinf.c,v 1.5 1995/05/12 04:57:25 jtc Exp $";
 #endif
 
+#include <float.h>
 #include <math.h>
 #include <math_private.h>
 
@@ -72,6 +73,11 @@ float __ieee754_asinf(float x)
 	    return (x-x)/(x-x);		/* asin(|x|>1) is NaN */
 	} else if (ix<0x3f000000) {	/* |x|<0.5 */
 	    if(ix<0x32000000) {		/* if |x| < 2**-27 */
+		if (fabsf (x) < FLT_MIN)
+		  {
+		    float force_underflow = x * x;
+		    math_force_eval (force_underflow);
+		  }
 		if(huge+x>one) return x;/* return x with inexact if x!=0*/
 	    } else {
 		t = x*x;
diff --git a/sysdeps/ieee754/ldbl-128/e_asinl.c b/sysdeps/ieee754/ldbl-128/e_asinl.c
index cd1a8da42c..353603d68c 100644
--- a/sysdeps/ieee754/ldbl-128/e_asinl.c
+++ b/sysdeps/ieee754/ldbl-128/e_asinl.c
@@ -59,6 +59,7 @@
  */
 
 
+#include <float.h>
 #include <math.h>
 #include <math_private.h>
 long double sqrtl (long double);
@@ -152,6 +153,11 @@ __ieee754_asinl (long double x)
     {
       if (ix < 0x3fc60000) /* |x| < 2**-57 */
 	{
+	  if (fabsl (x) < LDBL_MIN)
+	    {
+	      long double force_underflow = x * x;
+	      math_force_eval (force_underflow);
+	    }
 	  if (huge + x > one)
 	    return x;		/* return x with inexact if x!=0 */
 	}
diff --git a/sysdeps/ieee754/ldbl-128ibm/e_asinl.c b/sysdeps/ieee754/ldbl-128ibm/e_asinl.c
index dece11875b..00386db04e 100644
--- a/sysdeps/ieee754/ldbl-128ibm/e_asinl.c
+++ b/sysdeps/ieee754/ldbl-128ibm/e_asinl.c
@@ -59,6 +59,7 @@
  */
 
 
+#include <float.h>
 #include <math.h>
 #include <math_private.h>
 long double sqrtl (long double);
@@ -146,6 +147,11 @@ __ieee754_asinl (long double x)
     {
       if (a < 6.938893903907228e-18L) /* |x| < 2**-57 */
 	{
+	  if (fabsl (x) < LDBL_MIN)
+	    {
+	      long double force_underflow = x * x;
+	      math_force_eval (force_underflow);
+	    }
 	  if (huge + x > one)
 	    return x;		/* return x with inexact if x!=0 */
 	}
diff --git a/sysdeps/ieee754/ldbl-96/e_asinl.c b/sysdeps/ieee754/ldbl-96/e_asinl.c
index c1ffa3e0d9..2973bf071b 100644
--- a/sysdeps/ieee754/ldbl-96/e_asinl.c
+++ b/sysdeps/ieee754/ldbl-96/e_asinl.c
@@ -58,6 +58,7 @@
  */
 
 
+#include <float.h>
 #include <math.h>
 #include <math_private.h>
 
@@ -111,6 +112,11 @@ __ieee754_asinl (long double x)
     {				/* |x|<0.5 */
       if (ix < 0x3fde8000)
 	{			/* if |x| < 2**-33 */
+	  if (fabsl (x) < LDBL_MIN)
+	    {
+	      long double force_underflow = x * x;
+	      math_force_eval (force_underflow);
+	    }
 	  if (huge + x > one)
 	    return x;		/* return x with inexact if x!=0 */
 	}
diff --git a/sysdeps/x86_64/fpu/multiarch/e_asin.c b/sysdeps/x86_64/fpu/multiarch/e_asin.c
index e742a9c133..55865c02f3 100644
--- a/sysdeps/x86_64/fpu/multiarch/e_asin.c
+++ b/sysdeps/x86_64/fpu/multiarch/e_asin.c
@@ -1,5 +1,6 @@
 #ifdef HAVE_FMA4_SUPPORT
 # include <init-arch.h>
+# include <math.h>
 # include <math_private.h>
 
 extern double __ieee754_acos_sse2 (double);