Fix dbl-64 lrint for 64-bit long (bug 19095) [committed]

[Joseph Myers]

The dbl-64 implementation of lrint produces incorrect results for some
arguments with 64-bit long because a 32-bit (unsigned) low part of the
mantissa is shifted left, losing high bits in the process.  This patch
fixes this by casting to long int before shifting, as in lround (as
this case only applies for 64-bit long, there are no issues with
sign-extension).

Tested for mips64 (n64).  Committed.

2015-10-09  Joseph Myers  <joseph@codesourcery.com>

	[BZ #19095]
	* sysdeps/ieee754/dbl-64/s_lrint.c (__lrint): Cast low part of
	mantissa to long int before shifting left.

diff --git a/sysdeps/ieee754/dbl-64/s_lrint.c b/sysdeps/ieee754/dbl-64/s_lrint.c
index 39f95ad..d004594 100644
--- a/sysdeps/ieee754/dbl-64/s_lrint.c
+++ b/sysdeps/ieee754/dbl-64/s_lrint.c
@@ -61,7 +61,7 @@ __lrint (double x)
   else if (j0 < (int32_t) (8 * sizeof (long int)) - 1)
     {
       if (j0 >= 52)
-	result = ((long int) i0 << (j0 - 20)) | (i1 << (j0 - 52));
+	result = ((long int) i0 << (j0 - 20)) | ((long int) i1 << (j0 - 52));
       else
 	{
 #if defined FE_INVALID || defined FE_INEXACT

-- 
Joseph S. Myers
joseph@codesourcery.com