unbreak ieeefp emulation code wrt converting double to unsigned

long ints for alpha.  we've got only one instruction (cvttq) to
convert double-t to quadword, and float64_to_int64 did not take
into account the unsigned conversions
therefore, overflow always occured, and half of the unsigned range
(LONG_MAX .. ULONG_MAX) was broken
introduce roundAndPackInt64NoOverflow and float64_to_int64_no_overflow
for softfloat, that works with unsigned integers as well.  note
that this will return zero for nan/inf/oflow/uflow, raising exception
flag

perl is happy now
looked over by miod@
tested by naddy@, and by me on nick@'s alpha

1 files changed, 4 insertions, 1 deletions

diff --git a/sys/lib/libkern/softfloat.h b/sys/lib/libkern/softfloat.h
index cc2bd1fbad9..cc5ed44172e 100644
--- a/sys/lib/libkern/softfloat.h
+++ b/sys/lib/libkern/softfloat.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: softfloat.h,v 1.3 2008/06/26 05:42:20 ray Exp $	*/
+/*	$OpenBSD: softfloat.h,v 1.4 2008/10/07 22:06:29 martynas Exp $	*/
 /*	$NetBSD: softfloat.h,v 1.1 2001/04/26 03:10:48 ross Exp $	*/
 
 /* This is a derivative work. */
@@ -232,6 +232,9 @@ int float64_to_int32( float64 );
 int float64_to_int32_round_to_zero( float64 );
 #ifndef SOFTFLOAT_FOR_GCC /* __fix?fdi provided by libgcc2.c */
 int64_t float64_to_int64( float64 );
+#ifdef __alpha__
+int64_t float64_to_int64_no_overflow( float64 );
+#endif /* __alpha__ */
 int64_t float64_to_int64_round_to_zero( float64 );
 #endif
 float32 float64_to_float32( float64 );