Soft-fp fix:

- Fix handling of implicit 1 bit of fraction part when converting
from int to float.  We cannot handle it properly unless we
defer implicit 1 bit handling till after rounding as
rounding may move where the implicit 1 bit is in
the final converted integer.

include/math-emu/op-1.h

@@ -58,6 +58,7 @@
 #define _FP_FRAC_NEGP_1(X)	((_FP_WS_TYPE)X##_f < 0)
 #define _FP_FRAC_ZEROP_1(X)	(X##_f == 0)
 #define _FP_FRAC_OVERP_1(fs,X)	(X##_f & _FP_OVERFLOW_##fs)
+#define _FP_FRAC_CLEAR_OVERP_1(fs,X)	(X##_f &= ~_FP_OVERFLOW_##fs)
 #define _FP_FRAC_EQ_1(X, Y)	(X##_f == Y##_f)
 #define _FP_FRAC_GE_1(X, Y)	(X##_f >= Y##_f)
 #define _FP_FRAC_GT_1(X, Y)	(X##_f > Y##_f)

include/math-emu/op-2.h

@@ -114,6 +114,7 @@
 #define _FP_FRAC_NEGP_2(X)	((_FP_WS_TYPE)X##_f1 < 0)
 #define _FP_FRAC_ZEROP_2(X)	((X##_f1 | X##_f0) == 0)
 #define _FP_FRAC_OVERP_2(fs,X)	(_FP_FRAC_HIGH_##fs(X) & _FP_OVERFLOW_##fs)
+#define _FP_FRAC_CLEAR_OVERP_2(fs,X)	(_FP_FRAC_HIGH_##fs(X) &= ~_FP_OVERFLOW_##fs)
 #define _FP_FRAC_EQ_2(X, Y)	(X##_f1 == Y##_f1 && X##_f0 == Y##_f0)
 #define _FP_FRAC_GT_2(X, Y)	\
   (X##_f1 > Y##_f1 || (X##_f1 == Y##_f1 && X##_f0 > Y##_f0))

include/math-emu/op-4.h

@@ -132,6 +132,7 @@
 #define _FP_FRAC_ZEROP_4(X)     ((X##_f[0] | X##_f[1] | X##_f[2] | X##_f[3]) == 0)
 #define _FP_FRAC_NEGP_4(X)      ((_FP_WS_TYPE)X##_f[3] < 0)
 #define _FP_FRAC_OVERP_4(fs,X)  (_FP_FRAC_HIGH_##fs(X) & _FP_OVERFLOW_##fs)
+#define _FP_FRAC_CLEAR_OVERP_4(fs,X)  (_FP_FRAC_HIGH_##fs(X) &= ~_FP_OVERFLOW_##fs)
 
 #define _FP_FRAC_EQ_4(X,Y)				\
  (X##_f[0] == Y##_f[0] && X##_f[1] == Y##_f[1]		\

include/math-emu/op-common.h

@@ -97,10 +97,9 @@ do {								\
 	_FP_ROUND(wc, X);					\
 	if (_FP_FRAC_OVERP_##wc(fs, X))				\
 	  {							\
-	    _FP_FRAC_SRL_##wc(X, (_FP_WORKBITS+1));		\
+	    _FP_FRAC_CLEAR_OVERP_##wc(fs, X);			\
 	    X##_e++;						\
 	  }							\
-	else							\
 	_FP_FRAC_SRL_##wc(X, _FP_WORKBITS);			\
 	if (X##_e >= _FP_EXPMAX_##fs)				\
 	  {							\
@@ -780,9 +779,8 @@ do {									\
 		X##_e -= (_FP_W_TYPE_SIZE - rsize);			\
 	X##_e = rsize - X##_e - 1;					\
 									\
-	r &= ~((rtype)1 << X##_e);					\
 	if (_FP_FRACBITS_##fs < rsize && _FP_WFRACBITS_##fs < X##_e)	\
-	  __FP_FRAC_SRS_1(r, (X##_e - _FP_WFRACBITS_##fs), rsize);	\
+	  __FP_FRAC_SRS_1(r, (X##_e - _FP_WFRACBITS_##fs + 1), rsize);	\
 	_FP_FRAC_DISASSEMBLE_##wc(X, ((unsigned rtype)r), rsize);	\
 	if ((_FP_WFRACBITS_##fs - X##_e - 1) > 0)			\
 	  _FP_FRAC_SLL_##wc(X, (_FP_WFRACBITS_##fs - X##_e - 1));	\