soft-fp fix:

- When converting from integer to float, rounding is done
incorrectly due to off-by-one shift in conversion

include/math-emu/op-common.h

@@ -782,7 +782,7 @@ do {									\
 									\
 	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 + 1), rsize);	\
+	  __FP_FRAC_SRS_1(r, (X##_e - _FP_WFRACBITS_##fs), 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));	\