/* Floating point definitions for GDB.
- Copyright (C) 1986, 1988, 1989, 1990, 1991, 1992, 1993, 1994, 1995, 1996,
- 1997, 1998, 1999, 2000, 2001, 2003, 2005, 2006, 2007, 2008, 2009, 2010
- Free Software Foundation, Inc.
+ Copyright (C) 1986-2017 Free Software Foundation, Inc.
This file is part of GDB.
struct type;
struct floatformat;
-/* Setup definitions for host and target floating point formats. We need to
- consider the format for `float', `double', and `long double' for both target
- and host. We need to do this so that we know what kind of conversions need
- to be done when converting target numbers to and from the hosts DOUBLEST
- data type. */
-
-/* This is used to indicate that we don't know the format of the floating point
- number. Typically, this is useful for native ports, where the actual format
- is irrelevant, since no conversions will be taking place. */
-
-#include "floatformat.h" /* For struct floatformat */
-
/* Use `long double' if the host compiler supports it. (Note that this is not
necessarily any longer than `double'. On SunOS/gcc, it's the same as
double.) This is necessary because GDB internally converts all floating
extern const char *floatformat_mantissa (const struct floatformat *,
const bfd_byte *);
-/* Given TYPE, return its floatformat. TYPE_FLOATFORMAT() may return
- NULL. type_floatformat() detects that and returns a floatformat
- based on the type size when FLOATFORMAT is NULL. */
+extern std::string floatformat_to_string (const struct floatformat *fmt,
+ const gdb_byte *in,
+ const char *format = nullptr);
+extern bool floatformat_from_string (const struct floatformat *fmt,
+ gdb_byte *out, const std::string &in);
+
+/* Return the floatformat's total size in host bytes. */
-const struct floatformat *floatformat_from_type (const struct type *type);
+extern size_t floatformat_totalsize_bytes (const struct floatformat *fmt);
extern DOUBLEST extract_typed_floating (const void *addr,
const struct type *type);