[PATCH 4/4] gdb: MSVC "long double" is a 64-bit double

[Pedro Alves <pedro@palves.net>]

The GNU (MinGW) and MSVC ABIs disagree on "long double".

- GCC the 80-bit x87 extended type on x86 (stored in 96 bits on i386
  and 128 bits on amd64), and the 128-bit IEEE quad on AArch64.

- MSVC makes "long double" just an alias for "double", an IEEE 64-bit
  double.

GDB modeled only the GNU sizes, via the per-architecture defaults, so
with an MSVC program it got "long double" wrong.  E.g. on AMD64:

 (gdb) set osabi Windows-MSVC
 (gdb) print sizeof (long double)
 $1 = 16

when an MSVC "long double" should be 8 bytes.

Now that we can distinguish the GNU and MSVC flavors, set the smaller
size and IEEE-double format for the MSVC flavor.  This is done in the
shared windows_init_abi, so it covers i386, amd64 and AArch64 all at
once.

After:

 (gdb) set osabi Windows-MSVC
 (gdb) print sizeof (long double)
 $1 = 8

Change-Id: I0acb2e80027cb7b1fcfd1f39dd050a24438e1662
---
 gdb/windows-tdep.c | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/gdb/windows-tdep.c b/gdb/windows-tdep.c
index 78bf49b9c81..a1e44f8aa3f 100644
--- a/gdb/windows-tdep.c
+++ b/gdb/windows-tdep.c
@@ -17,6 +17,7 @@
 
 #include "windows-tdep.h"
 #include "extract-store-integer.h"
+#include "gdbtypes.h"
 #include "gdbsupport/gdb_obstack.h"
 #include "xml-support.h"
 #include "gdbarch.h"
@@ -1031,6 +1032,14 @@ windows_init_abi (struct gdbarch_info info, struct gdbarch *gdbarch)
   /* On Windows, "long"s are only 32bit.  */
   set_gdbarch_long_bit (gdbarch, 32);
 
+  /* With the MSVC ABI, "long double" is just an IEEE 64-bit double,
+     the same as "double".  */
+  if (info.osabi == GDB_OSABI_WINDOWS_MSVC)
+    {
+      set_gdbarch_long_double_bit (gdbarch, 64);
+      set_gdbarch_long_double_format (gdbarch, floatformats_ieee_double);
+    }
+
   /* Enable TLS support.  */
   set_gdbarch_fetch_tls_load_module_address (gdbarch,
 					     windows_tls_index_address);
-- 
2.54.0