[RFA]: fix ia64 long double support

[RFA]: fix ia64 long double support

[J. Johnston]

[-- Attachment #1: Type: text/plain, Size: 738 bytes --]

Fix to a couple of problems.  First of all, we aren't setting up a long double 
format for the architecture so it defaults to double when you try and print a 
long double value via gdb.

Although the ia64 has a special long double format it specifies, this format is 
only valid for register values.  Values in memory use the x86 extended float 
format which is 80 bits.  Register long doubles are 82 bits (2 more bits of 
exponent).

Anyway, this simple change fixes it so long doubles print correctly.

Ok to commit?

-- Jeff J.

2003-12-04  Jeff Johnston  <jjohnstn@redhat.com>

	* ia64-tdep.c (ia64_gdbarch_init): Set up the gdbarch long double
	format to be the i387 extended float format which is used for
	long double's in memory.

[-- Attachment #2: ia64-long-double.patch --]
[-- Type: text/plain, Size: 852 bytes --]

Index: ia64-tdep.c
===================================================================
RCS file: /cvs/src/src/gdb/ia64-tdep.c,v
retrieving revision 1.104
diff -u -p -r1.104 ia64-tdep.c
--- ia64-tdep.c	17 Nov 2003 21:38:36 -0000	1.104
+++ ia64-tdep.c	4 Dec 2003 22:29:43 -0000
@@ -3361,6 +3361,11 @@ ia64_gdbarch_init (struct gdbarch_info i
                0, "builtin_type_ia64_ext", NULL);
   TYPE_FLOATFORMAT (builtin_type_ia64_ext) = &floatformat_ia64_ext;
 
+  /* On the ia64, long double floats in memory are actually stored
+     in memory different than they are in registers.  In memory,
+     they match the x86 extended float format which is 80-bits.  */
+  set_gdbarch_long_double_format (gdbarch, &floatformat_i387_ext);
+
   set_gdbarch_short_bit (gdbarch, 16);
   set_gdbarch_int_bit (gdbarch, 32);
   set_gdbarch_long_bit (gdbarch, 64);

Re: [RFA]: fix ia64 long double support

[Richard Henderson]

On Thu, Dec 04, 2003 at 05:37:41PM -0500, J. Johnston wrote:
> 	* ia64-tdep.c (ia64_gdbarch_init): Set up the gdbarch long double
> 	format to be the i387 extended float format which is used for
> 	long double's in memory.

This is os specific.  HPUX uses ieee quad format.


r~

Re: [RFA]: fix ia64 long double support

[J. Johnston]

Richard Henderson wrote:
> On Thu, Dec 04, 2003 at 05:37:41PM -0500, J. Johnston wrote:
> 
>>	* ia64-tdep.c (ia64_gdbarch_init): Set up the gdbarch long double
>>	format to be the i387 extended float format which is used for
>>	long double's in memory.
> 
> 
> This is os specific.  HPUX uses ieee quad format.
> 
> 
> r~
> 

The Intel doc doesn't differentiate for OS.  Kevin, is this something we should 
account for by calling a function in a native tdep file?

-- Jeff J.

Re: [RFA]: fix ia64 long double support

[Kevin Buettner]

On Dec 4,  8:54pm, J. Johnston wrote:

> Richard Henderson wrote:
> > On Thu, Dec 04, 2003 at 05:37:41PM -0500, J. Johnston wrote:
> > 
> >>	* ia64-tdep.c (ia64_gdbarch_init): Set up the gdbarch long double
> >>	format to be the i387 extended float format which is used for
> >>	long double's in memory.
> > 
> > This is os specific.  HPUX uses ieee quad format.
> > 
> > r~
> > 
> 
> The Intel doc doesn't differentiate for OS.  Kevin, is this something we should 
> account for by calling a function in a native tdep file?

Hmm.  I suppose we could introduce an ia64-linux-tdep.c (with the
appropriate osabi machinery) to set this for linux.  If it was ever
made to work with hpux, there'd be an ia64-hpux-tdep.c file which
would set it to IEEE quad format.

If you feel up to this, go ahead.  Otherwise, just add a comment
to ia64-tdep.c which indicates that the setting is Linux specific
and that other OSes may use other types.  The comment should also
indicate the correct way that it should be handled (via an OS-specific
tdep.c file.)

Kevin

Re: [RFA]: fix ia64 long double support

[J. Johnston]

Kevin Buettner wrote:
> On Dec 4,  8:54pm, J. Johnston wrote:
> 
> 
>>Richard Henderson wrote:
>>
>>>On Thu, Dec 04, 2003 at 05:37:41PM -0500, J. Johnston wrote:
>>>
>>>
>>>>	* ia64-tdep.c (ia64_gdbarch_init): Set up the gdbarch long double
>>>>	format to be the i387 extended float format which is used for
>>>>	long double's in memory.
>>>
>>>This is os specific.  HPUX uses ieee quad format.
>>>
>>>r~
>>>
>>
>>The Intel doc doesn't differentiate for OS.  Kevin, is this something we should 
>>account for by calling a function in a native tdep file?
> 
> 
> Hmm.  I suppose we could introduce an ia64-linux-tdep.c (with the
> appropriate osabi machinery) to set this for linux.  If it was ever
> made to work with hpux, there'd be an ia64-hpux-tdep.c file which
> would set it to IEEE quad format.
> 
> If you feel up to this, go ahead.  Otherwise, just add a comment
> to ia64-tdep.c which indicates that the setting is Linux specific
> and that other OSes may use other types.  The comment should also
> indicate the correct way that it should be handled (via an OS-specific
> tdep.c file.)
> 
> Kevin
>

I have added a comment and checked the code in.

Thanks,

-- Jeff J.

Re: [RFA]: fix ia64 long double support

[Kevin Buettner]

On Dec 4,  5:37pm, J. Johnston wrote:

> Fix to a couple of problems.  First of all, we aren't setting up a long double 
> format for the architecture so it defaults to double when you try and print a 
> long double value via gdb.
> 
> Although the ia64 has a special long double format it specifies,
> this format is only valid for register values.  Values in memory use
> the x86 extended float format which is 80 bits.  Register long
> doubles are 82 bits (2 more bits of exponent).

So what happens when we try to print a long double in a register?

> Anyway, this simple change fixes it so long doubles print correctly.
> 
> Ok to commit?

Sure.  I'm guessing that we still won't always print long doubles correctly,
but it looks like this gets us closer...

Kevin

Re: [RFA]: fix ia64 long double support

[J. Johnston]

Kevin Buettner wrote:
> On Dec 4,  5:37pm, J. Johnston wrote:
> 
> 
>>Fix to a couple of problems.  First of all, we aren't setting up a long double 
>>format for the architecture so it defaults to double when you try and print a 
>>long double value via gdb.
>>
>>Although the ia64 has a special long double format it specifies,
>>this format is only valid for register values.  Values in memory use
>>the x86 extended float format which is 80 bits.  Register long
>>doubles are 82 bits (2 more bits of exponent).
> 
> 
> So what happens when we try to print a long double in a register?
> 

The routine ia64_register_type() returns builtin_type_ia64_ext for the 
floating-point registers which is the 82-bit register variant.

> 
>>Anyway, this simple change fixes it so long doubles print correctly.
>>
>>Ok to commit?
> 
> 
> Sure.  I'm guessing that we still won't always print long doubles correctly,
> but it looks like this gets us closer...
>
> Kevin
> 

Re: [RFA]: fix ia64 long double support

[Kevin Buettner]

On Dec 4,  8:52pm, J. Johnston wrote:

> > So what happens when we try to print a long double in a register?
> 
> The routine ia64_register_type() returns builtin_type_ia64_ext for the 
> floating-point registers which is the 82-bit register variant.

Okay, cool.

(The patch is still approved.)

Thanks,

Kevin