register type as signed or unsigned?

register type as signed or unsigned?

[ligang]

hello all,

I am porting GDB to a new target.
I am not aware of the meaning of register type.
You can do as follows:
set_gdbarch_register_type (gdbarch, builtin_type_int32);
or
set_gdbarch_register_type (gdbarch, builtin_type_uint32);

What is the real difference between the two situation?
Why should GDB specify the register type as signed or unsigned?
Dose it mean the former must use regcache_cooked_read_signed() and the 
latter must use
regcache_cooked_read_unsigned()?

Best regards
                     Ligang

Re: register type as signed or unsigned?

[Daniel Jacobowitz]

On Tue, Oct 10, 2006 at 04:39:18PM +0800, ligang@sunnorth.com.cn wrote:
> hello all,
> 
> I am porting GDB to a new target.
> I am not aware of the meaning of register type.
> You can do as follows:
> set_gdbarch_register_type (gdbarch, builtin_type_int32);
> or
> set_gdbarch_register_type (gdbarch, builtin_type_uint32);
> 
> What is the real difference between the two situation?
> Why should GDB specify the register type as signed or unsigned?
> Dose it mean the former must use regcache_cooked_read_signed() and the 
> latter must use
> regcache_cooked_read_unsigned()?

No; in fact, it doesn't make much difference.  You should use whichever
is "more natural" for your target instruction set; it will affect
"print $reg" and "info reg".

-- 
Daniel Jacobowitz
CodeSourcery

Re: register type as signed or unsigned?

[ligang]

Daniel Jacobowitz <drow@false.org> wrote on 2006-10-10 20:50:02:

> On Tue, Oct 10, 2006 at 04:39:18PM +0800, ligang@sunnorth.com.cn wrote:
> > hello all,
> > 
> > I am porting GDB to a new target.
> > I am not aware of the meaning of register type.
> > You can do as follows:
> > set_gdbarch_register_type (gdbarch, builtin_type_int32);
> > or
> > set_gdbarch_register_type (gdbarch, builtin_type_uint32);
> > 
> > What is the real difference between the two situation?
> > Why should GDB specify the register type as signed or unsigned?
> > Dose it mean the former must use regcache_cooked_read_signed() and the 

> > latter must use
> > regcache_cooked_read_unsigned()?
> 
> No; in fact, it doesn't make much difference.  You should use whichever
> is "more natural" for your target instruction set; it will affect
> "print $reg" and "info reg".

That is to say, whether builtin_type_int32 or builtin_type_uint32 is 
correct for GDB.
But how to understand "more natural" and how dose it affect "info reg"? 
Would you please give me some more hints?

> -- 
> Daniel Jacobowitz
> CodeSourcery

Re: register type as signed or unsigned?

[Jan Kratochvil]

On Wed, 11 Oct 2006 02:37:57 +0200, ligang@sunnorth.com.cn wrote:
> Daniel Jacobowitz <drow@false.org> wrote on 2006-10-10 20:50:02:
...
> > No; in fact, it doesn't make much difference.  You should use whichever
> > is "more natural" for your target instruction set; it will affect
> > "print $reg" and "info reg".
> 
> That is to say, whether builtin_type_int32 or builtin_type_uint32 is 
> correct for GDB.

If the register is used as address, use "uint32".
If it is used for general computation incl. offsets in memory, use "int32".
"uint32" should be IMO a safer bet but sometimes you may need `(int)$regname'.


> Would you please give me some more hints?

You may check the recent thread being affected by such register typing:
	http://sourceware.org/ml/gdb-patches/2006-09/msg00191.html

Failed now to simulate it on 32-bit only gdbhost/inferior.


Regards,
Jan

Re: register type as signed or unsigned?

[Paul Koning]

>>>>> "Jan" == Jan Kratochvil <jan.kratochvil@redhat.com> writes:

 Jan> On Wed, 11 Oct 2006 02:37:57 +0200, ligang@sunnorth.com.cn
 Jan> wrote:
 >> Daniel Jacobowitz <drow@false.org> wrote on 2006-10-10 20:50:02:
 Jan> ...
 >> > No; in fact, it doesn't make much difference.  You should use
 >> whichever > is "more natural" for your target instruction set; it
 >> will affect > "print $reg" and "info reg".
 >> 
 >> That is to say, whether builtin_type_int32 or builtin_type_uint32
 >> is correct for GDB.

 Jan> If the register is used as address, use "uint32".  

Only if addresses are unsigned.  That's the usual case, but MIPS has
the odd convention that addresses are viewed as signed.

    paul

Re: register type as signed or unsigned?

[Daniel Jacobowitz]

On Wed, Oct 11, 2006 at 02:48:12AM +0200, Jan Kratochvil wrote:
> On Wed, 11 Oct 2006 02:37:57 +0200, ligang@sunnorth.com.cn wrote:
> > Daniel Jacobowitz <drow@false.org> wrote on 2006-10-10 20:50:02:
> ...
> > > No; in fact, it doesn't make much difference.  You should use whichever
> > > is "more natural" for your target instruction set; it will affect
> > > "print $reg" and "info reg".
> > 
> > That is to say, whether builtin_type_int32 or builtin_type_uint32 is 
> > correct for GDB.
> 
> If the register is used as address, use "uint32".

In fact, if you know the register is always used as an address, use
a pointer type.

-- 
Daniel Jacobowitz
CodeSourcery

Re: register type as signed or unsigned?

[Mark Kettenis]

>  On Wed, 11 Oct 2006 02:37:57 +0200, ligang@sunnorth.com.cn wrote:
> > Daniel Jacobowitz <drow@false.org> wrote on 2006-10-10 20:50:02:
>  ...
> > > No; in fact, it doesn't make much difference.  You should use
> > whichever
> > > is "more natural" for your target instruction set; it will affect
> > > "print $reg" and "info reg".
> >
> > That is to say, whether builtin_type_int32 or builtin_type_uint32 is
> > correct for GDB.
>
>  If the register is used as address, use "uint32".

If the register is typically used as an address, builtin_type_void_data_ptr
is probably more appropriate.  For the instruction pointer you should use
builtin_type_void_func_ptr though.

>  If it is used for general computation incl. offsets in memory, use
>  "int32".
>  "uint32" should be IMO a safer bet but sometimes you may need
>  `(int)$regname'.

int32 is IMHO more useful since actually large unsigned numbers are rare.
But if your CPU only supports unsigned integer math, uint32 is the right
choice.

Mark