[patch/rfc] Don't include vector registers in ``info registers''

[patch/rfc] Don't include vector registers in ``info registers''

[Andrew Cagney]

Hello,

The attached patch modifies the generic ``info registsters'' command so 
that it precludes vector registers (in addition to floating-point 
registers).  The online doco indicates:

(gdb) help info registers
List of integer registers and their contents, for selected stack frame.
Register name as argument means describe only that register.
(gdb) help info all-registers
List of all registers and their contents, for selected stack frame.
Register name as argument means describe only that register.

I think the change makes the behavour a better match for both the 
documentation and what I think is the intent of the command.  Print a 
minimal set of registers.

It will eventually affect the i386 -- I've a patch to change the type of 
xmm [and mmx] registers to true vectors.  When that is in, ``info 
registers'' will stop displaying the xmm registers.

Thoughts?

enjoy,
Andrew

Re: [patch/rfc] Don't include vector registers in ``info registers''

[Andrew Cagney]

[-- Attachment #1: Type: text/plain, Size: 940 bytes --]

[With patch]


> Hello,
> 
> The attached patch modifies the generic ``info registsters'' command so that it precludes vector registers (in addition to floating-point registers).  The online doco indicates:
> 
> (gdb) help info registers
> List of integer registers and their contents, for selected stack frame.
> Register name as argument means describe only that register.
> (gdb) help info all-registers
> List of all registers and their contents, for selected stack frame.
> Register name as argument means describe only that register.
> 
> I think the change makes the behavour a better match for both the documentation and what I think is the intent of the command.  Print a minimal set of registers.
> 
> It will eventually affect the i386 -- I've a patch to change the type of xmm [and mmx] registers to true vectors.  When that is in, ``info registers'' will stop displaying the xmm registers.
> 
> Thoughts?
> 
> enjoy,
> Andrew


[-- Attachment #2: diffs --]
[-- Type: text/plain, Size: 1747 bytes --]

2002-08-08  Andrew Cagney  <cagney@redhat.com>

	* infcmd.c (do_registers_info): Rename parameter ``fpregs'' to
	``print_all''.  Only print vector registers when ``print_all''.

Index: infcmd.c
===================================================================
RCS file: /cvs/src/src/gdb/infcmd.c,v
retrieving revision 1.49
diff -u -r1.49 infcmd.c
--- infcmd.c	3 Jul 2002 20:36:54 -0000	1.49
+++ infcmd.c	8 Aug 2002 21:14:22 -0000
@@ -1549,9 +1549,9 @@
 #ifdef REGISTER_NAMES
 char *gdb_register_names[] = REGISTER_NAMES;
 #endif
-/* Print out the machine register regnum. If regnum is -1,
-   print all registers (fpregs == 1) or all non-float registers
-   (fpregs == 0).
+/* Print out the machine register regnum. If regnum is -1, print all
+   registers (all == 1) or all non-float and non-vector registers (all
+   == 0).
 
    For most machines, having all_registers_info() print the
    register(s) one per line is good enough. If a different format
@@ -1561,7 +1561,7 @@
    to provide that format.  */
 
 void
-do_registers_info (int regnum, int fpregs)
+do_registers_info (int regnum, int print_all)
 {
   register int i;
   int numregs = NUM_REGS + NUM_PSEUDO_REGS;
@@ -1570,11 +1570,17 @@
 
   for (i = 0; i < numregs; i++)
     {
-      /* Decide between printing all regs, nonfloat regs, or specific reg.  */
+      /* Decide between printing all regs, non-float / vector regs, or
+         specific reg.  */
       if (regnum == -1)
 	{
-	  if (TYPE_CODE (REGISTER_VIRTUAL_TYPE (i)) == TYPE_CODE_FLT && !fpregs)
-	    continue;
+	  if (!print_all)
+	    {
+	      if (TYPE_CODE (REGISTER_VIRTUAL_TYPE (i)) == TYPE_CODE_FLT)
+		continue;
+	      if (TYPE_VECTOR (REGISTER_VIRTUAL_TYPE (i)))
+		continue;
+	    }
 	}
       else
 	{

Re: [patch/rfc] Don't include vector registers in ``info registers''

[Daniel Jacobowitz]

On Thu, Aug 08, 2002 at 07:51:57PM -0400, Andrew Cagney wrote:
> Hello,
> 
> The attached patch modifies the generic ``info registsters'' command so 
> that it precludes vector registers (in addition to floating-point 
> registers).  The online doco indicates:
> 
> (gdb) help info registers
> List of integer registers and their contents, for selected stack frame.
> Register name as argument means describe only that register.
> (gdb) help info all-registers
> List of all registers and their contents, for selected stack frame.
> Register name as argument means describe only that register.
> 
> I think the change makes the behavour a better match for both the 
> documentation and what I think is the intent of the command.  Print a 
> minimal set of registers.
> 
> It will eventually affect the i386 -- I've a patch to change the type of 
> xmm [and mmx] registers to true vectors.  When that is in, ``info 
> registers'' will stop displaying the xmm registers.
> 
> Thoughts?

My only comment is that I would also like ``info registers'' not to
print kernel-mode or processor status registers, since that seems (to
me) to be appropriate; it should print registers for stack/pc and
computation primarily, not things like the MSR.

Might be harder to do generically though.

-- 
Daniel Jacobowitz                           Carnegie Mellon University
MontaVista Software                         Debian GNU/Linux Developer

Re: [patch/rfc] Don't include vector registers in ``info registers''

[Michael Snyder]

Daniel Jacobowitz wrote:
> 
> On Thu, Aug 08, 2002 at 07:51:57PM -0400, Andrew Cagney wrote:
> > Hello,
> >
> > The attached patch modifies the generic ``info registsters'' command so
> > that it precludes vector registers (in addition to floating-point
> > registers).  The online doco indicates:
> >
> > (gdb) help info registers
> > List of integer registers and their contents, for selected stack frame.
> > Register name as argument means describe only that register.
> > (gdb) help info all-registers
> > List of all registers and their contents, for selected stack frame.
> > Register name as argument means describe only that register.
> >
> > I think the change makes the behavour a better match for both the
> > documentation and what I think is the intent of the command.  Print a
> > minimal set of registers.
> >
> > It will eventually affect the i386 -- I've a patch to change the type of
> > xmm [and mmx] registers to true vectors.  When that is in, ``info
> > registers'' will stop displaying the xmm registers.
> >
> > Thoughts?
> 
> My only comment is that I would also like ``info registers'' not to
> print kernel-mode or processor status registers, since that seems (to
> me) to be appropriate; it should print registers for stack/pc and
> computation primarily, not things like the MSR.
> 
> Might be harder to do generically though.

That would mean that in order to see things like the status and
control registers, you would have to do info all-reg, which would
then spit out all your FP and VEC registers too.  Folks who need
the in-between registers might not like that.

Re: [patch/rfc] Don't include vector registers in ``info registers''

[Michael Snyder]

Andrew Cagney wrote:
> 
> Hello,
> 
> The attached patch modifies the generic ``info registsters'' command so
> that it precludes vector registers (in addition to floating-point
> registers).  The online doco indicates:
> 
> (gdb) help info registers
> List of integer registers and their contents, for selected stack frame.
> Register name as argument means describe only that register.
> (gdb) help info all-registers
> List of all registers and their contents, for selected stack frame.
> Register name as argument means describe only that register.
> 
> I think the change makes the behavour a better match for both the
> documentation and what I think is the intent of the command.  Print a
> minimal set of registers.

This matches what I implemented for at least one vector machine.
And my intuition agrees with yours -- info registers should produce
a display that fits in one screen, if possible.  Masses of registers
should be displayed some other way.


> It will eventually affect the i386 -- I've a patch to change the type of
> xmm [and mmx] registers to true vectors.  When that is in, ``info
> registers'' will stop displaying the xmm registers.
> 
> Thoughts?

Not having seen your patch (?), I wonder how you distinguished
vector registers from generic ones.

Re: [patch/rfc] Don't include vector registers in ``info registers''

[Elena Zannoni]

Ah, compare with what the rs6000-tdep.c has:
rs6000_do_registers_info().

I made the same change, but also added a skipping of the print of a
single vector register in decimal, it was taking up too much screen
real estate.

          /* If not a vector register, print it also in decimal.  */
	  if (!altivec_register_p (i))
	    {
	      printf_filtered ("\t");
	      val_print (REGISTER_VIRTUAL_TYPE (i), virtual_buffer, 0, 0,
			 gdb_stdout, 0, 1, 0, Val_pretty_default);
	    }

I also did something different: defining altivec_register_p(). I was
envisioning this predicate becoming an architecture method called
vector_register_p().

Another thing I've done in rs6000-tdep.c is the 'info vector' command
(even though it is called 'info altivec').

I'll be submitting shortly another powerpc variation (the e500) which
has vector registers, so I generalized things a bit in that file, wrt
vector registers.

Elena


Andrew Cagney writes:
 > [With patch]
 > 
 > 
 > > Hello,
 > > 
 > > The attached patch modifies the generic ``info registsters'' command so that it precludes vector registers (in addition to floating-point registers).  The online doco indicates:
 > > 
 > > (gdb) help info registers
 > > List of integer registers and their contents, for selected stack frame.
 > > Register name as argument means describe only that register.
 > > (gdb) help info all-registers
 > > List of all registers and their contents, for selected stack frame.
 > > Register name as argument means describe only that register.
 > > 
 > > I think the change makes the behavour a better match for both the documentation and what I think is the intent of the command.  Print a minimal set of registers.
 > > 
 > > It will eventually affect the i386 -- I've a patch to change the type of xmm [and mmx] registers to true vectors.  When that is in, ``info registers'' will stop displaying the xmm registers.
 > > 
 > > Thoughts?
 > > 
 > > enjoy,
 > > Andrew
 > 
 > 2002-08-08  Andrew Cagney  <cagney@redhat.com>
 > 
 > 	* infcmd.c (do_registers_info): Rename parameter ``fpregs'' to
 > 	``print_all''.  Only print vector registers when ``print_all''.
 > 
 > Index: infcmd.c
 > ===================================================================
 > RCS file: /cvs/src/src/gdb/infcmd.c,v
 > retrieving revision 1.49
 > diff -u -r1.49 infcmd.c
 > --- infcmd.c	3 Jul 2002 20:36:54 -0000	1.49
 > +++ infcmd.c	8 Aug 2002 21:14:22 -0000
 > @@ -1549,9 +1549,9 @@
 >  #ifdef REGISTER_NAMES
 >  char *gdb_register_names[] = REGISTER_NAMES;
 >  #endif
 > -/* Print out the machine register regnum. If regnum is -1,
 > -   print all registers (fpregs == 1) or all non-float registers
 > -   (fpregs == 0).
 > +/* Print out the machine register regnum. If regnum is -1, print all
 > +   registers (all == 1) or all non-float and non-vector registers (all
 > +   == 0).
 >  
 >     For most machines, having all_registers_info() print the
 >     register(s) one per line is good enough. If a different format
 > @@ -1561,7 +1561,7 @@
 >     to provide that format.  */
 >  
 >  void
 > -do_registers_info (int regnum, int fpregs)
 > +do_registers_info (int regnum, int print_all)
 >  {
 >    register int i;
 >    int numregs = NUM_REGS + NUM_PSEUDO_REGS;
 > @@ -1570,11 +1570,17 @@
 >  
 >    for (i = 0; i < numregs; i++)
 >      {
 > -      /* Decide between printing all regs, nonfloat regs, or specific reg.  */
 > +      /* Decide between printing all regs, non-float / vector regs, or
 > +         specific reg.  */
 >        if (regnum == -1)
 >  	{
 > -	  if (TYPE_CODE (REGISTER_VIRTUAL_TYPE (i)) == TYPE_CODE_FLT && !fpregs)
 > -	    continue;
 > +	  if (!print_all)
 > +	    {
 > +	      if (TYPE_CODE (REGISTER_VIRTUAL_TYPE (i)) == TYPE_CODE_FLT)
 > +		continue;
 > +	      if (TYPE_VECTOR (REGISTER_VIRTUAL_TYPE (i)))
 > +		continue;
 > +	    }
 >  	}
 >        else
 >  	{

Re: [patch/rfc] Don't include vector registers in ``info registers''

[Andrew Cagney]

> Another thing I've done in rs6000-tdep.c is the 'info vector' command
> (even though it is called 'info altivec').

Hmm, ``info vector-registers'' sounds good to me.

Andrew

Re: [patch/rfc] Don't include vector registers in ``info registers''

[Andrew Cagney]

> On Thu, Aug 08, 2002 at 07:51:57PM -0400, Andrew Cagney wrote:
> 
>> Hello,
>> 
>> The attached patch modifies the generic ``info registsters'' command so 
>> that it precludes vector registers (in addition to floating-point 
>> registers).  The online doco indicates:
>> 
>> (gdb) help info registers
>> List of integer registers and their contents, for selected stack frame.
>> Register name as argument means describe only that register.
>> (gdb) help info all-registers
>> List of all registers and their contents, for selected stack frame.
>> Register name as argument means describe only that register.
>> 
>> I think the change makes the behavour a better match for both the 
>> documentation and what I think is the intent of the command.  Print a 
>> minimal set of registers.
>> 
>> It will eventually affect the i386 -- I've a patch to change the type of 
>> xmm [and mmx] registers to true vectors.  When that is in, ``info 
>> registers'' will stop displaying the xmm registers.
>> 
>> Thoughts?
> 
> 
> My only comment is that I would also like ``info registers'' not to
> print kernel-mode or processor status registers, since that seems (to
> me) to be appropriate; it should print registers for stack/pc and
> computation primarily, not things like the MSR.

Yes.

There has been talk of adding a register attribute function.  I suspect, 
though, that the function is of more use to GUI's then the CLI.

> Might be harder to do generically though.

Yes.

I suspect the reality is that every architecture will eventually ends up 
with a custom info-registers function.   Simply because they want the 
output to be something useful.  Might even end up doing this for the i386.

I think we need to be careful to not fall into the trap of over 
engineering the default function.  Instead, encourage each target to do 
it custom but do provide some useful register print routines.

--

Thinking out loud.  Perhaphs the default ``info float[-registers]'' 
should scan for and print any floating-point registers and only print 
``no float'' if none are found.

enjoy,
Andrew

Re: [patch/rfc] Don't include vector registers in ``info registers''

[Andrew Cagney]

Not my day.

> It will eventually affect the i386 -- I've a patch to change the type of xmm [and mmx] registers to true vectors.  When that is in, ``info registers'' will stop displaying the xmm registers.

I figured out why it wasn't already affecting the i386.  That's been 
converted to the new vec128i already and the test is out-of-date :-(

> +	  if (!print_all)
> +	    {
> +	      if (TYPE_CODE (REGISTER_VIRTUAL_TYPE (i)) == TYPE_CODE_FLT)
> +		continue;
> +	      if (TYPE_VECTOR (REGISTER_VIRTUAL_TYPE (i)))
> +		continue;
> +	    }

This would need to read something like:

	if (TYPE_CODE (REGISTER_VIRTUAL_TYPE (i)) == struct or union
	  && first field's type was a vector)
	  continue;

What the heck, time for ``register_attribute()''.

Andrew

Re: [patch/rfc] Don't include vector registers in ``info registers''

[Andrew Cagney]

> Ah, compare with what the rs6000-tdep.c has:
> rs6000_do_registers_info().
> 
> I made the same change, but also added a skipping of the print of a
> single vector register in decimal, it was taking up too much screen
> real estate.
> 
>           /* If not a vector register, print it also in decimal.  */
> 	  if (!altivec_register_p (i))
> 	    {
> 	      printf_filtered ("\t");
> 	      val_print (REGISTER_VIRTUAL_TYPE (i), virtual_buffer, 0, 0,
> 			 gdb_stdout, 0, 1, 0, Val_pretty_default);
> 	    }

Hmm, yes, the vector register output is a bit long.

> I also did something different: defining altivec_register_p(). I was
> envisioning this predicate becoming an architecture method called
> vector_register_p().

Yes, I know GUI people are crying out for some sort of register 
attribute mechanism.  If/when it is added the current tests could 
quickly be replaced with more generic code that, by default, used the 
current heuristics.

> Another thing I've done in rs6000-tdep.c is the 'info vector' command
> (even though it is called 'info altivec').
> 
> I'll be submitting shortly another powerpc variation (the e500) which
> has vector registers, so I generalized things a bit in that file, wrt
> vector registers.

Hmm...

Andrew

Re: [patch/rfc] Don't include vector registers in ``info registers''

[Andrew Cagney]

> 2002-08-08  Andrew Cagney  <cagney@redhat.com>
> 
> 	* infcmd.c (do_registers_info): Rename parameter ``fpregs'' to
> 	``print_all''.  Only print vector registers when ``print_all''.
> 
I've checked this in.  `info registers', by default, excludes vector 
registers.

Andrew