two bugs in the arm simulator

two bugs in the arm simulator

[Janne Hellsten]

Hi list,

I have submitted two bug reports to the GDB bug database (PR #1957 and
PR #1900).  Both of these are issues in the ARM simulator.

Is anyone working on these?  I'm asking, because at least the Mac OSX
compilation issue should be easily fixable, and I might be able to do
this myself.  However, if someone's has already looked at this, then
it doesn't make much sense for me to start.

As I didn't get any response from the GNATS interface, it's difficult
for me to check whether or not these have been fixed in CVS.

Best regards,
Janne

Re: two bugs in the arm simulator

[Daniel Jacobowitz]

On Mon, Jul 04, 2005 at 12:06:23PM +0900, Janne Hellsten wrote:
> Hi list,
> 
> I have submitted two bug reports to the GDB bug database (PR #1957 and
> PR #1900).  Both of these are issues in the ARM simulator.
> 
> Is anyone working on these?  I'm asking, because at least the Mac OSX
> compilation issue should be easily fixable, and I might be able to do
> this myself.  However, if someone's has already looked at this, then
> it doesn't make much sense for me to start.
> 
> As I didn't get any response from the GNATS interface, it's difficult
> for me to check whether or not these have been fixed in CVS.

GNATS ought to have sent you something back.  It does have your
address.  I'm afraid PR 1900 is the least of the problems with the
simulator on 64-bit hosts; it appears to be pretty busted.  No one is
working on either AFAIK though I think there's another PR about the
busted simulator somewhere.

-- 
Daniel Jacobowitz
CodeSourcery, LLC

Re: two bugs in the arm simulator

[Janne Hellsten]

Hi,

> GNATS ought to have sent you something back.  It does have your
> address.  I'm afraid PR 1900 is the least of the problems with the
> simulator on 64-bit hosts; it appears to be pretty busted.  No one is
> working on either AFAIK though I think there's another PR about the
> busted simulator somewhere.

Yes, it did send back my reports, but nothing else.

I can probably fix PR 1957 myself, I'd like to get familiar with the
simulator.  The thing is that I'd like to extend GDB's ARM simulator
so that I can get it to report the number of cycles spent in some
program.  This could be used for simple benchmarking (to detect
regression over time).

I wonder if such an extension would be a big task?  I'm not familiar
with GDB's code at all.

BR,
Janne

Re: two bugs in the arm simulator

[Daniel Jacobowitz]

On Mon, Jul 04, 2005 at 03:20:36PM +0900, Janne Hellsten wrote:
> Hi,
> 
> > GNATS ought to have sent you something back.  It does have your
> > address.  I'm afraid PR 1900 is the least of the problems with the
> > simulator on 64-bit hosts; it appears to be pretty busted.  No one is
> > working on either AFAIK though I think there's another PR about the
> > busted simulator somewhere.
> 
> Yes, it did send back my reports, but nothing else.
> 
> I can probably fix PR 1957 myself, I'd like to get familiar with the
> simulator.  The thing is that I'd like to extend GDB's ARM simulator

In that case, you may want to file an FSF copyright assignment. 
Otherwise we can't accept substantive contributions.

> so that I can get it to report the number of cycles spent in some
> program.  This could be used for simple benchmarking (to detect
> regression over time).
> 
> I wonder if such an extension would be a big task?  I'm not familiar
> with GDB's code at all.

The GDB simulators aren't even a little bit cycle accurate,
unfortunately.

-- 
Daniel Jacobowitz
CodeSourcery, LLC

number of executed instructions from simulator (was Re: two bugs in the arm simulator)

[Janne Hellsten]

Hi,

> > so that I can get it to report the number of cycles spent in some
> > program.  This could be used for simple benchmarking (to detect
> > regression over time).
> >
> > I wonder if such an extension would be a big task?  I'm not familiar
> > with GDB's code at all.
> 
> The GDB simulators aren't even a little bit cycle accurate,
> unfortunately.

Yes, I am aware of this.  However, just getting the number of executed
instructions would be sufficient for my purposes.  I could use this in
our company's performance regression tester to check relative
performance figures (e.g., is last week's version faster than today's
version, etc.).

Best regards,
Janne

Re: two bugs in the arm simulator

[Janne Hellsten]

Hi,

> I have submitted two bug reports to the GDB bug database (PR #1957 and
> PR #1900).  Both of these are issues in the ARM simulator.

Looks like PR 1957 is already fixed in CVS.  I tested compiling both
gdb-6.3 and the CVS with GCC 4.0.0, and while the 6.3 version didn't
compile, the CVS version did.

As a work-around for PR 1900, I guess I could compile GDB into a
32-bit binary (on an x86_64 host).  Is this somehow possible?

Best regards,
Janne

Re: two bugs in the arm simulator

[Daniel Jacobowitz]

On Tue, Jul 05, 2005 at 05:01:11PM +0900, Janne Hellsten wrote:
> Hi,
> 
> > I have submitted two bug reports to the GDB bug database (PR #1957 and
> > PR #1900).  Both of these are issues in the ARM simulator.
> 
> Looks like PR 1957 is already fixed in CVS.  I tested compiling both
> gdb-6.3 and the CVS with GCC 4.0.0, and while the 6.3 version didn't
> compile, the CVS version did.
> 
> As a work-around for PR 1900, I guess I could compile GDB into a
> 32-bit binary (on an x86_64 host).  Is this somehow possible?

Sure.  Use CC='gcc -m32' if you have the appropriate libraries
installed from your vendor.

-- 
Daniel Jacobowitz
CodeSourcery, LLC