Failed to run example from "CUDA by Example"

233 views
Skip to first unread message

Conrado Miranda

unread,
Feb 20, 2012, 12:02:27 PM2/20/12
to gpuocelot
Hi there!

I just installed gpuocelot and I'm very satisfied with its current
status. Congratulations to you all.

As a test, I tried to run the examples from the book "CUDA by
Example", available at http://developer.nvidia.com/cuda-example-introduction-general-purpose-gpu-programming.

The first few examples ran perfectly, but the example named "ray.cu",
it failed while running with the following message:

terminate called after throwing an instance of
'executive::RuntimeException'
what(): Cvta instruction cannot convert from specified address
space to generic addresses
Aborted

The information that I can think right now that could be useful are:
- The code for "ray.cu" is available at http://pastebin.com/G8sH57V4.
- The configuration file is the one available at the source code and
put at http://pastebin.com/PTmwGZSN.
- The script I used to compile that example is the same I used for the
other examples that worked and is available at http://pastebin.com/kLMS935K.
- "OcelotConfig -v" outputs "2.1.1764" and I'm at revision 1773 of
SVN.

The example's code is correct, as it runs perfectly at a GTX580.

I'd be glad to provide any other information you guys need.

Thanks.

Gregory Diamos

unread,
Feb 24, 2012, 8:39:14 PM2/24/12
to gpuo...@googlegroups.com
Hi,

Sorry about this, I think the issue is that an optimization pass was
committed
to the trunk that wasn't ready for prime-time. Try disabling
hoistSpecialValues
in your configure.ocelot file. The sample file in the SVN trunk should
be updated
now.

Regards,

Greg

Conrado Miranda

unread,
Feb 24, 2012, 10:45:24 PM2/24/12
to gpuo...@googlegroups.com
Hi,

That example ran perfectly after that change. I'll finish running them
(may take a while, but I'll do my best) and report any other problem.

Thank you very much!

> --
> You received this message because you are subscribed to the Google Groups
> "gpuocelot" group.
> To post to this group, send email to gpuo...@googlegroups.com.
> To unsubscribe from this group, send email to
> gpuocelot+...@googlegroups.com.
> For more options, visit this group at
> http://groups.google.com/group/gpuocelot?hl=en.
>

Jimmy

unread,
Apr 10, 2012, 7:08:43 AM4/10/12
to gpuocelot
Hi,

I also ran these examples with gpuocelot and found that those using
lib_glut ran extremely slow. I could Ctrl+C to terminate them and get
traces with 32768 warps. Is that correct?

Thank you!

On Feb 25, 11:45 am, Conrado Miranda <miranda.conr...@gmail.com>
wrote:
> Hi,
>
> That example ran perfectly after that change. I'll finish running them
> (may take a while, but I'll do my best) and report any other problem.
>
> Thank you very much!
>
> On Fri, Feb 24, 2012 at 11:39 PM, Gregory Diamos
>
>
>
>
>
>
>
> <gregory.dia...@gatech.edu> wrote:
> > Hi,
>
> > Sorry about this, I think the issue is that an optimization pass was
> > committed
> > to the trunk that wasn't ready for prime-time.  Try disabling
> > hoistSpecialValues
> > in your configure.ocelot file. The sample file in the SVN trunk should be
> > updated
> > now.
>
> > Regards,
>
> > Greg
>
> > On 02/20/2012 09:02 AM, Conrado Miranda wrote:
>
> >> Hi there!
>
> >> I just installed gpuocelot and I'm very satisfied with its current
> >> status. Congratulations to you all.
>
> >> As a test, I tried to run the examples from the book "CUDA by
> >> Example", available at
> >>http://developer.nvidia.com/cuda-example-introduction-general-purpose....
>
> >> The first few examples ran perfectly, but the example named "ray.cu",
> >> it failed while running with the following message:
>
> >> terminate called after throwing an instance of
> >> 'executive::RuntimeException'
> >>   what():  Cvta instruction cannot convert from specified address
> >> space to generic addresses
> >> Aborted
>
> >> The information that I can think right now that could be useful are:
> >> - The code for "ray.cu" is available athttp://pastebin.com/G8sH57V4.
> >> - The configuration file is the one available at the source code and
> >> put athttp://pastebin.com/PTmwGZSN.

Gregory...@gatech.edu

unread,
Apr 10, 2012, 12:30:29 PM4/10/12
to gpuo...@googlegroups.com
Are you trying to generate traces?  You can turn most is them on/off in configure.ocelot.

Also expect at least a 40x slowdown compared to native execution on a CPU without tracing on.  It is an emulator.

Gregory Diamos


-----Original message-----
From: Jimmy <panjans...@gmail.com>
To:
gpuocelot <gpuo...@googlegroups.com>
Sent:
Tue, Apr 10, 2012 16:25:17 GMT+00:00
Subject:
[gpuocelot] Re: Failed to run example from "CUDA by Example"

Hi,

I also ran these examples with gpuocelot and found that those using
lib_glut ran extremely slow. I could Ctrl+C to terminate them and get
traces with 32768 warps. Is that correct?

Thank you!

On Feb 25, 11:45 am, Conrado Miranda
wrote:
> Hi,
>
> That example ran perfectly after that change. I'll finish running them
> (may take a while, but I'll do my best) and report any other problem.
>
> Thank you very much!
>
> On Fri, Feb 24, 2012 at 11:39 PM, Gregory Diamos
>
>
>
>
>
>
>

Jianming Pan

unread,
Apr 11, 2012, 11:29:29 AM4/11/12
to gpuo...@googlegroups.com
Yes, I'm trying to gen the traces. And I completed after set the bitmap down to 256x256.

Thank you very much!




--
Best regards,

Jianming Pan (潘健明)
Master Student
Institute of Microelectronics, Tsinghua Univ., 100084 Beijng, China
Faculty of Engineering, K.U.Leuven, B-3001 Leuven, Belgium

Gregory Diamos

unread,
Apr 12, 2012, 5:41:16 PM4/12/12
to gpuo...@googlegroups.com
I think that the X86 traces used to drive macsim will generate traces with entries for each warp.  Each of the trace
generators can use a different format, so some of them may be tiny and some of them may be very large.  You can
control which ones are enabled in the configure.ocelot file.

Greg
Reply all
Reply to author
Forward
0 new messages