Building Tensorflow from source with XLA enabled and without GPU

474 views
Skip to first unread message

Ankita Chandak

unread,
Jun 28, 2017, 5:31:47 PM6/28/17
to Discuss

Hi,
I built and installed tensorflow from source with XLA enabled and GPU disabled (basically I opted N for everything while configuring via ./config except XLA enabling as Y). There were lot of warnings regrding deprecated syntax while building. but the build was successful.
I am able to import tensorflow and run basic print command in session. But while I try to do some computation (for eg. simple addition) it gives me following error:
2017-06-28 15:09:22.366052: F tensorflow/compiler/xla/statusor.cc:41] Attempting to fetch value instead of handling error Not found: could not find registered computation placer for platform Executor -- check target linkage
Aborted


I did a bit of debugging and this error comes just after the call from client/sessions.py:1262 to pywrap_tensorflow:
tf_session.TF_Run(session, options,
feed_dict, fetch_list, target_list,
status, run_metadata)


so I believe it's because it is unable to link to _pywrap_tensorflow_internal.so.
Can you please provide any fix to this or is there something am doing wrong here?
This is blocking my further task so any kind of help is appreciated!


Thanks & Regards

benhurna...@gmail.com

unread,
Jul 3, 2017, 4:39:26 PM7/3/17
to Discuss
Same problem. Very hard to just compile TensorFlow and use it.

Annanay Agarwal

unread,
Jul 4, 2017, 1:42:11 AM7/4/17
to Ankita Chandak, Discuss
Hi Ankita,

Have you followed the instructions strictly according to tensorflow.org/install/install_sources ?
If so, then the problem should not persist.

The warnings are OK. If the build was successful, move to a folder apart from the root installation and then run the python script that uses tensorflow.

Thanks,

--
You received this message because you are subscribed to the Google Groups "Discuss" group.
To unsubscribe from this group and stop receiving emails from it, send an email to discuss+unsubscribe@tensorflow.org.
To post to this group, send email to dis...@tensorflow.org.
To view this discussion on the web visit https://groups.google.com/a/tensorflow.org/d/msgid/discuss/07a5e53c-a295-4fed-a5ef-7b5d23be26fc%40tensorflow.org.



--
Annanay Agarwal
Department of CSE
IIT Hyderabad

Toby Boyd

unread,
Jul 5, 2017, 12:59:37 PM7/5/17
to Annanay Agarwal, Ankita Chandak, Discuss

Note: Turning on JIT at the session level will not result in operations being compiled for the CPU. JIT compilation for CPU operations must be done via the manual method documented below. This decision was made due to the CPU backend being single-threaded.

Sharing, on the off chance you did not see the message above.  I did not want you to experience any additional frustration.  There are plans for more work on XLA and CPU as well as GPU in the near future and contributions are welcome.

Toby

To view this discussion on the web visit https://groups.google.com/a/tensorflow.org/d/msgid/discuss/CA%2BizcZRtP5QezAKsOSTPYRQSGEsYV4tv14hOy1t1HzsAg9n1eA%40mail.gmail.com.

ac2...@cornell.edu

unread,
Jul 5, 2017, 4:14:06 PM7/5/17
to Discuss, ankitac...@gmail.com
Yes I did!


On Tuesday, July 4, 2017 at 12:42:11 AM UTC-5, Annanay Agarwal wrote:
Hi Ankita,

Have you followed the instructions strictly according to tensorflow.org/install/install_sources ?
If so, then the problem should not persist.

The warnings are OK. If the build was successful, move to a folder apart from the root installation and then run the python script that uses tensorflow.

Thanks,
On Thu, Jun 29, 2017 at 3:01 AM, Ankita Chandak <ankitac...@gmail.com> wrote:

Hi,
I built and installed tensorflow from source with XLA enabled and GPU disabled (basically I opted N for everything while configuring via ./config except XLA enabling as Y). There were lot of warnings regrding deprecated syntax while building. but the build was successful.
I am able to import tensorflow and run basic print command in session. But while I try to do some computation (for eg. simple addition) it gives me following error:
2017-06-28 15:09:22.366052: F tensorflow/compiler/xla/statusor.cc:41] Attempting to fetch value instead of handling error Not found: could not find registered computation placer for platform Executor -- check target linkage
Aborted


I did a bit of debugging and this error comes just after the call from client/sessions.py:1262 to pywrap_tensorflow:
tf_session.TF_Run(session, options,
feed_dict, fetch_list, target_list,
status, run_metadata)


so I believe it's because it is unable to link to _pywrap_tensorflow_internal.so.
Can you please provide any fix to this or is there something am doing wrong here?
This is blocking my further task so any kind of help is appreciated!


Thanks & Regards

--
You received this message because you are subscribed to the Google Groups "Discuss" group.
To unsubscribe from this group and stop receiving emails from it, send an email to discuss+u...@tensorflow.org.

ac2...@cornell.edu

unread,
Jul 5, 2017, 4:16:17 PM7/5/17
to Discuss, cs14bte...@iith.ac.in, ankitac...@gmail.com
Yes I saw this note. But the issue isn't regrading JIT compilation I feel. Even if I don't enable XLA, I am getting the error. I did find workaround though. . I've to explicitly assign tf.device to '/cpu'. But I hope the issue gets resolved in upcoming releases.
To unsubscribe from this group and stop receiving emails from it, send an email to discuss+u...@tensorflow.org.
--
Annanay Agarwal
Department of CSE
IIT Hyderabad

--
You received this message because you are subscribed to the Google Groups "Discuss" group.
To unsubscribe from this group and stop receiving emails from it, send an email to discuss+u...@tensorflow.org.

To post to this group, send email to dis...@tensorflow.org.

Darren Garvey

unread,
Jul 5, 2017, 4:53:13 PM7/5/17
to ac2...@cornell.edu, ankitac...@gmail.com, Discuss, cs14bte...@iith.ac.in
This is a real issue I think. I spotted the same appear in master in the last couple of weeks, on a specific net, with XLA compiled in but not enabled. The same net worked fine with XLA disabled at compile time.

We should probably move this to a bug on github. I'll dig out the net and see what op is causing it in my case, if that's what it is.

-- 
Darren



To unsubscribe from this group and stop receiving emails from it, send an email to discuss+unsubscribe@tensorflow.org.

To post to this group, send email to dis...@tensorflow.org.

Toby Boyd

unread,
Jul 5, 2017, 5:00:20 PM7/5/17
to Darren Garvey, ac2...@cornell.edu, Ankita Chandak, Discuss, Annanay Agarwal
Opening an issue with the details and code sample would be great.

Thanks Darren.  

To view this discussion on the web visit https://groups.google.com/a/tensorflow.org/d/msgid/discuss/CAEW5AMC53MHOcFqnkLYT5PdFWASpvapz0tjCC6ea_5dbMdwKRg%40mail.gmail.com.

ac2...@cornell.edu

unread,
Jul 5, 2017, 5:12:22 PM7/5/17
to Discuss, darren...@gmail.com, ac2...@cornell.edu, ankitac...@gmail.com, cs14bte...@iith.ac.in
I've raised a PR on github already with the same title.

--
You received this message because you are subscribed to the Google Groups "Discuss" group.
To unsubscribe from this group and stop receiving emails from it, send an email to discuss+u...@tensorflow.org.
To post to this group, send email to dis...@tensorflow.org.
Reply all
Reply to author
Forward
0 new messages