New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Building issues #4368

Closed
fayeshine opened this Issue Sep 14, 2016 · 15 comments

Comments

Projects
None yet
@fayeshine
Contributor

fayeshine commented Sep 14, 2016

bazel build -c opt --config=cuda //tensorflow/tools/pip_package:build_pip_package
ERROR: The specified --crosstool_top '@local_config_cuda//crosstool:CROSSTOOL' is not a valid cc_toolchain_suite rule.
@jorgehatccrma

This comment has been minimized.

Show comment
Hide comment
@jorgehatccrma

jorgehatccrma Sep 15, 2016

I'm having the same issue, compiling on CentOS 7 using Cuda 7.5. If I compile without --config=cuda it builds and I can run it in CPU, but of course, I'd like to be able to run on GPU.

Any idea on what the problem could be?

jorgehatccrma commented Sep 15, 2016

I'm having the same issue, compiling on CentOS 7 using Cuda 7.5. If I compile without --config=cuda it builds and I can run it in CPU, but of course, I'd like to be able to run on GPU.

Any idea on what the problem could be?

@jorgehatccrma

This comment has been minimized.

Show comment
Hide comment
@jorgehatccrma

jorgehatccrma Sep 16, 2016

Could it be that Tensorflow relies on an older version of bazel to build for GPU (i.e. passing --config=cuda)? I say this because in this bazel changelog. In version 0.2.3 (2016-05-10), there's an entry that reads:

The key for the map to cc_toolchain_suite.toolchains is now a string of the form "cpu|compiler" (previously, it was just "cpu").

I then decided to compile bazel 0.2.2 (I was previously using 0.3.1) from source, and was finally able to successfully build with --config=cuda.

I guess something should be changed somwhere in tensorflow to use the new key format, so it can build with the latest bazel.

jorgehatccrma commented Sep 16, 2016

Could it be that Tensorflow relies on an older version of bazel to build for GPU (i.e. passing --config=cuda)? I say this because in this bazel changelog. In version 0.2.3 (2016-05-10), there's an entry that reads:

The key for the map to cc_toolchain_suite.toolchains is now a string of the form "cpu|compiler" (previously, it was just "cpu").

I then decided to compile bazel 0.2.2 (I was previously using 0.3.1) from source, and was finally able to successfully build with --config=cuda.

I guess something should be changed somwhere in tensorflow to use the new key format, so it can build with the latest bazel.

@blinkseb

This comment has been minimized.

Show comment
Hide comment
@blinkseb

blinkseb Sep 21, 2016

Same issue here. Using an older version of bazel works for r0.1.0, but I need a more recent version (we have multiple versions of gcc in our systems, and the most recent one is not in /usr/local/bin), which unfortunately requires bazel 0.3.1.

Any solution?

blinkseb commented Sep 21, 2016

Same issue here. Using an older version of bazel works for r0.1.0, but I need a more recent version (we have multiple versions of gcc in our systems, and the most recent one is not in /usr/local/bin), which unfortunately requires bazel 0.3.1.

Any solution?

@tkuebler

This comment has been minimized.

Show comment
Hide comment
@tkuebler

tkuebler Sep 22, 2016

Having this problem with both 0.2.3 and 0.3.1. Gonna try 0.2.2, but this is kinda bad, right? Anyone have an explanation of why? This is the first time I've used bazel.

tkuebler commented Sep 22, 2016

Having this problem with both 0.2.3 and 0.3.1. Gonna try 0.2.2, but this is kinda bad, right? Anyone have an explanation of why? This is the first time I've used bazel.

@davidzchen

This comment has been minimized.

Show comment
Hide comment
@davidzchen

davidzchen Sep 27, 2016

Member

Can you try again with TensorFlow at HEAD? #4285 made the CUDA CROSSTOOL file a proper cc_toolchain_suite.

Member

davidzchen commented Sep 27, 2016

Can you try again with TensorFlow at HEAD? #4285 made the CUDA CROSSTOOL file a proper cc_toolchain_suite.

@dantkz

This comment has been minimized.

Show comment
Hide comment
@dantkz

dantkz Oct 24, 2016

Contributor

This is still an issue for r0.10.

Contributor

dantkz commented Oct 24, 2016

This is still an issue for r0.10.

@Samurais

This comment has been minimized.

Show comment
Hide comment
@Samurais

Samurais Nov 9, 2016

I am using r0.10, when building tf with bazel build -c opt --config=cuda //tensorflow/cc:tutorials_example_trainer after configuring.

It throws an error.
image

Samurais commented Nov 9, 2016

I am using r0.10, when building tf with bazel build -c opt --config=cuda //tensorflow/cc:tutorials_example_trainer after configuring.

It throws an error.
image

@gunan

This comment has been minimized.

Show comment
Hide comment
@gunan

gunan Dec 21, 2016

Member

Is this still a problem at master or 0.12?

Member

gunan commented Dec 21, 2016

Is this still a problem at master or 0.12?

@gunan

This comment has been minimized.

Show comment
Hide comment
@gunan

gunan Jan 4, 2017

Member

Closing due to inactivity.
All our continuous builds and my local tests are clear of this problem. Please file a new issue if the problem persists.

Member

gunan commented Jan 4, 2017

Closing due to inactivity.
All our continuous builds and my local tests are clear of this problem. Please file a new issue if the problem persists.

@gunan gunan closed this Jan 4, 2017

@alexattia

This comment has been minimized.

Show comment
Hide comment
@alexattia

alexattia Jan 5, 2017

I am facing the same issue. Does someone know how to fix it ?

alexattia commented Jan 5, 2017

I am facing the same issue. Does someone know how to fix it ?

@gunan

This comment has been minimized.

Show comment
Hide comment
@gunan

gunan Jan 5, 2017

Member

Is it possible you forgot to run ./configure before bazel build?

Member

gunan commented Jan 5, 2017

Is it possible you forgot to run ./configure before bazel build?

@gaoshan2006

This comment has been minimized.

Show comment
Hide comment
@gaoshan2006

gaoshan2006 Feb 5, 2017

I meet the same issue. I built bazel( 0.4.3) from a old bazel ( 0.4.0). Then ./configure on Tensorflow, then use 0.4.3 bazel to compile. Got the issue ,
ERROR: The specified --crosstool_top '//third_party/gpus/crosstool:crosstool' is not a valid cc_toolchain_suite rule.

gaoshan2006 commented Feb 5, 2017

I meet the same issue. I built bazel( 0.4.3) from a old bazel ( 0.4.0). Then ./configure on Tensorflow, then use 0.4.3 bazel to compile. Got the issue ,
ERROR: The specified --crosstool_top '//third_party/gpus/crosstool:crosstool' is not a valid cc_toolchain_suite rule.

@gunan

This comment has been minimized.

Show comment
Hide comment
@gunan

gunan Feb 5, 2017

Member

Any bazel version older than 0.4.2 will not work, as there has been some backwards incompatible changes to our build files. Not sure what you mean with you built bazel 0.4.3 from 0.4.0.

Member

gunan commented Feb 5, 2017

Any bazel version older than 0.4.2 will not work, as there has been some backwards incompatible changes to our build files. Not sure what you mean with you built bazel 0.4.3 from 0.4.0.

@AntoineHus

This comment has been minimized.

Show comment
Hide comment
@AntoineHus

AntoineHus Apr 25, 2017

Hello I have the issue with my configuration since I updated cud from 7.5 to 8.0.
My tensorflow version is r0.10, cudnn version is 5 and hazel is 0.3.2.

Any idea how to solve this issue ?

AntoineHus commented Apr 25, 2017

Hello I have the issue with my configuration since I updated cud from 7.5 to 8.0.
My tensorflow version is r0.10, cudnn version is 5 and hazel is 0.3.2.

Any idea how to solve this issue ?

@zhm1995

This comment has been minimized.

Show comment
Hide comment
@zhm1995

zhm1995 Apr 9, 2018

Using a lower version of bazel can solve this problem (e.g bazel 0.2.2)

zhm1995 commented Apr 9, 2018

Using a lower version of bazel can solve this problem (e.g bazel 0.2.2)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment