Not able to build GPU custom op example

[alexminnaar]

Please make sure that this is a bug. As per our GitHub Policy, we only address code/doc bugs, performance issues, feature requests and build/installation issues on GitHub. tag:bug_template

System information

• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Ubuntu 16.04
• TensorFlow installed from (source or binary): binary
• TensorFlow version (use command below): 1.14.0
• Python version: 3.5
• CUDA/cuDNN version: 10.0
• GPU model and memory: GeForce GTX 980M

Describe the current behavior
Following the instructions here I navigate to tensorflow/tensorflow/examples/adding_an_op and run

nvcc -std=c++11 -c -o cuda_op_kernel.cu.o cuda_op_kernel.cu.cc  ${TF_CFLAGS[@]} -D GOOGLE_CUDA=1 -x cu -Xcompiler -fPIC

and I get the error

In file included from cuda_op_kernel.cu.cc:19:0:
/home/alex/.local/lib/python3.5/site-packages/tensorflow/include/tensorflow/core/util/gpu_kernel_helper.h:22:53: fatal error: third_party/gpus/cuda/include/cuda_fp16.h: No such file or directory
compilation terminated.

[jvishnuvardhan]

@alexminnaar

1. Could you mention with little more what were the steps you took before this error?

2. It is complaining cuda related kernel: No such file or directory
   Did you check the directory?
   Did you ran any TF code on GPU without the custom_op?

3. Please check these two resources 1 and 2.
   Thanks!

[alexminnaar]

@jvishnuvardhan sorry for the delay. Here are some more detailed steps.

1. Setup new venv and pip installed tensorflow-gpu
2. cloned the TensorFlow repo
3. ran TF_CFLAGS=( $(python -c 'import tensorflow as tf; print("".join(tf.sysconfig.get_compile_flags()))') )
4. Navigated to tensorflow/tensorflow/examples/adding_an_op within the cloned TensorFlow repo.
5. Followed the instructions outlined here i.e. nvcc -std=c++11 -c -o cuda_op_kernel.cu.o cuda_op_kernel.cu.cc ${TF_CFLAGS[@]} -D GOOGLE_CUDA=1 -x cu -Xcompiler -fPIC and received the error mentioned above.

[kzawisto]

@alexminnaar
I've stumbled across same problem.

The file you are looking for, third_party/gpus/cuda/include/cuda_fp16.h really comes from your CUDA installation i.e. $CUDA_HOME/include/cuda_fp16.h .

Somehow copy/link does not exist.

How about you go there:
/home/alex/.local/lib/python3.5/site-packages/tensorflow/include/
create directory third_party/gpus/cuda
and then symlink your CUDA include dir here (/usr/local/cuda/include on my Debian box)?

[sanjoy]

Relevant comment from @Artem-B: #34428 (comment)

[Saduf2019]

@alexminnaar
Please update as per above comment

[google-ml-butler]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you.

[google-ml-butler]

Closing as stale. Please reopen if you'd like to work on this further.

[google-ml-butler]

Are you satisfied with the resolution of your issue?
Yes
No