I'm trying to get a tensorflow C++ build (or tensorflow lite) for Windows that runs on GPU (WITHOUT using CUDA, it should work on AMD). My best guess was to use OpenCL and SYCL. I downloaded Codeplay's tensorflow-sycl repo and tried building it with this command after running configure.py:
bazel build --verbose_failures --jobs=6 --config=sycl --config=opt //tensorflow:tensorflow.dll
This causes the following error:
WARNING: The following rc files are no longer being read, please transfer their contents or import their path into one of the standard rc files:
c:\users\asus\desktop\tfcodeplay/tools/bazel.rc
ERROR: Config value 'monolithic' is not defined in any .rc file
After a little search on the internet I found out that I can copy the necessary contents of the tools/bazel.rc file into .tf_configure.bazelrc fixes the error. However, now I have this error:
ERROR: C:/users/asus/desktop/tfcodeplay/WORKSPACE:3:1: name 'http_archive' is not defined
ERROR: Error computing the main repository mapping: Encountered error while reading extension file 'closure/defs.bzl': no such package '@io_bazel_rules_closure//closure': error loading package 'external': Could not load //external package
Loading:
My specs:
- Bazel 6.1.2
- Intel Core I7-6700HQ
- Nvidia GeForce GTX 960m
- Windows 10
My question is this: Is it even possible to achieve a Tensorflow C++ build on Windows that runs on GPU without CUDA? If so, what am I doing wrong with this build? Do I need to be on an AMD device to get this build, if so is this error caused by me being on an Intel machine?
