Question

任务：
我正在尝试使用OpenCL使用bazel构建TensorFlow库，以便在GPU上运行操作而无需启用Tensorflow的GPU（不使用CUDA / NVIDIA）。
基本上想借助Android NDK和内置的opencl-tensorflow库在移动GPU上运行）

采取的步骤：
将/home/ashok/.cache/bazel/_bazel_ashok/337a0d8d7fbaddcaa8098e349718a6d6/external中的eigen_archive libray替换为＆eigen-opencl＆＃39; https://bitbucket.org/benoitsteiner/eigen-opencl的库，使用bazel构建TensorFlow。

面临的问题：＆＃39; half＆＃39;不是Eigen＆＃39;

的成员

INFO: Found 1 target...
ERROR: /home/ashok/Ashok/tensorflow-1.0.1/tensorflow/core/BUILD:814:1: C++ compilation of rule '//tensorflow/core:android_tensorflow_lib_lite' failed (Exit 1)
In file included from ./tensorflow/core/framework/allocator.h:25:0,
from ./tensorflow/core/framework/op_kernel.h:22,
from ./tensorflow/core/util/guarded_philox_random.h:19,
from tensorflow/core/util/guarded_philox_random.cc:16:
./tensorflow/core/framework/type_traits.h:69:52: error: 'half' is not a member of 'Eigen'
std::is_trivial::value || std::is_same<T, Eigen::half>::value ||
^
./tensorflow/core/framework/type_traits.h:69:52: error: 'half' is not a member of 'Eigen'
./tensorflow/core/framework/type_traits.h:69:63: error: template argument 2 is invalid
std::is_trivial::value || std::is_same<T, Eigen::half>::value ||
^
Target //tensorflow/examples/android:tensorflow_demo failed to build

请帮助解决问题

Answer 1

使用https://bitbucket.org/benoitsteiner/opencl代替https://bitbucket.org/benoitsteiner/eigen-opencl，因为它是最新的存储库

＆＃39;半＆＃39;不是Eigen＆＃39;的成员。

1 个答案: