diff options
| author |  A. Unique TensorFlower <nobody@tensorflow.org> | 2016-04-13 13:35:03 -0800 |
|---|---|---|
| committer |  TensorFlower Gardener <gardener@tensorflow.org> | 2016-04-13 14:42:11 -0700 |
| commit | 53b9ba7c821ad1a44b63be58a93d3cd1859aac3d (patch) | |
| tree | dca142f13e15476ee2c24c72d419cca37376e652 /tensorflow/core/kernels/bias_op_gpu.cu.cc | |
| parent | 021f88578a15986b8d5b3cc2db0291fc3ab4dde4 (diff) | |
Make TF_CALL_REAL_NUMBER_TYPES() and related macros include Eigen::half
in addition to float. Explicitly exempt GPU effects that rely on atomics,
for which we have no good solution for half yet.
Note that there are still ops that don't _declare_ half support (those
that use \u201callnumbertypes\u201d or similar do, those that use \u201cfloat, double\u201d
don't); these will be fixed in a forthcoming commit.
Change: 119786433
Diffstat (limited to 'tensorflow/core/kernels/bias_op_gpu.cu.cc')
| -rw-r--r-- | tensorflow/core/kernels/bias_op_gpu.cu.cc | 2 |
1 files changed, 1 insertions, 1 deletions
diff --git a/tensorflow/core/kernels/bias_op_gpu.cu.cc b/tensorflow/core/kernels/bias_op_gpu.cu.cc index 62c6ed31a0..364555f141 100644 --- a/tensorflow/core/kernels/bias_op_gpu.cu.cc +++ b/tensorflow/core/kernels/bias_op_gpu.cu.cc @@ -227,7 +227,7 @@ void BiasGradGPU<T>::compute(const GPUDevice& d, const T* output_backprop, template struct BiasGPU<T>; \ template struct BiasGradGPU<T>; -TF_CALL_GPU_NUMBER_TYPES(DEFINE_GPU_SPECS); +TF_CALL_GPU_NUMBER_TYPES_NO_HALF(DEFINE_GPU_SPECS); } // end namespace tensorflow |