4 Commits (e57a40b8a64a87370d7b9f46b86ec7887694229a)

Author	SHA1	Message	Date
QI JUN	61ec0b9516	Refine device context (#6433) ... There are mainly following fixes: - take `DeviceContext` as the template parameter of math functors and OpKernel instead of `Place` - remove `eigen_device` interface in base class `DeviceContext` - remove `GetEigenDevice` interface in `ExecutionContext` and base class `DeviceContext` - remove unused `platform::EigenDeviceConverter` - rename `REGISTER_OP_GPU_KERNEL` to `REGISTER_OP_CUDA_KERNEL` - rename `USE_GPU_ONLY_OP` to `USE_CUDA_ONLY_OP`	7 years ago
dangqingqing	91d4fc6941	Fix compling for softmax_with_cross_entropy_op.	7 years ago
Liu Yiqun	05ed8ee8ab	Add SoftmaxGradFunctor, and use SoftmaxGradFunctor in softmax_op instead.	8 years ago
caoying03	f1d5fb3b9a	support soft labels.	8 years ago