1 /* Copyright 2015 The TensorFlow Authors. All Rights Reserved. 2 3 Licensed under the Apache License, Version 2.0(the "License"); 4 you may not use this file except in compliance with the License. 5 You may obtain a copy of the License at 6 7 http://www.apache.org/licenses/LICENSE-2.0 8 9 Unless required by applicable law or agreed to in writing, software 10 distributed under the License is distributed on an "AS IS" BASIS, 11 WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 12 See the License for the specific language governing permissions and 13 limitations under the License. 14 ==============================================================================*/ 15 16 #ifdef INTEL_MKL 17 18 // See docs in ../ops/math_ops.cc. 19 20 #define EIGEN_USE_THREADS 21 #include <iostream> 22 #include <vector> 23 24 #include "tensorflow/core/kernels/cwise_ops_common.h" 25 26 #include "tensorflow/core/util/mkl_util.h" 27 28 namespace tensorflow { 29 30 typedef Eigen::ThreadPoolDevice CPUDevice; 31 32 template <typename Device, typename Functor> 33 class MklBinaryOp : public BinaryOp<Device, Functor> { 34 public: 35 explicit MklBinaryOp(OpKernelConstruction* context) 36 : BinaryOp<Device, Functor>(context) {} 37 38 void Compute(OpKernelContext* context) override { 39 auto in0 = context->input(0); 40 auto in1 = context->input(1); 41 VLOG(1) << "Shapes (start mklbinaryop compute): " 42 << in0.shape().DebugString() << " _and_ " 43 << in1.shape().DebugString(); 44 45 // Call the TensorFlow BinaryOp Compute method 46 BinaryOp<Device, Functor>::Compute(context); 47 48 auto out = context->mutable_output(0); 49 VLOG(1) << "Shapes (output): " << out->shape().DebugString(); 50 51 // Pass input shape through to output shape 52 ForwardMklMetaDataInToOut(context, 0, 0); 53 54 out = context->mutable_output(0); 55 VLOG(1) << "Shapes (output): " << out->shape().DebugString(); 56 } 57 }; 58 59 //---------- Registration macros for various element-wise ops ----------- 60 // We will need to redefine "REGISTER" to include the mkl_op_registry flag 61 #pragma push_macro("REGISTER") 62 #undef REGISTER 63 #define REGISTER(OP, D, N, F, T) \ 64 REGISTER_KERNEL_BUILDER(Name(N) \ 65 .Device(DEVICE_##D) \ 66 .TypeConstraint<T>("T") \ 67 .Label(mkl_op_registry::kMklOpLabel), \ 68 OP<D##Device, F<T>>); 69 70 REGISTER5(MklBinaryOp, CPU, "_MklAdd", functor::add, float, Eigen::half, double, 71 int32, int64); 72 REGISTER7(MklBinaryOp, CPU, "_MklSub", functor::sub, float, Eigen::half, double, 73 int32, int64, complex64, complex128); 74 REGISTER5(MklBinaryOp, CPU, "_MklMul", functor::mul, float, Eigen::half, double, 75 uint8, int32); 76 REGISTER5(MklBinaryOp, CPU, "_MklMaximum", functor::maximum, float, Eigen::half, 77 double, int32, int64); 78 REGISTER5(MklBinaryOp, CPU, "_MklSquaredDifference", 79 functor::squared_difference, float, Eigen::half, double, int32, 80 int64); 81 82 #undef REGISTER 83 #pragma pop_macro("REGISTER") 84 //----------------------------------------------------------------------- 85 86 } // end namespace tensorflow 87 88 #endif // INTEL_MKL 89
