1 /* Copyright 2017 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 #if GOOGLE_CUDA 17 18 #define EIGEN_USE_GPU 19 20 #include "tensorflow/core/framework/register_types.h" 21 #include "tensorflow/core/framework/tensor_types.h" 22 #include "tensorflow/core/kernels/cuda_device_array_gpu.h" 23 #include "tensorflow/core/util/cuda_kernel_helper.h" 24 25 namespace tensorflow { 26 27 using GPUDevice = Eigen::GpuDevice; 28 29 namespace { 30 31 template <typename T> 32 __global__ void DynamicStitchKernel(const int32 slice_size, 33 const int32 output_size, 34 CudaDeviceArrayStruct<int32> input_indices, 35 CudaDeviceArrayStruct<const T*> input_ptrs, 36 T* output) { 37 int32* data_indices = GetCudaDeviceArrayOnDevice(&input_indices); 38 const T** data_ptrs = GetCudaDeviceArrayOnDevice(&input_ptrs); 39 CUDA_1D_KERNEL_LOOP(output_index, output_size) { 40 const int32 slice_id = output_index / slice_size; 41 const int32 slice_offset = output_index % slice_size; 42 const int32 input_index = data_indices[slice_id]; 43 if (input_index != -1) { 44 output[output_index] = ldg(data_ptrs[input_index] + slice_offset); 45 } 46 } 47 } 48 49 } // namespace 50 51 template <typename T> 52 void DynamicStitchGPUImpl(const Eigen::GpuDevice& gpu_device, 53 const int32 slice_size, const int32 first_dim_size, 54 const CudaDeviceArrayStruct<int>& input_indices, 55 const CudaDeviceArrayStruct<const T*>& input_ptrs, 56 T* output) { 57 const int32 output_size = first_dim_size * slice_size; 58 auto config = GetCudaLaunchConfig(output_size, gpu_device); 59 60 DynamicStitchKernel<T> 61 <<<config.block_count, config.thread_per_block, 0, gpu_device.stream()>>>( 62 slice_size, output_size, input_indices, input_ptrs, output); 63 } 64 65 #define REGISTER_GPU(T) \ 66 template void DynamicStitchGPUImpl( \ 67 const Eigen::GpuDevice& gpu_device, const int32 slice_size, \ 68 const int32 first_dim_size, \ 69 const CudaDeviceArrayStruct<int32>& input_indices, \ 70 const CudaDeviceArrayStruct<const T*>& input_ptrs, T* output); 71 72 TF_CALL_GPU_NUMBER_TYPES(REGISTER_GPU); 73 TF_CALL_complex64(REGISTER_GPU); 74 TF_CALL_complex128(REGISTER_GPU); 75 TF_CALL_int64(REGISTER_GPU); 76 TF_CALL_int32(REGISTER_GPU) 77 78 #undef REGISTER_GPU 79 80 } // namespace tensorflow 81 #endif // GOOGLE_CUDA 82
