CwiseMul.cpp (2551B)
1 #include <iostream> 2 #define EIGEN_USE_SYCL 3 #include <unsupported/Eigen/CXX11/Tensor> 4 5 using Eigen::array; 6 using Eigen::SyclDevice; 7 using Eigen::Tensor; 8 using Eigen::TensorMap; 9 10 int main() 11 { 12 using DataType = float; 13 using IndexType = int64_t; 14 constexpr auto DataLayout = Eigen::RowMajor; 15 16 auto devices = Eigen::get_sycl_supported_devices(); 17 const auto device_selector = *devices.begin(); 18 Eigen::QueueInterface queueInterface(device_selector); 19 auto sycl_device = Eigen::SyclDevice(&queueInterface); 20 21 // create the tensors to be used in the operation 22 IndexType sizeDim1 = 3; 23 IndexType sizeDim2 = 3; 24 IndexType sizeDim3 = 3; 25 array<IndexType, 3> tensorRange = {{sizeDim1, sizeDim2, sizeDim3}}; 26 27 // initialize the tensors with the data we want manipulate to 28 Tensor<DataType, 3,DataLayout, IndexType> in1(tensorRange); 29 Tensor<DataType, 3,DataLayout, IndexType> in2(tensorRange); 30 Tensor<DataType, 3,DataLayout, IndexType> out(tensorRange); 31 32 // set up some random data in the tensors to be multiplied 33 in1 = in1.random(); 34 in2 = in2.random(); 35 36 // allocate memory for the tensors 37 DataType * gpu_in1_data = static_cast<DataType*>(sycl_device.allocate(in1.size()*sizeof(DataType))); 38 DataType * gpu_in2_data = static_cast<DataType*>(sycl_device.allocate(in2.size()*sizeof(DataType))); 39 DataType * gpu_out_data = static_cast<DataType*>(sycl_device.allocate(out.size()*sizeof(DataType))); 40 41 // 42 TensorMap<Tensor<DataType, 3, DataLayout, IndexType>> gpu_in1(gpu_in1_data, tensorRange); 43 TensorMap<Tensor<DataType, 3, DataLayout, IndexType>> gpu_in2(gpu_in2_data, tensorRange); 44 TensorMap<Tensor<DataType, 3, DataLayout, IndexType>> gpu_out(gpu_out_data, tensorRange); 45 46 // copy the memory to the device and do the c=a*b calculation 47 sycl_device.memcpyHostToDevice(gpu_in1_data, in1.data(),(in1.size())*sizeof(DataType)); 48 sycl_device.memcpyHostToDevice(gpu_in2_data, in2.data(),(in2.size())*sizeof(DataType)); 49 gpu_out.device(sycl_device) = gpu_in1 * gpu_in2; 50 sycl_device.memcpyDeviceToHost(out.data(), gpu_out_data,(out.size())*sizeof(DataType)); 51 sycl_device.synchronize(); 52 53 // print out the results 54 for (IndexType i = 0; i < sizeDim1; ++i) { 55 for (IndexType j = 0; j < sizeDim2; ++j) { 56 for (IndexType k = 0; k < sizeDim3; ++k) { 57 std::cout << "device_out" << "(" << i << ", " << j << ", " << k << ") : " << out(i,j,k) 58 << " vs host_out" << "(" << i << ", " << j << ", " << k << ") : " << in1(i,j,k) * in2(i,j,k) << "\n"; 59 } 60 } 61 } 62 printf("c=a*b Done\n"); 63 }