1 #define EIGEN_USE_THREADS 2 3 #include <string> 4 5 #include "tensor_benchmarks.h" 6 7 #define CREATE_THREAD_POOL(threads) \ 8 Eigen::ThreadPool pool(threads); \ 9 Eigen::ThreadPoolDevice device(&pool, threads); 10 11 12 // Contractions for number of threads ranging from 1 to 32 13 // Dimensions are Rows, Cols, Depth 14 #define BM_ContractionCPU(D1, D2, D3) \ 15 static void BM_##Contraction##_##D1##x##D2##x##D3(int iters, int Threads) { \ 16 StopBenchmarkTiming(); \ 17 CREATE_THREAD_POOL(Threads); \ 18 BenchmarkSuite<Eigen::ThreadPoolDevice, float> suite(device, D1, D2, D3); \ 19 suite.contraction(iters); \ 20 } \ 21 BENCHMARK_RANGE(BM_##Contraction##_##D1##x##D2##x##D3, 1, 32); 22 23 24 // Vector Matrix and Matrix Vector products 25 BM_ContractionCPU(1, 2000, 500); 26 BM_ContractionCPU(2000, 1, 500); 27 28 // Various skinny matrices 29 BM_ContractionCPU(250, 3, 512); 30 BM_ContractionCPU(1500, 3, 512); 31 32 BM_ContractionCPU(512, 800, 4); 33 BM_ContractionCPU(512, 80, 800); 34 BM_ContractionCPU(512, 80, 13522); 35 BM_ContractionCPU(1, 80, 13522); 36 37 BM_ContractionCPU(3200, 512, 4); 38 BM_ContractionCPU(3200, 512, 80); 39 BM_ContractionCPU(3200, 80, 512); 40
