#10: allow for different N for each benchmark

Author: cwschilly

src/benchmarks.cc

@@ -11,7 +11,7 @@ namespace benchmarks {
 
 template <>
 std::string typeToString<double>() {
-    return "double";
+    return "double ";
 }
 
 template <>
@@ -61,7 +61,7 @@ benchmark_results_t runBenchmarkLevel1(int N, int iters) {
 
     int rank = -1;
     MPI_Comm_rank(MPI_COMM_WORLD, &rank);
-    std::cout << "[level1 " << typeToString<T>() << "] rank: " << rank << ", total_time=" << total_time << std::endl;
+    std::cout << "[level1 " << typeToString<T>() << "] (N=" << N << ") rank: " << rank << ", total_time=" << total_time << std::endl;
 
     return std::make_tuple(iter_timings, total_time);
 }
@@ -101,7 +101,7 @@ benchmark_results_t runBenchmarkLevel2(int M, int N, int iters) {
 
     int rank = -1;
     MPI_Comm_rank(MPI_COMM_WORLD, &rank);
-    std::cout << "[level2 " << typeToString<T>() << "] rank: " << rank << ", total_time=" << total_time << std::endl;
+    std::cout << "[level2 " << typeToString<T>() << "] (M=" << M << ", N=" << N << ") rank: " << rank << ", total_time=" << total_time << std::endl;
 
     return std::make_tuple(iter_timings, total_time);
 }
@@ -142,7 +142,7 @@ benchmark_results_t runBenchmarkLevel3(int M, int N, int K, int iters) {
     int rank = -1;
     MPI_Comm_rank(MPI_COMM_WORLD, &rank);
 
-    std::cout << "[level3 " << typeToString<T>() << "] rank: " << rank << ", total_time=" << total_time << std::endl;
+    std::cout << "[level3 " << typeToString<T>() << "] (M=" << M << ", N=" << N << ", K=" << K << ") rank: " << rank << ", total_time=" << total_time << std::endl;
 
     return std::make_tuple(iter_timings, total_time);
 }
@@ -191,7 +191,7 @@ benchmark_results_t runBenchmarkDPOTRF(int N, int iters) {
     int rank = -1;
     MPI_Comm_rank(MPI_COMM_WORLD, &rank);
 
-    std::cout << "[dpotrf " << typeToString<T>() << "] rank: " << rank << ", total_time=" << total_time << std::endl;
+    std::cout << "[dpotrf " << typeToString<T>() << "] (N=" << N << ") rank: " << rank << ", total_time=" << total_time << std::endl;
 
     return std::make_tuple(iter_timings, total_time);
 }
@@ -212,9 +212,11 @@ benchmark_results_t runBenchmark(benchmark_types b, int M, int N, int K, int ite
     }
 }
 
-all_results_t runAllBenchmarks(int M, int N, int K, int iters) {
+all_results_t runAllBenchmarks(int M, int N1, int N2, int N3, int K, int iters) {
     all_results_t all_results;
+    int N;
     for (int i=0; i < benchmark_types::num_benchmarks; i++) {
+        N = i == 0 ? N1 : (i == 1 ? N2 : N3);
         auto b = static_cast<benchmark_types>(i);
         std::string benchmark_str = benchmarkToString(b);
         all_results[benchmark_str + "_double"] = runBenchmark<double>(b, M, N, K, iters);

src/benchmarks.h

@@ -37,7 +37,7 @@ benchmark_results_t runBenchmarkDPOTRF(int N, int iters);
 template <typename T>
 benchmark_results_t runBenchmark(benchmark_types b, int M, int N, int K, int iters);
 
-all_results_t runAllBenchmarks(int M, int N, int K, int iters);
+all_results_t runAllBenchmarks(int M, int N1, int N2, int N3, int K, int iters);
 
 void printBenchmarkOutput(all_results_t benchmark_results, int iters);
 

src/slow_node.cc

@@ -8,15 +8,41 @@
 
 static int iters = 100;
 static int M = 128;
-static int N = 128;
+static int N1 = 128;
+static int N2 = 128;
+static int N3 = 128;
 static int K = 128;
 
+/*
+ * USAGE: ./slow_node <iters> <M/K> | <N1> <N2> <N3>
+ *
+ *    M and K share the same value, given by the <M/K> slot.
+ *
+ *    If <N1> <N2> and <N3> are not provided, the value given for
+ *    M and K is also used for all N.
+ *
+ *    Similarly, if <N2> is not provided, the value given for <N1>
+ *    is used for all N, and so on with <N3>.
+ */
+
 int main(int argc, char** argv) {
+
   if (argc > 1) {
     iters = atoi(argv[1]) + 1; // add one iteration since we will drop the first one
-    M = N = K = atoi(argv[2]);
+    M = N1 = N2 = N3 = K = atoi(argv[2]);
+    if (argc > 3) {
+      N1 = N2 = N3 = atoi(argv[3]);
+      if (argc > 4) {
+        N2 = N3 = atoi(argv[4]);
+        if (argc > 5) {
+          N3 = atoi(argv[5]);
+        }
+      }
+    }
   }
-  std::cout << "iters: " << iters << ", M=N=K=" << M << std::endl;
+
+  std::cout << "iters: " << iters << ", M=K=" << M << std::endl;
+  std::cout << "N1=" << N1 << ", N2=" << N2 << ", N3=" << N3 << std::endl;
 
   MPI_Init(&argc, &argv);
   Kokkos::initialize(argc, argv);
@@ -27,7 +53,7 @@ int main(int argc, char** argv) {
 
   // Loop through all available benchmarks
   sensors::runSensorsAndReduceOutput(processor_name, "pre");
-  auto output = benchmarks::runAllBenchmarks(M, N, K, iters);
+  auto output = benchmarks::runAllBenchmarks(M, N1, N2, N3, K, iters);
   sensors::runSensorsAndReduceOutput(processor_name, "post");
   benchmarks::printBenchmarkOutput(output, iters);