3 files changed, 74 insertions, 30 deletions
diff --git a/GEMV/baselines/cpu/Makefile b/GEMV/baselines/cpu/Makefile
index c779651..6f369a7 100644
--- a/GEMV/baselines/cpu/Makefile
+++ b/GEMV/baselines/cpu/Makefile
@@ -1,7 +1,26 @@
-all:
-		gcc -o gemv -fopenmp gemv_openmp.c 
+.PHONY: all
+all: gemv
 
-clean:
-		rm gemv
+gemv: gemv_openmp.c
+	gcc -O2 -o gemv -fopenmp gemv_openmp.c
+
+gemv_O0: gemv_openmp.c
+	gcc -o gemv_O0 -fopenmp gemv_openmp.c
+
+gemv_O2: gemv_openmp.c
+	gcc -O2 -o gemv_O2 -fopenmp gemv_openmp.c
+
+.PHONY: run run_O0 run_O2
 
+run: gemv
+	./gemv
 
+run_O0: gemv_O0
+	./gemv_O0
+
+run_O2: gemv_O2
+	./gemv_O2
+
+.PHONY: clean
+clean:
+	rm -f gemv gemv_O0 gemv_O2
diff --git a/GEMV/baselines/cpu/gemv_openmp.c b/GEMV/baselines/cpu/gemv_openmp.c
index 307e03b..69c3ae1 100644
--- a/GEMV/baselines/cpu/gemv_openmp.c
+++ b/GEMV/baselines/cpu/gemv_openmp.c
@@ -5,43 +5,55 @@
 
 int main(int argc, char *argv[])
 {
-  const size_t rows = 20480;
-  const size_t cols = 8192;
+    const size_t rows = 20480;
+    const size_t cols = 8192;
 
-  double **A, *b, *x;
+    double **A, *b, *x;
 
-  b = (double*) malloc(sizeof(double)*rows);
-  x = (double*) malloc(sizeof(double)*cols);
+    b = (double*) malloc(sizeof(double)*rows);
+    x = (double*) malloc(sizeof(double)*cols);
 
-  allocate_dense(rows, cols, &A);
+    allocate_dense(rows, cols, &A);
 
-  make_hilbert_mat(rows,cols, &A);
+    make_hilbert_mat(rows,cols, &A);
+
+    Timer timer;
+    for (int i = 0; i < 100; i++) {
 
 #pragma omp parallel
-    {
+        {
 #pragma omp for
-    for (size_t i = 0; i < cols; i++) {
-      x[i] = (double) i+1 ;
-    }
+        for (size_t i = 0; i < cols; i++) {
+          x[i] = (double) i+1 ;
+        }
 
 #pragma omp for
-    for (size_t i = 0; i < rows; i++) {
-      b[i] = (double) 0.0;
-    }
-    }
-
-  Timer timer;
-  start(&timer, 0, 0);
-
-
-   gemv(A, x, rows, cols, &b);
-   
-   stop(&timer, 0);
+        for (size_t i = 0; i < rows; i++) {
+          b[i] = (double) 0.0;
+        }
+        }
 
+        unsigned int nr_threads = 0;
+#pragma omp parallel
+#pragma omp atomic
+        nr_threads++;
+
+        start(&timer, 0, 0);
+        gemv(A, x, rows, cols, &b);
+        stop(&timer, 0);
+        printf("[::] n_threads=%d e_type=%s n_elements=%d "
+            "| throughput_cpu_omp_MBps=%f\n",
+            nr_threads, "double", rows * cols,
+            rows * cols * sizeof(double) / timer.time[0]);
+        printf("[::] n_threads=%d e_type=%s n_elements=%d "
+            "| throughput_cpu_omp_MOpps=%f\n",
+            nr_threads, "double", rows * cols,
+            rows * cols / timer.time[0]);
+        printf("[::] n_threads=%d e_type=%s n_elements=%d |",
+            nr_threads, "double", rows * cols);
+        printall(&timer, 0);
+    }
 
-    printf("Kernel ");
-    print(&timer, 0, 1);
-    printf("\n");
 
 #if 0
   print_vec(x, rows);
diff --git a/GEMV/baselines/cpu/run.sh b/GEMV/baselines/cpu/run.sh
new file mode 100755
index 0000000..8f936a3
--- /dev/null
+++ b/GEMV/baselines/cpu/run.sh
@@ -0,0 +1,13 @@
+#!/bin/sh
+
+set -e
+
+echo "prim-benchmarks GEMV CPU (dfatool edition)"
+echo "Started at $(date)"
+echo "Revision $(git describe --always)"
+
+make
+
+for nr_threads in 1 2 4 6 8 12 16 20 24 32; do
+	OMP_NUM_THREADS=${nr_threads} timeout --foreground -k 1m 30m ./gemv || true
+done