pgpu debut TP3

TPs/TP3/CODE/AXPY/axpy.c

@@ -6,7 +6,7 @@ int main(int argc, char** argv)
 {
 	double alpha = 2;
 	double *X = NULL, *Y = NULL;
-  int N = 1000;
+	int N = 100000000;
 	if (argc > 1) N = atoi(argv[1]);
 
 	X = (double*) malloc(sizeof(double) * N);
@@ -18,6 +18,8 @@ int main(int argc, char** argv)
 		Y[i] = X[i] + i;
 	}
 
+	int runningOnGPU = 0;
+	#pragma omp target teams distribute parallel for map(to: X[:N]) map(tofrom: Y[:N]) map(from: runningOnGPU)
 	for(int i = 0; i < N; ++i)
 	{
 		Y[i] += alpha * X[i];

TPs/TP3/CODE/DGEMM/dgemm.c

@@ -19,21 +19,25 @@ double get_elapsedtime(void)
 void mult(int N, float *A, float* B, float* C)
 {
 	int i = 0, j = 0, k = 0;
-  int total_size = N*N;
+	//int total_size = N*N;
+	#pragma omp target teams distribute parallel for map(to: A) map(to: B) map(from: C)
 	for(i = 0; i < N; ++i)
 	{
 		for(j = 0; j < N; ++j)
 		{
-      float val = 0.;
 			for(k = 0; k < N; ++k)
 			{
-        val += A[i * N + k] * B[k * N + j];
+				C[i * N + j] += A[i * N + k] * B[k * N + j];
 			}
-      C[i * N + j] = val;
 		}
 	}
 }
 
+[[clang::optnone]]
+void donotopti(float *C){
+	return;
+}
+
 int main(int argc, char** argv)
 {
 	double t0 = 0., t1 = 0., duration = 0.;
@@ -77,5 +81,7 @@ int main(int argc, char** argv)
 
 	fprintf(stdout, "		 Elapsed Time : %f\n", duration);
 
+	donotopti(C);
+
 	return 0;
 }

TPs/TP3/CODE/ONGPU/Makefile

-CC=clang -fopenmp -fopenmp-targets=nvptx64-nvidia-cuda -Xopenmp-target -march=sm_61
+CC=clang -fopenmp -fopenmp-targets=nvptx64-nvidia-cuda
 CFLAGS=-O3 -Wall
 LDFLAGS=-lm -lrt
 

TPs/TP3/CODE/ONGPU/ongpu.c

@@ -17,5 +17,17 @@ int main()
 	else
 		printf("### Unable to use the GPU, using CPU! ###\n");
 
+	printf("Nb. devices: %d\n", omp_get_num_devices());
+
+	int team = 0;
+#pragma omp target map(from: team)
+	{
+#pragma omp teams
+		{
+			team = omp_get_num_teams();
+		}
+	}
+
+	printf("Nb. teams: %d\n", team);
 	return 0;
 }