Updated benchmark for GPU

elementary/benchmark/main.c

@@ -27,19 +27,28 @@ void init_chunk_elems(chunk_info_t *ci) {
 void compute_next_chunk_elems(struct dispatch_context *dc, struct futhark_context *fc, chunk_info_t *ci) {
     envelope_t *outer_envelope = get_outer_envelope(dc, fc, 1);
 
-    struct futhark_opaque_envelope_1d_i8 *fut_outer_envelope = futhark_outer_envelope_new(dc, fc, outer_envelope,
-                                                                                          futhark_restore_opaque_envelope_1d_i8,
-                                                                                          FUTHARK_I8);
+    double start0 = MPI_Wtime();
+    struct futhark_opaque_envelope_1d_i8 *fut_outer_envelope =
+            futhark_outer_envelope_new(dc, fc, outer_envelope, futhark_restore_opaque_envelope_1d_i8, FUTHARK_I8);
 
     struct futhark_i8_1d *fut_chunk_elems = futhark_new_i8_1d(fc, ci->data, ci->dimensions[1]);
     struct futhark_i8_1d *fut_next_chunk_elems;
 
     futhark_context_sync(fc);
+    double finish0 = MPI_Wtime();
+    printf("0. nce -> %f\n", finish0 - start0);
+
+    double start1 = MPI_Wtime();
     futhark_entry_next_chunk_elems(fc, &fut_next_chunk_elems, fut_chunk_elems, fut_outer_envelope);
     futhark_context_sync(fc);
+    double finish1 = MPI_Wtime();
+    printf("1. nce -> %f\n", finish1 - start1);
 
+    double start2 = MPI_Wtime();
     futhark_values_i8_1d(fc, fut_next_chunk_elems, ci->data);
     futhark_context_sync(fc);
+    double finish2 = MPI_Wtime();
+    printf("2. values_i8 -> %f\n", finish2 - start2);
 
     futhark_free_i8_1d(fc, fut_next_chunk_elems);
     futhark_free_i8_1d(fc, fut_chunk_elems);
@@ -61,7 +70,7 @@ int main(int argc, char *argv[]) {
     struct futhark_context_config *fut_config = futhark_context_config_new();
     int nb_devices = atoi(argv[1]);
 #if defined(FUTHARK_BACKEND_cuda) || defined(FUTHARK_BACKEND_opencl)
-    #if defined(FUTHARK_BACKEND_opencl)
+#if defined(FUTHARK_BACKEND_opencl)
     futhark_context_config_list_devices(fut_config);
 #endif
     char device[4] = {0};