void advance_eye_paths(void* buffers[], void* args_orig) {
const timeval start_time = my_WallClockTime();
// cl_args
const starpu_args args;
unsigned iteration;
starpu_codelet_unpack_args(args_orig, &args, &iteration);
// buffers
// hit point static info
HitPointPosition* const hit_points = reinterpret_cast<HitPointPosition* const>(STARPU_VECTOR_GET_PTR(buffers[0]));
// eye paths
EyePath* const eye_paths = reinterpret_cast<EyePath* const>(STARPU_VECTOR_GET_PTR(buffers[1]));
const unsigned eye_paths_count = STARPU_VECTOR_GET_NX(buffers[1]);
// seed buffer
Seed* const seed_buffer = reinterpret_cast<Seed* const>(STARPU_VECTOR_GET_PTR(buffers[2]));
advance_eye_paths_impl(hit_points, // hit_points_count,
eye_paths, eye_paths_count,
seed_buffer, // seed_buffer_count,
args.cpu_scene,
args.config->max_eye_path_depth,
starpu_combined_worker_get_size());
const timeval end_time = my_WallClockTime();
task_info("CPU", 0, starpu_combined_worker_get_size(), iteration, start_time, end_time, "(3) advance_eye_paths");
}
static void cpu_mult(void *descr[], STARPU_ATTRIBUTE_UNUSED void *arg)
{
TYPE *subA = (TYPE *)STARPU_MATRIX_GET_PTR(descr[0]);
TYPE *subB = (TYPE *)STARPU_MATRIX_GET_PTR(descr[1]);
TYPE *subC = (TYPE *)STARPU_MATRIX_GET_PTR(descr[2]);
unsigned nxC = STARPU_MATRIX_GET_NX(descr[2]);
unsigned nyC = STARPU_MATRIX_GET_NY(descr[2]);
unsigned nyA = STARPU_MATRIX_GET_NY(descr[0]);
unsigned ldA = STARPU_MATRIX_GET_LD(descr[0]);
unsigned ldB = STARPU_MATRIX_GET_LD(descr[1]);
unsigned ldC = STARPU_MATRIX_GET_LD(descr[2]);
int worker_size = starpu_combined_worker_get_size();
if (worker_size == 1)
{
/* Sequential CPU task */
CPU_GEMM("N", "N", nxC, nyC, nyA, (TYPE)1.0, subA, ldA, subB, ldB, (TYPE)0.0, subC, ldC);
}
else
{
/* Parallel CPU task */
unsigned rank = starpu_combined_worker_get_rank();
unsigned block_size = (nyC + worker_size - 1)/worker_size;
unsigned new_nyC = STARPU_MIN(nyC, block_size*(rank+1)) - block_size*rank;
STARPU_ASSERT(nyC = STARPU_MATRIX_GET_NY(descr[1]));
TYPE *new_subB = &subB[block_size*rank];
TYPE *new_subC = &subC[block_size*rank];
CPU_GEMM("N", "N", nxC, new_nyC, nyA, (TYPE)1.0, subA, ldA, new_subB, ldB, (TYPE)0.0, new_subC, ldC);
}
}
void func(void* buffers[], void *args) {
printf("%d\n", starpu_combined_worker_get_size());
}
