int main(int argc, char **argv) {
read_args(argc, argv);
counters timer;
start_measure(timer);
// declarations
Complex ioB(1.0, 1.0);
ioBuffer = cl::sycl::buffer<Complex,2>(cl::sycl::range<2> {M, N});
ioABuffer = cl::sycl::buffer<Complex,2>(cl::sycl::range<2> {M, N});
ioBBuffer = cl::sycl::buffer<Complex,1>(&ioB, cl::sycl::range<1> {1});
// initialization
for (size_t i = 0; i < M; ++i){
for (size_t j = 0; j < N; ++j){
float tmp = (float) (i*(j+2) + 10) / N;
Complex value(tmp, tmp);
cl::sycl::id<2> id = {i, j};
ioBuffer.get_access<cl::sycl::access::mode::write>()[id] = value;
ioABuffer.get_access<cl::sycl::access::mode::write>()[id] = value;
}
}
// our work
coef_var2D<0, 0> c1;
coef_var2D<1, 0> c2;
coef_var2D<0, 1> c3;
coef_var2D<-1, 0> c4;
coef_var2D<0, -1> c5;
auto st = c1+c2+c3+c4+c5;
input_var2D<Complex, &ioABuffer, &ioBBuffer, &fdl_in, &fac> work_in;
output_2D<Complex, &ioBuffer, &fdl_out> work_out;
auto op_work = work_out << st << work_in;
auto st_id = c1.toStencil();
input_var2D<Complex, &ioBuffer, &ioBBuffer, &fdl_in, &fac_id> copy_in;
output_2D<Complex, &ioABuffer, &fdl_out> copy_out;
auto op_copy = copy_out << st_id << copy_in;
end_init(timer);
auto begin_op = counters::clock_type::now();
// compute result with "gpu"
{
cl::sycl::queue myQueue;
for (unsigned int i = 0; i < NB_ITER; ++i){
//op_work.doComputation(myQueue);
op_work.doLocalComputation(myQueue);
op_copy.doComputation(myQueue);
}
}
auto end_op = counters::clock_type::now();
timer.stencil_time = std::chrono::duration_cast<counters::duration_type>(end_op - begin_op);
// loading time is not watched
end_measure(timer);
return 0;
}
int main(int argc, char **argv) {
read_args(argc, argv);
struct counters timer;
start_measure(timer);
// declarations
float tab_var = 1.0;
float *ioB = &tab_var;
ioBuffer = cl::sycl::buffer<float,2>(cl::sycl::range<2> {M, N});
ioABuffer = cl::sycl::buffer<float,2>(cl::sycl::range<2> {M, N});
ioBBuffer = cl::sycl::buffer<float,1>(ioB, cl::sycl::range<1> {1});
#if DEBUG_STENCIL
float *a_test = (float *) malloc(sizeof(float)*M*N);
float *b_test = (float *) malloc(sizeof(float)*M*N);
#endif
// initialization
for (size_t i = 0; i < M; ++i){
for (size_t j = 0; j < N; ++j){
float value = ((float) i*(j+2) + 10) / N;
cl::sycl::id<2> id = {i, j};
ioBuffer.get_access<cl::sycl::access::mode::write, cl::sycl::access::target::host_buffer>()[id] = value;
ioABuffer.get_access<cl::sycl::access::mode::write, cl::sycl::access::target::host_buffer>()[id] = value;
#if DEBUG_STENCIL
a_test[i*N+j] = value;
b_test[i*N+j] = value;
#endif
}
}
// our work
coef_var2D<0, 0> c1;
coef_var2D<1, 0> c2;
coef_var2D<0, 1> c3;
coef_var2D<-1, 0> c4;
coef_var2D<0, -1> c5;
auto st = c1+c2+c3+c4+c5;
input_var2D<float, &ioABuffer, &ioBBuffer, &fdl_in, &fac> work_in;
output_2D<float, &ioBuffer, &fdl_out> work_out;
auto op_work = work_out << st << work_in;
auto st_id = c1.toStencil();
input_var2D<float, &ioBuffer, &ioBBuffer, &fdl_in, &fac_id> copy_in;
output_2D<float, &ioABuffer, &fdl_out> copy_out;
auto op_copy = copy_out << st_id << copy_in;
end_init(timer);
struct op_time time_op;
begin_op(time_op);
// compute result with "gpu"
{
cl::sycl::queue myQueue;
for (unsigned int i = 0; i < NB_ITER; ++i){
//op_work.doComputation(myQueue);
op_work.doLocalComputation(myQueue);
op_copy.doComputation(myQueue);
}
}
end_op(time_op, timer.stencil_time);
// loading time is not watched
end_measure(timer);
#if DEBUG_STENCIL
// get the gpu result
auto C = (ioABuffer).get_access<cl::sycl::access::mode::read, cl::sycl::access::target::host_buffer>();
ute_and_are(a_test,b_test,C);
free(a_test);
free(b_test);
#endif
return 0;
}
