1
Fork 0
mirror of https://github.com/Steffo99/unimore-hpc-assignments.git synced 2024-11-23 08:34:23 +00:00
hpc-2022-g3/atax/atax.cu

132 lines
2.5 KiB
Text
Raw Normal View History

2022-11-11 12:23:45 +00:00
#include <stdio.h>
#include <unistd.h>
#include <string.h>
#include <math.h>
/* Include polybench common header. */
#include "polybench.hu"
2022-11-11 12:23:45 +00:00
/* Include benchmark-specific header. */
/* Default data type is double, default size is 4000. */
2022-11-28 13:37:37 +00:00
#include "atax.hu"
2022-11-11 12:23:45 +00:00
// Workaround for the editor not finding M_PI
// It is exclusive to the GNU C compiler
// https://www.gnu.org/software/libc/manual/html_node/Mathematical-Constants.html
#ifndef M_PI
#define M_PI 3.141
#endif
/**
* Initialize the arrays to be used in the computation:
*
* - `x` is filled with multiples of `M_PI`;
* - `A` is filled with sample data.
*
* To be called on the CPU (uses the `__host__` qualifier).
*/
2022-11-29 10:27:51 +00:00
__host__ static void init_array(DATA_TYPE** A, DATA_TYPE* x)
2022-11-11 12:23:45 +00:00
{
2022-11-29 13:23:24 +00:00
for (int i = 0; i < NY; i++)
{
x[i] = i * M_PI;
}
2022-11-29 13:23:24 +00:00
for (int i = 0; i < NX; i++)
{
for (int j = 0; j < NY; j++)
{
2022-11-29 10:27:51 +00:00
A[i][j] = ((DATA_TYPE)i * (j + 1)) / NX;
}
}
2022-11-11 12:23:45 +00:00
}
2022-11-29 13:23:24 +00:00
/**
* Print the given array.
*
* Cannot be parallelized, as the elements of the array should be
*
* To be called on the CPU (uses the `__host__` qualifier).
*/
2022-11-29 13:23:24 +00:00
__host__ static void print_array(DATA_TYPE* y)
2022-11-11 12:23:45 +00:00
{
2022-11-29 13:23:24 +00:00
for (int i = 0; i < NX; i++)
{
fprintf(stderr, DATA_PRINTF_MODIFIER, y[i]);
}
fprintf(stderr, "\n");
2022-11-11 12:23:45 +00:00
}
/**
* Compute ATAX.
*
* Parallelizing this is the goal of the assignment.
*
* Currently to be called on the CPU (uses the `__host__` qualifier), but we may probably want to change that soon.
*/
2022-11-29 10:27:51 +00:00
__host__ static void kernel_atax(DATA_TYPE** A, DATA_TYPE* x, DATA_TYPE* y)
{
2022-11-29 13:23:24 +00:00
for (int i = 0; i < NY; i++)
{
y[i] = 0;
}
2022-11-29 13:23:24 +00:00
for (int i = 0; i < NX; i++)
{
DATA_TYPE tmp = 0;
2022-11-29 13:23:24 +00:00
for (int j = 0; j < NY; j++)
{
tmp += A[i][j] * x[j];
}
2022-11-29 13:23:24 +00:00
for (int j = 0; j < NY; j++)
{
y[j] = y[j] + A[i][j] * tmp;
}
}
2022-11-11 12:23:45 +00:00
}
2022-11-29 13:23:24 +00:00
/**
* The main function of the benchmark, which sets up tooling to measure the time spent computing `kernel_atax`.
*
* We should probably avoid editing this.
*/
__host__ int main(int argc, char **argv)
2022-11-11 12:23:45 +00:00
{
2022-11-29 13:23:24 +00:00
DATA_TYPE **A = new DATA_TYPE*[NX];
DATA_TYPE x[NY], y[NX];
for(size_t i=0; i<NX; i++)
{
A[i] = new DATA_TYPE[NY];
}
#ifdef POLYBENCH_INCLUDE_INIT
polybench_start_instruments;
#endif
2022-11-29 13:23:24 +00:00
2022-11-29 10:27:51 +00:00
init_array(A, x);
#ifndef POLYBENCH_INCLUDE_INIT
polybench_start_instruments;
#endif
2022-11-29 10:27:51 +00:00
kernel_atax(A, x, y);
polybench_stop_instruments;
polybench_print_instruments;
/* Prevent dead-code elimination. All live-out data must be printed by the function call in argument. */
2022-11-29 10:27:51 +00:00
polybench_prevent_dce(print_array(y));
return 0;
2022-11-11 12:23:45 +00:00
}