mirror of
https://github.com/Steffo99/unimore-hpc-assignments.git
synced 2024-11-26 10:04:23 +00:00
157 lines
3.2 KiB
Text
157 lines
3.2 KiB
Text
#include <stdio.h>
|
|
#include <unistd.h>
|
|
#include <string.h>
|
|
#include <math.h>
|
|
|
|
/* Include polybench common header. */
|
|
#include "polybench.hu"
|
|
|
|
/* Include benchmark-specific header. */
|
|
/* Default data type is double, default size is 4000. */
|
|
#include "atax.hu"
|
|
|
|
// 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).
|
|
*/
|
|
__host__ static void init_array(DATA_TYPE** A, DATA_TYPE* X, DATA_TYPE* Y)
|
|
{
|
|
/* X = [ 3.14, 6.28, 9.42, ... ] */
|
|
for (unsigned int y = 0; y < NY; y++)
|
|
{
|
|
X[y] = y * M_PI;
|
|
}
|
|
|
|
/* Y = [ 0.00, 0.00, 0.00, ... ] */
|
|
for (unsigned int x = 0; x < NY; x++)
|
|
{
|
|
Y[x] = 0;
|
|
}
|
|
|
|
/*
|
|
* A = [
|
|
* [ 0, 0, 0, 0, ... ],
|
|
* [ 1 / NX, 2 / NX, 3 / NX, 4 / NX, ... ],
|
|
* [ 2 / NX, 4 / NX, 6 / NX, 8 / NX, ... ],
|
|
* [ 3 / NX, 6 / NX, 9 / NX, 12 / NX, ... ],
|
|
* ...
|
|
* ]
|
|
*/
|
|
for (unsigned int x = 0; x < NX; x++)
|
|
{
|
|
for (unsigned int y = 0; y < NY; y++)
|
|
{
|
|
A[x][y] = (DATA_TYPE)(x * (y + 1)) / NX;
|
|
}
|
|
}
|
|
}
|
|
|
|
|
|
/**
|
|
* 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).
|
|
*/
|
|
__host__ static void print_array(DATA_TYPE* Y)
|
|
{
|
|
for (unsigned int x = 0; x < NX; x++)
|
|
{
|
|
fprintf(stderr, DATA_PRINTF_MODIFIER, Y[x]);
|
|
}
|
|
fprintf(stderr, "\n");
|
|
}
|
|
|
|
|
|
/**
|
|
* Compute ATAX :
|
|
* - A is the input matrix
|
|
* - X is an input vector
|
|
* - Y is the result vector
|
|
*
|
|
* In particular:
|
|
* ```
|
|
* A * (A * X) = Y
|
|
* ```
|
|
* Wait, there's no transposition here?!?
|
|
*
|
|
* 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.
|
|
*/
|
|
__host__ static void kernel_atax(DATA_TYPE** A, DATA_TYPE* X, DATA_TYPE* Y)
|
|
{
|
|
for (unsigned int x = 0; x < NX; x++)
|
|
{
|
|
DATA_TYPE tmp = 0;
|
|
|
|
for (unsigned int y = 0; y < NY; y++)
|
|
{
|
|
tmp += A[x][y] * X[y];
|
|
}
|
|
|
|
for (unsigned int y = 0; y < NY; y++)
|
|
{
|
|
Y[y] += A[x][y] * tmp;
|
|
}
|
|
}
|
|
}
|
|
|
|
|
|
/**
|
|
* 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)
|
|
{
|
|
#ifndef POLYBENCH_USE_CUDA
|
|
|
|
// A[NX][NY]
|
|
DATA_TYPE** A = new DATA_TYPE*[NX] {};
|
|
for(unsigned int x = 0; x < NX; x++)
|
|
{
|
|
A[x] = new DATA_TYPE[NY] {};
|
|
}
|
|
|
|
DATA_TYPE* x = new DATA_TYPE[NY] {};
|
|
DATA_TYPE* y = new DATA_TYPE[NX] {};
|
|
|
|
#ifdef POLYBENCH_INCLUDE_INIT
|
|
polybench_start_instruments;
|
|
#endif
|
|
|
|
init_array(A, x, y);
|
|
|
|
#ifndef POLYBENCH_INCLUDE_INIT
|
|
polybench_start_instruments;
|
|
#endif
|
|
|
|
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. */
|
|
polybench_prevent_dce(print_array(y));
|
|
|
|
#else
|
|
|
|
|
|
|
|
#endif
|
|
|
|
return 0;
|
|
}
|