naive-matmul.c

inline
float* h_loc(float* h_B, int width, int i, int j)
{
	return h_B + i*width + j;
}

void hostMatrixMultiply(
	float* h_C,
	float* h_A,
	float* h_B,
	int m,
	int n,
	int r)
{

	clock_t start = clock(), diff;

	for(int i=0; i<m; i++)
	{
		for(int k=0; k<r; k++)
		{
			float sum = 0;
			for(int j=0; j<n; j++)
			{
				sum += *h_loc(h_A, n, i, j) * *h_loc(h_B, r, j, k);
			}

			*h_loc(h_C, r, i, k) = sum;
		}
	}

	diff = clock() - start;

	int msec = diff * 1000 / CLOCKS_PER_SEC;
	printf("hostMatrixMultiply: %i ms\n", msec);

}