#define TS 32
#define WPT 8
#define RTS (TS/WPT)

__kernel void sgemm(__global float *A, __global float *B, __global float *C, int M, int N, int K) {
  const int row = get_local_id(0); // row index of C
  const int col = get_local_id(1); // column index of C
	const int global_row = TS * get_group_id(0) + row;
	const int global_col = TS * get_group_id(1) + col;

	__local float subA[TS][TS];
	__local float subB[TS][TS];

	float accum[WPT];
	for (int w = 0; w < WPT; w++) {
		accum[w] = 0.0f;
	}

	const int num_tiles = (K+TS-1)/TS;
	for (int t = 0; t < num_tiles; t++) {
		for (int w = 0; w < WPT; w++) {
			const int tiledRow = TS*t + row;
			const int tiledCol = TS*t + col;
			/*if 			(M <= global_row) subA[row + w*RTS][col] = 0.0f;
			else*/

			if (K <= tiledCol) subA[row + w*RTS][col] = 0.0f;
			else if (M <= (global_row + w*RTS)) subA[row + w*RTS][col] = 0.0f;

			else 
				subA[row + w*RTS][col] = A[(global_row + w*RTS)*K + tiledCol];

			if			(N <= global_col) subB[row + w*RTS][col] = 0.0f;
			//else if	(K <= tiledRow) subB[row + w*RTS][col] = 0.0f;
			else if	(K <= (tiledRow + w*RTS)) subB[row + w*RTS][col] = 0.0f;

			else 
				subB[row + w*RTS][col] = B[(tiledRow + w*RTS)*N + global_col];
		}
		barrier(CLK_LOCAL_MEM_FENCE);

		for (int k = 0; k < TS; k++) {
			for (int w = 0; w < WPT; w++) {
				accum[w] += subA[row + w*RTS][k] * subB[k][col];
			}
		}
		barrier(CLK_LOCAL_MEM_FENCE);
	}
	for (int w = 0; w < WPT; w++) {
		if				(M <= (global_row + w*RTS));// C[(global_row + w*RTS)*N + global_col] = 0.0f;
		else if		(N <= global_col);// C[(global_row + w*RTS)*N + global_col] = 0.0f;
		else 
			C[(global_row + w*RTS)*N + global_col] = accum[w];
	}
}