GPU ядро умножения матриц

GPU ядро умножения матриц

(продолжение)

As[ty][tx] = A[a + wA * ty + tx]; 

// Load the matrices from global memory to shared memory;

Bs[ty][tx] = B[b + wB * ty + tx]; 

// each thread loads one element of each matrix 

__syncthreads(); 

// Synchronize to make sure the matrices are loaded

y

()

y

// Multiply the two matrices together; 
// each thread computes one element 
// of the block sub-matrix

//

for

(

int

k = 0; k < BLOCK_SIZE; ++k) 

Csub += As[ty][k] * Bs[k][tx]; 

// Synchronize to make sure that the preceding 

// Synchronize to make sure that the preceding 
// computation is done before loading two new 
// sub-matrices of A and B in the next iteration

__syncthreads(); 
} 

} 

// Write the block sub-matrix to global memory; 
// each thread writes one element 

int

c = wB * BLOCK_SIZE * by + BLOCK_SIZE * bx; 

C[c + wB * ty + tx] = Csub; 

C[c + wB * ty + tx] = Csub; 
}