{
// Block index
int bx = blockIdx.x;
int by = blockIdx.
// Thread index
int tx = threadIdx.x;
int ty = threadIdx.
// Index of the first sub-matrix of A processed by the block
int aBegin = wA * BLOCK_SIZE *
// Index of the last sub-matrix of A processed by the block
int aEnd = aBegin + wA -
// Step size used to iterate through the sub-matrices of A
int aStep =
// Index of the first sub-matrix of B processed by the block
int bBegin = BLOCK_SIZE *
// Step size used to iterate through the sub-matrices of B
int bStep = BLOCK_SIZE *
// Csub is used to store the element of the block sub-matrix
// that is computed by the thread
float CSub[2] = { 0,0
// Declaration of the shared memory arrays used to
// store the sub-matrix of A
__shared__ float Aa[BLOCK_SIZE][BLOCK_SIZE];
__shared__ float Ab[BLOCK_SIZE][BLOCK_SIZE];
// Declaration of the shared memory arrays used to
// store the sub-matrix of B
__shared__ float Ba[BLOCK_SIZE][2 * BLOCK_SIZE];
__shared__ float Bb[BLOCK_SIZE][2 * BLOCK_SIZE];
// Initial load
Aa[ty][tx] = A[aBegin + wA * ty + tx];
Ba[ty][tx] = B[bBegin + wB * ty + tx];
Ba[ty][tx + BLOCK_SIZE] = B[bBegin + wB * ty + tx + BLOCK_SIZE];
// Synchronize to make sure that initial matrices are loaded
__syncthreads();
// Loop over all the sub-matrices of A and B
// required to compute the block sub-matrix
for (int a = aBegin, b = bBegin;
a <= aEnd;
a += aStep, b += bStep)
{
if (bx % 2 !=
{
break;
}
//
// Declaration of the
Ab[ty][tx] = Aa[ty][tx];
Bb[ty][tx] = Ba[ty][tx];
Bb[ty][tx + BLOCK_SIZE] = Ba[ty][tx + BLOCK_SIZE];
//
__shared__ float As[BLOCK_SIZE][BLOCK_SIZE];
// Declaration of the shared memory array Bs used to
// store the sub-matrix of B
__shared__ float Bs[BLOCK_SIZE][2 * BLOCK_SIZE];
__syncthreads();
// Load the matrices from device memory
// to shared memory; each thread loads
//
Aa[ty][tx] = A[a + wA * ty +
Bs[ty][tx]
Ba[ty][tx] = B[b + wB * ty +
Bs[ty][tx
Ba[ty][tx + BLOCK_SIZE] = B[b
__syncthreads();
// Multiply the two matrices together;
// each thread computes one element
// of the block sub-matrix
#pragma unroll
for (int k = 0; k < BLOCK_SIZE; ++k)
{
CSub[0] +=
CSub[1] +=
}
// 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 device memory;
// each thread writes one element
int c = wB * BLOCK_SIZE * by + BLOCK_SIZE * bx;
C[c + wB * ty + tx] = CSub[0];
C[c + wB * ty + tx + BLOCK_SIZE] = CSub[1];
}