status_t GetCheckSum(uint64 blockIndex, CheckSum& checkSum)
{
ASSERT(blockIndex < fBlockCount);
MutexLocker locker(fLock);
CheckSumBlock* block;
status_t error = _GetBlock(
fBlockCount + blockIndex / kCheckSumsPerBlock, block);
if (error != B_OK)
return error;
checkSum = block->checkSums[blockIndex % kCheckSumsPerBlock];
return B_OK;
}
status_t SetCheckSum(uint64 blockIndex, const CheckSum& checkSum)
{
ASSERT(blockIndex < fBlockCount);
MutexLocker locker(fLock);
CheckSumBlock* block;
status_t error = _GetBlock(
fBlockCount + blockIndex / kCheckSumsPerBlock, block);
if (error != B_OK)
return error;
block->checkSums[blockIndex % kCheckSumsPerBlock] = checkSum;
block->dirty = true;
return B_OK;
}
void strsm_llnu_lu( unsigned int SPE_id, LS_ShaderInfo *info, Operation_t *myop, Functions_t *funcs )
{
_DBGP_UINT( 1 );
// Make room for python meta data
PyArrayObject *A = (PyArrayObject *)info->LS_shaderMemory;
PyArrayObject *B = (PyArrayObject *)info->LS_shaderMemory+sizeof( PyArrayObject );
PyArrayObject *alpha = (PyArrayObject *)info->LS_shaderMemory+sizeof( PyArrayObject )*2;
PyArrayObject *cb = (PyArrayObject *)info->LS_shaderMemory+sizeof( PyArrayObject )*3;
// Get meta data
_GetPyArrayMetaData( myop->obj[0], A, 0 );
_GetPyArrayMetaData( myop->obj[1], B, 0 );
_GetPyArrayMetaData( myop->scalar[0], alpha, 0 );
_GetPyArrayMetaData( myop->scalar[1], cb, 0 );
_Wait(0); // WAIT!!!
_DBGP_UINT( 2 );
// Make room for data block arrays
char **blocks_A = (char**)( ( (unsigned int)( cb + sizeof( PyArrayObject )) + 127 ) &~ 127 );
char **blocks_B = (char**)( blocks_A + _Pad16bytes( A->numberOfBlocks ) );
char **blocks_alpha = (char**)( blocks_B + _Pad16bytes( B->numberOfBlocks ) );
char **blocks_cb = (char**)( blocks_alpha + _Pad16bytes( alpha->numberOfBlocks ) );
// Local vars
unsigned int idx = 0, nextidx;
unsigned int i, j, k, l, row, col;
// unsigned int c = 0;
//
// unsigned int Afourth = A->blockSize / 4;
// unsigned int xfourth = x->blockSize / 4;
// unsigned int AAfourth = A->blockSize*A->blockSize / 4;
// unsigned int t = 0;
// Uploadet scalars
float alpha_f, beta_f;
unsigned int CB;
_GetScalar( alpha, blocks_alpha, &alpha_f, 1 );
_GetScalar( cb, blocks_cb, &CB, 1 );
_DBGP_UINT( 3 );
vector float alphav = spu_splats(alpha_f);
// Fetch pointers to first __ datablocks for vectors
_GetBlockPointers( B, blocks_B, B->numberOfBlocks, _BTAG );
_GetBlockPointers( A, blocks_A, A->numberOfBlocks, _BTAG );
_Wait(_BTAG);
vector float *BKJ, *BIJ, aik;
float *AIK;
unsigned int x;
int ain, ain2;
_DBGP_UINT( 4000 );
_DBGP_UINT( CB );
// Same A, get it once
_GetBlock( blocks_A, info->LS_blockDataArea[1], CB*A->numberOfBlocksXDim+CB, A->blockSize*A->blockSize, idx );
_Wait(idx);
_DBGP_UINT( CB*B->numberOfBlocksXDim+1+CB+SPE_id );
// Get first block of B
_GetBlock( blocks_B, info->LS_blockDataArea[0], CB*B->numberOfBlocksXDim+1+CB+SPE_id, B->blockSize*B->blockSize, idx );
_Wait(idx);
_DBGP_UINT( 5000 );
//funcs->printuint(CB + SPE_id + 1);
//funcs->printuint(B->numberOfBlocksXDim-myop->num_SPES);
// For each block column
// BLOCK >
for( col = CB + SPE_id + 1 ; col < B->numberOfBlocksXDim-myop->num_SPES ; col += myop->num_SPES )
{
_DBGP_UINT( 100000 + col );
nextidx = idx ^ 1;
//funcs->printuint(CB*A->numberOfBlocksXDim+col+myop->num_SPES);
_GetBlock( blocks_B, info->LS_blockDataArea[nextidx*3], CB*B->numberOfBlocksXDim+col+myop->num_SPES, B->blockSize*B->blockSize, nextidx );
_Wait( idx );
for( l = 1 ; l < B->blockSize ; l++ )
{
_DBGP_UINT( 200000 + l );
BKJ = (vector float*)( info->LS_blockDataArea[idx*3] + (l-1) % B->blockSize * B->blockSize * 4 );
// BIJ[i] = BIJ[i] - BKJ[i] * aik;
_compute( B->blockSize, l, l-1, BKJ, (float*)info->LS_blockDataArea[idx*3], (float*)info->LS_blockDataArea[1] );
//_compute2( B->blockSize, l, l-1, BKJ, (float*)info->LS_blockDataArea[idx*3], (float*)info->LS_blockDataArea[1], funcs );
}
// Return the result to the PPE
_PutAsync( (unsigned int)blocks_B[CB*A->numberOfBlocksXDim+col], B->blockSize * B->blockSize * 4, info->LS_blockDataArea[idx*3], idx );
idx = nextidx;
}
_DBGP_UINT( 100000 + col );
_Wait( idx );
for( l = 1 ; l < B->blockSize ; l++ )
{
BKJ = (vector float*)( info->LS_blockDataArea[idx*3] + (l-1) % B->blockSize * B->blockSize * 4 );
//_compute2( B->blockSize, l, l-1, BKJ, (float*)info->LS_blockDataArea[idx*3], (float*)info->LS_blockDataArea[1], funcs );
_compute( B->blockSize, l, l-1, BKJ, (float*)info->LS_blockDataArea[idx*3], (float*)info->LS_blockDataArea[1] );
}
//funcs->printuint( 678 );
//_PrintMatrix( B, info->LS_blockDataArea[idx*3], funcs );
// Return the result to the PPE
_PutAsync( (unsigned int)blocks_B[CB*A->numberOfBlocksXDim+col], B->blockSize * B->blockSize * 4, info->LS_blockDataArea[idx*3], idx );
}
