void operator()(const blocked_range3d<int>& r) const{
int begin1 = r.rows().begin(), end1 = r.rows().end();
int begin2 = r.cols().begin(), end2 = r.cols().end();
int begin3 = r.pages().begin(), end3 = r.pages().end();
for (int i = begin1; i < end1; i++){
for (int j = begin2; j < end2; j++){
/*
for(int k = begin3; k < end3; k++){
dst[i * src2w + j] += A[k] * B[k];
}
*/
/*
const double *vec1, *vec2;
vec1 = &(A[i * src1w]);
vec2 = &(B[j * src1w]);
ScalarMultiplicator mul(vec1, vec2);
parallel_reduce(r.pages(), mul);
dst[i * src2w + j] = mul.Result();
*/
/*
const double *vec1, *vec2;
vec1 = &(A[i * src1w]);
vec2 = &(B[j * src1w]);
dst[i * src2w + j] = TBBSclMlt(vec1, vec2, src1w);
*/
const double *vec1, *vec2;
vec1 = &(A[i * src1w + begin3]);
vec2 = &(B[j * src1w + begin3]);
dst[i * src2w + j] += TBBSclMlt(vec1, vec2, end3 - begin3);
}
}
}
void operator()( const blocked_range3d<size_t>& r ) const
{
ContextPtr context = new Context( *m_parentContext, Context::Borrowed );
Context::Scope scope( context.get() );
const Box2i operationWindow( V2i( r.rows().begin()+m_dataWindow.min.x, r.cols().begin()+m_dataWindow.min.y ), V2i( r.rows().end()+m_dataWindow.min.x-1, r.cols().end()+m_dataWindow.min.y-1 ) );
V2i minTileOrigin = ImagePlug::tileOrigin( operationWindow.min );
V2i maxTileOrigin = ImagePlug::tileOrigin( operationWindow.max );
size_t imageStride = m_dataWindow.size().x + 1;
for( size_t channelIndex = r.pages().begin(); channelIndex < r.pages().end(); ++channelIndex )
{
context->set( ImagePlug::channelNameContextName, m_channelNames[channelIndex] );
float *channelBegin = m_imageChannelData[channelIndex];
for( int tileOriginY = minTileOrigin.y; tileOriginY <= maxTileOrigin.y; tileOriginY += m_tileSize )
{
for( int tileOriginX = minTileOrigin.x; tileOriginX <= maxTileOrigin.x; tileOriginX += m_tileSize )
{
context->set( ImagePlug::tileOriginContextName, V2i( tileOriginX, tileOriginY ) );
Box2i tileBound( V2i( tileOriginX, tileOriginY ), V2i( tileOriginX + m_tileSize - 1, tileOriginY + m_tileSize - 1 ) );
Box2i b = boxIntersection( tileBound, operationWindow );
size_t tileStrideSize = sizeof(float) * ( b.size().x + 1 );
ConstFloatVectorDataPtr tileData = m_channelDataPlug->getValue();
const float *tileDataBegin = &(tileData->readable()[0]);
for( int y = b.min.y; y<=b.max.y; y++ )
{
const float *tilePtr = tileDataBegin + (y - tileOriginY) * m_tileSize + (b.min.x - tileOriginX);
float *channelPtr = channelBegin + ( m_dataWindow.size().y - ( y - m_dataWindow.min.y ) ) * imageStride + (b.min.x - m_dataWindow.min.x);
std::memcpy( channelPtr, tilePtr, tileStrideSize );
}
}
}
}
}
