void
Grid::openPredictionGrid(QString fname)
{
fstream file_ip((fname.toStdString()).c_str(), ios::in);
if (!file_ip) {
cerr << "Can't open input file " << endl;
return;
}
int x_size = 0;
int y_size = 0;
int index = 0;
int splitIndex = 0;
QString qFname = "";
string intBufferStr = "";
string line = "";
string listFname = "";
string vecIndex = "";
cout << "doing file stuff" << endl;
// get the stored size of the vector
getline(file_ip,intBufferStr);
istringstream intBuffer(intBufferStr);
intBuffer >> x_size;
getline(file_ip, intBufferStr);
istringstream intBufferTwo(intBufferStr);
intBufferTwo >> y_size;
cout << "x_size = " << x_size << endl;
cout << "y_size = " << y_size << endl;
cout << "Before reset grid" << endl;
resetGrid();
cout << "After reset grid" << endl;
int i = 1;
while(!file_ip.eof())
{
getline(file_ip,line);
cout << "Line is " << line << endl;
splitIndex = line.find_first_of(',');
vecIndex = line.substr(0,splitIndex);
listFname = line.substr(splitIndex + 1);
qFname = listFname.c_str();
istringstream buffer(vecIndex);
buffer >> index;
cout << "index = " << index << endl;
cout << "filename = " << qFname << endl;
addTrack(index % som_height, index / som_height, qFname);
}
}
void AudioDataOutput::processBuffer(GstPad*, GstBuffer* buffer, gpointer gThat)
{
// TODO emit endOfMedia
AudioDataOutput *that = reinterpret_cast<AudioDataOutput*>(gThat);
// determine the number of channels
GstStructure* structure = gst_caps_get_structure (GST_BUFFER_CAPS(buffer), 0);
gst_structure_get_int (structure, "channels", &that->m_channels);
if (that->m_channels > 2 || that->m_channels < 0) {
qWarning() << Q_FUNC_INFO << ": Number of channels not supported: " << that->m_channels;
return;
}
gint16 *data = reinterpret_cast<gint16*>(GST_BUFFER_DATA(buffer));
guint size = GST_BUFFER_SIZE(buffer) / sizeof(gint16);
that->m_pendingData.reserve(that->m_pendingData.size() + size);
for (uint i=0; i<size; i++) {
// 8 bit? interleaved? yay for lacking documentation!
that->m_pendingData.append(data[i]);
}
while (that->m_pendingData.size() > that->m_dataSize * that->m_channels) {
if (that->m_channels == 1) {
QVector<qint16> intBuffer(that->m_dataSize);
memcpy(intBuffer.data(), that->m_pendingData.constData(), that->m_dataSize * sizeof(qint16));
that->convertAndEmit(intBuffer, intBuffer);
int newSize = that->m_pendingData.size() - that->m_dataSize;
memmove(that->m_pendingData.data(), that->m_pendingData.constData() + that->m_dataSize, newSize * sizeof(qint16));
that->m_pendingData.resize(newSize);
} else {
QVector<qint16> left(that->m_dataSize), right(that->m_dataSize);
for (int i=0; i<that->m_dataSize; i++) {
left[i] = that->m_pendingData[i*2];
right[i] = that->m_pendingData[i*2+1];
}
that->m_pendingData.resize(that->m_pendingData.size() - that->m_dataSize*2);
that->convertAndEmit(left, right);
}
}
}
inline void fillIntTest()
{
TEST_INIT;
b3FillCL* fillCL = new b3FillCL(g_context,g_device,g_queue);
int maxSize=1024*256;
b3OpenCLArray<int> intBuffer(g_context,g_queue,maxSize);
intBuffer.resize(maxSize);
#define NUM_TESTS 7
int dx = maxSize/NUM_TESTS;
for (int iter=0;iter<NUM_TESTS;iter++)
{
int size = b3Min( 11+dx*iter, maxSize );
int value = 2;
int offset=0;
fillCL->execute(intBuffer,value,size,offset);
b3AlignedObjectArray<int> hostBuf2;
hostBuf2.resize(size);
fillCL->executeHost(hostBuf2,value,size,offset);
b3AlignedObjectArray<int> hostBuf;
intBuffer.copyToHost(hostBuf);
for(int i=0; i<size; i++)
{
TEST_ASSERT( hostBuf[i] == hostBuf2[i] );
TEST_ASSERT( hostBuf[i] == hostBuf2[i] );
}
}
delete fillCL;
TEST_REPORT( "fillIntTest" );
}
void octree::sort_dust(tree_structure &tree)
{
if(tree.n_dust == 0) return;
devContext.startTiming(execStream->s());
//Start reduction to get the boundary's of the dust
boundaryReduction.set_arg<int>(0, &tree.n_dust);
boundaryReduction.set_arg<cl_mem>(1, tree.dust_pos.p());
boundaryReduction.set_arg<cl_mem>(2, devMemRMIN.p());
boundaryReduction.set_arg<cl_mem>(3, devMemRMAX.p());
boundaryReduction.setWork(tree.n, NTHREAD_BOUNDARY, NBLOCK_BOUNDARY); //256 threads and 120 blocks in total
boundaryReduction.execute(execStream->s());
devMemRMIN.d2h(); //Need to be defined and initialized somewhere outside this function
devMemRMAX.d2h(); //Need to be defined and initialized somewhere outside this function
real4 r_min = make_real4(+1e10, +1e10, +1e10, +1e10);
real4 r_max = make_real4(-1e10, -1e10, -1e10, -1e10);
//Reduce the blocks, done on host since its
//A faster and B we need the results anyway
for (int i = 0; i < 120; i++) {
r_min.x = std::min(r_min.x, devMemRMIN[i].x);
r_min.y = std::min(r_min.y, devMemRMIN[i].y);
r_min.z = std::min(r_min.z, devMemRMIN[i].z);
r_max.x = std::max(r_max.x, devMemRMAX[i].x);
r_max.y = std::max(r_max.y, devMemRMAX[i].y);
r_max.z = std::max(r_max.z, devMemRMAX[i].z);
}
LOG("Found dust boundarys, number of particles %d : \n", tree.n_dust);
LOG("min: %f\t%f\t%f\tmax: %f\t%f\t%f \n", r_min.x,r_min.y,r_min.z,r_max.x,r_max.y,r_max.z);
//Compute the boundarys of the dust, needed to get the PH key
real size = 1.001f*std::max(r_max.z - r_min.z,
std::max(r_max.y - r_min.y, r_max.x - r_min.x));
float4 corner = make_real4(0.5f*(r_min.x + r_max.x) - 0.5f*size,
0.5f*(r_min.y + r_max.y) - 0.5f*size,
0.5f*(r_min.z + r_max.z) - 0.5f*size, size);
float domain_fac = size/(1 << MAXLEVELS);
corner.w = domain_fac;
//Compute the keys
my_dev::dev_mem<uint4> srcValues(devContext);
//The generalBuffer1 has size uint*4*N*3
//this buffer gets part: 0-uint*4*N
srcValues.cmalloc_copy(tree.generalBuffer1, tree.n_dust, 0);
//Compute the keys directly into srcValues
build_key_list.set_arg<cl_mem>(0, srcValues.p());
build_key_list.set_arg<cl_mem>(1, tree.dust_pos.p());
build_key_list.set_arg<int>(2, &tree.n_dust);
build_key_list.set_arg<real4>(3, &corner);
build_key_list.setWork(tree.n_dust, 128); //128 threads per block
build_key_list.execute(execStream->s());
// If srcValues and buffer are different, then the original values
// are preserved, if they are the same srcValues will be overwritten
gpuSort(devContext, srcValues, tree.dust_key,srcValues, tree.n_dust, 32, 3, tree);
//Sort the relevant properties.Note we can optimize this
//further as done with the normal particles
my_dev::dev_mem<real4> real4Buffer1(devContext);
my_dev::dev_mem<real4> real4Buffer2(devContext);
my_dev::dev_mem<real4> real4Buffer3(devContext);
int memOffset1 = real4Buffer1.cmalloc_copy(tree.generalBuffer1,tree.n_dust, 0);
memOffset1 = real4Buffer2.cmalloc_copy(tree.generalBuffer1,tree.n_dust, memOffset1);
memOffset1 = real4Buffer3.cmalloc_copy(tree.generalBuffer1,tree.n_dust, memOffset1);
dataReorderCombined.set_arg<int>(0, &tree.n_dust);
dataReorderCombined.set_arg<cl_mem>(1, tree.dust_key.p());
dataReorderCombined.setWork(tree.n_dust, 512);
//Position, velocity and acc0
dataReorderCombined.set_arg<cl_mem>(2, tree.dust_pos.p());
dataReorderCombined.set_arg<cl_mem>(3, real4Buffer1.p());
dataReorderCombined.set_arg<cl_mem>(4, tree.dust_vel.p());
dataReorderCombined.set_arg<cl_mem>(5, real4Buffer2.p());
dataReorderCombined.set_arg<cl_mem>(6, tree.dust_acc0.p());
dataReorderCombined.set_arg<cl_mem>(7, real4Buffer3.p());
dataReorderCombined.execute(execStream->s());
tree.dust_pos.copy(real4Buffer1, tree.n_dust);
tree.dust_vel.copy(real4Buffer2, tree.n_dust);
tree.dust_acc0.copy(real4Buffer3, tree.n_dust);
my_dev::dev_mem<int> intBuffer(devContext);
my_dev::dev_mem<float2> float2Buffer(devContext);
my_dev::dev_mem<int> sortPermutation(devContext);
memOffset1 = float2Buffer.cmalloc_copy (tree.generalBuffer1,tree.n_dust, 0);
memOffset1 = sortPermutation.cmalloc_copy(tree.generalBuffer1,tree.n_dust, memOffset1);
memOffset1 = intBuffer.cmalloc_copy (tree.generalBuffer1,tree.n_dust, memOffset1);
dataReorderF2.set_arg<int>(0, &tree.n_dust);
dataReorderF2.set_arg<cl_mem>(1, tree.dust_key.p());
dataReorderF2.set_arg<cl_mem>(2, float2Buffer.p()); //Place holder, dust has no time
dataReorderF2.set_arg<cl_mem>(3, float2Buffer.p()); //Reuse as destination1
dataReorderF2.set_arg<cl_mem>(4, tree.dust_ids.p());
dataReorderF2.set_arg<cl_mem>(5, sortPermutation.p()); //Reuse as destination2
dataReorderF2.setWork(tree.n_dust, 512);
dataReorderF2.execute(execStream->s());
tree.dust_ids.copy(sortPermutation, sortPermutation.get_size());
devContext.stopTiming("DustSortReorder", -1, execStream->s());
}
