MultiBlockManagement3D extend( MultiBlockManagement3D const& management,
Box3D addedBulk, Box3D addedUniqueBulk )
{
std::vector<plint> newIds;
SparseBlockStructure3D resultStructure =
extend( management.getSparseBlockStructure(), addedBulk, addedBulk, newIds );
std::vector<plint> mpiProcesses(newIds.size()), localThreads(newIds.size());
for (pluint iNew=0; iNew<newIds.size(); ++iNew) {
// Default-attribute the newly created blocks to the main process.
mpiProcesses[iNew] = global::mpi().bossId();
localThreads[iNew] = 0;
}
return MultiBlockManagement3D (
resultStructure,
management.getThreadAttribution().extend (
newIds, mpiProcesses, localThreads ),
management.getEnvelopeWidth(),
management.getRefinementLevel() );
}
int SobelFilterImage::runCLKernels()
{
cl_int status;
cl::size_t<3> origin;
origin[0] = 0;
origin[1] = 0;
origin[2] = 0;
cl::size_t<3> region;
region[0] = width;
region[1] = height;
region[2] = 1;
cl::Event writeEvt;
status = commandQueue.enqueueWriteImage(
inputImage2D,
CL_TRUE,
origin,
region,
0,
0,
inputImageData,
NULL,
&writeEvt);
CHECK_OPENCL_ERROR(status,
"CommandQueue::enqueueWriteImage failed. (inputImage2D)");
status = commandQueue.flush();
CHECK_OPENCL_ERROR(status, "cl::CommandQueue.flush failed.");
cl_int eventStatus = CL_QUEUED;
while(eventStatus != CL_COMPLETE)
{
status = writeEvt.getInfo<cl_int>(
CL_EVENT_COMMAND_EXECUTION_STATUS,
&eventStatus);
CHECK_OPENCL_ERROR(status,
"cl:Event.getInfo(CL_EVENT_COMMAND_EXECUTION_STATUS) failed.");
}
// Set appropriate arguments to the kernel
// input buffer image
status = kernel.setArg(0, inputImage2D);
CHECK_OPENCL_ERROR(status, "Kernel::setArg() failed. (inputImageBuffer)");
// outBuffer imager
status = kernel.setArg(1, outputImage2D);
CHECK_OPENCL_ERROR(status, "Kernel::setArg() failed. (outputImageBuffer)");
/*
* Enqueue a kernel run call.
*/
cl::NDRange globalThreads(width, height);
if (blockSizeX > width)
blockSizeX = width;
cl::NDRange localThreads(blockSizeX, blockSizeY);
cl::Event ndrEvt;
status = commandQueue.enqueueNDRangeKernel(
kernel,
cl::NullRange,
globalThreads,
localThreads,
0,
&ndrEvt);
CHECK_OPENCL_ERROR(status, "CommandQueue::enqueueNDRangeKernel() failed.");
status = commandQueue.flush();
CHECK_OPENCL_ERROR(status, "cl::CommandQueue.flush failed.");
eventStatus = CL_QUEUED;
while(eventStatus != CL_COMPLETE)
{
status = ndrEvt.getInfo<cl_int>(
CL_EVENT_COMMAND_EXECUTION_STATUS,
&eventStatus);
CHECK_OPENCL_ERROR(status,
"cl:Event.getInfo(CL_EVENT_COMMAND_EXECUTION_STATUS) failed.");
}
// Enqueue Read Image
origin[0] = 0;
origin[1] = 0;
origin[2] = 0;
region[0] = width;
region[1] = height;
region[2] = 1;
// Enqueue readBuffer
cl::Event readEvt;
status = commandQueue.enqueueReadImage(
outputImage2D,
CL_FALSE,
origin,
region,
0,
0,
outputImageData,
NULL,
&readEvt);
CHECK_OPENCL_ERROR(status, "CommandQueue::enqueueReadImage failed.");
status = commandQueue.flush();
CHECK_OPENCL_ERROR(status, "cl::CommandQueue.flush failed.");
eventStatus = CL_QUEUED;
while(eventStatus != CL_COMPLETE)
{
status = readEvt.getInfo<cl_int>(
CL_EVENT_COMMAND_EXECUTION_STATUS,
&eventStatus);
CHECK_OPENCL_ERROR(status,
"cl:Event.getInfo(CL_EVENT_COMMAND_EXECUTION_STATUS) failed.");
}
return 0;
}
int
GaussianNoise::runCLKernels()
{
cl_int status;
cl_int eventStatus = CL_QUEUED;
cl::Event writeEvt;
status = commandQueue.enqueueWriteBuffer(
inputImageBuffer,
CL_FALSE,
0,
width * height * pixelSize,
inputImageData,
NULL,
&writeEvt);
CHECK_OPENCL_ERROR(status, "CommandQueue::enqueueWriteBuffer() failed. (inputImageBuffer)");
status = commandQueue.flush();
CHECK_OPENCL_ERROR(status, "cl::CommandQueue.flush failed.");
eventStatus = CL_QUEUED;
while(eventStatus != CL_COMPLETE)
{
status = writeEvt.getInfo<cl_int>(
CL_EVENT_COMMAND_EXECUTION_STATUS,
&eventStatus);
CHECK_OPENCL_ERROR(status, "cl:Event.getInfo(CL_EVENT_COMMAND_EXECUTION_STATUS) failed.");
}
// Set appropriate arguments to the kernel
// input buffer image
status = kernel.setArg(0, inputImageBuffer);
CHECK_OPENCL_ERROR(status, "Kernel::setArg() failed. (inputImageBuffer)");
// outBuffer imager
status = kernel.setArg(1, outputImageBuffer);
CHECK_OPENCL_ERROR(status, "Kernel::setArg() failed. (outputImageBuffer)");
// factor
status = kernel.setArg(2, factor);
CHECK_OPENCL_ERROR(status, "Kernel::setArg() failed. (factor)");
/*
* Enqueue a kernel run call.
*/
// Each thread calculates 2 gaussian numbers
cl::NDRange globalThreads(width / 2, height);
cl::NDRange localThreads(blockSizeX, blockSizeY);
cl::Event ndrEvt;
status = commandQueue.enqueueNDRangeKernel(
kernel,
cl::NullRange,
globalThreads,
localThreads,
NULL,
&ndrEvt);
CHECK_OPENCL_ERROR(status, "CommandQueue::enqueueNDRangeKernel() failed.");
status = commandQueue.flush();
CHECK_OPENCL_ERROR(status, "cl::CommandQueue.flush failed.");
eventStatus = CL_QUEUED;
while(eventStatus != CL_COMPLETE)
{
status = ndrEvt.getInfo<cl_int>(
CL_EVENT_COMMAND_EXECUTION_STATUS,
&eventStatus);
CHECK_OPENCL_ERROR(status, "cl:Event.getInfo(CL_EVENT_COMMAND_EXECUTION_STATUS) failed.");
}
// Enqueue readBuffer
cl::Event readEvt;
status = commandQueue.enqueueReadBuffer(
outputImageBuffer,
CL_FALSE,
0,
width * height * pixelSize,
outputImageData,
NULL,
&readEvt);
CHECK_OPENCL_ERROR(status, "CommandQueue::enqueueReadBuffer failed. (outputImageBuffer)");
status = commandQueue.flush();
CHECK_OPENCL_ERROR(status, "cl::CommandQueue.flush failed.");
eventStatus = CL_QUEUED;
while(eventStatus != CL_COMPLETE)
{
status = readEvt.getInfo<cl_int>(
CL_EVENT_COMMAND_EXECUTION_STATUS,
&eventStatus);
CHECK_OPENCL_ERROR(status, "cl:Event.getInfo(CL_EVENT_COMMAND_EXECUTION_STATUS) failed.");
}
return SDK_SUCCESS;
}
int
HDRToneMapping::runCLKernels()
{
cl_int status;
cl_int eventStatus = CL_QUEUED;
if (deviceType.compare("gpu") == 0)
{
cl::Event mapEvt;
void *mapPtr = commandQueue.enqueueMapBuffer(inputImageBuffer,
CL_FALSE,
CL_MAP_WRITE,
0,
width * height * numChannels * sizeof(cl_float),
NULL,
&mapEvt,
&status);
CHECK_OPENCL_ERROR(status, "CommandQueue::enqueueWriteBuffer() failed. (inputImageBuffer)");
status = commandQueue.flush();
CHECK_OPENCL_ERROR(status, "cl::CommandQueue.flush failed.");
eventStatus = CL_QUEUED;
while(eventStatus != CL_COMPLETE)
{
status = mapEvt.getInfo<cl_int>(
CL_EVENT_COMMAND_EXECUTION_STATUS,
&eventStatus);
CHECK_OPENCL_ERROR(status, "cl:Event.getInfo(CL_EVENT_COMMAND_EXECUTION_STATUS) failed.");
}
memcpy(mapPtr,
input,
height * width * numChannels * sizeof(cl_float));
cl::Event unMapEvt;
status = commandQueue.enqueueUnmapMemObject(inputImageBuffer,
mapPtr,
NULL,
&unMapEvt);
CHECK_OPENCL_ERROR(status, "commandQueue.enqueueUnmapObject failed");
status = commandQueue.flush();
CHECK_OPENCL_ERROR(status, "cl::CommandQueue.flush failed.");
eventStatus = CL_QUEUED;
while(eventStatus != CL_COMPLETE)
{
status = unMapEvt.getInfo<cl_int>(
CL_EVENT_COMMAND_EXECUTION_STATUS,
&eventStatus);
CHECK_OPENCL_ERROR(status, "cl:Event.getInfo(CL_EVENT_COMMAND_EXECUTION_STATUS) failed.");
}
}
// input buffer image
status = kernel.setArg(0, inputImageBuffer);
CHECK_OPENCL_ERROR(status, "Kernel::setArg() failed. (inputImageBuffer)");
// outBuffer imager
status = kernel.setArg(1, outputImageBuffer);
CHECK_OPENCL_ERROR(status, "Kernel::setArg() failed. (outputImageBuffer)");
// Average Luminance
status = kernel.setArg(2, averageLuminance);
CHECK_OPENCL_ERROR(status, "Kernel::setArg() failed. (averageLuminance)");
// Gamma
status = kernel.setArg(3, gammaPattanaik);
CHECK_OPENCL_ERROR(status, "Kernel::setArg() failed. (gammaPattanaik)");
// C
status = kernel.setArg(4, cPattanaik);
CHECK_OPENCL_ERROR(status, "Kernel::setArg() failed. (cPattanaik)");
// Delta
status = kernel.setArg(5, deltaPattanaik);
CHECK_OPENCL_ERROR(status, "Kernel::setArg() failed. (deltaPattanaik)");
// Width
status = kernel.setArg(6, width);
CHECK_OPENCL_ERROR(status, "Kernel::setArg() failed. (width)");
// numChannels
status = kernel.setArg(7, numChannels);
CHECK_OPENCL_ERROR(status, "Kernel::setArg() failed. (numChannels)");
// Height
status = kernel.setArg(8, height);
CHECK_OPENCL_ERROR(status, "Kernel::setArg() failed. (height)");
/*
* Enqueue a kernel run call.
*/
cl::NDRange globalThreads(width, height);
cl::NDRange localThreads(blockSizeX, blockSizeY);
cl::Event ndrEvt;
status = commandQueue.enqueueNDRangeKernel(
kernel,
cl::NullRange,
globalThreads,
localThreads,
NULL,
&ndrEvt);
CHECK_OPENCL_ERROR(status, "CommandQueue::enqueueNDRangeKernel() failed.");
status = commandQueue.flush();
CHECK_OPENCL_ERROR(status, "cl::CommandQueue.flush failed.");
eventStatus = CL_QUEUED;
while(eventStatus != CL_COMPLETE)
{
status = ndrEvt.getInfo<cl_int>(
CL_EVENT_COMMAND_EXECUTION_STATUS,
&eventStatus);
CHECK_OPENCL_ERROR(status, "cl:Event.getInfo(CL_EVENT_COMMAND_EXECUTION_STATUS) failed.");
}
if (deviceType.compare("gpu") == 0)
{
// Enqueue readBuffer
cl::Event readEvt;
status = commandQueue.enqueueReadBuffer(
outputImageBuffer,
CL_FALSE,
0,
width * height * numChannels * sizeof(cl_float),
output,
NULL,
&readEvt);
CHECK_OPENCL_ERROR(status, "CommandQueue::enqueueReadBuffer failed. (outputImageBuffer)");
status = commandQueue.flush();
CHECK_OPENCL_ERROR(status, "cl::CommandQueue.flush failed.");
eventStatus = CL_QUEUED;
while(eventStatus != CL_COMPLETE)
{
status = readEvt.getInfo<cl_int>(
CL_EVENT_COMMAND_EXECUTION_STATUS,
&eventStatus);
CHECK_OPENCL_ERROR(status, "cl:Event.getInfo(CL_EVENT_COMMAND_EXECUTION_STATUS) failed.");
}
}
return SDK_SUCCESS;
}
