int
PrefixSum::setupCL(void)
{
cl_int status = 0;
cl_device_type dType;
if(deviceType.compare("cpu") == 0)
{
dType = CL_DEVICE_TYPE_CPU;
}
else //deviceType = "gpu"
{
dType = CL_DEVICE_TYPE_GPU;
if(isThereGPU() == false)
{
std::cout << "GPU not found. Falling back to CPU device" << std::endl;
dType = CL_DEVICE_TYPE_CPU;
}
}
/*
* Have a look at the available platforms and pick either
* the AMD one if available or a reasonable default.
*/
cl_platform_id platform = NULL;
int retValue = sampleCommon->getPlatform(platform, platformId, isPlatformEnabled());
CHECK_ERROR(retValue, SDK_SUCCESS, "sampleCommon->getPlatform() failed");
// Display available devices.
retValue = sampleCommon->displayDevices(platform, dType);
CHECK_ERROR(retValue, SDK_SUCCESS, "sampleCommon::displayDevices() failed");
/*
* If we could find our platform, use it. Otherwise use just available platform.
*/
cl_context_properties cps[3] =
{
CL_CONTEXT_PLATFORM,
(cl_context_properties)platform,
0
};
context = clCreateContextFromType(
cps,
dType,
NULL,
NULL,
&status);
CHECK_OPENCL_ERROR(status, "clCreateContextFromType failed.");
status = sampleCommon->getDevices(context, &devices, deviceId, isDeviceIdEnabled());
CHECK_ERROR(status, SDK_SUCCESS, "sampleCommon::getDevices() failed");
//Set device info of given cl_device_id
status = deviceInfo.setDeviceInfo(devices[deviceId]);
CHECK_ERROR(status, SDK_SUCCESS, "SDKDeviceInfo::setDeviceInfo() failed");
{
// The block is to move the declaration of prop closer to its use
cl_command_queue_properties prop = 0;
commandQueue = clCreateCommandQueue(
context,
devices[deviceId],
prop,
&status);
CHECK_OPENCL_ERROR(status, "clCreateCommandQueue failed.");
}
// Set Presistent memory only for AMD platform
cl_mem_flags inMemFlags = CL_MEM_READ_ONLY;
if(isAmdPlatform())
inMemFlags |= CL_MEM_USE_PERSISTENT_MEM_AMD;
inputBuffer = clCreateBuffer(
context,
inMemFlags,
sizeof(cl_float) * length,
NULL,
&status);
CHECK_OPENCL_ERROR(status, "clCreateBuffer failed. (inputBuffer)");
outputBuffer = clCreateBuffer(
context,
CL_MEM_WRITE_ONLY | CL_MEM_ALLOC_HOST_PTR,
sizeof(cl_float) * length,
NULL,
&status);
CHECK_OPENCL_ERROR(status, "clCreateBuffer failed. (outputBuffer)");
// create a CL program using the kernel source
streamsdk::buildProgramData buildData;
buildData.kernelName = std::string("PrefixSum_Kernels.cl");
buildData.devices = devices;
buildData.deviceId = deviceId;
buildData.flagsStr = std::string("");
if(isLoadBinaryEnabled())
buildData.binaryName = std::string(loadBinary.c_str());
if(isComplierFlagsSpecified())
buildData.flagsFileName = std::string(flags.c_str());
retValue = sampleCommon->buildOpenCLProgram(program, context, buildData);
CHECK_ERROR(retValue, SDK_SUCCESS, "sampleCommon::buildOpenCLProgram() failed");
// get a kernel object handle for a kernel with the given name
kernel = clCreateKernel(program, "prefixSum", &status);
CHECK_OPENCL_ERROR(status, "clCreateKernel failed.");
return SDK_SUCCESS;
}
int
DwtHaar1D::setupCL(void)
{
cl_int status = 0;
cl_device_type dType;
if(deviceType.compare("cpu") == 0)
{
dType = CL_DEVICE_TYPE_CPU;
}
else //deviceType = "gpu"
{
dType = CL_DEVICE_TYPE_GPU;
if(isThereGPU() == false)
{
std::cout << "GPU not found. Falling back to CPU device" << std::endl;
dType = CL_DEVICE_TYPE_CPU;
}
}
/*
* Have a look at the available platforms and pick either
* the AMD one if available or a reasonable default.
*/
cl_platform_id platform = NULL;
int retValue = sampleCommon->getPlatform(platform, platformId, isPlatformEnabled());
CHECK_ERROR(retValue, SDK_SUCCESS, "sampleCommon::getPlatform() failed");
// Display available devices.
retValue = sampleCommon->displayDevices(platform, dType);
CHECK_ERROR(retValue, SDK_SUCCESS, "sampleCommon::displayDevices() failed");
// If we could find our platform, use it. Otherwise use just available platform.
cl_context_properties cps[3] =
{
CL_CONTEXT_PLATFORM,
(cl_context_properties)platform,
0
};
context = clCreateContextFromType(cps,
dType,
NULL,
NULL,
&status);
CHECK_OPENCL_ERROR(status, "clCreateContextFromType failed.");
// getting device on which to run the sample
status = sampleCommon->getDevices(context, &devices, deviceId, isDeviceIdEnabled());
CHECK_ERROR(status, SDK_SUCCESS, "sampleCommon::getDevices() failed");
commandQueue = clCreateCommandQueue(context,
devices[deviceId],
0,
&status);
CHECK_OPENCL_ERROR(status, "clCreateCommandQueue failed.");
//Set device info of given cl_device_id
retValue = deviceInfo.setDeviceInfo(devices[deviceId]);
CHECK_ERROR(retValue, 0, "SDKDeviceInfo::setDeviceInfo() failed");
// Set Presistent memory only for AMD platform
cl_mem_flags inMemFlags = CL_MEM_READ_ONLY;
if(isAmdPlatform())
inMemFlags |= CL_MEM_USE_PERSISTENT_MEM_AMD;
inDataBuf = clCreateBuffer(context,
inMemFlags,
sizeof(cl_float) * signalLength,
NULL,
&status);
CHECK_OPENCL_ERROR(status, "clCreateBuffer failed. (inDataBuf)");
dOutDataBuf = clCreateBuffer(context,
CL_MEM_WRITE_ONLY,
signalLength * sizeof(cl_float),
NULL,
&status);
CHECK_OPENCL_ERROR(status, "clCreateBuffer failed. (dOutDataBuf)");
dPartialOutDataBuf = clCreateBuffer(context,
CL_MEM_WRITE_ONLY,
signalLength * sizeof(cl_float),
NULL,
&status);
CHECK_OPENCL_ERROR(status, "clCreateBuffer failed. (dPartialOutDataBuf)");
// create a CL program using the kernel source
streamsdk::buildProgramData buildData;
buildData.kernelName = std::string("DwtHaar1DCPPKernel_Kernels.cl");
buildData.devices = devices;
buildData.deviceId = deviceId;
buildData.flagsStr = std::string("-x clc++ ");
if(isLoadBinaryEnabled())
buildData.binaryName = std::string(loadBinary.c_str());
if(isComplierFlagsSpecified())
buildData.flagsFileName = std::string(flags.c_str());
retValue = sampleCommon->buildOpenCLProgram(program, context, buildData);
CHECK_ERROR(retValue, 0, "sampleCommon::buildOpenCLProgram() failed");
// get a kernel object handle for a kernel with the given name
kernel = clCreateKernel(program, "dwtHaar1D", &status);
CHECK_OPENCL_ERROR(status, "clCreateKernel failed.");
status = kernelInfo.setKernelWorkGroupInfo(kernel,devices[deviceId]);
CHECK_ERROR(status, SDK_SUCCESS, " setKernelWorkGroupInfo() failed");
return SDK_SUCCESS;
}
int
EigenValue::setupCL(void)
{
cl_int status = 0;
cl_device_type dType;
if(deviceType.compare("cpu") == 0)
{
dType = CL_DEVICE_TYPE_CPU;
}
else //deviceType = "gpu"
{
dType = CL_DEVICE_TYPE_GPU;
if(isThereGPU() == false)
{
std::cout << "GPU not found. Falling back to CPU device" << std::endl;
dType = CL_DEVICE_TYPE_CPU;
}
}
/*
* Have a look at the available platforms and pick either
* the AMD one if available or a reasonable default.
*/
cl_platform_id platform = NULL;
int retValue = sampleCommon->getPlatform(platform, platformId, isPlatformEnabled());
CHECK_ERROR(retValue, SDK_SUCCESS, "sampleCommon::getPlatform() failed");
// Display available devices.
retValue = sampleCommon->displayDevices(platform, dType);
CHECK_ERROR(retValue, SDK_SUCCESS, "sampleCommon::displayDevices() failed");
// If we could find our platform, use it. Otherwise use just available platform.
cl_context_properties cps[3] =
{
CL_CONTEXT_PLATFORM,
(cl_context_properties)platform,
0
};
context = clCreateContextFromType(
cps,
dType,
NULL,
NULL,
&status);
CHECK_OPENCL_ERROR(status, "clCreateContextFromType failed.");
// getting device on which to run the sample
status = sampleCommon->getDevices(context, &devices, deviceId, isDeviceIdEnabled());
CHECK_ERROR(status, 0, "sampleCommon::getDevices() failed");
{
// The block is to move the declaration of prop closer to its use
cl_command_queue_properties prop = 0;
commandQueue = clCreateCommandQueue(
context,
devices[deviceId],
prop,
&status);
CHECK_OPENCL_ERROR(status, "clCreateCommandQueue failed.");
}
// Set Presistent memory only for AMD platform
cl_mem_flags inMemFlags = CL_MEM_READ_ONLY;
if(isAmdPlatform())
inMemFlags |= CL_MEM_USE_PERSISTENT_MEM_AMD;
// cl mem to store the diagonal elements of the matrix
diagonalBuffer = clCreateBuffer(
context,
inMemFlags,
sizeof(cl_float) * length,
NULL,
&status);
CHECK_OPENCL_ERROR(status, "clCreateBuffer failed. (diagonalBuffer)");
// cl mem to store the number of eigenvalues in each interval
numEigenValuesIntervalBuffer = clCreateBuffer(
context,
CL_MEM_READ_WRITE | CL_MEM_ALLOC_HOST_PTR,
sizeof(cl_uint) * length,
NULL,
&status);
CHECK_OPENCL_ERROR(status, "clCreateBuffer failed. (diagonalBuffer)");
// cl mem to store the offDiagonal elements of the matrix
offDiagonalBuffer = clCreateBuffer(
context,
inMemFlags,
sizeof(cl_float) * (length-1),
NULL,
&status);
CHECK_OPENCL_ERROR(status, "clCreateBuffer failed. (offDiagonalBuffer)");
// cl mem to store the eigenvalue intervals
for(int i = 0 ; i < 2 ; ++ i)
{
eigenIntervalBuffer[i] = clCreateBuffer(
context,
CL_MEM_READ_WRITE | CL_MEM_ALLOC_HOST_PTR,
sizeof(cl_uint) * length * 2,
NULL,
&status);
CHECK_OPENCL_ERROR(status, "clCreateBuffer failed. (eigenIntervalBuffer)");
}
// create a CL program using the kernel source
streamsdk::buildProgramData buildData;
buildData.kernelName = std::string("EigenValue_Kernels.cl");
buildData.devices = devices;
buildData.deviceId = deviceId;
buildData.flagsStr = std::string("-x clc++");
if(isLoadBinaryEnabled())
buildData.binaryName = std::string(loadBinary.c_str());
if(isComplierFlagsSpecified())
buildData.flagsFileName = std::string(flags.c_str());
retValue = sampleCommon->buildOpenCLProgram(program, context, buildData);
CHECK_ERROR(retValue, 0, "sampleCommon::buildOpenCLProgram() failed");
// get a kernel object handle for a kernel with the given name
kernel[0] = clCreateKernel(program, "calNumEigenValueInterval", &status);
if(sampleCommon->checkVal(
status,
CL_SUCCESS,
"clCreateKernel failed."))
return SDK_FAILURE;
// get a kernel object handle for a kernel with the given name
kernel[1] = clCreateKernel(program, "recalculateEigenIntervals", &status);
if(sampleCommon->checkVal(
status,
CL_SUCCESS,
"clCreateKernel failed."))
return SDK_FAILURE;
return SDK_SUCCESS;
}
int
FluidSimulation2D::setupCL()
{
cl_int status = CL_SUCCESS;
cl_device_type dType;
if(deviceType.compare("cpu") == 0)
{
dType = CL_DEVICE_TYPE_CPU;
}
else //deviceType = "gpu"
{
dType = CL_DEVICE_TYPE_GPU;
if(isThereGPU() == false)
{
std::cout << "GPU not found. Falling back to CPU device" << std::endl;
dType = CL_DEVICE_TYPE_CPU;
}
}
/*
* Have a look at the available platforms and pick either
* the AMD one if available or a reasonable default.
*/
cl_platform_id platform = NULL;
int retValue = sampleCommon->getPlatform(platform, platformId, isPlatformEnabled());
CHECK_ERROR(retValue, SDK_SUCCESS, "sampleCommon::getPlatform() failed");
// Display available devices.
retValue = sampleCommon->displayDevices(platform, dType);
CHECK_ERROR(retValue, SDK_SUCCESS, "sampleCommon::displayDevices() failed");
// If we could find our platform, use it. Otherwise use just available platform.
cl_context_properties cps[3] =
{
CL_CONTEXT_PLATFORM,
(cl_context_properties)platform,
0
};
context = clCreateContextFromType(
cps,
dType,
NULL,
NULL,
&status);
CHECK_OPENCL_ERROR( status, "clCreateContextFromType failed.");
// getting device on which to run the sample
status = sampleCommon->getDevices(context, &devices, deviceId, isDeviceIdEnabled());
CHECK_ERROR(status, SDK_SUCCESS, "sampleCommon::getDevices() failed");
{
// The block is to move the declaration of prop closer to its use
cl_command_queue_properties prop = 0;
commandQueue = clCreateCommandQueue(
context,
devices[deviceId],
prop,
&status);
CHECK_OPENCL_ERROR( status, "clCreateCommandQueue failed.");
}
//Set device info of given cl_device_id
retValue = deviceInfo.setDeviceInfo(devices[deviceId]);
CHECK_ERROR(retValue, 0, "SDKDeviceInfo::setDeviceInfo() failed");
std::string buildOptions = std::string("");
// Check if cl_khr_fp64 extension is supported
if(strstr(deviceInfo.extensions, "cl_khr_fp64"))
{
buildOptions.append("-D KHR_DP_EXTENSION");
}
else
{
// Check if cl_amd_fp64 extension is supported
if(!strstr(deviceInfo.extensions, "cl_amd_fp64"))
{
reqdExtSupport = false;
OPENCL_EXPECTED_ERROR("Device does not support cl_amd_fp64 extension!");
}
}
/*
* Create and initialize memory objects
*/
size_t temp = dims[0] * dims[1];
d_if0 = clCreateBuffer(context,
CL_MEM_READ_WRITE,
sizeof(cl_double) * temp,
0,
&status);
CHECK_OPENCL_ERROR(status, "clCreateBuffer failed. (d_if0)");
status = clEnqueueWriteBuffer(commandQueue,
d_if0,
1,
0,
sizeof(cl_double) * temp,
h_if0,
0, 0, 0);
CHECK_OPENCL_ERROR(status, "clEnqueueWriteBuffer failed. (d_if0)");
d_if1234 = clCreateBuffer(context,
CL_MEM_READ_WRITE,
sizeof(cl_double4) * temp,
0,
&status);
CHECK_OPENCL_ERROR(status, "clCreateBuffer failed. (d_if1234)");
status = clEnqueueWriteBuffer(commandQueue,
d_if1234,
1,
0,
sizeof(cl_double4) * temp,
h_if1234,
0, 0, 0);
CHECK_OPENCL_ERROR(status, "clEnqueueWriteBuffer failed. (d_if1234)");
d_if5678 = clCreateBuffer(context,
CL_MEM_READ_WRITE,
sizeof(cl_double4) * temp,
0,
&status);
CHECK_OPENCL_ERROR(status, "clCreateBuffer failed. (d_if5678)");
status = clEnqueueWriteBuffer(commandQueue,
d_if5678,
1,
0,
sizeof(cl_double4) * temp,
h_if5678,
0, 0, 0);
CHECK_OPENCL_ERROR(status, "clEnqueueWriteBuffer failed. (d_if5678)");
d_of0 = clCreateBuffer(context,
CL_MEM_READ_WRITE,
sizeof(cl_double) * temp,
0,
&status);
CHECK_OPENCL_ERROR(status, "clCreateBuffer failed. (d_of0)");
d_of1234 = clCreateBuffer(context,
CL_MEM_READ_WRITE,
sizeof(cl_double4) * temp,
0,
&status);
CHECK_OPENCL_ERROR(status, "clCreateBuffer failed. (d_of1234)");
d_of5678 = clCreateBuffer(context,
CL_MEM_READ_WRITE,
sizeof(cl_double4) * temp,
0,
&status);
CHECK_OPENCL_ERROR(status, "clCreateBuffer failed. (d_of5678)");
status = clEnqueueCopyBuffer(commandQueue,
d_if0,
d_of0,
0, 0, sizeof(cl_double) * temp,
0, 0, 0);
CHECK_OPENCL_ERROR(status, "clEnqueueCopyBuffer failed. (d_if0->d_of0)");
status = clEnqueueCopyBuffer(commandQueue,
d_if1234,
d_of1234,
0, 0, sizeof(cl_double4) * temp,
0, 0, 0);
CHECK_OPENCL_ERROR(status, "clEnqueueCopyBuffer failed. (d_if1234->d_of1234)");
status = clEnqueueCopyBuffer(commandQueue,
d_if5678,
d_of5678,
0, 0, sizeof(cl_double4) * temp,
0, 0, 0);
CHECK_OPENCL_ERROR(status, "clEnqueueCopyBuffer failed. (d_if5678->d_of5678)");
status = clFinish(commandQueue);
CHECK_OPENCL_ERROR(status, "clFinish failed.");
// Set Presistent memory only for AMD platform
cl_mem_flags inMemFlags = CL_MEM_READ_ONLY;
if(isAmdPlatform())
inMemFlags |= CL_MEM_USE_PERSISTENT_MEM_AMD;
//Constant arrays
type = clCreateBuffer(context,
inMemFlags,
sizeof(cl_bool) * temp,
0,
&status);
CHECK_OPENCL_ERROR(status, "clCreateBuffer failed. (type)");
weight = clCreateBuffer(context,
CL_MEM_READ_ONLY,
sizeof(cl_double) * 9,
0,
&status);
CHECK_OPENCL_ERROR(status, "clCreateBuffer failed. (weight)");
status = clEnqueueWriteBuffer(commandQueue,
weight,
1, 0, sizeof(cl_double) * 9,
w,
0, 0, 0);
CHECK_OPENCL_ERROR(status, "clEnqueueWriteBuffer failed. (weight)");
velocity = clCreateBuffer(context,
CL_MEM_WRITE_ONLY | CL_MEM_ALLOC_HOST_PTR,
sizeof(cl_double2) * temp,
0, &status);
CHECK_OPENCL_ERROR(status, "clCreateBuffer failed. (velocity)");
// create a CL program using the kernel source
streamsdk::buildProgramData buildData;
buildData.kernelName = std::string("FluidSimulation2D_Kernels.cl");
buildData.devices = devices;
buildData.deviceId = deviceId;
buildData.flagsStr = std::string("");
if(isLoadBinaryEnabled())
buildData.binaryName = std::string(loadBinary.c_str());
if(isComplierFlagsSpecified())
buildData.flagsFileName = std::string(flags.c_str());
retValue = sampleCommon->buildOpenCLProgram(program, context, buildData);
CHECK_ERROR(retValue, 0, "sampleCommon::buildOpenCLProgram() failed");
// get a kernel object handle for a kernel with the given name
kernel = clCreateKernel(
program,
"lbm",
&status);
CHECK_OPENCL_ERROR(status, "clCreateKernel failed.");
return SDK_SUCCESS;
}
int
Lucas::setupCL (void)
{
cl_int status = 0;
cl_device_type dType;
if (deviceType.compare ("cpu") == 0)
dType = CL_DEVICE_TYPE_CPU;
else //deviceType = "gpu"
{
dType = CL_DEVICE_TYPE_GPU;
if (isThereGPU () == false)
{
std::cout << "GPU not found. Falling back to CPU device" << std::
endl;
dType = CL_DEVICE_TYPE_CPU;
}
}
/*
* Have a look at the available platforms and pick either
* the AMD one if available or a reasonable default.
*/
status = cl::Platform::get (&platforms);
CHECK_OPENCL_ERROR (status, "Platform::get() failed.");
std::vector < cl::Platform >::iterator i;
if (platforms.size () > 0)
{
if (isPlatformEnabled ())
{
i = platforms.begin () + platformId;
}
else
{
for (i = platforms.begin (); i != platforms.end (); ++i)
{
if (!strcmp ((*i).getInfo < CL_PLATFORM_VENDOR > ().c_str (),
"Advanced Micro Devices, Inc."))
{
break;
}
}
}
}
cl_context_properties cps[3] = {
CL_CONTEXT_PLATFORM,
(cl_context_properties) (*i) (),
0
};
if (NULL == (*i) ())
{
sampleCommon->error ("NULL platform found so Exiting Application.");
return SDK_FAILURE;
}
context = cl::Context (dType, cps, NULL, NULL, &status);
CHECK_OPENCL_ERROR (status, "Context::Context() failed.");
devices = context.getInfo < CL_CONTEXT_DEVICES > ();
CHECK_OPENCL_ERROR (status, "Context::getInfo() failed.");
std::cout << "Platform :" << (*i).getInfo < CL_PLATFORM_VENDOR >
().c_str () << "\n";
int deviceCount = (int) devices.size ();
int j = 0;
for (std::vector < cl::Device >::iterator i = devices.begin ();
i != devices.end (); ++i, ++j)
{
std::cout << "Device " << j << " : ";
std::string deviceName = (*i).getInfo < CL_DEVICE_NAME > ();
std::cout << deviceName.c_str () << "\n";
}
std::cout << "\n";
if (deviceCount == 0)
{
std::cerr << "No device available\n";
return SDK_FAILURE;
}
if (sampleCommon->validateDeviceId (deviceId, deviceCount))
{
sampleCommon->error ("sampleCommon::validateDeviceId() failed");
return SDK_FAILURE;
}
std::string extensions =
devices[deviceId].getInfo < CL_DEVICE_EXTENSIONS > ();
std::string buildOptions = std::string ("");
// Check if cl_khr_fp64 extension is supported
if (strstr (extensions.c_str (), "cl_khr_fp64"))
{
buildOptions.append ("-D KHR_DP_EXTENSION");
}
else
{
// Check if cl_amd_fp64 extension is supported
if (!strstr (extensions.c_str (), "cl_amd_fp64"))
{
OPENCL_EXPECTED_ERROR
("Device does not support cl_amd_fp64 extension!");
}
}
cl_uint localMemType;
// Get device specific information
status = devices[deviceId].getInfo<cl_uint>(
CL_DEVICE_LOCAL_MEM_TYPE,
&localMemType);
CHECK_OPENCL_ERROR(status, "Device::getInfo CL_DEVICE_LOCAL_MEM_TYPE) failed.");
// If scratchpad is available then update the flag
if(localMemType != CL_LOCAL)
OPENCL_EXPECTED_ERROR ("Device does not support local memory.");
// Get Device specific Information
status = devices[deviceId].getInfo<size_t>(
CL_DEVICE_MAX_WORK_GROUP_SIZE,
&maxWorkGroupSize);
CHECK_OPENCL_ERROR(status, "Device::getInfo(CL_DEVICE_MAX_WORK_GROUP_SIZE) failed.");
if(threads > maxWorkGroupSize)
OPENCL_EXPECTED_ERROR ("Device does not support threads.");
status = devices[deviceId].getInfo<cl_uint>(
CL_DEVICE_MAX_WORK_ITEM_DIMENSIONS,
&maxDimensions);
CHECK_OPENCL_ERROR(status, "Device::getInfo(CL_DEVICE_MAX_WORK_ITEM_DIMENSIONS) failed.");
maxWorkItemSizes = (size_t*)malloc(maxDimensions * sizeof(size_t));
std::vector<size_t> workItems = devices[deviceId].getInfo<CL_DEVICE_MAX_WORK_ITEM_SIZES>();
for(cl_uint i = 0; i < maxDimensions; ++i)
maxWorkItemSizes[i] = workItems[i];
status = devices[deviceId].getInfo<cl_ulong>(
CL_DEVICE_LOCAL_MEM_SIZE,
&totalLocalMemory);
CHECK_OPENCL_ERROR(status, "Device::getInfo(CL_DEVICE_LOCAL_MEM_SIZES) failed.");
// Set command queue properties
cl_command_queue_properties prop = 0;
if (!eAppGFLOPS)
prop |= CL_QUEUE_PROFILING_ENABLE;
commandQueue = cl::CommandQueue (context, devices[deviceId], prop, &status);
CHECK_OPENCL_ERROR (status, "CommandQueue::CommandQueue() failed.");
// Set Presistent memory only for AMD platform
cl_mem_flags inMemFlags = CL_MEM_READ_ONLY;
if (isAmdPlatform ())
inMemFlags |= CL_MEM_USE_PERSISTENT_MEM_AMD;
device.push_back (devices[deviceId]);
// create a CL program using the kernel source
streamsdk::SDKFile kernelFile;
std::string kernelPath = sampleCommon->getPath ();
kernelPath.append ("Kernels.cl");
if (!kernelFile.open (kernelPath.c_str ()))
{
std::cout << "Failed to load kernel file : " << kernelPath <<
std::endl;
return SDK_FAILURE;
}
cl::Program::Sources programSource (1,
std::make_pair (kernelFile.
source ().data (),
kernelFile.
source ().size ()));
program = cl::Program (context, programSource, &status);
CHECK_OPENCL_ERROR (status, "Program::Program(Source) failed.");
std::string flagsStr = std::string ("");
status = program.build (device, flagsStr.c_str ());
if (status != CL_SUCCESS)
{
if (status == CL_BUILD_PROGRAM_FAILURE)
{
std::string str =
program.getBuildInfo < CL_PROGRAM_BUILD_LOG > (devices[deviceId]);
std::cout << " \n\t\t\tBUILD LOG\n";
std::cout << " ************************************************\n";
std::cout << str << std::endl;
std::cout << " ************************************************\n";
}
}
CHECK_OPENCL_ERROR (status, "Program::build() failed.");
// Create kernel
// If local memory is present then use the specific kernel
mul_kernel = cl::Kernel (program, "mul_Kernel", &status);
CHECK_OPENCL_ERROR (status, "cl::Kernel failed.");
status = mul_kernel.getWorkGroupInfo < cl_ulong > (devices[deviceId],
CL_KERNEL_LOCAL_MEM_SIZE,
&usedLocalMemory);
CHECK_OPENCL_ERROR (status,
"Kernel::getWorkGroupInfo(CL_KERNEL_LOCAL_MEM_SIZE) failed"
".(usedLocalMemory)");
// Create normalize_kernel
// If local memory is present then use the specific kernel
normalize_kernel = cl::Kernel (program, "normalize_Kernel", &status);
CHECK_OPENCL_ERROR (status, "cl::Kernel failed.");
// Create normalize2_kernel
// If local memory is present then use the specific kernel
normalize2_kernel = cl::Kernel (program, "normalize2_Kernel", &status);
CHECK_OPENCL_ERROR (status, "cl::Kernel failed.");
return SDK_SUCCESS;
}
int
BoxFilterSeparable::setupCL()
{
cl_int status = 0;
cl_device_type dType;
if(deviceType.compare("cpu") == 0)
{
dType = CL_DEVICE_TYPE_CPU;
}
else //deviceType = "gpu"
{
dType = CL_DEVICE_TYPE_GPU;
if(isThereGPU() == false)
{
std::cout << "GPU not found. Falling back to CPU device" << std::endl;
dType = CL_DEVICE_TYPE_CPU;
}
}
/*
* Have a look at the available platforms and pick either
* the AMD one if available or a reasonable default.
*/
cl_platform_id platform = NULL;
int retValue = sampleCommon->getPlatform(platform, platformId, isPlatformEnabled());
CHECK_ERROR(retValue, SDK_SUCCESS, "sampleCommon::getPlatform() failed");
// Display available devices.
retValue = sampleCommon->displayDevices(platform, dType);
CHECK_ERROR(retValue, SDK_SUCCESS, "sampleCommon::displayDevices() failed");
// If we could find our platform, use it. Otherwise use just available platform.
cl_context_properties cps[3] =
{
CL_CONTEXT_PLATFORM,
(cl_context_properties)platform,
0
};
context = clCreateContextFromType(
cps,
dType,
NULL,
NULL,
&status);
CHECK_OPENCL_ERROR( status, "clCreateContextFromType failed.");
// getting device on which to run the sample
status = sampleCommon->getDevices(context, &devices, deviceId, isDeviceIdEnabled());
CHECK_ERROR(status, SDK_SUCCESS, "sampleCommon::getDevices() failed");
{
// The block is to move the declaration of prop closer to its use
cl_command_queue_properties prop = 0;
commandQueue = clCreateCommandQueue(
context,
devices[deviceId],
prop,
&status);
CHECK_OPENCL_ERROR( status, "clCreateCommandQueue failed.");
}
//Set device info of given cl_device_id
retValue = deviceInfo.setDeviceInfo(devices[deviceId]);
CHECK_ERROR(retValue, 0, "SDKDeviceInfo::setDeviceInfo() failed");
// Create and initialize memory objects
// Set Presistent memory only for AMD platform
cl_mem_flags inMemFlags = CL_MEM_READ_ONLY;
if(isAmdPlatform())
inMemFlags |= CL_MEM_USE_PERSISTENT_MEM_AMD;
// Create memory object for input Image
inputImageBuffer = clCreateBuffer(
context,
inMemFlags,
width * height * pixelSize,
NULL,
&status);
CHECK_OPENCL_ERROR(status, "clCreateBuffer failed. (inputImageBuffer)");
// Create memory object for temp Image
tempImageBuffer = clCreateBuffer(
context,
CL_MEM_READ_WRITE,
width * height * pixelSize,
0,
&status);
CHECK_OPENCL_ERROR(status, "clCreateBuffer failed. (tempImageBuffer)");
// Create memory objects for output Image
outputImageBuffer = clCreateBuffer(context,
CL_MEM_WRITE_ONLY,
width * height * pixelSize,
NULL,
&status);
CHECK_OPENCL_ERROR(status, "clCreateBuffer failed. (outputImageBuffer)");
// create a CL program using the kernel source
streamsdk::buildProgramData buildData;
buildData.kernelName = std::string("BoxFilter_Kernels.cl");
buildData.devices = devices;
buildData.deviceId = deviceId;
buildData.flagsStr = std::string("");
if(isLoadBinaryEnabled())
buildData.binaryName = std::string(loadBinary.c_str());
if(isComplierFlagsSpecified())
buildData.flagsFileName = std::string(flags.c_str());
retValue = sampleCommon->buildOpenCLProgram(program, context, buildData);
CHECK_ERROR(retValue, 0, "sampleCommon::buildOpenCLProgram() failed");
// get a kernel object handle for a kernel with the given name
verticalKernel = clCreateKernel(program,
"box_filter_vertical",
&status);
CHECK_OPENCL_ERROR(status, "clCreateKernel failed. (vertical)");
#ifdef USE_LDS
horizontalKernel = clCreateKernel(program,
"box_filter_horizontal_local",
&status);
#else
horizontalKernel = clCreateKernel(program,
"box_filter_horizontal",
&status);
#endif
CHECK_OPENCL_ERROR(status, "clCreateKernel failed. (horizontal)");
status = kernelInfoH.setKernelWorkGroupInfo(horizontalKernel, devices[deviceId]);
CHECK_ERROR(status, SDK_SUCCESS, "setKErnelWorkGroupInfo() failed");
status = kernelInfoV.setKernelWorkGroupInfo(verticalKernel, devices[deviceId]);
CHECK_ERROR(status, SDK_SUCCESS, "setKErnelWorkGroupInfo() failed");
if((blockSizeX * blockSizeY) > kernelInfoV.kernelWorkGroupSize)
{
if(!quiet)
{
std::cout << "Out of Resources!" << std::endl;
std::cout << "Group Size specified : "
<< blockSizeX * blockSizeY << std::endl;
std::cout << "Max Group Size supported on the kernel : "
<< kernelInfoV.kernelWorkGroupSize << std::endl;
std::cout << "Falling back to " << kernelInfoV.kernelWorkGroupSize << std::endl;
}
// Three possible cases
if(blockSizeX > kernelInfoV.kernelWorkGroupSize)
{
blockSizeX = kernelInfoV.kernelWorkGroupSize;
blockSizeY = 1;
}
}
return SDK_SUCCESS;
}
int
MatrixMulDouble::setupCL(void)
{
cl_int status = 0;
cl_device_type dType;
if(deviceType.compare("cpu") == 0)
dType = CL_DEVICE_TYPE_CPU;
else //deviceType = "gpu"
{
dType = CL_DEVICE_TYPE_GPU;
if(isThereGPU() == false)
{
std::cout << "GPU not found. Falling back to CPU device" << std::endl;
dType = CL_DEVICE_TYPE_CPU;
}
}
/*
* Have a look at the available platforms and pick either
* the AMD one if available or a reasonable default.
*/
status = cl::Platform::get(&platforms);
CHECK_OPENCL_ERROR(status, "Platform::get() failed.");
std::vector<cl::Platform>::iterator i;
if(platforms.size() > 0)
{
if(isPlatformEnabled())
{
i = platforms.begin() + platformId;
}
else
{
for(i = platforms.begin(); i != platforms.end(); ++i)
{
if(!strcmp((*i).getInfo<CL_PLATFORM_VENDOR>().c_str(),
"Advanced Micro Devices, Inc."))
{
break;
}
}
}
}
cl_context_properties cps[3] =
{
CL_CONTEXT_PLATFORM,
(cl_context_properties)(*i)(),
0
};
if(NULL == (*i)())
{
sampleCommon->error("NULL platform found so Exiting Application.");
return SDK_FAILURE;
}
context = cl::Context(dType, cps, NULL, NULL, &status);
CHECK_OPENCL_ERROR(status, "Context::Context() failed.");
devices = context.getInfo<CL_CONTEXT_DEVICES>();
CHECK_OPENCL_ERROR(status, "Context::getInfo() failed.");
std::cout << "Platform :" << (*i).getInfo<CL_PLATFORM_VENDOR>().c_str() << "\n";
int deviceCount = (int)devices.size();
int j = 0;
for (std::vector<cl::Device>::iterator i = devices.begin(); i != devices.end(); ++i, ++j)
{
std::cout << "Device " << j << " : ";
std::string deviceName = (*i).getInfo<CL_DEVICE_NAME>();
std::cout << deviceName.c_str() << "\n";
}
std::cout << "\n";
if (deviceCount == 0)
{
std::cerr << "No device available\n";
return SDK_FAILURE;
}
if(sampleCommon->validateDeviceId(deviceId, deviceCount))
{
sampleCommon->error("sampleCommon::validateDeviceId() failed");
return SDK_FAILURE;
}
std::string extensions = devices[deviceId].getInfo<CL_DEVICE_EXTENSIONS>();
std::string buildOptions = std::string("");
// Check if cl_khr_fp64 extension is supported
if(strstr(extensions.c_str(), "cl_khr_fp64"))
{
buildOptions.append("-D KHR_DP_EXTENSION");
}
else
{
// Check if cl_amd_fp64 extension is supported
if(!strstr(extensions.c_str(), "cl_amd_fp64"))
{
OPENCL_EXPECTED_ERROR("Device does not support cl_amd_fp64 extension!");
}
}
cl_uint localMemType;
// Get device specific information
status = devices[deviceId].getInfo<cl_uint>(
CL_DEVICE_LOCAL_MEM_TYPE,
&localMemType);
CHECK_OPENCL_ERROR(status, "Device::getInfo CL_DEVICE_LOCAL_MEM_TYPE) failed.");
// If scratchpad is available then update the flag
if(localMemType == CL_LOCAL)
lds = true;
// Get Device specific Information
status = devices[deviceId].getInfo<size_t>(
CL_DEVICE_MAX_WORK_GROUP_SIZE,
&maxWorkGroupSize);
CHECK_OPENCL_ERROR(status, "Device::getInfo(CL_DEVICE_MAX_WORK_GROUP_SIZE) failed.");
status = devices[deviceId].getInfo<cl_uint>(
CL_DEVICE_MAX_WORK_ITEM_DIMENSIONS,
&maxDimensions);
CHECK_OPENCL_ERROR(status, "Device::getInfo(CL_DEVICE_MAX_WORK_ITEM_DIMENSIONS) failed.");
maxWorkItemSizes = (size_t*)malloc(maxDimensions * sizeof(size_t));
std::vector<size_t> workItems = devices[deviceId].getInfo<CL_DEVICE_MAX_WORK_ITEM_SIZES>();
for(cl_uint i = 0; i < maxDimensions; ++i)
maxWorkItemSizes[i] = workItems[i];
status = devices[deviceId].getInfo<cl_ulong>(
CL_DEVICE_LOCAL_MEM_SIZE,
&totalLocalMemory);
CHECK_OPENCL_ERROR(status, "Device::getInfo(CL_DEVICE_LOCAL_MEM_SIZES) failed.");
// Set command queue properties
cl_command_queue_properties prop = 0;
if(!eAppGFLOPS)
prop |= CL_QUEUE_PROFILING_ENABLE;
commandQueue = cl::CommandQueue(context, devices[deviceId], prop, &status);
CHECK_OPENCL_ERROR(status, "CommandQueue::CommandQueue() failed.");
// Set Presistent memory only for AMD platform
cl_mem_flags inMemFlags = CL_MEM_READ_ONLY;
if(isAmdPlatform())
inMemFlags |= CL_MEM_USE_PERSISTENT_MEM_AMD;
// Create buffer for matrix A
inputBufA = cl::Buffer(
context,
inMemFlags,
sizeof(cl_double) * widthA * heightA,
NULL,
&status);
CHECK_OPENCL_ERROR(status, "cl::Buffer failed. (inputBufA)");
// Create buffer for matrix B
inputBufB = cl::Buffer(
context,
inMemFlags,
sizeof(cl_double) * widthB * heightB,
NULL,
&status);
CHECK_OPENCL_ERROR(status, "cl::Buffer failed. (inputBufB)");
outputBuf = cl::Buffer(
context,
CL_MEM_WRITE_ONLY | CL_MEM_ALLOC_HOST_PTR,
sizeof(cl_double) * heightA * widthB,
NULL,
&status);
CHECK_OPENCL_ERROR(status, "cl::Buffer failed. (outputBuf)");
device.push_back(devices[deviceId]);
// create a CL program using the kernel source
streamsdk::SDKFile kernelFile;
std::string kernelPath = sampleCommon->getPath();
if(isLoadBinaryEnabled())
{
kernelPath.append(loadBinary.c_str());
if(!kernelFile.readBinaryFromFile(kernelPath.c_str()))
{
std::cout << "Failed to load kernel file : " << kernelPath << std::endl;
return SDK_FAILURE;
}
cl::Program::Binaries programBinary(1,std::make_pair(
(const void*)kernelFile.source().data(),
kernelFile.source().size()));
program = cl::Program(context, device, programBinary, NULL, &status);
CHECK_OPENCL_ERROR(status, "Program::Program(Binary) failed.");
}
else
{
kernelPath.append("MatrixMulDouble_Kernels.cl");
if(!kernelFile.open(kernelPath.c_str()))
{
std::cout << "Failed to load kernel file : " << kernelPath << std::endl;
return SDK_FAILURE;
}
cl::Program::Sources programSource(
1,
std::make_pair(kernelFile.source().data(),
kernelFile.source().size()));
program = cl::Program(context, programSource, &status);
CHECK_OPENCL_ERROR(status, "Program::Program(Source) failed.");
}
std::string flagsStr = std::string("");
// Get build options if any
flagsStr.append(buildOptions.c_str());
// Get additional options
if(isComplierFlagsSpecified())
{
streamsdk::SDKFile flagsFile;
std::string flagsPath = sampleCommon->getPath();
flagsPath.append(flags.c_str());
if(!flagsFile.open(flagsPath.c_str()))
{
std::cout << "Failed to load flags file: " << flagsPath << std::endl;
return SDK_FAILURE;
}
flagsFile.replaceNewlineWithSpaces();
const char * flags = flagsFile.source().c_str();
flagsStr.append(flags);
}
if(flagsStr.size() != 0)
std::cout << "Build Options are : " << flagsStr.c_str() << std::endl;
status = program.build(device, flagsStr.c_str());
if(status != CL_SUCCESS)
{
if(status == CL_BUILD_PROGRAM_FAILURE)
{
std::string str = program.getBuildInfo<CL_PROGRAM_BUILD_LOG>(devices[deviceId]);
std::cout << " \n\t\t\tBUILD LOG\n";
std::cout << " ************************************************\n";
std::cout << str << std::endl;
std::cout << " ************************************************\n";
}
}
CHECK_OPENCL_ERROR(status, "Program::build() failed.");
// Create kernel
// If local memory is present then use the specific kernel
if(lds)
kernel = cl::Kernel(program, "mmmKernel_local", &status);
else
kernel = cl::Kernel(program, "mmmKernel", &status);
CHECK_OPENCL_ERROR(status, "cl::Kernel failed.");
status = kernel.getWorkGroupInfo<cl_ulong>(
devices[deviceId],
CL_KERNEL_LOCAL_MEM_SIZE,
&usedLocalMemory);
CHECK_OPENCL_ERROR(status, "Kernel::getWorkGroupInfo(CL_KERNEL_LOCAL_MEM_SIZE) failed"
".(usedLocalMemory)");
availableLocalMemory = totalLocalMemory - usedLocalMemory;
if(lds)
neededLocalMemory = (blockSize * 4) * (blockSize * 4) * sizeof(cl_double);
else
neededLocalMemory = 0;
if(neededLocalMemory > availableLocalMemory)
{
std::cout << "Unsupported: Insufficient local memory on device." << std::endl;
return SDK_FAILURE;
}
// Check group size against group size returned by kernel
kernelWorkGroupSize = kernel.getWorkGroupInfo<CL_KERNEL_WORK_GROUP_SIZE>(devices[deviceId], &status);
CHECK_OPENCL_ERROR(status, "Kernel::getWorkGroupInfo() failed.");
if((cl_uint)(blockSize * blockSize) > kernelWorkGroupSize)
{
if(kernelWorkGroupSize >= 64)
blockSize = 8;
else if(kernelWorkGroupSize >= 32)
blockSize = 4;
else
{
std::cout << "Out of Resources!" << std::endl;
std::cout << "Group Size specified : " << blockSize * blockSize << std::endl;
std::cout << "Max Group Size supported on the kernel : "
<< kernelWorkGroupSize<<std::endl;
return SDK_FAILURE;
}
}
if(blockSize > maxWorkItemSizes[0] ||
blockSize > maxWorkItemSizes[1] ||
blockSize * blockSize > maxWorkGroupSize)
{
sampleCommon->error("Unsupported: Device does not support requested number of work items.");
return SDK_FAILURE;
}
return SDK_SUCCESS;
}
int
GaussianNoise::setupCL()
{
cl_int err = CL_SUCCESS;
cl_device_type dType;
if(deviceType.compare("cpu") == 0)
{
dType = CL_DEVICE_TYPE_CPU;
}
else //deviceType = "gpu"
{
dType = CL_DEVICE_TYPE_GPU;
if(isThereGPU() == false)
{
std::cout << "GPU not found. Falling back to CPU device" << std::endl;
dType = CL_DEVICE_TYPE_CPU;
}
}
/*
* Have a look at the available platforms and pick either
* the AMD one if available or a reasonable default.
*/
err = cl::Platform::get(&platforms);
CHECK_OPENCL_ERROR(err, "Platform::get() failed.");
std::vector<cl::Platform>::iterator i;
if(platforms.size() > 0)
{
if(isPlatformEnabled())
{
i = platforms.begin() + platformId;
}
else
{
for(i = platforms.begin(); i != platforms.end(); ++i)
{
if(!strcmp((*i).getInfo<CL_PLATFORM_VENDOR>().c_str(),
"Advanced Micro Devices, Inc."))
{
break;
}
}
}
}
cl_context_properties cps[3] =
{
CL_CONTEXT_PLATFORM,
(cl_context_properties)(*i)(),
0
};
context = cl::Context(dType, cps, NULL, NULL, &err);
CHECK_OPENCL_ERROR(err, "Context::Context() failed.");
devices = context.getInfo<CL_CONTEXT_DEVICES>();
CHECK_OPENCL_ERROR(err, "Context::getInfo() failed.");
std::cout << "Platform :" << (*i).getInfo<CL_PLATFORM_VENDOR>().c_str() << "\n";
int deviceCount = (int)devices.size();
int j = 0;
for (std::vector<cl::Device>::iterator i = devices.begin(); i != devices.end(); ++i, ++j)
{
std::cout << "Device " << j << " : ";
std::string deviceName = (*i).getInfo<CL_DEVICE_NAME>();
std::cout << deviceName.c_str() << "\n";
}
std::cout << "\n";
if (deviceCount == 0)
{
std::cerr << "No device available\n";
return SDK_FAILURE;
}
if(sampleCommon->validateDeviceId(deviceId, deviceCount))
{
sampleCommon->error("sampleCommon::validateDeviceId() failed");
return SDK_FAILURE;
}
commandQueue = cl::CommandQueue(context, devices[deviceId], 0, &err);
CHECK_OPENCL_ERROR(err, "CommandQueue::CommandQueue() failed.");
/*
* Create and initialize memory objects
*/
// Set Presistent memory only for AMD platform
cl_mem_flags inMemFlags = CL_MEM_READ_ONLY;
if(isAmdPlatform())
inMemFlags |= CL_MEM_USE_PERSISTENT_MEM_AMD;
// Create memory object for input Image
inputImageBuffer = cl::Buffer(context,
inMemFlags,
width * height * pixelSize,
0,
&err);
CHECK_OPENCL_ERROR(err, "Buffer::Buffer() failed. (inputImageBuffer)");
// Create memory object for output Image
outputImageBuffer = cl::Buffer(context,
CL_MEM_WRITE_ONLY,
width * height * pixelSize,
NULL,
&err);
CHECK_OPENCL_ERROR(err, "Buffer::Buffer() failed. (outputImageBuffer)");
device.push_back(devices[deviceId]);
// create a CL program using the kernel source
streamsdk::SDKFile kernelFile;
std::string kernelPath = sampleCommon->getPath();
if(isLoadBinaryEnabled())
{
kernelPath.append(loadBinary.c_str());
if(!kernelFile.readBinaryFromFile(kernelPath.c_str()))
{
std::cout << "Failed to load kernel file : " << kernelPath << std::endl;
return SDK_FAILURE;
}
cl::Program::Binaries programBinary(1,std::make_pair(
(const void*)kernelFile.source().data(),
kernelFile.source().size()));
program = cl::Program(context, device, programBinary, NULL, &err);
CHECK_OPENCL_ERROR(err, "Program::Program(Binary) failed.");
}
else
{
kernelPath.append("GaussianNoise_Kernels.cl");
if(!kernelFile.open(kernelPath.c_str()))
{
std::cout << "Failed to load kernel file : " << kernelPath << std::endl;
return SDK_FAILURE;
}
cl::Program::Sources programSource(1,
std::make_pair(kernelFile.source().data(),
kernelFile.source().size()));
program = cl::Program(context, programSource, &err);
CHECK_OPENCL_ERROR(err, "Program::Program(Source) failed.");
}
std::string flagsStr = std::string("");
// Get additional options
if(isComplierFlagsSpecified())
{
streamsdk::SDKFile flagsFile;
std::string flagsPath = sampleCommon->getPath();
flagsPath.append(flags.c_str());
if(!flagsFile.open(flagsPath.c_str()))
{
std::cout << "Failed to load flags file: " << flagsPath << std::endl;
return SDK_FAILURE;
}
flagsFile.replaceNewlineWithSpaces();
const char * flags = flagsFile.source().c_str();
flagsStr.append(flags);
}
if(flagsStr.size() != 0)
std::cout << "Build Options are : " << flagsStr.c_str() << std::endl;
err = program.build(device, flagsStr.c_str());
if(err != CL_SUCCESS)
{
if(err == CL_BUILD_PROGRAM_FAILURE)
{
std::string str = program.getBuildInfo<CL_PROGRAM_BUILD_LOG>(devices[deviceId]);
std::cout << " \n\t\t\tBUILD LOG\n";
std::cout << " ************************************************\n";
std::cout << str << std::endl;
std::cout << " ************************************************\n";
}
}
CHECK_OPENCL_ERROR(err, "Program::build() failed.");
// Create kernel
kernel = cl::Kernel(program, "gaussian_transform", &err);
CHECK_OPENCL_ERROR(err, "Kernel::Kernel() failed.");
// Check group size against group size returned by kernel
kernelWorkGroupSize = kernel.getWorkGroupInfo<CL_KERNEL_WORK_GROUP_SIZE>(devices[deviceId], &err);
CHECK_OPENCL_ERROR(err, "Kernel::getWorkGroupInfo() failed.");
if((blockSizeX * blockSizeY) > kernelWorkGroupSize)
{
if(!quiet)
{
std::cout << "Out of Resources!" << std::endl;
std::cout << "Group Size specified : "
<< blockSizeX * blockSizeY << std::endl;
std::cout << "Max Group Size supported on the kernel : "
<< kernelWorkGroupSize << std::endl;
std::cout << "Falling back to " << kernelWorkGroupSize << std::endl;
}
if(blockSizeX > kernelWorkGroupSize)
{
blockSizeX = kernelWorkGroupSize;
blockSizeY = 1;
}
}
return SDK_SUCCESS;
}
int
HDRToneMapping::setupCL()
{
cl_int err = CL_SUCCESS;
cl_device_type dType;
if(deviceType.compare("cpu") == 0)
{
dType = CL_DEVICE_TYPE_CPU;
}
else //deviceType = "gpu"
{
dType = CL_DEVICE_TYPE_GPU;
if(isThereGPU() == false)
{
std::cout << "GPU not found. Falling back to CPU device" << std::endl;
dType = CL_DEVICE_TYPE_CPU;
}
}
/*
* Have a look at the available platforms and pick either
* the AMD one if available or a reasonable default.
*/
err = cl::Platform::get(&platforms);
CHECK_OPENCL_ERROR(err, "Platform::get() failed.");
std::vector<cl::Platform>::iterator i;
if(platforms.size() > 0)
{
if(isPlatformEnabled())
{
i = platforms.begin() + platformId;
}
else
{
for(i = platforms.begin(); i != platforms.end(); ++i)
{
if(!strcmp((*i).getInfo<CL_PLATFORM_VENDOR>().c_str(),
"Advanced Micro Devices, Inc."))
{
break;
}
}
}
}
cl_context_properties cps[3] =
{
CL_CONTEXT_PLATFORM,
(cl_context_properties)(*i)(),
0
};
context = cl::Context(dType, cps, NULL, NULL, &err);
CHECK_OPENCL_ERROR(err, "Context::Context() failed.");
devices = context.getInfo<CL_CONTEXT_DEVICES>();
CHECK_OPENCL_ERROR(err, "Context::getInfo() failed.");
std::cout << "Platform :" << (*i).getInfo<CL_PLATFORM_VENDOR>().c_str() << "\n";
int deviceCount = (int)devices.size();
int j = 0;
for (std::vector<cl::Device>::iterator i = devices.begin(); i != devices.end(); ++i, ++j)
{
std::cout << "Device " << j << " : ";
std::string deviceName = (*i).getInfo<CL_DEVICE_NAME>();
std::cout << deviceName.c_str() << "\n";
}
std::cout << "\n";
if (deviceCount == 0)
{
std::cout << "No device available\n";
return SDK_FAILURE;
}
if(sampleCommon->validateDeviceId(deviceId, deviceCount) != SDK_SUCCESS)
{
std::cout << "sampleCommon::validateDeviceId() failed";
return SDK_FAILURE;
}
// Get Device specific Information
err = devices[deviceId].getInfo<size_t>(
CL_DEVICE_MAX_WORK_GROUP_SIZE,
&maxWorkGroupSize);
CHECK_OPENCL_ERROR(err, "Device::getInfo(CL_DEVICE_MAX_WORK_GROUP_SIZE) failed.");
err = devices[deviceId].getInfo<cl_uint>(
CL_DEVICE_MAX_WORK_ITEM_DIMENSIONS,
&maxDimensions);
CHECK_OPENCL_ERROR(err, "Device::getInfo(CL_DEVICE_MAX_WORK_ITEM_DIMENSIONS) failed.");
maxWorkItemSizes = (size_t*)malloc(maxDimensions * sizeof(size_t));
std::vector<size_t> workItems = devices[deviceId].getInfo<CL_DEVICE_MAX_WORK_ITEM_SIZES>();
for(cl_uint i = 0; i < maxDimensions; ++i)
maxWorkItemSizes[i] = workItems[i];
err = devices[deviceId].getInfo<cl_ulong>(
CL_DEVICE_LOCAL_MEM_SIZE,
&totalLocalMemory);
CHECK_OPENCL_ERROR(err, "Device::getInfo(CL_DEVICE_LOCAL_MEM_SIZES) failed.");
commandQueue = cl::CommandQueue(context, devices[deviceId], 0, &err);
CHECK_OPENCL_ERROR(err, "CommandQueue::CommandQueue() failed.");
/*
* Create and initialize memory objects
*/
// Set Presistent memory only for AMD platform
cl_mem_flags inMemFlags = CL_MEM_READ_ONLY;
if(isAmdPlatform())
inMemFlags |= CL_MEM_USE_PERSISTENT_MEM_AMD;
// Create memory object for input Image
/**
* We use CL_MEM_USE_HOST_PTR for CPU as the CPU device is running the kernel
* on the actual buffer provided by the application
*/
if (dType == CL_DEVICE_TYPE_CPU)
{
inputImageBuffer = cl::Buffer(context,
CL_MEM_USE_HOST_PTR | CL_MEM_READ_ONLY,
width * height * numChannels * sizeof(cl_float),
input,
&err);
CHECK_OPENCL_ERROR(err, "Buffer::Buffer() failed. (inputImageBuffer)");
// Create memory object for output Image
outputImageBuffer = cl::Buffer(context,
CL_MEM_WRITE_ONLY | CL_MEM_USE_HOST_PTR,
width * height * numChannels * sizeof(cl_float),
output,
&err);
CHECK_OPENCL_ERROR(err, "Buffer::Buffer() failed. (outputImageBuffer)");
}
else if (dType == CL_DEVICE_TYPE_GPU)
{
inputImageBuffer = cl::Buffer(context,
inMemFlags,
width * height * numChannels * sizeof(cl_float),
0,
&err);
CHECK_OPENCL_ERROR(err, "Buffer::Buffer() failed. (inputImageBuffer)");
// Create memory object for output Image
outputImageBuffer = cl::Buffer(context,
CL_MEM_WRITE_ONLY,
width * height * numChannels * sizeof(cl_float),
NULL,
&err);
CHECK_OPENCL_ERROR(err, "Buffer::Buffer() failed. (outputImageBuffer)");
}
device.push_back(devices[deviceId]);
// create a CL program using the kernel source
streamsdk::SDKFile kernelFile;
std::string kernelPath = sampleCommon->getPath();
if(isLoadBinaryEnabled())
{
kernelPath.append(loadBinary.c_str());
if(kernelFile.readBinaryFromFile(kernelPath.c_str()))
{
std::cout << "Failed to load kernel file : " << kernelPath << std::endl;
return SDK_FAILURE;
}
cl::Program::Binaries programBinary(1,std::make_pair(
(const void*)kernelFile.source().data(),
kernelFile.source().size()));
program = cl::Program(context, device, programBinary, NULL, &err);
CHECK_OPENCL_ERROR(err, "Program::Program(Binary) failed.");
}
else
{
kernelPath.append("HDRToneMapping_Kernels.cl");
if(!kernelFile.open(kernelPath.c_str()))
{
std::cout << "Failed to load kernel file : " << kernelPath << std::endl;
return SDK_FAILURE;
}
cl::Program::Sources programSource(1,
std::make_pair(kernelFile.source().data(),
kernelFile.source().size()));
program = cl::Program(context, programSource, &err);
CHECK_OPENCL_ERROR(err, "Program::Program(Source) failed.");
}
std::string flagsStr = std::string("");
// Get additional options
if(isComplierFlagsSpecified())
{
streamsdk::SDKFile flagsFile;
std::string flagsPath = sampleCommon->getPath();
flagsPath.append(flags.c_str());
if(!flagsFile.open(flagsPath.c_str()))
{
std::cout << "Failed to load flags file: " << flagsPath << std::endl;
return SDK_FAILURE;
}
flagsFile.replaceNewlineWithSpaces();
const char * flags = flagsFile.source().c_str();
flagsStr.append(flags);
}
if(flagsStr.size() != 0)
std::cout << "Build Options are : " << flagsStr.c_str() << std::endl;
err = program.build(device, flagsStr.c_str());
if(err != CL_SUCCESS)
{
if(err == CL_BUILD_PROGRAM_FAILURE)
{
std::string str = program.getBuildInfo<CL_PROGRAM_BUILD_LOG>(devices[deviceId]);
std::cout << " \n\t\t\tBUILD LOG\n";
std::cout << " ************************************************\n";
std::cout << str << std::endl;
std::cout << " ************************************************\n";
}
}
CHECK_OPENCL_ERROR(err, "Program::build() failed.");
// Create kernel
kernel = cl::Kernel(program, "toneMappingPattanaik", &err);
CHECK_OPENCL_ERROR(err, "Kernel::Kernel() failed.");
// Check group size against group size returned by kernel
kernelWorkGroupSize = kernel.getWorkGroupInfo<CL_KERNEL_WORK_GROUP_SIZE>(devices[deviceId], &err);
CHECK_OPENCL_ERROR(err, "Kernel::getWorkGroupInfo() failed.");
/**
* For CPU device the kernel work group size is 1024.
* Workgroup creation/replacement is an overhead -
* avoid workgroups with small number of workitems (we pay more for replacing a WG than running more WI in a for loop).
*/
if (kernelWorkGroupSize >= 1024)
{
blockSizeX = 32;
blockSizeY = 32;
}
if((cl_uint)(blockSizeX * blockSizeY) > kernelWorkGroupSize)
{
if(kernelWorkGroupSize >= 64)
{
blockSizeX = 8;
blockSizeY = 8;
}
else if(kernelWorkGroupSize >= 32)
{
blockSizeX = 4;
blockSizeY = 4;
}
else
{
std::cout << "Out of Resources!" << std::endl;
std::cout << "Group Size specified : " << blockSizeX * blockSizeY << std::endl;
std::cout << "Max Group Size supported on the kernel : "
<< kernelWorkGroupSize<<std::endl;
return SDK_FAILURE;
}
}
if(blockSizeX > maxWorkItemSizes[0] ||
blockSizeY > maxWorkItemSizes[1] ||
blockSizeX * blockSizeY > maxWorkGroupSize)
{
std::cout << "Unsupported: Device does not support requested number of work items." << std::endl;
return SDK_FAILURE;
}
return SDK_SUCCESS;
}
