int
ComputeBench::setupCL(void)
{
cl_int status = 0;
cl_device_type dType;
if (sampleArgs->deviceType.compare("cpu") == 0) {
dType = CL_DEVICE_TYPE_CPU;
} else //deviceType = "gpu"
{
dType = CL_DEVICE_TYPE_GPU;
if (sampleArgs->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 = getPlatform(platform, sampleArgs->platformId, sampleArgs->isPlatformEnabled());
CHECK_ERROR(retValue, SDK_SUCCESS, "getPlatform() failed");
// Display available devices.
retValue = displayDevices(platform, dType);
CHECK_ERROR(retValue, SDK_SUCCESS, "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 = getDevices(context, &devices, sampleArgs->deviceId,
sampleArgs->isDeviceIdEnabled());
CHECK_ERROR(status, SDK_SUCCESS, "getDevices() failed");
//Set device info of given cl_device_id
retValue = deviceInfo.setDeviceInfo(devices[sampleArgs->deviceId]);
CHECK_ERROR(retValue, SDK_SUCCESS, "SDKDeviceInfo::setDeviceInfo() failed");
std::string deviceStr(deviceInfo.deviceVersion);
size_t vStart = deviceStr.find(" ", 0);
size_t vEnd = deviceStr.find(" ", vStart + 1);
std::string vStrVal = deviceStr.substr(vStart + 1, vEnd - vStart - 1);
// OpenCL 1.1 has inbuilt support for vec3 data types
if (vec3 == true) {
OPENCL_EXPECTED_ERROR("Device doesn't support built-in 3 component vectors!");
}
// The block is to move the declaration of prop closer to its use
/* Note: Using deprecated clCreateCommandQueue as CL_QUEUE_PROFILING_ENABLE flag not currently working
***with clCreateCommandQueueWithProperties*/
cl_command_queue_properties prop = 0;
prop |= CL_QUEUE_PROFILING_ENABLE;
commandQueue = clCreateCommandQueue(context,
devices[sampleArgs->deviceId],
prop,
&status);
CHECK_OPENCL_ERROR(status, "clCreateCommandQueue failed.");
if (sampleArgs->isLoadBinaryEnabled()) {
// Always assuming kernel was dumped for vector-width 1
if (vectorSize != 0) {
std::cout <<
"Ignoring specified vector-width. Assuming kernel was dumped for vector-width 1"
<< std::endl;
}
vectorSize = 1;
} else {
// If vector-size is not specified in the command-line, choose the preferred size for the device
if (vectorSize == 0) {
vectorSize = deviceInfo.preferredFloatVecWidth;
} else if (vectorSize == 3) {
//Make vectorSize as 4 if -v option is 3.
//This memory alignment is required as per OpenCL for type3 vectors
vec3 = true;
vectorSize = 4;
} else if ((1 != vectorSize) && (2 != vectorSize) && (4 != vectorSize) &&
(8 != vectorSize) && (16 != vectorSize)) {
std::cout << "The vectorsize can only be one of 1,2,3(4),4,8,16!" << std::endl;
return SDK_FAILURE;
}
}
outputKadd = clCreateBuffer(context, CL_MEM_WRITE_ONLY, sizeof (cl_float) * vectorSize * length, 0, &status);
CHECK_OPENCL_ERROR(status, "clCreateBuffer failed. (outputKadd)");
// create a CL program using the kernel source
char buildOption[512];
if (vectorSize == 1) {
sprintf(buildOption, "-D DATATYPE=uint -D DATATYPE2=uint4 ");
//sprintf(buildOption, "-D DATATYPE=float -D DATATYPE2=float4 ");
} else {
sprintf(buildOption, "-D DATATYPE=uint%d -D DATATYPE2=uint%d ", (vec3 == true) ? 3 : vectorSize, (vec3 == true) ? 3 : vectorSize);
//sprintf(buildOption, "-D DATATYPE=float%d -D DATATYPE2=float%d ", (vec3 == true) ? 3 : vectorSize, (vec3 == true) ? 3 : vectorSize);
}
strcat(buildOption, "-D IDXTYPE=uint ");
// create a CL program using the kernel source
buildProgramData buildData;
buildData.kernelName = std::string("ComputeBench.cl");
buildData.devices = devices;
buildData.deviceId = sampleArgs->deviceId;
buildData.flagsStr = std::string(buildOption);
if (sampleArgs->isLoadBinaryEnabled()) {
buildData.binaryName = std::string(sampleArgs->loadBinary.c_str());
}
if (sampleArgs->isComplierFlagsSpecified()) {
buildData.flagsFileName = std::string(sampleArgs->flags.c_str());
}
retValue = buildOpenCLProgram(program, context, buildData);
CHECK_ERROR(retValue, SDK_SUCCESS, "buildOpenCLProgram() failed");
// Global memory bandwidth from read-single access
kernel[0] = clCreateKernel(program, "Kadd", &status);
CHECK_OPENCL_ERROR(status, "clCreateKernel failed.(Kadd)");
return SDK_SUCCESS;
}
int AtomicCounters::setupCL(void) {
cl_int status = 0;
cl_device_type dType;
if (sampleArgs->deviceType.compare("cpu") == 0) {
dType = CL_DEVICE_TYPE_CPU;
} else // deviceType = "gpu"
{
dType = CL_DEVICE_TYPE_GPU;
if (sampleArgs->isThereGPU() == false) {
std::cout << "GPU not found. Falling back to CPU" << std::endl;
dType = CL_DEVICE_TYPE_CPU;
}
}
cl_platform_id platform = NULL;
int retValue = getPlatform(platform, sampleArgs->platformId,
sampleArgs->isPlatformEnabled());
CHECK_ERROR(retValue, SDK_SUCCESS, "getPlatform() failed.");
// Display available devices.
retValue = displayDevices(platform, dType);
CHECK_ERROR(retValue, SDK_SUCCESS, "displayDevices() failed.");
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 = getDevices(context, &devices, sampleArgs->deviceId,
sampleArgs->isDeviceIdEnabled());
CHECK_ERROR(status, SDK_SUCCESS, "getDevices() failed ");
// Set device info of given cl_device_id
retValue = deviceInfo.setDeviceInfo(devices[sampleArgs->deviceId]);
CHECK_ERROR(retValue, SDK_SUCCESS, "SDKDeviceInfo::setDeviceInfo() failed");
// Check device extensions
if (!strstr(deviceInfo.extensions, "cl_ext_atomic_counters_32")) {
OPENCL_EXPECTED_ERROR(
"Device does not support cl_ext_atomic_counters_32 extension!");
}
if (!strstr(deviceInfo.extensions, "cl_khr_local_int32_base_atomics")) {
OPENCL_EXPECTED_ERROR(
"Device does not support cl_khr_local_int32_base_atomics extension!");
}
// Get OpenCL device version
std::string deviceVersionStr = std::string(deviceInfo.deviceVersion);
size_t vStart = deviceVersionStr.find(" ", 0);
size_t vEnd = deviceVersionStr.find(" ", vStart + 1);
std::string vStrVal = deviceVersionStr.substr(vStart + 1, vEnd - vStart - 1);
// Check of OPENCL_C_VERSION if device version is 1.1 or later
#ifdef CL_VERSION_1_1
if (deviceInfo.openclCVersion) {
// Exit if OpenCL C device version is 1.0
deviceVersionStr = std::string(deviceInfo.openclCVersion);
vStart = deviceVersionStr.find(" ", 0);
vStart = deviceVersionStr.find(" ", vStart + 1);
vEnd = deviceVersionStr.find(" ", vStart + 1);
vStrVal = deviceVersionStr.substr(vStart + 1, vEnd - vStart - 1);
if (vStrVal.compare("1.0") <= 0) {
OPENCL_EXPECTED_ERROR(
"Unsupported device! Required CL_DEVICE_OPENCL_C_VERSION as 1.1");
}
} else {
OPENCL_EXPECTED_ERROR(
"Unsupported device! Required CL_DEVICE_OPENCL_C_VERSION as 1.1");
}
#else
OPENCL_EXPECTED_ERROR(
"Unsupported device! Required CL_DEVICE_OPENCL_C_VERSION as 1.1");
#endif
// Setup application data
if (setupAtomicCounters() != SDK_SUCCESS) {
return SDK_FAILURE;
}
cl_command_queue_properties props = CL_QUEUE_PROFILING_ENABLE;
commandQueue = clCreateCommandQueue(context, devices[sampleArgs->deviceId],
props, &status);
CHECK_OPENCL_ERROR(status, "clCreateCommandQueue failed(commandQueue)");
// Set Persistent memory only for AMD platform
cl_mem_flags inMemFlags = CL_MEM_READ_ONLY;
if (sampleArgs->isAmdPlatform()) {
inMemFlags |= CL_MEM_USE_PERSISTENT_MEM_AMD;
}
// Create buffer for input array
inBuf = clCreateBuffer(context, inMemFlags, length * sizeof(cl_uint), NULL,
&status);
CHECK_OPENCL_ERROR(status, "clCreateBuffer failed.(inBuf)");
// Set up data for input array
cl_event writeEvt;
status =
clEnqueueWriteBuffer(commandQueue, inBuf, CL_FALSE, 0,
length * sizeof(cl_uint), input, 0, NULL, &writeEvt);
CHECK_OPENCL_ERROR(status, "clEnqueueWriteBuffer(inBuf) failed..");
status = clFlush(commandQueue);
CHECK_OPENCL_ERROR(status, "clFlush(commandQueue) failed.");
counterOutBuf = clCreateBuffer(context, CL_MEM_READ_WRITE, sizeof(cl_uint),
NULL, &status);
CHECK_OPENCL_ERROR(status, "clCreateBuffer failed.(counterOutBuf).");
globalOutBuf = clCreateBuffer(context, CL_MEM_READ_WRITE, sizeof(cl_uint),
NULL, &status);
CHECK_OPENCL_ERROR(status, "clCreateBuffer failed.(globalOutBuf).");
// create a CL program using the kernel source
buildProgramData buildData;
buildData.kernelName = std::string("AtomicCounters_Kernels.cl");
buildData.devices = devices;
buildData.deviceId = sampleArgs->deviceId;
buildData.flagsStr = std::string("");
if (sampleArgs->isLoadBinaryEnabled()) {
buildData.binaryName = std::string(sampleArgs->loadBinary.c_str());
}
if (sampleArgs->isComplierFlagsSpecified()) {
buildData.flagsFileName = std::string(sampleArgs->flags.c_str());
}
retValue = buildOpenCLProgram(program, context, buildData);
CHECK_ERROR(retValue, SDK_SUCCESS, "buildOpenCLProgram() failed");
// ConstantBuffer bandwidth from single access
counterKernel = clCreateKernel(program, "atomicCounters", &status);
CHECK_OPENCL_ERROR(status, "clCreateKernel failed.(counterKernel).");
globalKernel = clCreateKernel(program, "globalAtomics", &status);
CHECK_OPENCL_ERROR(status, "clCreateKernel(globalKernel) failed.");
status = kernelInfoC.setKernelWorkGroupInfo(counterKernel,
devices[sampleArgs->deviceId]);
CHECK_OPENCL_ERROR(status, "kernelInfo.setKernelWorkGroupInfo failed");
status = kernelInfoG.setKernelWorkGroupInfo(globalKernel,
devices[sampleArgs->deviceId]);
CHECK_OPENCL_ERROR(status, "kernelInfo.setKernelWorkGroupInfo failed");
if (counterWorkGroupSize > kernelInfoC.kernelWorkGroupSize) {
if (!sampleArgs->quiet) {
std::cout << "Out of Resources!" << std::endl;
std::cout << "Group Size specified : " << counterWorkGroupSize
<< std::endl;
std::cout << "Max Group Size supported on the kernel(readKernel) : "
<< kernelInfoC.kernelWorkGroupSize << std::endl;
std::cout << "Falling back to " << kernelInfoC.kernelWorkGroupSize
<< std::endl;
}
counterWorkGroupSize = kernelInfoC.kernelWorkGroupSize;
}
if (globalWorkGroupSize > kernelInfoG.kernelWorkGroupSize) {
if (!sampleArgs->quiet) {
std::cout << "Out of Resources!" << std::endl;
std::cout << "Group Size specified : " << globalWorkGroupSize
<< std::endl;
std::cout << "Max Group Size supported on the kernel(writeKernel) : "
<< kernelInfoG.kernelWorkGroupSize << std::endl;
std::cout << "Falling back to " << kernelInfoG.kernelWorkGroupSize
<< std::endl;
}
globalWorkGroupSize = kernelInfoG.kernelWorkGroupSize;
}
// Wait for event and release event
status = waitForEventAndRelease(&writeEvt);
CHECK_OPENCL_ERROR(status, "waitForEventAndRelease(writeEvt) failed.");
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
GlobalMemoryBandwidth::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");
//Set device info of given cl_device_id
retValue = deviceInfo.setDeviceInfo(devices[deviceId]);
CHECK_ERROR(retValue, SDK_SUCCESS, "SDKDeviceInfo::setDeviceInfo() failed");
std::string deviceStr(deviceInfo.deviceVersion);
size_t vStart = deviceStr.find(" ", 0);
size_t vEnd = deviceStr.find(" ", vStart + 1);
std::string vStrVal = deviceStr.substr(vStart + 1, vEnd - vStart - 1);
#ifdef CL_VERSION_1_1
if(vStrVal.compare("1.0") > 0)
{
char openclVersion[1024];
status = clGetDeviceInfo(devices[deviceId],
CL_DEVICE_OPENCL_C_VERSION,
sizeof(openclVersion),
openclVersion,
0);
CHECK_OPENCL_ERROR(status, "clGetDeviceInfo failed.");
std::string tempStr(openclVersion);
size_t dotPos = tempStr.find_first_of(".");
size_t spacePos = tempStr.find_last_of(" ");
tempStr = tempStr.substr(dotPos + 1, spacePos - dotPos);
int minorVersion = atoi(tempStr.c_str());
// OpenCL 1.1 has inbuilt support for vec3 data types
if(minorVersion < 1 && vec3 == true)
{
OPENCL_EXPECTED_ERROR("Device doesn't support built-in 3 component vectors!");
}
}
else
{
// OpenCL 1.1 has inbuilt support for vec3 data types
if(vec3 == true)
{
OPENCL_EXPECTED_ERROR("Device doesn't support built-in 3 component vectors!");
}
}
#else
// OpenCL 1.1 has inbuilt support for vec3 data types
if(vec3 == true)
{
OPENCL_EXPECTED_ERROR("Device doesn't support built-in 3 component vectors!");
}
#endif
{
// The block is to move the declaration of prop closer to its use
cl_command_queue_properties prop = 0;
prop |= CL_QUEUE_PROFILING_ENABLE;
commandQueue = clCreateCommandQueue(context,
devices[deviceId],
prop,
&status);
CHECK_OPENCL_ERROR(status, "clCreateCommandQueue failed.");
}
cl_uint sizeElement = vectorSize * sizeof(cl_float);
cl_uint readLength = length + (NUM_READS * 1024 / sizeElement) + EXTRA_BYTES;
cl_uint size = readLength * vectorSize * sizeof(cl_float);
// Create input buffer
inputBuffer = clCreateBuffer(context,
CL_MEM_READ_ONLY,
size,
0,
&status);
CHECK_OPENCL_ERROR(status, "clCreateBuffer failed. (inputBuffer)");
// Write data to buffer
status = clEnqueueWriteBuffer(commandQueue,
inputBuffer,
1,
0,
size,
input,
0,
0,
0);
CHECK_OPENCL_ERROR(status, "clEnqueueWriteBuffer failed. (inputBuffer)");
outputBufferReadSingle = clCreateBuffer(context,
CL_MEM_WRITE_ONLY,
sizeof(cl_float) * vectorSize * length,
0,
&status);
CHECK_OPENCL_ERROR(status, "clCreateBuffer failed. (outputBufferReadSingle)");
// Write data to buffer
status = clEnqueueWriteBuffer(commandQueue,
outputBufferReadSingle,
CL_TRUE,
0,
sizeof(cl_float) * vectorSize * length,
outputReadSingle,
0,
NULL,
NULL);
CHECK_OPENCL_ERROR(status, "clEnqueueWriteBuffer failed. (outputBufferReadSingle)");
outputBufferReadLinear = clCreateBuffer(context,
CL_MEM_WRITE_ONLY,
sizeof(cl_float) * vectorSize * length,
0,
&status);
CHECK_OPENCL_ERROR(status, "clCreateBuffer failed. (outputBufferReadLinear)");
// Write data to buffer
status = clEnqueueWriteBuffer(commandQueue,
outputBufferReadLinear,
CL_TRUE,
0,
sizeof(cl_float) * vectorSize * length,
outputReadLinear,
0,
NULL,
NULL);
CHECK_OPENCL_ERROR(status, "clEnqueueWriteBuffer failed. (outputBufferReadLinear)");
outputBufferReadLU = clCreateBuffer(context,
CL_MEM_WRITE_ONLY,
sizeof(cl_float) * vectorSize * length,
0,
&status);
CHECK_OPENCL_ERROR(status, "clCreateBuffer failed. (outputBufferReadLU)");
// Write data to buffer
status = clEnqueueWriteBuffer(commandQueue,
outputBufferReadLU,
CL_TRUE,
0,
sizeof(cl_float) * vectorSize * length,
outputReadLU,
0,
NULL,
NULL);
CHECK_OPENCL_ERROR(status, "clEnqueueWriteBuffer failed. (outputBufferReadLU)");
outputBufferWriteLinear = clCreateBuffer(context,
CL_MEM_WRITE_ONLY,
size,
0,
&status);
CHECK_OPENCL_ERROR(status, "clCreateBuffer failed. (outputBufferWriteLinear)");
// Write data to buffer
status = clEnqueueWriteBuffer(commandQueue,
outputBufferWriteLinear,
CL_TRUE,
0,
size,
outputWriteLinear,
0,
NULL,
NULL);
CHECK_OPENCL_ERROR(status, "clEnqueueWriteBuffer failed. (outputBufferWriteLinear)");
// create a CL program using the kernel source
char buildOption[128];
if(vectorSize == 1)
sprintf(buildOption, "-D DATATYPE=float -D OFFSET=%d ", OFFSET);
else
sprintf(buildOption, "-D DATATYPE=float%d -D OFFSET=%d ", (vec3 == true) ? 3 : vectorSize, OFFSET);
// create a CL program using the kernel source
streamsdk::buildProgramData buildData;
buildData.kernelName = std::string("GlobalMemoryBandwidth_Kernels.cl");
buildData.devices = devices;
buildData.deviceId = deviceId;
buildData.flagsStr = std::string(buildOption);
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");
// Global memory bandwidth from read-single access
kernel[0] = clCreateKernel(program, "read_single", &status);
CHECK_OPENCL_ERROR(status, "clCreateKernel failed.(read_single)");
// Global memory bandwidth from read-linear access
kernel[1] = clCreateKernel(program, "read_linear", &status);
CHECK_OPENCL_ERROR(status, "clCreateKernel failed.(read_linear)");
// Global memory bandwidth from read-linear access
kernel[2] = clCreateKernel(program, "read_linear_uncached", &status);
CHECK_OPENCL_ERROR(status, "clCreateKernel failed.(read_linear_uncached)");
// Global memory bandwidth from write-linear access
kernel[3] = clCreateKernel(program, "write_linear", &status);
CHECK_OPENCL_ERROR(status, "clCreateKernel failed.(GlobalBandwidth_write_linear)");
return SDK_SUCCESS;
}
int
MatrixMulDouble::setupCL(void)
{
cl_int status = 0;
cl_device_type dType;
if(sampleArgs->deviceType.compare("cpu") == 0)
{
dType = CL_DEVICE_TYPE_CPU;
}
else //deviceType = "gpu"
{
dType = CL_DEVICE_TYPE_GPU;
if(sampleArgs->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(sampleArgs->isPlatformEnabled())
{
i = platforms.begin() + sampleArgs->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)())
{
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(validateDeviceId(sampleArgs->deviceId, deviceCount))
{
error("validateDeviceId() failed");
return SDK_FAILURE;
}
std::string extensions =
devices[sampleArgs->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[sampleArgs->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[sampleArgs->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[sampleArgs->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[sampleArgs->deviceId].getInfo<CL_DEVICE_MAX_WORK_ITEM_SIZES>();
for(cl_uint i = 0; i < maxDimensions; ++i)
{
maxWorkItemSizes[i] = workItems[i];
}
status = devices[sampleArgs->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[sampleArgs->deviceId], prop,
&status);
CHECK_OPENCL_ERROR(status, "CommandQueue::CommandQueue() failed.");
// Set Persistent memory only for AMD platform
cl_mem_flags inMemFlags = CL_MEM_READ_ONLY;
if(sampleArgs->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[sampleArgs->deviceId]);
// create a CL program using the kernel source
SDKFile kernelFile;
std::string kernelPath = getPath();
if(sampleArgs->isLoadBinaryEnabled())
{
kernelPath.append(sampleArgs->loadBinary.c_str());
if(kernelFile.readBinaryFromFile(kernelPath.c_str()) != SDK_SUCCESS)
{
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(sampleArgs->isComplierFlagsSpecified())
{
SDKFile flagsFile;
std::string flagsPath = getPath();
flagsPath.append(sampleArgs->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[sampleArgs->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[sampleArgs->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[sampleArgs->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)
{
error("Unsupported: Device does not support requested number of work items.");
return SDK_FAILURE;
}
return SDK_SUCCESS;
}
int
Histogram::setupCL(void)
{
cl_int status = 0;
cl_device_type dType;
if(sampleArgs->deviceType.compare("cpu") == 0)
{
dType = CL_DEVICE_TYPE_CPU;
}
else //sampleArgs->deviceType = "gpu"
{
dType = CL_DEVICE_TYPE_GPU;
if(sampleArgs->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 = getPlatform(platform, sampleArgs->platformId,
sampleArgs->isPlatformEnabled());
CHECK_ERROR(retValue, SDK_SUCCESS, "getPlatform() failed");
// Display available devices.
retValue = displayDevices(platform, dType);
CHECK_ERROR(retValue, SDK_SUCCESS, "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 = getDevices(context, &devices, sampleArgs->deviceId,
sampleArgs->isDeviceIdEnabled());
CHECK_ERROR(status, SDK_SUCCESS, "getDevices() failed");
// Create command queue
commandQueue = clCreateCommandQueue(context,
devices[sampleArgs->deviceId],
0,
&status);
CHECK_OPENCL_ERROR(status, "clCreateCommandQueue failed.");
//Set device info of given cl_device_id
retValue = deviceInfo.setDeviceInfo(devices[sampleArgs->deviceId]);
CHECK_ERROR(retValue, SDK_SUCCESS, "SDKDeviceInfo::setDeviceInfo() failed");
if(scalar && vector)//if both options are specified
{
std::cout<<"Ignoring --scalar and --vector option and using the default vector width of the device"<<std::endl;
vectorWidth = deviceInfo.preferredFloatVecWidth;
}
else if(scalar)
{
vectorWidth = 1;
}
else if(vector)
{
vectorWidth = 4;
}
else //if no option is specified.
{
vectorWidth = deviceInfo.preferredFloatVecWidth;
}
if(!sampleArgs->quiet)
{
if(vectorWidth == 1)
{
std::cout<<"Selecting scalar kernel\n"<<std::endl;
}
else
{
std::cout<<"Selecting vector kernel\n"<<std::endl;
}
}
subHistgCnt = (width * height) / (groupSize * groupIterations);
// Check if byte-addressable store is supported
if(!strstr(deviceInfo.extensions, "cl_khr_byte_addressable_store"))
{
byteRWSupport = false;
OPENCL_EXPECTED_ERROR("Device does not support cl_khr_byte_addressable_store extension!");
}
dataBuf = clCreateBuffer(
context,
CL_MEM_READ_ONLY,
sizeof(cl_uint) * width * height,
NULL,
&status);
CHECK_OPENCL_ERROR(status, "clCreateBuffer failed. (dataBuf)");
midDeviceBinBuf = clCreateBuffer(
context,
CL_MEM_WRITE_ONLY,
sizeof(cl_uint) * binSize * subHistgCnt,
NULL,
&status);
CHECK_OPENCL_ERROR(status, "clCreateBuffer failed. (midDeviceBinBuf)");
// create a CL program using the kernel source
buildProgramData buildData;
buildData.kernelName = std::string("Histogram_Kernels.cl");
buildData.devices = devices;
buildData.deviceId = sampleArgs->deviceId;
buildData.flagsStr = std::string("");
if(sampleArgs->isLoadBinaryEnabled())
{
buildData.binaryName = std::string(sampleArgs->loadBinary.c_str());
}
if(sampleArgs->isComplierFlagsSpecified())
{
buildData.flagsFileName = std::string(sampleArgs->flags.c_str());
}
retValue = buildOpenCLProgram(program, context, buildData);
CHECK_ERROR(retValue, 0, "buildOpenCLProgram() failed");
// get a kernel object handle for a kernel with the given name
const char *kernelName = (vectorWidth == 4)? "histogram256_vector":
"histogram256_scalar";
kernel = clCreateKernel(program, kernelName, &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
LDSBandwidth::setupCL(void)
{
cl_int status = 0;
cl_device_type dType;
if(sampleArgs->deviceType.compare("cpu") == 0)
{
dType = CL_DEVICE_TYPE_CPU;
}
else //deviceType = "gpu"
{
dType = CL_DEVICE_TYPE_GPU;
if(sampleArgs->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 = getPlatform(platform, sampleArgs->platformId,
sampleArgs->isPlatformEnabled());
CHECK_ERROR(retValue, SDK_SUCCESS, "getPlatform() failed");
// Display available devices.
retValue = displayDevices(platform, dType);
CHECK_ERROR(retValue, SDK_SUCCESS, "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 = getDevices(context, &devices, sampleArgs->deviceId,
sampleArgs->isDeviceIdEnabled());
CHECK_ERROR(status, SDK_SUCCESS, "getDevices() failed");
//Set device info of given cl_device_id
retValue = deviceInfo.setDeviceInfo(devices[sampleArgs->deviceId]);
CHECK_ERROR(retValue, SDK_SUCCESS, "SDKDeviceInfo::setDeviceInfo() failed");
std::string deviceStr(deviceInfo.deviceVersion);
size_t vStart = deviceStr.find(" ", 0);
size_t vEnd = deviceStr.find(" ", vStart + 1);
std::string vStrVal = deviceStr.substr(vStart + 1, vEnd - vStart - 1);
#ifdef CL_VERSION_1_1
if(vStrVal.compare("1.0") > 0)
{
char openclVersion[1024];
status = clGetDeviceInfo(devices[sampleArgs->deviceId],
CL_DEVICE_OPENCL_C_VERSION,
sizeof(openclVersion),
openclVersion,
0);
CHECK_OPENCL_ERROR(status, "clGetDeviceInfo failed.");
std::string tempStr(openclVersion);
size_t dotPos = tempStr.find_first_of(".");
size_t spacePos = tempStr.find_last_of(" ");
tempStr = tempStr.substr(dotPos + 1, spacePos - dotPos);
int minorVersion = atoi(tempStr.c_str());
// OpenCL 1.1 has inbuilt support for vec3 data types
if(minorVersion < 1 && vec3 == true)
{
OPENCL_EXPECTED_ERROR("Device doesn't support built-in 3 component vectors!");
}
}
else
{
// OpenCL 1.1 has inbuilt support for vec3 data types
if(vec3 == true)
{
OPENCL_EXPECTED_ERROR("Device doesn't support built-in 3 component vectors!");
}
}
#else
// OpenCL 1.1 has inbuilt support for vec3 data types
if(vec3 == true)
{
OPENCL_EXPECTED_ERROR("Device doesn't support built-in 3 component vectors!");
}
#endif
{
// The block is to move the declaration of prop closer to its use
cl_command_queue_properties prop = 0;
prop |= CL_QUEUE_PROFILING_ENABLE;
commandQueue = clCreateCommandQueue(context,
devices[sampleArgs->deviceId],
prop,
&status);
CHECK_OPENCL_ERROR(status, "clCreateCommandQueue failed.");
}
outputBuffer = clCreateBuffer(context,
CL_MEM_WRITE_ONLY,
sizeof(cl_float) * vectorSize * length,
0,
&status);
CHECK_OPENCL_ERROR(status, "clCreateBuffer failed. (outputBuffer)");
// create a CL program using the kernel source
char buildOption[64];
if(vectorSize == 1)
{
sprintf(buildOption, "-D DATATYPE=float ");
}
else
{
sprintf(buildOption, "-D DATATYPE=float%d ", vec3 == true ? 3 : vectorSize);
}
buildProgramData buildData;
buildData.kernelName = std::string("LDSBandwidth_Kernels.cl");
buildData.devices = devices;
buildData.deviceId = sampleArgs->deviceId;
buildData.flagsStr = std::string(buildOption);
if(sampleArgs->isLoadBinaryEnabled())
{
buildData.binaryName = std::string(sampleArgs->loadBinary.c_str());
}
if(sampleArgs->isComplierFlagsSpecified())
{
buildData.flagsFileName = std::string(sampleArgs->flags.c_str());
}
retValue = buildOpenCLProgram(program, context, buildData);
CHECK_ERROR(retValue, SDK_SUCCESS, "buildOpenCLProgram() failed");
// ConstantBuffer bandwidth from single access
kernel[0] = clCreateKernel(program, "LDSBandwidth_single", &status);
CHECK_OPENCL_ERROR(status, "clCreateKernel failed.(LDSBandwidth_single)");
// ConstantBuffer bandwidth from linear access
kernel[1] = clCreateKernel(program, "LDSBandwidth_linear", &status);
CHECK_OPENCL_ERROR(status, "clCreateKernel failed.(LDSBandwidth_linear)");
kernel[2] = clCreateKernel(program, "LDSBandwidth_single_verify", &status);
CHECK_OPENCL_ERROR(status,
"clCreateKernel failed.(LDSBandwidth_single_verify)");
kernel[3] = clCreateKernel(program, "LDSBandwidth_linear_verify", &status);
CHECK_OPENCL_ERROR(status,
"clCreateKernel failed.(LDSBandwidth_linear_verify)");
kernel[4] = clCreateKernel(program, "LDSBandwidth_write_linear", &status);
CHECK_OPENCL_ERROR(status,
"clCreateKernel failed.(LDSBandwidth_linear_verify)");
kernel[5] = clCreateKernel(program, "LDSBandwidth_write_linear_verify",
&status);
CHECK_OPENCL_ERROR(status,
"clCreateKernel failed.(LDSBandwidth_linear_verify)");
return SDK_SUCCESS;
}
int
ImageOverlap::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");
status = deviceInfo.setDeviceInfo(devices[deviceId]);
CHECK_OPENCL_ERROR(status, "deviceInfo.setDeviceInfo failed");
if(!deviceInfo.imageSupport)
{
OPENCL_EXPECTED_ERROR(" Expected Error: Device does not support Images");
}
blockSizeX = deviceInfo.maxWorkGroupSize<GROUP_SIZE?deviceInfo.maxWorkGroupSize:GROUP_SIZE;
// Create command queue
cl_command_queue_properties prop = 0;
for(int i=0;i<3;i++)
{
commandQueue[i] = clCreateCommandQueue(
context,
devices[deviceId],
prop,
&status);
CHECK_OPENCL_ERROR(status,"clCreateCommandQueuefailed.");
}
// Create and initialize image objects
// Create map image
mapImage = clCreateImage(context,
CL_MEM_READ_ONLY | CL_MEM_USE_HOST_PTR,
&imageFormat,
&image_desc,
mapImageData,
&status);
CHECK_OPENCL_ERROR(status,"clCreateBuffer failed. (mapImage)");
int color[4] = {0,0,80,255};
size_t origin[3] = {300,300,0};
size_t region[3] = {100,100,1};
status = clEnqueueFillImage(commandQueue[0], mapImage, color, origin, region, NULL, NULL, &eventlist[0]);
// Create fill image
fillImage = clCreateImage(context,
CL_MEM_READ_ONLY | CL_MEM_USE_HOST_PTR,
&imageFormat,
&image_desc,
fillImageData,
&status);
CHECK_OPENCL_ERROR(status,"clCreateBuffer failed. (fillImage)");
color[0] = 80;
color[1] = 0;
color[2] = 0;
color[3] = 0;
origin[0] = 50;
origin[1] = 50;
status = clEnqueueFillImage(commandQueue[1], fillImage, color, origin, region, NULL, NULL, &eventlist[1]);
//Create output image
outputImage = clCreateImage(context,
CL_MEM_WRITE_ONLY | CL_MEM_ALLOC_HOST_PTR,
&imageFormat,
&image_desc,
NULL,
&status);
CHECK_OPENCL_ERROR(status,"clCreateBuffer failed. (outputImage)");
// create a CL program using the kernel source
streamsdk::buildProgramData buildData;
buildData.kernelName = std::string("ImageOverlap_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
kernelOverLap = clCreateKernel(program, "OverLap", &status);
CHECK_OPENCL_ERROR(status,"clCreateKernel failed.(OverLap)");
return SDK_SUCCESS;
}
