BLARGG_EXPORT int fex_err_code( fex_err_t err )
{
int code = err_code( err );
return (code >= 0 ? code : fex_err_generic);
}
int main(int argc, char** argv)
{
if (argc != 2)
{
std::cout << "Usage: ./pi_vocl num\n"
<< "\twhere num = 1, 4 or 8\n";
return EXIT_FAILURE;
}
int vector_size = atoi(argv[1]);
// Define some vector size specific constants
unsigned int ITERS, WGS;
if (vector_size == 1)
{
ITERS = 262144;
WGS = 8;
}
else if (vector_size == 4)
{
ITERS = 262144 / 4;
WGS = 32;
}
else if (vector_size == 8)
{
ITERS = 262144 / 8;
WGS = 64;
}
else
{
std::cerr << "Invalid vector size\n";
return EXIT_FAILURE;
}
// Set some default values:
// Default number of steps (updated later to device preferable)
unsigned int in_nsteps = INSTEPS;
// Default number of iterations
unsigned int niters = ITERS;
unsigned int work_group_size = WGS;
try
{
// Create context, queue and build program
cl::Context context(DEVICE);
cl::CommandQueue queue(context);
cl::Program program(context, util::loadProgram("../pi_vocl.cl"), true);
cl::Kernel kernel;
// Now that we know the size of the work_groups, we can set the number of work
// groups, the actual number of steps, and the step size
unsigned int nwork_groups = in_nsteps/(work_group_size*niters);
// Get the max work group size for the kernel pi on our device
unsigned int max_size;
std::vector<cl::Device> devices = context.getInfo<CL_CONTEXT_DEVICES>();
if (vector_size == 1)
{
kernel = cl::Kernel(program, "pi");
max_size = kernel.getWorkGroupInfo<CL_KERNEL_WORK_GROUP_SIZE>(devices[0]);
}
else if (vector_size == 4)
{
kernel = cl::Kernel(program, "pi_vec4");
max_size = kernel.getWorkGroupInfo<CL_KERNEL_WORK_GROUP_SIZE>(devices[0]);
}
else if (vector_size == 8)
{
kernel = cl::Kernel(program, "pi_vec8");
max_size = kernel.getWorkGroupInfo<CL_KERNEL_WORK_GROUP_SIZE>(devices[0]);
}
if (max_size > work_group_size)
{
work_group_size = max_size;
nwork_groups = in_nsteps/(nwork_groups*niters);
}
if (nwork_groups < 1)
{
nwork_groups = devices[0].getInfo<CL_DEVICE_MAX_COMPUTE_UNITS>();
work_group_size = in_nsteps/(nwork_groups*niters);
}
unsigned int nsteps = work_group_size * niters * nwork_groups;
float step_size = 1.0f / (float) nsteps;
// Vector to hold partial sum
std::vector<float> h_psum(nwork_groups);
std::cout << nwork_groups << " work groups of size " << work_group_size << ".\n"
<< nsteps << " Integration steps\n";
cl::Buffer d_partial_sums(context, CL_MEM_WRITE_ONLY, sizeof(float) * nwork_groups);
// Start the timer
util::Timer timer;
// Execute the kernel over the entire range of our 1d input data et
// using the maximum number of work group items for this device
cl::NDRange global(nwork_groups * work_group_size);
cl::NDRange local(work_group_size);
kernel.setArg(0, niters);
kernel.setArg(1, step_size);
cl::LocalSpaceArg localmem = cl::Local(sizeof(float) * work_group_size);
kernel.setArg(2, localmem);
kernel.setArg(3, d_partial_sums);
queue.enqueueNDRangeKernel(kernel, cl::NullRange, global, local);
cl::copy(queue, d_partial_sums, h_psum.begin(), h_psum.end());
// Complete the sum and compute the final integral value
float pi_res = 0.0;
for (std::vector<float>::iterator x = h_psum.begin(); x != h_psum.end(); x++)
pi_res += *x;
pi_res *= step_size;
// Stop the timer
double rtime = static_cast<double>(timer.getTimeMilliseconds()) / 1000.;
std::cout << "The calculation ran in " << rtime << " seconds\n"
<< " pi = " << pi_res << " for " << nsteps << " steps\n";
return EXIT_SUCCESS;
}
catch (cl::Error err)
{
std::cout << "Exception\n";
std::cerr
<< "ERROR: "
<< err.what()
<< "("
<< err_code(err.err())
<< ")"
<< std::endl;
return EXIT_FAILURE;
}
}
/*
* Returns a new instance of an LDM proxy. Can take a while because it
* establishes a connection to the LDM.
*
* Arguments:
* host Identifier of the host on which an LDM server is
* running.
* instance Pointer to a pointer to the new instance. "*instance"
* is set upon successful return.
* Returns:
* 0 Success. "*instance" is set.
* LP_SYSTEM System error. "log_start()" called.
* LP_TIMEDOUT Connection attempt timed-out. "log_start()" called.
* LP_HOSTUNREACH Host is unreachable. "log_start()" called.
* LP_RPC_ERROR RPC error. "log_start()" called.
* LP_LDM_ERROR LDM error. "log_start()" called.
*/
LdmProxyStatus
lp_new(
const char* const host,
LdmProxy** const instance)
{
LdmProxyStatus status = 0; /* success */
size_t nbytes = sizeof(LdmProxy);
LdmProxy* proxy = (LdmProxy*)malloc(nbytes);
if (NULL == proxy) {
log_serror("Couldn't allocate %lu bytes for new LdmProxy", nbytes);
status = LP_SYSTEM;
}
else {
proxy->host = strdup(host);
if (NULL == proxy->host) {
LOG_SERROR1("Couldn't duplicate string \"%s\"", host);
status = LP_SYSTEM;
}
else {
CLIENT* clnt = NULL;
ErrorObj* error = ldm_clnttcp_create_vers(host, LDM_PORT, 6,
&clnt, NULL, NULL);
if (!error) {
proxy->version = 6;
proxy->hiya = my_hiya_6;
proxy->send = my_send_6;
proxy->flush = my_flush_6;
}
else if (LDM_CLNT_BAD_VERSION == err_code(error)) {
/* Couldn't connect due to protocol version. */
err_free(error);
error = ldm_clnttcp_create_vers(host, LDM_PORT, 5,
&clnt, NULL, NULL);
if (!error) {
proxy->version = 5;
proxy->hiya = my_hiya_5;
proxy->send = my_send_5;
proxy->flush = my_flush_5;
}
}
if (error) {
LOG_START1("%s", err_message(error));
err_free(error);
free(proxy->host);
status = convertStatus(error);
}
else {
proxy->clnt = clnt;
proxy->rpcTimeout = rpcTimeout;
}
} /* "proxy->host" allocated */
if (LP_OK == status) {
*instance = proxy;
}
else {
free(proxy);
}
} /* "proxy" allocated */
return status;
}
int main(void)
{
std::vector<float> h_a(LENGTH); // a vector
std::vector<float> h_b(LENGTH); // b vector
std::vector<float> h_c(LENGTH, 0xdeadbeef); // c = a + b, from compute device
cl::Buffer d_a; // device memory used for the input a vector
cl::Buffer d_b; // device memory used for the input b vector
cl::Buffer d_c; // device memory used for the output c vector
// Fill vectors a and b with random float values
int count = LENGTH;
for(int i = 0; i < count; i++)
{
h_a[i] = rand() / (float)RAND_MAX;
h_b[i] = rand() / (float)RAND_MAX;
}
try
{
// Create a context
cl::Context context(DEVICE);
// Load in kernel source, creating a program object for the context
cl::Program program(context, util::loadProgram("vadd.cl"), true);
// Get the command queue
cl::CommandQueue queue(context);
// Create the kernel functor
cl::make_kernel<cl::Buffer, cl::Buffer, cl::Buffer, int> vadd(program, "vadd");
d_a = cl::Buffer(context, h_a.begin(), h_a.end(), true);
d_b = cl::Buffer(context, h_b.begin(), h_b.end(), true);
d_c = cl::Buffer(context, CL_MEM_WRITE_ONLY, sizeof(float) * LENGTH);
util::Timer timer;
vadd(
cl::EnqueueArgs(
queue,
cl::NDRange(count)),
d_a,
d_b,
d_c,
count);
queue.finish();
double rtime = static_cast<double>(timer.getTimeMilliseconds()) / 1000.0;
printf("\nThe kernels ran in %lf seconds\n", rtime);
cl::copy(queue, d_c, h_c.begin(), h_c.end());
// Test the results
int correct = 0;
float tmp;
for(int i = 0; i < count; i++) {
tmp = h_a[i] + h_b[i]; // expected value for d_c[i]
tmp -= h_c[i]; // compute errors
if(tmp*tmp < TOL*TOL) { // correct if square deviation is less
correct++; // than tolerance squared
}
else {
printf(
" tmp %f h_a %f h_b %f h_c %f \n",
tmp,
h_a[i],
h_b[i],
h_c[i]);
}
}
// summarize results
printf(
"vector add to find C = A+B: %d out of %d results were correct.\n",
correct,
count);
}
catch (cl::Error err) {
std::cout << "Exception\n";
std::cerr
<< "ERROR: "
<< err.what()
<< "("
<< err_code(err.err())
<< ")"
<< std::endl;
}
}
int main(int argc, char** argv)
{
cl_int err; // error code returned from OpenCL calls
size_t dataSize = sizeof(float) * LENGTH;
float* h_a = (float *)malloc(dataSize); // a vector
float* h_b = (float *)malloc(dataSize); // b vector
float* h_c = (float *)malloc(dataSize); // c vector (result)
float* h_d = (float *)malloc(dataSize); // d vector (result)
float* h_e = (float *)malloc(dataSize); // e vector
float* h_f = (float *)malloc(dataSize); // f vector (result)
float* h_g = (float *)malloc(dataSize); // g vector
unsigned int correct; // number of correct results
size_t global; // global domain size
cl_device_id device_id; // compute device id
cl_context context; // compute context
cl_command_queue commands; // compute command queue
cl_program program; // compute program
cl_kernel ko_vadd; // compute kernel
cl_mem d_a; // device memory used for the input a vector
cl_mem d_b; // device memory used for the input b vector
cl_mem d_c; // device memory used for the output c vector
cl_mem d_d; // device memory used for the output d vector
cl_mem d_e; // device memory used for the input e vector
cl_mem d_f; // device memory used for the output f vector
cl_mem d_g; // device memory used for the input g vector
// Fill vectors a and b with random float values
int i = 0;
for(i = 0; i < LENGTH; i++){
h_a[i] = rand() / (float)RAND_MAX;
h_b[i] = rand() / (float)RAND_MAX;
h_e[i] = rand() / (float)RAND_MAX;
h_g[i] = rand() / (float)RAND_MAX;
}
// Set up platform and GPU device
cl_uint numPlatforms;
// Find number of platforms
err = clGetPlatformIDs(0, NULL, &numPlatforms);
checkError(err, "Finding platforms");
if (numPlatforms == 0)
{
printf("Found 0 platforms!\n");
return EXIT_FAILURE;
}
// Get all platforms
cl_platform_id Platform[numPlatforms];
err = clGetPlatformIDs(numPlatforms, Platform, NULL);
checkError(err, "Getting platforms");
// Secure a GPU
for (i = 0; i < numPlatforms; i++)
{
err = clGetDeviceIDs(Platform[i], DEVICE, 1, &device_id, NULL);
if (err == CL_SUCCESS)
{
break;
}
}
if (device_id == NULL)
checkError(err, "Getting device");
err = output_device_info(device_id);
checkError(err, "Outputting device info");
// Create a compute context
context = clCreateContext(0, 1, &device_id, NULL, NULL, &err);
checkError(err, "Creating context");
// Create a command queue
commands = clCreateCommandQueue(context, device_id, 0, &err);
checkError(err, "Creating command queue");
// Create the compute program from the source buffer
program = clCreateProgramWithSource(context, 1, (const char **) & KernelSource, NULL, &err);
checkError(err, "Creating program");
// Build the program
err = clBuildProgram(program, 0, NULL, NULL, NULL, NULL);
if (err != CL_SUCCESS)
{
size_t len;
char buffer[2048];
printf("Error: Failed to build program executable!\n%s\n", err_code(err));
clGetProgramBuildInfo(program, device_id, CL_PROGRAM_BUILD_LOG, sizeof(buffer), buffer, &len);
printf("%s\n", buffer);
return EXIT_FAILURE;
}
// Create the compute kernel from the program
ko_vadd = clCreateKernel(program, "vadd", &err);
checkError(err, "Creating kernel");
// Create the input (a, b, e, g) arrays in device memory
// NB: we copy the host pointers here too
d_a = clCreateBuffer(context, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR, dataSize, h_a, &err);
checkError(err, "Creating buffer d_a");
d_b = clCreateBuffer(context, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR, dataSize, h_b, &err);
checkError(err, "Creating buffer d_b");
d_e = clCreateBuffer(context, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR, dataSize, h_e, &err);
checkError(err, "Creating buffer d_e");
d_g = clCreateBuffer(context, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR, dataSize, h_g, &err);
checkError(err, "Creating buffer d_g");
// Create the output arrays in device memory
d_c = clCreateBuffer(context, CL_MEM_READ_WRITE, dataSize, NULL, &err);
checkError(err, "Creating buffer d_c");
d_d = clCreateBuffer(context, CL_MEM_READ_WRITE, dataSize, NULL, &err);
checkError(err, "Creating buffer d_d");
d_f = clCreateBuffer(context, CL_MEM_WRITE_ONLY, dataSize, NULL, &err);
checkError(err, "Creating buffer d_f");
const int count = LENGTH;
// Enqueue kernel - first time
// Set the arguments to our compute kernel
err = clSetKernelArg(ko_vadd, 0, sizeof(cl_mem), &d_a);
err |= clSetKernelArg(ko_vadd, 1, sizeof(cl_mem), &d_b);
err |= clSetKernelArg(ko_vadd, 2, sizeof(cl_mem), &d_c);
err |= clSetKernelArg(ko_vadd, 3, sizeof(unsigned int), &count);
checkError(err, "Setting kernel arguments");
// Execute the kernel over the entire range of our 1d input data set
// letting the OpenCL runtime choose the work-group size
global = count;
err = clEnqueueNDRangeKernel(commands, ko_vadd, 1, NULL, &global, NULL, 0, NULL, NULL);
checkError(err, "Enqueueing kernel 1st time");
// Enqueue kernel - second time
// Set different arguments to our compute kernel
err = clSetKernelArg(ko_vadd, 0, sizeof(cl_mem), &d_e);
err |= clSetKernelArg(ko_vadd, 1, sizeof(cl_mem), &d_c);
err |= clSetKernelArg(ko_vadd, 2, sizeof(cl_mem), &d_d);
checkError(err, "Setting kernel arguments");
// Enqueue the kernel again
err = clEnqueueNDRangeKernel(commands, ko_vadd, 1, NULL, &global, NULL, 0, NULL, NULL);
checkError(err, "Enqueueing kernel 2nd time");
// Enqueue kernel - third time
// Set different (again) arguments to our compute kernel
err = clSetKernelArg(ko_vadd, 0, sizeof(cl_mem), &d_g);
err |= clSetKernelArg(ko_vadd, 1, sizeof(cl_mem), &d_d);
err |= clSetKernelArg(ko_vadd, 2, sizeof(cl_mem), &d_f);
checkError(err, "Setting kernel arguments");
// Enqueue the kernel again
err = clEnqueueNDRangeKernel(commands, ko_vadd, 1, NULL, &global, NULL, 0, NULL, NULL);
checkError(err, "Enqueueing kernel 3rd time");
// Read back the result from the compute device
err = clEnqueueReadBuffer( commands, d_f, CL_TRUE, 0, sizeof(float) * count, h_f, 0, NULL, NULL );
checkError(err, "Reading back d_f");
// Test the results
correct = 0;
float tmp;
for(i = 0; i < count; i++)
{
tmp = h_a[i] + h_b[i] + h_e[i] + h_g[i]; // assign element i of a+b+e+g to tmp
tmp -= h_f[i]; // compute deviation of expected and output result
if(tmp*tmp < TOL*TOL) // correct if square deviation is less than tolerance squared
correct++;
else {
printf(" tmp %f h_a %f h_b %f h_e %f h_g %f h_f %f\n",tmp, h_a[i], h_b[i], h_e[i], h_g[i], h_f[i]);
}
}
// summarize results
printf("C = A+B+E+G: %d out of %d results were correct.\n", correct, count);
// cleanup then shutdown
clReleaseMemObject(d_a);
clReleaseMemObject(d_b);
clReleaseMemObject(d_c);
clReleaseMemObject(d_d);
clReleaseMemObject(d_e);
clReleaseMemObject(d_f);
clReleaseMemObject(d_g);
clReleaseProgram(program);
clReleaseKernel(ko_vadd);
clReleaseCommandQueue(commands);
clReleaseContext(context);
free(h_a);
free(h_b);
free(h_c);
free(h_d);
free(h_e);
free(h_f);
free(h_g);
return 0;
}
static int err_code_pkt(DUSBVirtualPacket *pkt)
{
return err_code((((uint16_t)pkt->data[0]) << 8) | pkt->data[1]);
}
/*
* Attempts to connect to an upstream LDM using a range of LDM versions. The
* versions are tried, in order, from highest to lowest. This function returns
* on the first successful attempt. If the host is unknown or the RPC call
* times-out, then the version-loop is prematurely terminated and this function
* returns immediately.
*
* The client is responsible for freeing the client resources set by this
* function on success. Calls exitIfDone() after potentially lengthy
* operations.
*
* Arguments:
* upName The name of the upstream LDM host.
* port The port on which to connect.
* version Program version.
* *client Pointer to CLIENT structure. Set on success.
* *socket The socket used for the connection. May be NULL.
* *upAddr The IP address of the upstream LDM host. Set on
* success. May be NULL.
* Returns:
* NULL Success. *vers_out, *client, *sock_out, and *upAddr
* set.
* !NULL Error. "*client" is not set. err_code(RETURN_VALUE):
* LDM_CLNT_UNKNOWN_HOST Unknown upstream host.
* LDM_CLNT_TIMED_OUT Call to upstream host timed-out.
* LDM_CLNT_BAD_VERSION Upstream LDM isn't given version.
* LDM_CLNT_NO_CONNECT Other connection-related error.
* LDM_CLNT_SYSTEM_ERROR A fatal system-error occurred.
*/
ErrorObj*
ldm_clnttcp_create_vers(
const char* const upName,
const unsigned port,
unsigned const version,
CLIENT** const client,
int* const socket,
struct sockaddr_in* upAddr)
{
ErrorObj* error;
struct sockaddr_in addr;
assert(upName != NULL);
assert(client != NULL);
/*
* Get the IP address of the upstream LDM. This is a potentially
* lengthy operation.
*/
(void)exitIfDone(0);
error = ldm_clnt_addr(upName, &addr);
if (error) {
error = ERR_NEW1(LDM_CLNT_UNKNOWN_HOST, error,
"Couldn't get IP address of host %s", upName);
}
else {
int sock;
int errCode;
CLIENT* clnt = NULL;
/*
* Connect to the remote port. This is a potentially lengthy
* operation.
*/
(void)exitIfDone(0);
error = ldm_clnt_tcp_create(&addr, version, port, &clnt, &sock);
if (error) {
errCode = err_code(error);
if (LDM_CLNT_NO_CONNECT != errCode) {
error =
ERR_NEW3(errCode, error,
"Couldn't connect to LDM %d on %s "
"using port %d",
version, upName, port);
}
else {
err_log_and_free(
ERR_NEW3(0, error,
"Couldn't connect to LDM %d on %s using port "
"%d",
version, upName, port),
ERR_INFO);
/*
* Connect using the portmapper. This is a
* potentially lengthy operation.
*/
(void)exitIfDone(0);
error = ldm_clnt_tcp_create(&addr, version, 0, &clnt, &sock);
if (error) {
error =
ERR_NEW2(err_code(error), error,
"Couldn't connect to LDM on %s "
"using either port %d or portmapper",
upName, port);
} /* portmapper failure */
} /* non-fatal port failure */
} /* port failure */
else {
assert(!error);
/*
* Success. Set the return arguments.
*/
*client = clnt;
if (socket)
*socket = sock;
if (upAddr)
*upAddr = addr;
} /* clnt != NULL */
} /* got upstream IP address */
return error;
}
void TransformExport::Export()
{
ModelExporter & modelExporter = ModelExporter::GetExporter();
modelExporter.ResetTimeline();
MStatus stat;
MItDependencyNodes itDep(MFn::kTransform,&stat);
while (!itDep.isDone())
{
MObject obj = itDep.item();
MFnTransform transform(obj, &stat);
err_code(stat);
MString cmd = MString("reference -q -f ") + transform.name();
MString file_id;
stat = MGlobal::executeCommand( cmd, file_id );
if( stat == MS::kSuccess )
{
itDep.next();
continue;
}
MString transformName = transform.name(&stat);
err_code(stat);
unsigned int parentCount = transform.parentCount(&stat);
err_code(stat);
unsigned int childCount = transform.childCount(&stat);
err_code(stat);
bool doExport = false;
if (childCount == 0)
{
doExport = true; // Tip JOINTS
}
for (unsigned int child = 0 ; child < childCount ; child++)
{
MObject childObj = transform.child(child,&stat);
err_code(stat);
MFn::Type childType = childObj.apiType();
if ( modelExporter.CheckChildType(childType) )
{
doExport = true;
break;
}
}
if (!doExport)
{
itDep.next();
continue;
}
TransformData * pTransformData = new TransformData();
pTransformData->name = transformName.asChar();
MObject parentObj = transform.parent(0,&stat);
MFn::Type parentType = parentObj.apiType();
if ( modelExporter.CheckParentType(parentType) )
{
MFnDagNode parentDagNode(parentObj, &stat);
MString parentName = parentDagNode.name(&stat);
if (parentName.length() > 0)
pTransformData->parentName = parentName.asChar();
}
if (parentType != MFn::kJoint && obj.apiType() == MFn::kJoint)
{
modelExporter.mSkeletonRoot = transform.name().asChar();
}
MVector translate = transform.getTranslation(MSpace::kTransform, &stat);
err_code(stat);
MVector pivot = transform.rotatePivotTranslation(MSpace::kTransform, &stat);
err_code(stat);
pTransformData->tx = (float)translate.x;
pTransformData->ty = (float)translate.y;
pTransformData->tz = (float)translate.z;
pTransformData->px = (float)pivot.x;
pTransformData->py = (float)pivot.y;
pTransformData->pz = (float)pivot.z;
//double rx,ry,rz,rw;
MQuaternion quat;
stat = transform.getRotation(quat);
err_code(stat);
if (transform.object().hasFn(MFn::kJoint))
{
err_code(stat);
MFnIkJoint joint(transform.object(), &stat);
err_code(stat);
MQuaternion RO;
MQuaternion R;
MQuaternion JO;
MQuaternion IS; // We dont have time for this.
stat = joint.getScaleOrientation(RO);
err_code(stat);
stat = joint.getRotation(R);
err_code(stat);
stat = joint.getOrientation(JO);
err_code(stat);
quat = RO*R*JO;
}
//stat = transform.getRotationQuaternion(rx,ry,rz,rw, MSpace::kTransform);
//err_code(stat);
pTransformData->rx = -(float)quat.x;
pTransformData->ry = -(float)quat.y;
pTransformData->rz = -(float)quat.z;
pTransformData->rw = (float)quat.w;
double scale[3];
stat = transform.getScale(scale);
err_code(stat);
pTransformData->sx = (float)scale[0];
pTransformData->sy = (float)scale[1];
pTransformData->sz = (float)scale[2];
pTransformData->index = modelExporter.mTransformCount;
modelExporter.mSceneTransforms.push_back(pTransformData);
modelExporter.mSceneTransformsTable[transformName.asChar()] = modelExporter.mTransformCount;
modelExporter.mTransformCount++;
stat = itDep.next();
err_code(stat);
}
vector<TransformData*>::iterator nodesIter = modelExporter.mSceneTransforms.begin();
while (nodesIter != modelExporter.mSceneTransforms.end())
{
TransformData * pTransformData = *nodesIter;
if (pTransformData->parentName.length() > 0)
pTransformData->parent = modelExporter.mSceneTransformsTable[pTransformData->parentName];
nodesIter++;
}
WriteTransforms();
}
int main(void) {
std::vector<float> h_a(count); // a vector
std::vector<float> h_b(count); // b vector
std::vector<float> h_c(count, 0xdeadbeef); // c = a + b
cl::Buffer d_a; // device memory used for the input a vector
cl::Buffer d_b; // device memory used for the input b vector
cl::Buffer d_c; // device memory used for the output c vector
// Fill vectors a and b with random float values
for (size_t i = 0; i < count; i++) {
h_a[i] = rand_r(&seed) / static_cast<float>(UINT32_MAX);
h_b[i] = rand_r(&seed) / static_cast<float>(UINT32_MAX);
}
try {
// Create a context
cl::Context context(DEVICE);
// Load in kernel source, creating a program object for the context
cl::Program program(context, util::loadProgram("vadd.cl"), true);
// Get the command queue
cl::CommandQueue queue(context);
// Create the kernel functor
auto vadd = cl::make_kernel<cl::Buffer, cl::Buffer,
cl::Buffer, int>(program, "vadd");
d_a = cl::Buffer(context, begin(h_a), end(h_a), true);
d_b = cl::Buffer(context, begin(h_b), end(h_b), true);
d_c = cl::Buffer(context, CL_MEM_WRITE_ONLY, sizeof(float) * count);
util::Timer timer;
vadd(cl::EnqueueArgs(queue, cl::NDRange(count)),
d_a, d_b, d_c, static_cast<int>(count));
queue.finish();
double rtime = static_cast<double>(timer.getTimeMilliseconds()) / 1000.0;
printf("\nThe kernels ran in %lf seconds\n", rtime);
cl::copy(queue, d_c, begin(h_c), end(h_c)); // NOLINT
// Test the results
int correct = 0;
float tmp;
for (size_t i = 0; i < count; i++) {
tmp = h_a[i] + h_b[i]; // expected value for d_c[i]
tmp -= h_c[i]; // compute errors
if (tmp * tmp < tolerance * tolerance)
correct++;
else
printf(" tmp %f h_a %f h_b %f h_c %f \n", tmp, h_a[i], h_b[i], h_c[i]);
}
// summarize results
printf("vector add to find C = A+B: %d out of %zu results were correct.\n",
correct, count);
} catch (cl::Error err) {
std::cout << "Exception\n";
std::cerr << "ERROR: " << err.what() << "(" << err_code(err.err()) << ")\n";
}
}
int main(void)
{
float *h_psum; // vector to hold partial sum
int in_nsteps = INSTEPS; // default number of steps (updated later to device preferable)
int niters = ITERS; // number of iterations
int nsteps;
float step_size;
size_t nwork_groups;
size_t max_size, work_group_size = 8;
float pi_res;
cl_mem d_partial_sums;
char *kernelsource = getKernelSource("../pi_ocl.cl"); // Kernel source
cl_int err;
cl_device_id device_id; // compute device id
cl_context context; // compute context
cl_command_queue commands; // compute command queue
cl_program program; // compute program
cl_kernel kernel_pi; // compute kernel
// Set up OpenCL context. queue, kernel, etc.
cl_uint numPlatforms;
// Find number of platforms
err = clGetPlatformIDs(0, NULL, &numPlatforms);
if (err != CL_SUCCESS || numPlatforms <= 0)
{
printf("Error: Failed to find a platform!\n%s\n",err_code(err));
return EXIT_FAILURE;
}
// Get all platforms
cl_platform_id Platform[numPlatforms];
err = clGetPlatformIDs(numPlatforms, Platform, NULL);
if (err != CL_SUCCESS || numPlatforms <= 0)
{
printf("Error: Failed to get the platform!\n%s\n",err_code(err));
return EXIT_FAILURE;
}
// Secure a device
for (int i = 0; i < numPlatforms; i++)
{
err = clGetDeviceIDs(Platform[i], DEVICE, 1, &device_id, NULL);
if (err == CL_SUCCESS)
break;
}
if (device_id == NULL)
{
printf("Error: Failed to create a device group!\n%s\n",err_code(err));
return EXIT_FAILURE;
}
// Output information
err = output_device_info(device_id);
// Create a compute context
context = clCreateContext(0, 1, &device_id, NULL, NULL, &err);
if (!context)
{
printf("Error: Failed to create a compute context!\n%s\n", err_code(err));
return EXIT_FAILURE;
}
// Create a command queue
commands = clCreateCommandQueue(context, device_id, 0, &err);
if (!commands)
{
printf("Error: Failed to create a command commands!\n%s\n", err_code(err));
return EXIT_FAILURE;
}
// Create the compute program from the source buffer
program = clCreateProgramWithSource(context, 1, (const char **) & kernelsource, NULL, &err);
if (!program)
{
printf("Error: Failed to create compute program!\n%s\n", err_code(err));
return EXIT_FAILURE;
}
// Build the program
err = clBuildProgram(program, 0, NULL, NULL, NULL, NULL);
if (err != CL_SUCCESS)
{
size_t len;
char buffer[2048];
printf("Error: Failed to build program executable!\n%s\n", err_code(err));
clGetProgramBuildInfo(program, device_id, CL_PROGRAM_BUILD_LOG, sizeof(buffer), buffer, &len);
printf("%s\n", buffer);
return EXIT_FAILURE;
}
// Create the compute kernel from the program
kernel_pi = clCreateKernel(program, "pi", &err);
if (!kernel_pi || err != CL_SUCCESS)
{
printf("Error: Failed to create compute kernel!\n%s\n", err_code(err));
return EXIT_FAILURE;
}
// Find kernel work-group size
err = clGetKernelWorkGroupInfo (kernel_pi, device_id, CL_KERNEL_WORK_GROUP_SIZE, sizeof(size_t), &work_group_size, NULL);
if (err != CL_SUCCESS)
{
printf("Error: Failed to get kernel work-group info\n%s\n", err_code(err));
return EXIT_FAILURE;
}
// Now that we know the size of the work-groups, we can set the number of
// work-groups, the actual number of steps, and the step size
nwork_groups = in_nsteps/(work_group_size*niters);
if (nwork_groups < 1)
{
err = clGetDeviceInfo(device_id, CL_DEVICE_MAX_COMPUTE_UNITS, sizeof(size_t), &nwork_groups, NULL);
work_group_size = in_nsteps / (nwork_groups * niters);
}
nsteps = work_group_size * niters * nwork_groups;
step_size = 1.0f/(float)nsteps;
h_psum = calloc(sizeof(float), nwork_groups);
if (!h_psum)
{
printf("Error: could not allocate host memory for h_psum\n");
return EXIT_FAILURE;
}
printf(" %ld work-groups of size %ld. %d Integration steps\n",
nwork_groups,
work_group_size,
nsteps);
d_partial_sums = clCreateBuffer(context, CL_MEM_WRITE_ONLY, sizeof(float) * nwork_groups, NULL, &err);
if (err != CL_SUCCESS)
{
printf("Error: Failed to create buffer\n%s\n", err_code(err));
return EXIT_FAILURE;
}
// Set kernel arguments
err = clSetKernelArg(kernel_pi, 0, sizeof(int), &niters);
err |= clSetKernelArg(kernel_pi, 1, sizeof(float), &step_size);
err |= clSetKernelArg(kernel_pi, 2, sizeof(float) * work_group_size, NULL);
err |= clSetKernelArg(kernel_pi, 3, sizeof(cl_mem), &d_partial_sums);
if (err != CL_SUCCESS)
{
printf("Error: Failed to set kernel arguments!\n");
return EXIT_FAILURE;
}
// Execute the kernel over the entire range of our 1D input data set
// using the maximum number of work items for this device
size_t global = nwork_groups * work_group_size;
size_t local = work_group_size;
double rtime = wtime();
err = clEnqueueNDRangeKernel(
commands,
kernel_pi,
1, NULL,
&global,
&local,
0, NULL, NULL);
if (err != CL_SUCCESS)
{
printf("Error: Failed to execute kernel\n%s\n", err_code(err));
return EXIT_FAILURE;
}
err = clEnqueueReadBuffer(
commands,
d_partial_sums,
CL_TRUE,
0,
sizeof(float) * nwork_groups,
h_psum,
0, NULL, NULL);
if (err != CL_SUCCESS)
{
printf("Error: Failed to read buffer\n%s\n", err_code(err));
return EXIT_FAILURE;
}
// complete the sum and compute the final integral value on the host
pi_res = 0.0f;
for (unsigned int i = 0; i < nwork_groups; i++)
{
pi_res += h_psum[i];
}
pi_res *= step_size;
rtime = wtime() - rtime;
printf("\nThe calculation ran in %lf seconds\n", rtime);
printf(" pi = %f for %d steps\n", pi_res, nsteps);
// clean up
clReleaseMemObject(d_partial_sums);
clReleaseProgram(program);
clReleaseKernel(kernel_pi);
clReleaseCommandQueue(commands);
clReleaseContext(context);
free(kernelsource);
free(h_psum);
}
int main(int argc, char *argv[])
{
// Declare variables - y', L
// Load from file? Declare within host?
Type y[K];
Type L[K*K];
Type R[K];
Type m[K];
Complex Xml[K];
int check_result;
cl_mem input_y;
cl_mem input_L;
cl_mem output_xml;
// OpenCL-specific variables
cl_device_id device_id;
cl_platform_id platform_id;
cl_context context;
cl_command_queue commands;
cl_program program;
cl_kernel SDkernel;
cl_int dev_type;
cl_int error;
cl_event event;
FILE *kernel;
char *kernelSRC;
size_t global[2];
size_t local[2];
if(argc > 1)
{
kernel = argv[1];
}
else
printf("\nError - must specify arguments\n");
//--------------------------------------------------------------------------------
// Create a context, queue and device.
//--------------------------------------------------------------------------------
cl_uint numPlatforms;
// Find number of platforms
err = clGetPlatformIDs(0, NULL, &numPlatforms);
if (err != CL_SUCCESS || numPlatforms <= 0)
{
printf("Error: Failed to find a platform!\n%s\n",err_code(err));
return EXIT_FAILURE;
}
// Get all platforms
cl_platform_id Platform[numPlatforms];
err = clGetPlatformIDs(numPlatforms, Platform, NULL);
if (err != CL_SUCCESS || numPlatforms <= 0)
{
printf("Error: Failed to get the platform!\n%s\n",err_code(err));
return EXIT_FAILURE;
}
// Secure a device
for (int i = 0; i < numPlatforms; i++)
{
err = clGetDeviceIDs(Platform[i], DEVICE, 1, &device_id, NULL);
if (err == CL_SUCCESS)
break;
}
if (device_id == NULL)
{
printf("Error: Failed to create a device group!\n%s\n",err_code(err));
return EXIT_FAILURE;
}
// Create a compute context
context = clCreateContext(0, 1, &device_id, NULL, NULL, &error);
if (!context)
{
printf("Error: Failed to create a compute context!\n%s\n", err_code(error));
return EXIT_FAILURE;
}
// Create a command queue
commands = clCreateCommandQueue(context, device_id, 0, &error);
if (!commands)
{
printf("Error: Failed to create a command commands!\n%s\n", err_code(error));
return EXIT_FAILURE;
}
// Create buffers for each argument of kernel
// Need to add for R^2 and m
input_y = clCreateBuffer(context, CL_MEM_READ_ONLY, sizeof(Type) * K, y, &err);
input_L = clCreateBuffer(context, CL_MEM_READ_ONLY, sizeof(Type) * K * K, L, &err);
output_xml = clCreateBuffer(context, CL_MEM_WRITE_ONLY, sizeof(Complex) * K, Xml, &err);
if (err != CL_SUCCESS)
{
printf("Error: failed to create buffer\n%s\n", err_code(err));
return EXIT_FAILURE;
}
// Create the compute program from the source buffer
program = clCreateProgramWithSource(context, 1, (const char **)&kernelSRC, NULL, &error);
if (err != CL_SUCCESS)
{
printf("Error: could not create program\n%s\n", err_code(err));
return EXIT_FAILURE;
}
// Build the program
err = clBuildProgram(program, 0, NULL, NULL, NULL, NULL);
if (err != CL_SUCCESS)
{
size_t len;
char buffer[2048];
printf("Error: Failed to build program executable!\n%s\n", err_code(err));
clGetProgramBuildInfo(program, device_id, CL_PROGRAM_BUILD_LOG, sizeof(buffer), buffer, &len);
printf("%s\n", buffer);
return EXIT_FAILURE;
}
// Create the compute kernel from the program
kernel = clCreateKernel(program, "SD", &err);
if (!kernel || err != CL_SUCCESS)
{
printf("Error: Failed to create compute kernel!\n%s\n", err_code(err));
return EXIT_FAILURE;
}
err = clEnqueueWriteBuffer(commands, input_y, CL_TRUE, 0, sizeof(Type)*K, y, 0, NULL, NULL);
err = clEnqueueWriteBuffer(commands, input_L, CL_TRUE, 0, sizeof(Type)*K*K, L, 0, NULL, NULL);
if(err != CL_SUCCESS)
{
printf("Error: could not write buffer\nError code %d\n", err);
return EXIT_FAILURE;
}
err = clSetKernelArg(SDkernel, 0, sizeof(cl_mem), &input_y);
err |= clSetKernelArg(SDkernel, 1, sizeof(cl_mem), &input_L);
err |= clSetKernelArg(SDkernel, 2, sizeof(cl_mem), &input_R);
err |= clSetKernelArg(SDkernel, 3, sizeof(cl_mem), &output_m);
err |= clSetKernelArg(SDkernel, 4, sizeof(cl_mem), &output_xml);
if(err != CL_SUCCESS)
{
printf("Error: could not set kernel arguments\nError code %d\n", err);
return EXIT_FAILURE;
}
global[0] = K;
global[1] = K;
local[0] = K;
local[1] = 1;
err = clEnqueueNDRangeKernel(commands,
kernel,
2,
NULL,
(size_t*)&global,
(size_t*)&local,
0,
NULL,
&event);
if(err != CL_SUCCESS)
{
printf("Error: could not set ND range\nError code %d\n", err);
return EXIT_FAILURE;
}
clEnqueueReadBuffer(commands, output_m, CL_TRUE, 0, sizeof(Type)*K, m, 0, NULL, NULL);
clEnqueueReadBuffer(commands, output_xml, CL_TRUE, 0, sizeof(Type)*K, m, 0, NULL, NULL);
clReleaseMemObject(input_y);
clReleaseMemObject(input_L);
clReleaseMemObject(output_xml);
clReleaseProgram(program);
clReleaseKernel(SDkernel);
clReleaseCommandQueue(commands);
clReleaseContext(context);
return EXIT_SUCCESS;
}
int main(void)
{
std::vector<float> h_a(LENGTH); // a vector
std::vector<float> h_b(LENGTH); // b vector
std::vector<float> h_c (LENGTH); // c vector
std::vector<float> h_r (LENGTH, 0xdeadbeef); // d vector (result)
cl::Buffer d_a; // device memory used for the input a vector
cl::Buffer d_b; // device memory used for the input b vector
cl::Buffer d_c; // device memory used for the input c vector
cl::Buffer d_r; // device memory used for the output r vector
// Fill vectors a and b with random float values
int count = LENGTH;
for(int i = 0; i < count; i++)
{
h_a[i] = rand() / (float)RAND_MAX;
h_b[i] = rand() / (float)RAND_MAX;
h_c[i] = rand() / (float)RAND_MAX;
}
try
{
// Create a context
cl::Context context(DEVICE);
// Load in kernel source, creating a program object for the context
cl::Program program(context, util::loadProgram("vadd_abc.cl"), true);
// Get the command queue
cl::CommandQueue queue(context);
// Create the kernel functor
auto vadd = cl::make_kernel<cl::Buffer, cl::Buffer, cl::Buffer, cl::Buffer, int>(program, "vadd");
d_a = cl::Buffer(context, begin(h_a), end(h_a), true);
d_b = cl::Buffer(context, begin(h_b), end(h_b), true);
d_c = cl::Buffer(context, begin(h_c), end(h_c), true);
d_r = cl::Buffer(context, CL_MEM_WRITE_ONLY, sizeof(float) * LENGTH);
vadd(
cl::EnqueueArgs(
queue,
cl::NDRange(count)),
d_a,
d_b,
d_c,
d_r,
count);
cl::copy(queue, d_r, begin(h_r), end(h_r));
// Test the results
int correct = 0;
float tmp;
for(int i = 0; i < count; i++)
{
tmp = h_a[i] + h_b[i] + h_c[i]; // assign element i of a+b+c to tmp
tmp -= h_r[i]; // compute deviation of expected and output result
if(tmp*tmp < TOL*TOL) // correct if square deviation is less than tolerance squared
correct++;
else {
printf(" tmp %f h_a %f h_b %f h_c %f h_r %f \n",tmp, h_a[i], h_b[i], h_c[i], h_r[i]);
}
}
// summarize results
printf("R = A+B+C: %d out of %d results were correct.\n", correct, count);
}
catch (cl::Error err) {
std::cout << "Exception\n";
std::cerr
<< "ERROR: "
<< err.what()
<< "("
<< err_code(err.err())
<< ")"
<< std::endl;
}
}
int main(void) {
BOOL_T err = CO_FALSE; /* error flag */
UNSIGNED8 temp = 0;
uword time_lo = 0, time_hi = 0;
/*--- hardware initialization e.g SIO, Chip-Selects, ...----------*/
iniDevice();
IO_vInit();
USIC0_vInit();
USIC1_vInit();
USIC2_vInit();
#if HW_KPS == 1
USIC3_vInit();
#endif
//RTC init
DAVE_vUnlockProtecReg();
RTC_vInit();
NOP(); /* one cycle delay */
NOP(); /* one cycle delay */
DAVE_vLockProtecReg();
lNodeId = (~IO_uwReadPort(P2))&0xFF;
if (lNodeId > 127) { lNodeId = 127; }
temp = (~IO_uwReadPort(P10))&0x0F;
switch(temp){
case 0: bitRate = 10; break;
case 1: bitRate = 20; break;
case 2: bitRate = 50; break;
case 3: bitRate = 125; break;
case 4: bitRate = 250; break;
case 5: bitRate = 500; break;
case 0x9:
case 0xE:
case 0xF:
bitRate = 125;
if (((~IO_uwReadPort(P2)) & 0xFF) == 0){ lNodeId = 127; }
break;
default: bitRate = 500; break;
}
initCan(bitRate);
init_Library();
initTimer();
Start_CAN();
ENABLE_CPU_INTERRUPTS();
RTC_vSetTime(0,0);
sw_init();
//PRINTF("loop\n");
while (err == CO_FALSE) {
FlushMbox(); /* Do the CANopen job */
myRTC_ulGetTime(&time_lo, &time_hi);
//odczyt temperatury wewn.
status_wewn[2] = ad7814_read(U2C0, SPI_CS3);
sw_loop();
//sygnalizacja na diodach
if (blink(time_lo, ch1_led_st)) { //St1
IO_vResetPin(IO_P0_0_LED_ST1);
} else {
IO_vSetPin(IO_P0_0_LED_ST1);
}
if (blink(time_lo, ch2_led_st)) { //St2
IO_vResetPin(IO_P4_1_LED_ST2);
} else {
IO_vSetPin(IO_P4_1_LED_ST2);
}
if (blink(time_lo, ch1_led_err)) { //Err1
IO_vResetPin(IO_P4_2_LED_ERR1);
} else {
IO_vSetPin(IO_P4_2_LED_ERR1);
}
if (blink(time_lo, ch2_led_err)) { //Err2
IO_vResetPin(IO_P4_0_LED_ERR2);
} else {
IO_vSetPin(IO_P4_0_LED_ERR2);
}
//w funkcji err_code dopisac cykliczna informacje o kilku bledach (numer bledu jako bity w slowie "dev_led_st" a nie wartosc)
if (err_code(time_lo, dev_led_st)) { //err code - Yellow diode
IO_vResetPin(IO_P2_8_LED_C);
} else {
IO_vSetPin(IO_P2_8_LED_C);
}
/* give a chance to finish the loop */
err = endLoop();
}
PRINTF("\nSTOP\n");
Stop_CAN();
DISABLE_CPU_INTERRUPTS();
releaseTimer();
ResetIntMask();
deinit_Library();
return 0;
}
int main(void)
{
int Mdim, Ndim, Pdim; // A[N][P], B[P][M], C[N][M]
int szA, szB, szC; // Number of elements in each matrix
double start_time; // Starting time
double run_time; // Timing
util::Timer timer; // Timing
Ndim = ORDER;
Pdim = ORDER;
Mdim = ORDER;
szA = Ndim * Pdim;
szB = Pdim * Mdim;
szC = Ndim * Mdim;
std::vector<float> h_A(szA); // Host memory for Matrix A
std::vector<float> h_B(szB); // Host memory for Matrix B
std::vector<float> h_C(szC); // Host memory for Matrix C
cl::Buffer d_a, d_b, d_c; // Matrices in device memory
initmat(Mdim, Ndim, Pdim, h_A, h_B, h_C);
timer.reset();
printf("\n===== Sequential, matrix mult (dot prod), order %d on host CPU ======\n",ORDER);
for(int i = 0; i < COUNT; i++)
{
zero_mat(Ndim, Mdim, h_C);
start_time = static_cast<double>(timer.getTimeMilliseconds()) / 1000.0;
seq_mat_mul_sdot(Mdim, Ndim, Pdim, h_A, h_B, h_C);
run_time = static_cast<double>(timer.getTimeMilliseconds()) / 1000.0 - start_time;
results(Mdim, Ndim, Pdim, h_C, run_time);
}
try
{
//--------------------------------------------------------------------------------
// Create a context and queue for DEVICE
//--------------------------------------------------------------------------------
cl::Context context(DEVICE);
cl::CommandQueue queue(context);
//--------------------------------------------------------------------------------
// Setup the buffers, initialize matrices, and write them into global memory
//--------------------------------------------------------------------------------
// Reset A, B and C matrices (just to play it safe)
initmat(Mdim, Ndim, Pdim, h_A, h_B, h_C);
d_a = cl::Buffer(context, begin(h_A), end(h_A), true);
d_b = cl::Buffer(context, begin(h_B), end(h_B), true);
d_c = cl::Buffer(context, CL_MEM_WRITE_ONLY, sizeof(float) * szC);
//--------------------------------------------------------------------------------
// OpenCL matrix multiplication ... Naive
//--------------------------------------------------------------------------------
// Create the compute program from the source buffer
cl::Program program(context, kernelsource, true);
// Create the compute kernel from the program
auto naive_mmul = cl::make_kernel<int, int, int, cl::Buffer, cl::Buffer, cl::Buffer>(program, "mmul");
printf("\n===== OpenCL, matrix mult, C(i,j) per work item, order %d ======\n",Ndim);
// Do the multiplication COUNT times
for (int i = 0; i < COUNT; i++)
{
zero_mat(Ndim, Mdim, h_C);
start_time = static_cast<double>(timer.getTimeMilliseconds()) / 1000.0;
// Execute the kernel over the entire range of C matrix elements ... computing
// a dot product for each element of the product matrix. The local work
// group size is set to NULL ... so I'm telling the OpenCL runtime to
// figure out a local work group size for me.
cl::NDRange global(Ndim, Mdim);
naive_mmul(cl::EnqueueArgs(queue, global),
Mdim, Ndim, Pdim, d_a, d_b, d_c);
queue.finish();
run_time = static_cast<double>(timer.getTimeMilliseconds()) / 1000.0 - start_time;
cl::copy(queue, d_c, begin(h_C), end(h_C));
results(Mdim, Ndim, Pdim, h_C, run_time);
} // end for loop
} catch (cl::Error err)
{
std::cout << "Exception\n";
std::cerr << "ERROR: "
<< err.what()
<< "("
<< err_code(err.err())
<< ")"
<< std::endl;
}
return EXIT_SUCCESS;
}
BLARGG_EXPORT const char* fex_err_details( fex_err_t err )
{
// If we don't have error code assigned, return entire string
return (err_code( err ) >= 0 ? blargg_err_details( err ) : blargg_err_str( err ));
}
int main(int argc, char *argv[])
{
int N; // A[N][N], B[N][N], C[N][N]
int size; // Number of elements in each matrix
double start_time; // Starting time
double run_time; // Timing
util::Timer timer; // Timing
N = ORDER;
size = N * N;
std::vector<float> h_A(size); // Host memory for Matrix A
std::vector<float> h_B(size); // Host memory for Matrix B
std::vector<float> h_C(size); // Host memory for Matrix C
cl::Buffer d_a, d_b, d_c; // Matrices in device memory
//--------------------------------------------------------------------------------
// Create a context and queue
//--------------------------------------------------------------------------------
try
{
cl_uint deviceIndex = 0;
parseArguments(argc, argv, &deviceIndex);
// Get list of devices
std::vector<cl::Device> devices;
unsigned numDevices = getDeviceList(devices);
// Check device index in range
if (deviceIndex >= numDevices)
{
std::cout << "Invalid device index (try '--list')\n";
return EXIT_FAILURE;
}
cl::Device device = devices[deviceIndex];
std::string name;
getDeviceName(device, name);
std::cout << "\nUsing OpenCL device: " << name << "\n";
std::vector<cl::Device> chosen_device;
chosen_device.push_back(device);
cl::Context context(chosen_device);
cl::CommandQueue queue(context, device);
//--------------------------------------------------------------------------------
// Run sequential matmul
//--------------------------------------------------------------------------------
initmat(N, h_A, h_B, h_C);
timer.reset();
printf("\n===== Sequential, matrix mult (dot prod), order %d on host CPU ======\n",N);
for(int i = 0; i < COUNT; i++)
{
zero_mat(N, h_C);
start_time = static_cast<double>(timer.getTimeMilliseconds()) / 1000.0;
seq_mat_mul_sdot(N, h_A, h_B, h_C);
run_time = static_cast<double>(timer.getTimeMilliseconds()) / 1000.0 - start_time;
results(N, h_C, run_time);
}
//--------------------------------------------------------------------------------
// Setup the buffers, initialize matrices, and write them into global memory
//--------------------------------------------------------------------------------
// Reset A, B and C matrices (just to play it safe)
initmat(N, h_A, h_B, h_C);
d_a = cl::Buffer(context, h_A.begin(), h_A.end(), true);
d_b = cl::Buffer(context, h_B.begin(), h_B.end(), true);
d_c = cl::Buffer(context, CL_MEM_WRITE_ONLY, sizeof(float) * size);
//--------------------------------------------------------------------------------
// OpenCL matrix multiplication ... Naive
//--------------------------------------------------------------------------------
// Create the compute program from the source buffer
cl::Program program(context, kernelsource, true);
// Create the compute kernel from the program
cl::make_kernel<int, cl::Buffer, cl::Buffer, cl::Buffer> naive_mmul(program, "mmul");
printf("\n===== OpenCL, matrix mult, C(i,j) per work item, order %d ======\n",N);
// Do the multiplication COUNT times
for (int i = 0; i < COUNT; i++)
{
zero_mat(N, h_C);
start_time = static_cast<double>(timer.getTimeMilliseconds()) / 1000.0;
// Execute the kernel over the entire range of C matrix elements ... computing
// a dot product for each element of the product matrix. The local work
// group size is set to NULL ... so I'm telling the OpenCL runtime to
// figure out a local work group size for me.
cl::NDRange global(N, N);
naive_mmul(cl::EnqueueArgs(queue, global),
N, d_a, d_b, d_c);
queue.finish();
run_time = static_cast<double>(timer.getTimeMilliseconds()) / 1000.0 - start_time;
cl::copy(queue, d_c, h_C.begin(), h_C.end());
results(N, h_C, run_time);
} // end for loop
} catch (cl::Error err)
{
std::cout << "Exception\n";
std::cerr << "ERROR: "
<< err.what()
<< "("
<< err_code(err.err())
<< ")"
<< std::endl;
}
return EXIT_SUCCESS;
}
int main(void)
{
try
{
// Discover number of platforms
std::vector<cl::Platform> platforms;
cl::Platform::get(&platforms);
std::cout << "\nNumber of OpenCL plaforms: " << platforms.size() << std::endl;
// Investigate each platform
std::cout << "\n-------------------------" << std::endl;
for (std::vector<cl::Platform>::iterator plat = platforms.begin(); plat != platforms.end(); plat++)
{
std::string s;
plat->getInfo(CL_PLATFORM_NAME, &s);
std::cout << "Platform: " << s << std::endl;
plat->getInfo(CL_PLATFORM_VENDOR, &s);
std::cout << "\tVendor: " << s << std::endl;
plat->getInfo(CL_PLATFORM_VERSION, &s);
std::cout << "\tVersion: " << s << std::endl;
// Discover number of devices
std::vector<cl::Device> devices;
plat->getDevices(CL_DEVICE_TYPE_ALL, &devices);
std::cout << "\n\tNumber of devices: " << devices.size() << std::endl;
// Investigate each device
for (std::vector<cl::Device>::iterator dev = devices.begin(); dev != devices.end(); dev++ )
{
std::cout << "\t-------------------------" << std::endl;
dev->getInfo(CL_DEVICE_NAME, &s);
std::cout << "\t\tName: " << s << std::endl;
dev->getInfo(CL_DEVICE_OPENCL_C_VERSION, &s);
std::cout << "\t\tVersion: " << s << std::endl;
int i;
dev->getInfo(CL_DEVICE_MAX_COMPUTE_UNITS, &i);
std::cout << "\t\tMax. Compute Units: " << i << std::endl;
size_t size;
dev->getInfo(CL_DEVICE_LOCAL_MEM_SIZE, &size);
std::cout << "\t\tLocal Memory Size: " << size/1024 << " KB" << std::endl;
dev->getInfo(CL_DEVICE_GLOBAL_MEM_SIZE, &size);
std::cout << "\t\tGlobal Memory Size: " << size/(1024*1024) << " MB" << std::endl;
dev->getInfo(CL_DEVICE_MAX_MEM_ALLOC_SIZE, &size);
std::cout << "\t\tMax Alloc Size: " << size/(1024*1024) << " MB" << std::endl;
dev->getInfo(CL_DEVICE_MAX_WORK_GROUP_SIZE, &size);
std::cout << "\t\tMax Work-group Total Size: " << size << std::endl;
std::vector<size_t> d;
dev->getInfo(CL_DEVICE_MAX_WORK_ITEM_SIZES, &d);
std::cout << "\t\tMax Work-group Dims: (";
for (std::vector<size_t>::iterator st = d.begin(); st != d.end(); st++)
std::cout << *st << " ";
std::cout << "\x08)" << std::endl;
std::cout << "\t-------------------------" << std::endl;
}
std::cout << "\n-------------------------\n";
}
}
catch (cl::Error err)
{
std::cout << "OpenCL Error: " << err.what() << " returned " << err_code(err.err()) << std::endl;
std::cout << "Check cl.h for error codes." << std::endl;
system("pause");
exit(-1);
}
system("pause");
return 0;
}
int main(int argc, char** argv)
{
if (argc != 2)
{
printf("Usage: ./pi_vocl num\n");
printf("\twhere num = 1, 4 or 8\n");
return EXIT_FAILURE;
}
int vector_size = atoi(argv[1]);
// Define some vector size specific constants
unsigned int ITERS, WGS;
if (vector_size == 1)
{
ITERS = 262144;
WGS = 8;
}
else if (vector_size == 4)
{
ITERS = 262144 / 4;
WGS = 32;
}
else if (vector_size == 8)
{
ITERS = 262144 / 8;
WGS = 64;
}
else
{
fprintf(stderr, "Invalid vector size\n");
return EXIT_FAILURE;
}
// Set some default values:
// Default number of steps (updated later to device preferable)
unsigned int in_nsteps = INSTEPS;
// Defaultl number of iterations
unsigned int niters = ITERS;
unsigned int work_group_size = WGS;
// Create context, queue and build program
cl_int err;
cl_context context;
cl_device_id device;
cl_command_queue queue;
cl_program program;
cl_kernel kernel;
// Find number of platforms
cl_uint numPlatforms;
err = clGetPlatformIDs(0, NULL, &numPlatforms);
if (err != CL_SUCCESS) die(err_code(err), __LINE__, __FILE__);
// Get all platforms
cl_platform_id platforms[numPlatforms];
err = clGetPlatformIDs(numPlatforms, platforms, NULL);
if (err != CL_SUCCESS) die(err_code(err), __LINE__, __FILE__);
// Secure a device
for (int i = 0; i < numPlatforms; i++)
{
err = clGetDeviceIDs(platforms[i], DEVICE, 1, &device, NULL);
if (err == CL_SUCCESS)
break;
}
if (device == NULL) die(err_code(err), __LINE__, __FILE__);
// Create a compute context
context = clCreateContext(0, 1, &device, NULL, NULL, &err);
if (err != CL_SUCCESS) die(err_code(err), __LINE__, __FILE__);
// Create a command queue
queue = clCreateCommandQueue(context, device, 0, &err);
if (err != CL_SUCCESS) die(err_code(err), __LINE__, __FILE__);
// Create the compute program from the source buffer
char *kernel_source = getKernelSource("../pi_vocl.cl");
program = clCreateProgramWithSource(context, 1, (const char**)&kernel_source, NULL, &err);
if (err != CL_SUCCESS) die(err_code(err), __LINE__, __FILE__);
// Build the program
err = clBuildProgram(program, 0, NULL, NULL, NULL, NULL);
if (err != CL_SUCCESS)
{
size_t len;
char buffer[2048];
clGetProgramBuildInfo(program, device, CL_PROGRAM_BUILD_LOG, sizeof(buffer), buffer, &len);
printf("%s\n", buffer);
die(err_code(err), __LINE__, __FILE__);
}
if (vector_size == 1)
{
kernel = clCreateKernel(program, "pi", &err);
if (err != CL_SUCCESS) die(err_code(err), __LINE__, __FILE__);
}
else if (vector_size == 4)
{
kernel = clCreateKernel(program, "pi_vec4", &err);
if (err != CL_SUCCESS) die(err_code(err), __LINE__, __FILE__);
}
else if (vector_size == 8)
{
kernel = clCreateKernel(program, "pi_vec8", &err);
if (err != CL_SUCCESS) die(err_code(err), __LINE__, __FILE__);
}
// Now that we know the size of the work_groups, we can set the number of work
// groups, the actual number of steps, and the step size
unsigned int nwork_groups = in_nsteps/(work_group_size*niters);
// Get the max work group size for the kernel pi on our device
size_t max_size;
err = clGetKernelWorkGroupInfo(kernel, device, CL_KERNEL_WORK_GROUP_SIZE, sizeof(max_size), &max_size, NULL);
if (err != CL_SUCCESS) die(err_code(err), __LINE__, __FILE__);
if (max_size > work_group_size)
{
work_group_size = max_size;
nwork_groups = in_nsteps/(nwork_groups*niters);
}
if (nwork_groups < 1)
{
err = clGetDeviceInfo(device, CL_DEVICE_MAX_COMPUTE_UNITS, sizeof(nwork_groups), &nwork_groups, NULL);
if (err != CL_SUCCESS) die(err_code(err), __LINE__, __FILE__);
work_group_size = in_nsteps/(nwork_groups*niters);
}
unsigned int nsteps = work_group_size * niters * nwork_groups;
float step_size = 1.0f / (float) nsteps;
// Array to hold partial sum
float *h_psum = (float*)calloc(nwork_groups, sizeof(float));
printf("%d work groups of size %d.\n", nwork_groups, work_group_size);
printf(" %u Integration steps\n", nsteps);
cl_mem d_partial_sums = clCreateBuffer(context, CL_MEM_WRITE_ONLY, sizeof(float) * nwork_groups, NULL, &err);
if (err != CL_SUCCESS) die(err_code(err), __LINE__, __FILE__);
// Execute the kernel over the entire range of our 1d input data et
// using the maximum number of work group items for this device
const size_t global = nwork_groups * work_group_size;
const size_t local = work_group_size;
err = clSetKernelArg(kernel, 0, sizeof(int), &niters);
if (err != CL_SUCCESS) die(err_code(err), __LINE__, __FILE__);
err = clSetKernelArg(kernel, 1, sizeof(float), &step_size);
if (err != CL_SUCCESS) die(err_code(err), __LINE__, __FILE__);
err = clSetKernelArg(kernel, 2, sizeof(float) * work_group_size, NULL);
if (err != CL_SUCCESS) die(err_code(err), __LINE__, __FILE__);
err = clSetKernelArg(kernel, 3, sizeof(cl_mem), &d_partial_sums);
if (err != CL_SUCCESS) die(err_code(err), __LINE__, __FILE__);
// Start the timer
double rtime = wtime();
err = clEnqueueNDRangeKernel(
queue, kernel,
1, NULL, &global, &local,
0, NULL, NULL);
if (err != CL_SUCCESS) die(err_code(err), __LINE__, __FILE__);
err = clEnqueueReadBuffer(queue, d_partial_sums, CL_TRUE, 0,
sizeof(float) * nwork_groups, h_psum, 0, NULL, NULL);
if (err != CL_SUCCESS) die(err_code(err), __LINE__, __FILE__);
// complete the sum and compute the final integral value on the host
float pi_res = 0.0f;
for (unsigned int i = 0; i < nwork_groups; i++)
{
pi_res += h_psum[i];
}
pi_res *= step_size;
rtime = wtime() - rtime;
printf("\nThe calculation ran in %lf seconds\n", rtime);
printf(" pi = %f for %u steps\n", pi_res, nsteps);
free(h_psum);
free(kernel_source);
}
int dusb_dissect_cmd_data(CalcModel model, FILE *f, const uint8_t * data, uint32_t len, uint32_t vtl_size, uint16_t vtl_type)
{
int ret = dusb_check_cmd_data(model, data, len, vtl_size, vtl_type);
(void)vtl_size;
if (ret)
{
return ret;
}
switch (vtl_type)
{
case DUSB_VPKT_PING:
{
uint16_t arg1 = (((uint16_t)(data[0])) << 8) | ((uint16_t)(data[1]));
uint16_t arg2 = (((uint16_t)(data[2])) << 8) | ((uint16_t)(data[3]));
uint16_t arg3 = (((uint16_t)(data[4])) << 8) | ((uint16_t)(data[5]));
uint16_t arg4 = (((uint16_t)(data[6])) << 8) | ((uint16_t)(data[7]));
uint16_t arg5 = (((uint16_t)(data[8])) << 8) | ((uint16_t)(data[9]));
fprintf(f, "Set mode: { %u, %u, %u, %u, 0x%04X }\n", arg1, arg2, arg3, arg4, arg5);
}
break;
case DUSB_VPKT_PARM_REQ:
{
uint16_t npids = (((uint16_t)(data[0])) << 8) | ((uint16_t)(data[1]));
uint16_t i;
if (len == 2U + npids * 2)
{
data += 2;
fprintf(f, "Requested %u (%X) parameter IDs:\n", npids, npids);
for (i = 0; i < npids; i++)
{
uint16_t pid = (((uint16_t)(data[0])) << 8) | ((uint16_t)(data[1]));
data += 2;
fprintf(f, "\t%04X (%s)\n", pid, dusb_cmd_param_type2name(pid));
}
fputc('\n', f);
}
}
break;
case DUSB_VPKT_PARM_DATA:
{
uint16_t nparams = (((uint16_t)(data[0])) << 8) | ((uint16_t)(data[1]));
uint16_t i;
uint32_t additional_size = 0;
if (len >= 2U + 3 * nparams)
{
data += 2;
fprintf(f, "Received %u (%X) parameter values:\n", nparams, nparams);
for (i = 0; i < nparams; i++)
{
uint16_t pid = (((uint16_t)(data[0])) << 8) | ((uint16_t)(data[1]));
uint8_t ok;
data += 2;
ok = !(*data++);
fprintf(f, "\t%04X (%s): ", pid, dusb_cmd_param_type2name(pid));
if (ok)
{
uint16_t j;
uint16_t size = (((uint16_t)(data[0])) << 8) | ((uint16_t)(data[1]));
data += 2;
additional_size += size + 2;
if (len < 2U + 3 * nparams + additional_size)
{
break;
}
fprintf(f, "OK, size %04X\n\t\t", size);
for (j = 0; j < size;)
{
fprintf(f, "%02X ", *data++);
if (!(++j & 15))
{
fprintf(f, "\n\t\t");
}
}
fputc('\n', f);
}
else
{
fputs("NOK !\n", f);
}
}
}
// else do nothing.
}
break;
case DUSB_VPKT_PARM_SET:
{
uint16_t id = (((uint16_t)(data[0])) << 8) | ((uint16_t)(data[1]));
uint16_t size = (((uint16_t)(data[2])) << 8) | ((uint16_t)(data[3]));
uint16_t i;
data += 4;
if (len == 4U + size)
{
fprintf(f, "Sending value of size %04X for parameter %04X\n\t", size, id);
for (i = 0; i < size; i++)
{
fprintf(f, "%02X ", *data++);
if (!(++i & 15))
{
fprintf(f, "\n\t");
}
}
}
// else do nothing.
}
break;
case DUSB_VPKT_OS_BEGIN:
{
uint32_t size = (((uint32_t)data[7]) << 24) | (((uint32_t)data[8]) << 16) | (((uint32_t)data[9]) << 8) | (((uint32_t)data[10]) << 0);
fprintf(f, "Size: %lu / %08lX\n", (unsigned long)size, (unsigned long)size);
}
break;
case DUSB_VPKT_OS_ACK:
{
uint32_t size = (((uint32_t)data[0]) << 24) | (((uint32_t)data[1]) << 16) | (((uint32_t)data[2]) << 8) | (((uint32_t)data[3]) << 0);
fprintf(f, "Chunk size: %lu / %08lX\n", (unsigned long)size, (unsigned long)size);
}
break;
case DUSB_VPKT_OS_HEADER:
case DUSB_VPKT_OS_DATA:
{
if (model == CALC_TI83PCE_USB || model == CALC_TI84PCE_USB)
{
uint32_t addr = (((uint32_t)data[3]) << 24) | (((uint32_t)data[2]) << 16) | (((uint32_t)data[1]) << 8) | (((uint32_t)data[0]) << 0);
fprintf(f, "Address: %08lX\n", (unsigned long)addr);
}
else if (model != CALC_TI89T_USB)
{
uint16_t addr = (((uint16_t)data[0]) << 8) | (((uint32_t)data[1]) << 0);
fprintf(f, "Address: %04X\tPage: %02X\tFlag: %02X\n", addr, data[2], data[3]);
}
// else do nothing.
}
break;
case DUSB_VPKT_DELAY_ACK:
{
uint32_t delay = (((uint32_t)data[0]) << 24) | (((uint32_t)data[1]) << 16) | (((uint32_t)data[2]) << 8) | (data[3] << 0);
fprintf(f, "Delay: %lu\n", (unsigned long)delay);
}
break;
case DUSB_VPKT_ERROR:
{
int err = err_code((((uint16_t)data[0]) << 8) | (((uint32_t)data[1]) << 0));
fprintf(f, "Error code: %u (%04X)\n", err, err);
}
break;
// Nothing to do.
case DUSB_VPKT_VAR_CNTS:
case DUSB_VPKT_MODE_SET:
case DUSB_VPKT_EOT_ACK:
case DUSB_VPKT_DATA_ACK:
case DUSB_VPKT_EOT:
break;
// TODO
case DUSB_VPKT_DIR_REQ:
case DUSB_VPKT_VAR_HDR:
case DUSB_VPKT_RTS:
case DUSB_VPKT_VAR_REQ:
case DUSB_VPKT_MODIF_VAR:
case DUSB_VPKT_EXECUTE:
{
fputs("(no extra dissection performed for now)\n", f);
}
break;
default:
{
fputs("(not performing extra dissection on unknown vpkt type)\n", f);
}
break;
}
return ret;
}
int main(int argc, char *argv[])
{
float *h_psum; // vector to hold partial sum
int in_nsteps = INSTEPS; // default number of steps (updated later to device preferable)
int niters = ITERS; // number of iterations
int nsteps;
float step_size;
size_t nwork_groups;
size_t max_size, work_group_size = 8;
float pi_res;
cl_mem d_partial_sums;
char *kernelsource = getKernelSource("../pi_ocl.cl"); // Kernel source
cl_int err;
cl_device_id device; // compute device id
cl_context context; // compute context
cl_command_queue commands; // compute command queue
cl_program program; // compute program
cl_kernel kernel_pi; // compute kernel
// Set up OpenCL context, queue, kernel, etc.
cl_uint deviceIndex = 0;
parseArguments(argc, argv, &deviceIndex);
// Get list of devices
cl_device_id devices[MAX_DEVICES];
unsigned numDevices = getDeviceList(devices);
// Check device index in range
if (deviceIndex >= numDevices)
{
printf("Invalid device index (try '--list')\n");
return EXIT_FAILURE;
}
device = devices[deviceIndex];
char name[MAX_INFO_STRING];
getDeviceName(device, name);
printf("\nUsing OpenCL device: %s\n", name);
// Create a compute context
context = clCreateContext(0, 1, &device, NULL, NULL, &err);
checkError(err, "Creating context");
// Create a command queue
commands = clCreateCommandQueue(context, device, 0, &err);
checkError(err, "Creating command queue");
// Create the compute program from the source buffer
program = clCreateProgramWithSource(context, 1, (const char **) & kernelsource, NULL, &err);
checkError(err, "Creating program");
// Build the program
err = clBuildProgram(program, 0, NULL, NULL, NULL, NULL);
if (err != CL_SUCCESS)
{
size_t len;
char buffer[2048];
printf("Error: Failed to build program executable!\n%s\n", err_code(err));
clGetProgramBuildInfo(program, device, CL_PROGRAM_BUILD_LOG, sizeof(buffer), buffer, &len);
printf("%s\n", buffer);
return EXIT_FAILURE;
}
// Create the compute kernel from the program
kernel_pi = clCreateKernel(program, "pi", &err);
checkError(err, "Creating kernel");
// Find kernel work-group size
err = clGetKernelWorkGroupInfo (kernel_pi, device, CL_KERNEL_WORK_GROUP_SIZE, sizeof(size_t), &work_group_size, NULL);
checkError(err, "Getting kernel work group info");
// Now that we know the size of the work-groups, we can set the number of
// work-groups, the actual number of steps, and the step size
nwork_groups = in_nsteps/(work_group_size*niters);
if (nwork_groups < 1)
{
err = clGetDeviceInfo(device, CL_DEVICE_MAX_COMPUTE_UNITS, sizeof(size_t), &nwork_groups, NULL);
checkError(err, "Getting device compute unit info");
work_group_size = in_nsteps / (nwork_groups * niters);
}
nsteps = work_group_size * niters * nwork_groups;
step_size = 1.0f/(float)nsteps;
printf("nsteps:%d\n", nsteps);
printf("niters:%d\n", niters);
printf("work_group_size:%zd\n", work_group_size);
printf("n work groups:%ld\n", nwork_groups);
printf("step_size:%f\n", step_size);
h_psum = calloc(sizeof(float), nwork_groups);
if (!h_psum)
{
printf("Error: could not allocate host memory for h_psum\n");
return EXIT_FAILURE;
}
printf(" %ld work-groups of size %ld. %d Integration steps\n",
nwork_groups,
work_group_size,
nsteps);
d_partial_sums = clCreateBuffer(context, CL_MEM_WRITE_ONLY, sizeof(float) * nwork_groups, NULL, &err);
checkError(err, "Creating buffer d_partial_sums");
// Set kernel arguments
err = clSetKernelArg(kernel_pi, 0, sizeof(int), &niters);
err |= clSetKernelArg(kernel_pi, 1, sizeof(float), &step_size);
err |= clSetKernelArg(kernel_pi, 2, sizeof(float) * work_group_size, NULL);
err |= clSetKernelArg(kernel_pi, 3, sizeof(cl_mem), &d_partial_sums);
checkError(err, "Settin kernel args");
// Execute the kernel over the entire range of our 1D input data set
// using the maximum number of work items for this device
size_t global = nsteps / niters;
size_t local = work_group_size;
double rtime = wtime();
err = clEnqueueNDRangeKernel(
commands,
kernel_pi,
1, NULL,
&global,
&local,
0, NULL, NULL);
checkError(err, "Enqueueing kernel");
err = clEnqueueReadBuffer(
commands,
d_partial_sums,
CL_TRUE,
0,
sizeof(float) * nwork_groups,
h_psum,
0, NULL, NULL);
checkError(err, "Reading back d_partial_sums");
// complete the sum and compute the final integral value on the host
pi_res = 0.0f;
for (unsigned int i = 0; i < nwork_groups; i++)
{
pi_res += h_psum[i];
}
pi_res *= step_size;
rtime = wtime() - rtime;
printf("\nThe calculation ran in %lf seconds\n", rtime);
printf(" pi = %f for %d steps\n", pi_res, nsteps);
// clean up
clReleaseMemObject(d_partial_sums);
clReleaseProgram(program);
clReleaseKernel(kernel_pi);
clReleaseCommandQueue(commands);
clReleaseContext(context);
free(kernelsource);
free(h_psum);
}
int main(int argc, char *argv[])
{
float *h_A; // A matrix
float *h_B; // B matrix
float *h_C; // C = A*B matrix
int N; // A[N][N], B[N][N], C[N][N]
int size; // number of elements in each matrix
cl_mem d_a, d_b, d_c; // Matrices in device memory
double start_time; // Starting time
double run_time; // timing data
char * kernelsource; // kernel source string
cl_int err; // error code returned from OpenCL calls
cl_device_id device; // compute device id
cl_context context; // compute context
cl_command_queue commands; // compute command queue
cl_program program; // compute program
cl_kernel kernel; // compute kernel
N = ORDER;
size = N * N;
h_A = (float *)malloc(size * sizeof(float));
h_B = (float *)malloc(size * sizeof(float));
h_C = (float *)malloc(size * sizeof(float));
//--------------------------------------------------------------------------------
// Create a context, queue and device.
//--------------------------------------------------------------------------------
cl_uint deviceIndex = 0;
parseArguments(argc, argv, &deviceIndex);
// Get list of devices
cl_device_id devices[MAX_DEVICES];
unsigned numDevices = getDeviceList(devices);
// Check device index in range
if (deviceIndex >= numDevices)
{
printf("Invalid device index (try '--list')\n");
return EXIT_FAILURE;
}
device = devices[deviceIndex];
char name[MAX_INFO_STRING];
getDeviceName(device, name);
printf("\nUsing OpenCL device: %s\n", name);
// Create a compute context
context = clCreateContext(0, 1, &device, NULL, NULL, &err);
checkError(err, "Creating context");
// Create a command queue
commands = clCreateCommandQueue(context, device, 0, &err);
checkError(err, "Creating command queue");
//--------------------------------------------------------------------------------
// Run sequential version on the host
//--------------------------------------------------------------------------------
initmat(N, h_A, h_B, h_C);
printf("\n===== Sequential, matrix mult (dot prod), order %d on host CPU ======\n",ORDER);
for(int i = 0; i < COUNT; i++)
{
zero_mat(N, h_C);
start_time = wtime();
seq_mat_mul_sdot(N, h_A, h_B, h_C);
run_time = wtime() - start_time;
results(N, h_C, run_time);
}
//--------------------------------------------------------------------------------
// Setup the buffers, initialize matrices, and write them into global memory
//--------------------------------------------------------------------------------
// Reset A, B and C matrices (just to play it safe)
initmat(N, h_A, h_B, h_C);
d_a = clCreateBuffer(context, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR,
sizeof(float) * size, h_A, &err);
checkError(err, "Creating buffer d_a");
d_b = clCreateBuffer(context, CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR,
sizeof(float) * size, h_B, &err);
checkError(err, "Creating buffer d_b");
d_c = clCreateBuffer(context, CL_MEM_WRITE_ONLY,
sizeof(float) * size, NULL, &err);
checkError(err, "Creating buffer d_c");
//--------------------------------------------------------------------------------
// OpenCL matrix multiplication ... Naive
//--------------------------------------------------------------------------------
kernelsource = getKernelSource("../C_elem.cl");
// Create the comput program from the source buffer
program = clCreateProgramWithSource(context, 1, (const char **) & kernelsource, NULL, &err);
checkError(err, "Creating program with C_elem.cl");
free(kernelsource);
// Build the program
err = clBuildProgram(program, 0, NULL, NULL, NULL, NULL);
if (err != CL_SUCCESS)
{
size_t len;
char buffer[2048];
printf("Error: Failed to build program executable!\n%s\n", err_code(err));
clGetProgramBuildInfo(program, device, CL_PROGRAM_BUILD_LOG, sizeof(buffer), buffer, &len);
printf("%s\n", buffer);
return EXIT_FAILURE;
}
// Create the compute kernel from the program
kernel = clCreateKernel(program, "mmul", &err);
checkError(err, "Creating kernel from C_elem.cl");
printf("\n===== OpenCL, matrix mult, C(i,j) per work item, order %d ======\n",N);
// Do the multiplication COUNT times
for (int i = 0; i < COUNT; i++)
{
zero_mat(N, h_C);
err = clSetKernelArg(kernel, 0, sizeof(int), &N);
err |= clSetKernelArg(kernel, 1, sizeof(cl_mem), &d_a);
err |= clSetKernelArg(kernel, 2, sizeof(cl_mem), &d_b);
err |= clSetKernelArg(kernel, 3, sizeof(cl_mem), &d_c);
checkError(err, "Setting kernel args");
start_time = wtime();
// Execute the kernel over the entire range of C matrix elements ... computing
// a dot product for each element of the product matrix. The local work
// group size is set to NULL ... so I'm telling the OpenCL runtime to
// figure out a local work group size for me.
const size_t global[2] = {N, N};
err = clEnqueueNDRangeKernel(
commands,
kernel,
2, NULL,
global, NULL,
0, NULL, NULL);
checkError(err, "Enqueueing kernel");
err = clFinish(commands);
checkError(err, "Waiting for kernel to finish");
run_time = wtime() - start_time;
err = clEnqueueReadBuffer(
commands, d_c, CL_TRUE, 0,
sizeof(float) * size, h_C,
0, NULL, NULL);
checkError(err, "Copying back d_c");
results(N, h_C, run_time);
} // end for loop
//--------------------------------------------------------------------------------
// OpenCL matrix multiplication ... C row per work item
//--------------------------------------------------------------------------------
kernelsource = getKernelSource("../C_row.cl");
// Create the comput program from the source buffer
program = clCreateProgramWithSource(context, 1, (const char **) & kernelsource, NULL, &err);
checkError(err, "Creating program with C_row.cl");
free(kernelsource);
// Build the program
err = clBuildProgram(program, 0, NULL, NULL, NULL, NULL);
if (err != CL_SUCCESS)
{
size_t len;
char buffer[2048];
printf("Error: Failed to build program executable!\n%s\n", err_code(err));
clGetProgramBuildInfo(program, device, CL_PROGRAM_BUILD_LOG, sizeof(buffer), buffer, &len);
printf("%s\n", buffer);
return EXIT_FAILURE;
}
// Create the compute kernel from the program
kernel = clCreateKernel(program, "mmul", &err);
checkError(err, "Creating kernel from C_row.cl");
printf("\n===== OpenCL, matrix mult, C row per work item, order %d ======\n",N);
// Do the multiplication COUNT times
for (int i = 0; i < COUNT; i++)
{
zero_mat(N, h_C);
err = clSetKernelArg(kernel, 0, sizeof(int), &N);
err |= clSetKernelArg(kernel, 1, sizeof(cl_mem), &d_a);
err |= clSetKernelArg(kernel, 2, sizeof(cl_mem), &d_b);
err |= clSetKernelArg(kernel, 3, sizeof(cl_mem), &d_c);
checkError(err, "Setting kernel args");
start_time = wtime();
// Execute the kernel over the rows of the C matrix ... computing
// a dot product for each element of the product matrix.
const size_t global = N;
err = clEnqueueNDRangeKernel(
commands,
kernel,
1, NULL,
&global, NULL,
0, NULL, NULL);
checkError(err, "Enqueueing kernel");
err = clFinish(commands);
checkError(err, "Waiting for kernel to finish");
run_time = wtime() - start_time;
err = clEnqueueReadBuffer(
commands, d_c, CL_TRUE, 0,
sizeof(float) * size, h_C,
0, NULL, NULL);
checkError(err, "Reading back d_c");
results(N, h_C, run_time);
} // end for loop
//--------------------------------------------------------------------------------
// OpenCL matrix multiplication ... C row per work item, A row in pivate memory
//--------------------------------------------------------------------------------
kernelsource = getKernelSource("../C_row_priv.cl");
// Create the comput program from the source buffer
program = clCreateProgramWithSource(context, 1, (const char **) & kernelsource, NULL, &err);
checkError(err, "Creating program from C_row_priv.cl");
free(kernelsource);
// Build the program
err = clBuildProgram(program, 0, NULL, NULL, NULL, NULL);
if (err != CL_SUCCESS)
{
size_t len;
char buffer[2048];
printf("Error: Failed to build program executable!\n%s\n", err_code(err));
clGetProgramBuildInfo(program, device, CL_PROGRAM_BUILD_LOG, sizeof(buffer), buffer, &len);
printf("%s\n", buffer);
return EXIT_FAILURE;
}
// Create the compute kernel from the program
kernel = clCreateKernel(program, "mmul", &err);
checkError(err, "Creating kernel from C_row_priv.cl");
printf("\n===== OpenCL, matrix mult, C row, A row in priv mem, order %d ======\n",N);
// Do the multiplication COUNT times
for (int i = 0; i < COUNT; i++)
{
zero_mat(N, h_C);
err = clSetKernelArg(kernel, 0, sizeof(int), &N);
err |= clSetKernelArg(kernel, 1, sizeof(cl_mem), &d_a);
err |= clSetKernelArg(kernel, 2, sizeof(cl_mem), &d_b);
err |= clSetKernelArg(kernel, 3, sizeof(cl_mem), &d_c);
checkError(err, "Setting kernel args");
start_time = wtime();
// Execute the kernel over the rows of the C matrix ... computing
// a dot product for each element of the product matrix.
const size_t global = N;
const size_t local = ORDER / 16;
err = clEnqueueNDRangeKernel(
commands,
kernel,
1, NULL,
&global, &local,
0, NULL, NULL);
checkError(err, "Enqueueing kernel");
err = clFinish(commands);
checkError(err, "Waiting for kernel to finish");
run_time = wtime() - start_time;
err = clEnqueueReadBuffer(
commands, d_c, CL_TRUE, 0,
sizeof(float) * size, h_C,
0, NULL, NULL);
checkError(err, "Reading back d_c");
results(N, h_C, run_time);
} // end for loop
//--------------------------------------------------------------------------------
// Clean up!
//--------------------------------------------------------------------------------
free(h_A);
free(h_B);
free(h_C);
clReleaseMemObject(d_a);
clReleaseMemObject(d_b);
clReleaseMemObject(d_c);
clReleaseProgram(program);
clReleaseKernel(kernel);
clReleaseCommandQueue(commands);
clReleaseContext(context);
return EXIT_SUCCESS;
}
int main(int ac, char *av[])
{
char myname[HOSTNAMESIZE];
char *progname = av[0];
char *logfname;
unsigned version;
prod_class_t clss;
prod_spec spec;
int seq_start = 0;
int status;
ErrorObj* error;
unsigned remotePort = LDM_PORT;
logfname = "-";
remote = "localhost";
(void)set_timestamp(&clss.from);
clss.to = TS_ENDT;
clss.psa.psa_len = 1;
clss.psa.psa_val = &spec;
spec.feedtype = DEFAULT_FEEDTYPE;
spec.pattern = ".*";
{
extern int optind;
extern char *optarg;
int ch;
int logmask = (LOG_MASK(LOG_ERR) | LOG_MASK(LOG_WARNING) |
LOG_MASK(LOG_NOTICE));
while ((ch = getopt(ac, av, "vxl:h:f:P:s:")) != EOF)
switch (ch) {
case 'v':
logmask |= LOG_MASK(LOG_INFO);
break;
case 'x':
logmask |= LOG_MASK(LOG_DEBUG);
break;
case 'l':
logfname = optarg;
break;
case 'h':
remote = optarg;
break;
case 'f':
spec.feedtype = atofeedtypet(optarg);
if(spec.feedtype == NONE)
{
fprintf(stderr, "Unknown feedtype \"%s\"\n", optarg);
usage(progname);
}
break;
case 'P': {
char* suffix = "";
long port;
errno = 0;
port = strtol(optarg, &suffix, 0);
if (0 != errno || 0 != *suffix ||
0 >= port || 0xffff < port) {
(void)fprintf(stderr, "%s: invalid port %s\n",
av[0], optarg);
usage(av[0]);
}
remotePort = (unsigned)port;
break;
}
case 's':
seq_start = atoi(optarg);
break;
case '?':
usage(progname);
break;
}
ac -= optind; av += optind;
if(ac < 1) usage(progname);
(void) setulogmask(logmask);
}
/*
* Set up error logging
*/
(void) openulog(ubasename(progname), LOG_NOTIME, LOG_LDM, logfname);
/*
* register exit handler
*/
if(atexit(cleanup) != 0)
{
serror("atexit");
exit(1);
}
/*
* set up signal handlers
*/
set_sigactions();
(void) strcpy(myname, ghostname());
/*
* Contact the server.
*/
error = ldm_clnttcp_create_vers(remote, remotePort, SIX, &clnt,
NULL, NULL);
(void)exitIfDone(1);
if (!error) {
version = SIX;
hiya = my_hiya_6;
send_product = send_product_6;
nullproc = nullproc_6;
}
else if (LDM_CLNT_BAD_VERSION == err_code(error)) {
err_free(error);
error = ldm_clnttcp_create_vers(remote, remotePort, FIVE, &clnt,
NULL, NULL);
(void)exitIfDone(1);
if (!error) {
version = FIVE;
hiya = my_hiya_5;
send_product = send_product_5;
nullproc = NULL;
}
}
if (error) {
err_log(error, ERR_FAILURE);
err_free(error);
status = 1;
}
else {
udebug("version %u", version);
status = ldmsend(clnt, &clss, myname, seq_start, ac, av);
}
return status != 0;
}
int main(int argc, char *argv[])
{
static const struct option long_opts[] = {
/* commands */
{"inject", required_argument, NULL, c_INJECT},
{"remove", required_argument, NULL, c_REMOVE},
{"hexdump", required_argument, NULL, c_HEXDUMP},
/* options */
{"output", required_argument, NULL, o_OUTPUT},
/* flags */
{"force", no_argument, NULL, f_FORCE},
{"p8", no_argument, NULL, f_P8},
{"verbose", no_argument, NULL, f_VERBOSE},
{"help", no_argument, NULL, f_HELP},
{0, 0, 0, 0}
};
static const char *short_opts = "I:R:H:o:fpvh";
int rc = EXIT_FAILURE;
if (argc == 1)
usage(args.short_name, false), exit(rc);
int opt = 0, idx = 0;
while ((opt = getopt_long(argc, argv, short_opts, long_opts,
&idx)) != -1)
if (process_argument(&args, opt, optarg) < 0)
goto error;
/* getopt_long doesn't know what to do with orphans, */
/* so we'll scoop them up here, and deal with them later */
while (optind < argc)
if (process_option(&args, argv[optind++]) < 0)
goto error;
if (args.verbose == f_VERBOSE)
args_dump(&args);
if (validate_args(&args) < 0)
goto error;
if (process_args(&args) < 0)
goto error;
rc = EXIT_SUCCESS;
if (false) {
err_t *err;
error:
err = err_get();
assert(err != NULL);
fprintf(stderr, "%s: %s : %s(%d) : (code=%d) %.*s\n",
program_invocation_short_name,
err_type_name(err), err_file(err), err_line(err),
err_code(err), err_size(err), (char *)err_data(err));
}
return rc;
}
int main(int argc, char * argv[])
{
MStatus stat;
// initialise the maya library - This basically starts up Maya
if(MLibrary::initialize(argv[0], true) != MS::kSuccess)
{
err_stop("[ERROR] Maya failed to initialise\n");
}
std::string fileName = argv[1];
cout << "Loading file: " << fileName;
//std::string scale = "0.01";
std::string scale = "1.0";
std::string todo = "all";
std::string target = "tegra"; // or tegra
if (argc > 2)
{
scale = argv[2];
}
if (argc > 3)
{
todo = argv[3];
}
if (argc > 4)
{
target = argv[4];
}
Globals & globalFlags = Globals::GetGlobals();
globalFlags.TARGET = target;
globalFlags.SCALE = (float)atof(scale.c_str());
char txt[128]={0};
sprintf(txt,"Scale: %f\n",globalFlags.SCALE);
err_info(txt);
for (unsigned int i=0;i<fileName.length();i++)
if (fileName[i] == '\\')
fileName[i] = '/';
err_info("Opening file: " + fileName);
stat = MFileIO::open(fileName.c_str());
err_code(stat);
int dot = fileName.rfind(".");
int slash = fileName.rfind("/");
string folder = fileName;
folder.erase(slash, folder.length() - slash);
fileName.erase(dot, fileName.length() - dot);
fileName.erase(0, slash+1);
if (todo == "model" || todo == "all")
{
ModelExporter & modelExport = ModelExporter::GetExporter();
modelExport.Export(folder, fileName);
AnimationExport & animationExport = AnimationExport::GetExporter();
animationExport.Export(folder, fileName);
UVAnimationExport & uvAnimationExport = UVAnimationExport::GetExporter();
uvAnimationExport.Export(folder, fileName);
}
if (todo == "locator" || todo == "all")
{
LocatorExport & locatorExport = LocatorExport::GetExporter();
locatorExport.Export(folder, fileName);
}
if (todo == "ref" || todo == "all")
{
RefSceneExport & refExport = RefSceneExport::GetExporter();
refExport.Export(folder, fileName);
}
if (todo == "stop_circle" || todo == "all")
{
StopCircleExport & stopCircleExport = StopCircleExport::GetExporter();
stopCircleExport.Export(folder, fileName);
}
if (todo == "launcher" || todo == "all")
{
BotLauncherExport & launcherExport = BotLauncherExport::GetExporter();
launcherExport.Export(folder, fileName);
}
if (todo == "curve" || todo == "all")
{
CurveExport & curveExport = CurveExport::GetExporter();
curveExport.Export(folder, fileName);
}
if (todo == "wire" || todo == "all")
{
WireExport & wireExport = WireExport::GetExporter();
wireExport.Export(folder, fileName);
}
if (todo == "collision_mesh" || todo == "all")
{
CollisionMeshExport & cllisionMeshExport = CollisionMeshExport::GetExporter();
cllisionMeshExport.Export(folder, fileName);
}
if (todo == "graph_mesh" || todo == "all")
{
GraphExport & graphExport = GraphExport::GetExporter();
graphExport.Export(folder, fileName);
}
if (todo == "doors" || todo == "all")
{
DoorsExport & doorsExport = DoorsExport::GetExporter();
doorsExport.Export(folder, fileName);
}
if (todo == "physics" || todo == "all")
{
PhysicsExport & physicsExport = PhysicsExport::GetExporter();
physicsExport.Export(folder, fileName);
}
return 0;
}
int main(void)
{
float *h_psum; // vector to hold partial sum
int in_nsteps = INSTEPS; // default number of steps (updated later to device prefereable)
int niters = ITERS; // number of iterations
int nsteps;
float step_size;
::size_t nwork_groups;
::size_t max_size, work_group_size = 8;
float pi_res;
cl::Buffer d_partial_sums;
try
{
// Create a context
cl::Context context(DEVICE);
// Create the program object
cl::Program program(context, util::loadProgram("pi_ocl.cl"), true);
// Get the command queue
cl::CommandQueue queue(context);
// Create the kernel object for quering information
cl::Kernel ko_pi(program, "pi");
// Get the device we are using
std::vector<cl::Device> devices = context.getInfo<CL_CONTEXT_DEVICES>();
cl::Device device = devices[0];
// Get the work group size
work_group_size = ko_pi.getWorkGroupInfo<CL_KERNEL_WORK_GROUP_SIZE>(device);
//printf("wgroup_size = %lu\n", work_group_size);
auto pi = cl::make_kernel<int, float, cl::LocalSpaceArg, cl::Buffer>(program, "pi");
// Now that we know the size of the work_groups, we can set the number of work
// groups, the actual number of steps, and the step size
nwork_groups = in_nsteps/(work_group_size*niters);
if ( nwork_groups < 1) {
nwork_groups =
device.getInfo<CL_DEVICE_MAX_COMPUTE_UNITS>();
work_group_size=in_nsteps / (nwork_groups*niters);
}
nsteps = work_group_size * niters * nwork_groups;
step_size = 1.0f/static_cast<float>(nsteps);
std::vector<float> h_psum(nwork_groups);
printf(
" %d work groups of size %d. %d Integration steps\n",
(int)nwork_groups,
(int)work_group_size,
nsteps);
d_partial_sums = cl::Buffer(context, CL_MEM_WRITE_ONLY, sizeof(float) * nwork_groups);
util::Timer timer;
// Execute the kernel over the entire range of our 1d input data set
// using the maximum number of work group items for this device
pi(
cl::EnqueueArgs(
queue,
cl::NDRange(nwork_groups * work_group_size),
cl::NDRange(work_group_size)),
niters,
step_size,
cl::Local(sizeof(float) * work_group_size),
d_partial_sums);
cl::copy(queue, d_partial_sums, begin(h_psum), end(h_psum));
// complete the sum and compute final integral value
pi_res = 0.0f;
for (unsigned int i = 0; i< nwork_groups; i++) {
pi_res += h_psum[i];
}
pi_res = pi_res * step_size;
//rtime = wtime() - rtime;
double rtime = static_cast<double>(timer.getTimeMilliseconds()) / 1000.;
printf("\nThe calculation ran in %lf seconds\n", rtime);
printf(" pi = %f for %d steps\n", pi_res, nsteps);
}
catch (cl::Error err) {
std::cout << "Exception\n";
std::cerr
<< "ERROR: "
<< err.what()
<< "("
<< err_code(err.err())
<< ")"
<< std::endl;
}
}
int main(void) {
//###############################################
//
// Declare variables for OpenCL
//
//###############################################
int err; // error code returned from OpenCL calls
size_t global; // global domain size
cl_device_id device_id; // compute device id
cl_context context; // compute context
cl_command_queue commands; // compute command queue
cl_program program; // compute program
cl_kernel ko_calculate_imagerowdots_iterations; // compute kernel
cl_kernel ko_calculate_colorrow; // compute kernel
cl_mem d_a; // device memory used for the input a vector
cl_mem d_b; // device memory
int i;
//###############################################
//
// Set values for mandelbrot
//
//###############################################
//plane section values
float x_ebene_min = -1;
float y_ebene_min = -1;
float x_ebene_max = 2;
float y_ebene_max = 1;
//monitor resolution values
const long x_mon = 640;
const long y_mon = 480;
//Iterations
long itr = 100;
//abort condition
float abort_value = 2;
//Number of images per second
long fps = 24;
//video duration in seconds
long video_duration = 3;
//zoom speed in percentage
float reduction = 5;
//zoom dot
my_complex_t zoom_dot;
//###############################################
//
// Set up platform and GPU device
//
//###############################################
cl_uint numPlatforms;
// Find number of platforms
err = clGetPlatformIDs(0, NULL, &numPlatforms);
checkError(err, "Finding platforms");
if (numPlatforms == 0) {
printf("Found 0 platforms!\n");
return EXIT_FAILURE;
}
// Get all platforms
cl_platform_id Platform[numPlatforms];
err = clGetPlatformIDs(numPlatforms, Platform, NULL);
checkError(err, "Getting platforms");
// Secure a GPU
for (i = 0; i < numPlatforms; i++) {
err = clGetDeviceIDs(Platform[i], DEVICE, 1, &device_id, NULL);
if (err == CL_SUCCESS) {
break;
}
}
if (device_id == NULL)
checkError(err, "Finding a device");
err = output_device_info(device_id);
checkError(err, "Printing device output");
//###############################################
//
// Create context, command queue and kernel
//
//###############################################
// Create a compute context
context = clCreateContext(0, 1, &device_id, NULL, NULL, &err);
checkError(err, "Creating context");
// Create a command queue
commands = clCreateCommandQueue(context, device_id, 0, &err);
checkError(err, "Creating command queue");
//Read Kernel source
FILE *fp;
char *source_str;
size_t source_size, program_size;
fp = fopen("./kernel/calculate_iterations.cl", "r");
if (!fp) {
printf("Failed to load kernel\n");
return 1;
}
fseek(fp, 0, SEEK_END);
program_size = ftell(fp);
rewind(fp);
source_str = (char*) malloc(program_size + 1);
source_str[program_size] = '\0';
fread(source_str, sizeof(char), program_size, fp);
fclose(fp);
// Create the compute program from the source buffer
program = clCreateProgramWithSource(context, 1, (const char **) &source_str,
NULL, &err);
checkError(err, "Creating program");
// Build the program
err = clBuildProgram(program, 0, NULL, NULL, NULL, NULL);
if (err != CL_SUCCESS) {
size_t len;
char buffer[2048];
printf("Error: Failed to build program executable!\n%s\n",
err_code(err));
clGetProgramBuildInfo(program, device_id, CL_PROGRAM_BUILD_LOG,
sizeof(buffer), buffer, &len);
printf("%s\n", buffer);
// Determine the size of the log
size_t log_size;
clGetProgramBuildInfo(program, device_id, CL_PROGRAM_BUILD_LOG, 0, NULL,
&log_size);
// Allocate memory for the log
char *log = (char *) malloc(log_size);
// Get the log
clGetProgramBuildInfo(program, device_id, CL_PROGRAM_BUILD_LOG,
log_size, log, NULL);
// Print the log
printf("%s\n", log);
return EXIT_FAILURE;
}
// Create the compute kernel from the program
ko_calculate_imagerowdots_iterations = clCreateKernel(program,
"calculate_imagerowdots_iterations", &err);
checkError(err, "Creating kernel");
// Create the compute kernel from the program
ko_calculate_colorrow = clCreateKernel(program, "calculate_colorrow", &err);
checkError(err, "Creating kernel");
int number_images = 0;
do {
//Get memory for image
long* h_image = (long*) calloc(x_mon * y_mon, sizeof(long));
unsigned char* h_image_pixel = (unsigned char*) calloc(
x_mon * y_mon * 3, sizeof(unsigned char));
//###############################################
//###############################################
//
// Loop to calculate image dot iterations
//
//###############################################
//###############################################
float y_value = y_ebene_max;
float delta_y = delta(y_ebene_min, y_ebene_max, y_mon);
for (int row = 0; row < y_mon; ++row) {
//###############################################
//
// Create and write buffer
//
//###############################################
//Get memory for row
long* h_image_row = (long*) calloc(x_mon, sizeof(long)); // a vector
d_a = clCreateBuffer(context, CL_MEM_READ_WRITE,
sizeof(long) * x_mon,
NULL, &err);
checkError(err, "Creating buffer d_a");
// Write a vector into compute device memory
err = clEnqueueWriteBuffer(commands, d_a, CL_TRUE, 0,
sizeof(long) * x_mon, h_image_row, 0, NULL, NULL);
checkError(err, "Copying h_a to device at d_a");
//###############################################
//
// Set the arguments to our compute kernel
//
//###############################################
err = clSetKernelArg(ko_calculate_imagerowdots_iterations, 0,
sizeof(float), &x_ebene_min);
err |= clSetKernelArg(ko_calculate_imagerowdots_iterations, 1,
sizeof(float), &x_ebene_max);
err |= clSetKernelArg(ko_calculate_imagerowdots_iterations, 2,
sizeof(float), &y_value);
err |= clSetKernelArg(ko_calculate_imagerowdots_iterations, 3,
sizeof(long), &x_mon);
err |= clSetKernelArg(ko_calculate_imagerowdots_iterations, 4,
sizeof(float), &abort_value);
err |= clSetKernelArg(ko_calculate_imagerowdots_iterations, 5,
sizeof(long), &itr);
err |= clSetKernelArg(ko_calculate_imagerowdots_iterations, 6,
sizeof(cl_mem), &d_a);
checkError(err, "Setting kernel arguments");
/*__kernel void calculate_imagerowdots_iterations(const float x_min, const float x_max,
const float y_value, const long x_mon, const float abort_value, const long itr,
__global long * imagerow)*/
// Execute the kernel over the entire range of our 1d input data set
// letting the OpenCL runtime choose the work-group size
global = x_mon;
err = clEnqueueNDRangeKernel(commands,
ko_calculate_imagerowdots_iterations, 1, NULL, &global,
NULL, 0,
NULL, NULL);
checkError(err, "Enqueueing kernel");
// Wait for the commands to complete
err = clFinish(commands);
checkError(err, "Waiting for kernel to finish");
// Read back the results from the compute device
err = clEnqueueReadBuffer(commands, d_a, CL_TRUE, 0,
sizeof(long) * x_mon, h_image_row, 0, NULL, NULL);
if (err != CL_SUCCESS) {
printf("Error: Failed to read output array!\n%s\n",
err_code(err));
exit(1);
}
//reduce y
y_value -= delta_y;
//cope row to image
memcpy(h_image + row * x_mon, h_image_row, sizeof(long) * x_mon);
free(h_image_row);
}
// for (i = 0; i < x_mon * y_mon; ++i) {
// printf("%ld ", h_image[i]);
// }
// fflush(stdout);
//###############################################
//###############################################
//
// End of loop to calculate image dot iterations
//
//###############################################
//###############################################
//###############################################
//###############################################
//
// Beginn color calculation
//
//###############################################
//###############################################
for (int row = 0; row < y_mon; ++row) {
//Get memory for row
long* h_image_row = (long*) calloc(x_mon, sizeof(long)); // a vector
memcpy(h_image_row, h_image + row * x_mon, sizeof(long) * x_mon);
d_a = clCreateBuffer(context, CL_MEM_READ_ONLY,
sizeof(long) * x_mon,
NULL, &err);
checkError(err, "Creating buffer d_a");
// Write a vector into compute device memory
err = clEnqueueWriteBuffer(commands, d_a, CL_TRUE, 0,
sizeof(long) * x_mon, h_image_row, 0, NULL, NULL);
checkError(err, "Copying h_image_row to device at d_a");
unsigned char* h_imagepixel_row = (unsigned char*) calloc(x_mon * 3,
sizeof(unsigned char)); // a vector
d_b = clCreateBuffer(context, CL_MEM_READ_WRITE,
sizeof(unsigned char) * x_mon * 3,
NULL, &err);
checkError(err, "Creating buffer d_b");
// Write a vector into compute device memory
err = clEnqueueWriteBuffer(commands, d_b, CL_TRUE, 0,
sizeof(unsigned char) * x_mon * 3, h_imagepixel_row, 0,
NULL, NULL);
checkError(err, "Copying h_imagepixel_row to device at d_b");
//###############################################
//
// Set the arguments to our compute kernel
//
//###############################################
err = clSetKernelArg(ko_calculate_colorrow, 0, sizeof(long),
&x_mon);
err |= clSetKernelArg(ko_calculate_colorrow, 1, sizeof(long), &itr);
err |= clSetKernelArg(ko_calculate_colorrow, 2, sizeof(cl_mem),
&d_a);
err |= clSetKernelArg(ko_calculate_colorrow, 3, sizeof(cl_mem),
&d_b);
checkError(err, "Setting kernel arguments");
/*__kernel void calculate_colorrow(const long width, long itr, long * imagerowvalues,
unsigned char * imagerow)*/
// Execute the kernel over the entire range of our 1d input data set
// letting the OpenCL runtime choose the work-group size
global = x_mon;
err = clEnqueueNDRangeKernel(commands, ko_calculate_colorrow, 1,
NULL, &global, NULL, 0,
NULL, NULL);
checkError(err, "Enqueueing kernel");
// Wait for the commands to complete
err = clFinish(commands);
checkError(err, "Waiting for kernel to finish");
// Read back the results from the compute device
err = clEnqueueReadBuffer(commands, d_b, CL_TRUE, 0,
sizeof(unsigned char) * x_mon * 3, h_imagepixel_row, 0,
NULL, NULL);
if (err != CL_SUCCESS) {
printf("Error: Failed to read output array!\n%s\n",
err_code(err));
exit(1);
}
memcpy(h_image_pixel + row * x_mon * 3, h_imagepixel_row,
sizeof(unsigned char) * x_mon * 3);
free(h_image_row);
free(h_imagepixel_row);
}
if (number_images == 0) {
zoom_dot = find_dot_to_zoom(x_ebene_min, x_ebene_max, y_ebene_min,
y_ebene_max, h_image, y_mon, x_mon, itr);
}
reduce_plane_section_focus_dot(&x_ebene_min, &x_ebene_max, &y_ebene_min,
&y_ebene_max, reduction, zoom_dot);
// save the image
char filename[50];
sprintf(filename, "img-%d.bmp", number_images);
safe_image_to_bmp(x_mon, y_mon, h_image_pixel, filename);
free(h_image);
free(h_image_pixel);
number_images++;
itr = (long) (itr + itr * reduction / 100);
printf("%d\n", number_images);
fflush(stdout);
} while (number_images < (fps * video_duration));
//###############################################
//
// cleanup then shutdown
//
//###############################################
clReleaseMemObject(d_a);
clReleaseMemObject(d_b);
clReleaseProgram(program);
clReleaseKernel(ko_calculate_imagerowdots_iterations);
clReleaseCommandQueue(commands);
clReleaseContext(context);
return 0;
}