xdl_int XdevLComputeProgramCL::execute(const XdevLComputeExecuteParameter& parameter) {
auto queueCL = static_cast<XdevLComputeDeviceQueueCL*>(parameter.queue);
auto kernelCL = static_cast<XdevLComputeKernelCL*>(parameter.kernel);
cl_int ret;
// ret = clEnqueueTask(queueCL->getCommandQueue(), kernelCL->getKernel(), 0, nullptr, nullptr);
// if(CL_SUCCESS != ret) {
// XDEVL_MODULEX_ERROR(XdevLComputeProgramCL, "clEnqueueTask failed: " << clErrorAsString(ret) << std::endl);
// return RET_FAILED;
// }
// std::size_t workgroup_size;
// ret = clGetKernelWorkGroupInfo(kernelCL->getKernel(), m_deviceId, CL_KERNEL_WORK_GROUP_SIZE, sizeof(std::size_t), &workgroup_size, nullptr);
// if(CL_SUCCESS != ret) {
// XDEVL_MODULEX_ERROR(XdevLComputeProgramCL, "clGetKernelWorkGroupInfo failed: " << clErrorAsString(ret) << std::endl);
// return RET_FAILED;
// }
ret = clEnqueueNDRangeKernel(queueCL->getCommandQueue(), kernelCL->getKernel(), parameter.global.size(), nullptr, parameter.global.data(), nullptr, 0, nullptr, nullptr);
if(CL_SUCCESS != ret) {
XDEVL_MODULEX_ERROR(XdevLComputeProgramCL, "clEnqueueNDRangeKernel failed: " << clErrorAsString(ret) << std::endl);
return RET_FAILED;
}
return RET_SUCCESS;
}
xdl_int XdevLComputeDeviceBufferCL::download(XdevLComputeDeviceQueue* queue, xdl_uint size, xdl_uint8* data) {
auto queueCL = static_cast<XdevLComputeDeviceQueueCL*>(queue);
cl_int ret = clEnqueueReadBuffer(queueCL->getCommandQueue(), m_memory, CL_TRUE, 0, size, data, 0, nullptr, nullptr);
if(CL_SUCCESS != ret) {
XDEVL_MODULEX_ERROR(XdevLComputeDeviceBufferCL,"clEnqueueWriteBuffer failed: " << clErrorAsString(ret) << std::endl);
return RET_FAILED;
}
return RET_SUCCESS;
}
void CLMatrixLoader::enqueue()
{
m_src->waitEndOfEvaluation();
//enqueue transfer execution
cl_int error;
error = getCommandQueue()->enqueueWriteBuffer (
*(m_result->getValues()),
CL_FALSE,
0,
sizeof ( cl_float ) * m_src->getWidth() * m_src->getHeight(),
m_src->getValues(),
0,
&(getEndOfEvaluation())
);
if ( error != CL_SUCCESS )
throw ( CLError ( error, "failed to enqueue memory transfer" ) );
}
static void
handleExecuteCommandAlarm (const AsyncAlarmResult *result) {
Queue *queue = getCommandQueue(0);
if (queue) {
int command = dequeueCommand(queue);
if (command != EOF) {
const CommandHandlerLevel *chl = commandHandlerStack;
while (chl) {
if (chl->handleCommand(command, chl->handlerData)) break;
chl = chl->previousLevel;
}
}
if (getQueueSize(queue) > 0) setExecuteCommandAlarm(result->data);
}
}
void CLMatrixUnloader::enqueue()
{
// prepare dependencies
ClAlgorithm* dependency = (ClAlgorithm*)(m_src->getParentAlgorithm());
m_dependenciesEvents.push_back(dependency->getEndOfEvaluation());
//enqueue transfer execution
cl_int error;
error = getCommandQueue()->enqueueReadBuffer(
*(m_src->getValues().get()),
CL_FALSE,
0,
sizeof ( cl_float ) * m_src->getWidth() * m_src->getHeight(),
m_result->getValues(),
&m_dependenciesEvents,
&(getEndOfEvaluation())
);
if ( error != CL_SUCCESS )
throw ( CLError ( error, "failed to enqueue memory transfer" ) );
}
int
enqueueCommand (int command) {
if (command != EOF) {
Queue *queue = getCommandQueue(1);
if (queue) {
CommandQueueItem *item = malloc(sizeof(CommandQueueItem));
if (item) {
item->command = command;
if (enqueueItem(queue, item)) {
if (getQueueSize(queue) == 1) setExecuteCommandAlarm(NULL);
return 1;
}
free(item);
}
}
}
return 0;
}
void compare_images::init_value_profiler()
{
v_profiler->init(getCommandQueue(),getContext());
}