result MultiFrame::InitSortKernel(const int &linear) {
sort_ = ClKernel(device_id_, "Finalise");
const int alpha_size = 16;
const cl_int2 top_left = {0, 0};
sort_.SetNumberedArg(1, sizeof(int), ®ion_width_);
sort_.SetNumberedArg(2, sizeof(cl_int2), &top_left);
sort_.SetNumberedArg(3, sizeof(int), &linear);
sort_.SetNumberedArg(4, sizeof(int), &alpha_size);
sort_.SetNumberedArg(5, sizeof(int), &alpha_set_size_);
sort_.SetNumberedArg(6, sizeof(cl_mem), g_devices[device_id_].buffers_.ptr(alpha_));
sort_.SetNumberedArg(7, sizeof(cl_mem), g_devices[device_id_].buffers_.ptr(dest_plane_));
if (sort_.arguments_valid()) {
sort_.set_work_dim(2);
const size_t set_local_work_size[2] = {8, 16};
// height is increased to offset the fact that 8 work items collaborate on one pixel
const size_t set_scalar_global_size[2] = {region_width_, region_height_ << 3};
const size_t set_scalar_item_size[2] = {1, 1};
sort_.set_local_work_size(set_local_work_size);
sort_.set_scalar_global_size(set_scalar_global_size);
sort_.set_scalar_item_size(set_scalar_item_size);
return FILTER_OK;
}
return FILTER_KERNEL_ARGUMENT_ERROR;
}
result MultiFrame::InitFilterKernel(
const int &sample_expand,
const int &linear,
const int &correction,
const int &balanced) {
filter_ = ClKernel(device_id_, "NLMMultiFrameFourPixel");
filter_.SetNumberedArg(FILTER_ARG_WIDTH, sizeof(int), &width_);
filter_.SetNumberedArg(FILTER_ARG_HEIGHT, sizeof(int), &height_);
filter_.SetNumberedArg(FILTER_ARG_H, sizeof(float), &h_);
filter_.SetNumberedArg(FILTER_ARG_SAMPLE_EXPAND, sizeof(int), &sample_expand);
filter_.SetNumberedArg(FILTER_ARG_G_GAUSSIAN, sizeof(cl_mem), g_devices[device_id_].buffers_.ptr(g_gaussian));
filter_.SetNumberedArg(FILTER_ARG_LINEAR, sizeof(int), &linear);
filter_.SetNumberedArg(FILTER_ARG_ALPHA_SET_SIZE, sizeof(int), &alpha_set_size_);
filter_.SetNumberedArg(FILTER_ARG_REGION_ALPHA, sizeof(cl_mem), g_devices[device_id_].buffers_.ptr(alpha_));
if (filter_.arguments_valid()) {
filter_.set_work_dim(2);
const size_t set_local_work_size[2] = {8, 16};
// height is increased to offset the fact that 8 work items collaborate on one pixel
const size_t set_scalar_global_size[2] = {region_width_, region_height_ << 3};
const size_t set_scalar_item_size[2] = {1, 1};
filter_.set_local_work_size(set_local_work_size);
filter_.set_scalar_global_size(set_scalar_global_size);
filter_.set_scalar_item_size(set_scalar_item_size);
return FILTER_OK;
}
return FILTER_KERNEL_ARGUMENT_ERROR;
}
void
OsdClKernelDispatcher::BindVertexBuffer(OsdVertexBuffer *vertex, OsdVertexBuffer *varying) {
if (vertex)
_currentVertexBuffer = dynamic_cast<OsdClVertexBuffer *>(vertex);
else
_currentVertexBuffer = NULL;
if (varying)
_currentVaryingBuffer = dynamic_cast<OsdClVertexBuffer *>(varying);
else
_currentVaryingBuffer = NULL;
int numVertexElements = vertex ? vertex->GetNumElements() : 0;
int numVaryingElements = varying ? varying->GetNumElements() : 0;
if (_currentVertexBuffer) {
_currentVertexBuffer->Map();
}
if (_currentVaryingBuffer) {
_currentVaryingBuffer->Map();
}
// find cl kernel from registry (create it if needed)
std::vector<ClKernel>::iterator it =
std::find_if(kernelRegistry.begin(), kernelRegistry.end(),
ClKernel::Match(numVertexElements, numVaryingElements));
if (it != kernelRegistry.end()) {
_clKernel = &(*it);
} else {
kernelRegistry.push_back(ClKernel());
_clKernel = &kernelRegistry.back();
_clKernel->Compile(_clContext, numVertexElements, numVaryingElements);
}
}
