// ----------------------------------------------------------------------------
// Generates a opacity map from the transfer function specification
// ----------------------------------------------------------------------------
void mapOpacity(Image3D<float> & map, std::list<std::shared_ptr<Node>> transfer)
{
float samples = static_cast<float>(map.width()) - 1;
auto buffer = map.data();
memset(buffer, 0, map.size()*sizeof(float));
auto iter = transfer.begin();
auto curr = *iter;
iter++;
while (iter != transfer.end())
{
auto next = *iter;
size_t x1 = static_cast<size_t>(curr->density * samples);
if (x1 >= map.width()) x1 = map.width()-1;
size_t x2 = static_cast<size_t>(next->density * samples);
if (x2 >= map.width()) x2 = map.width()-1;
float y1 = curr->material->opticalThickness;
float y2 = next->material->opticalThickness;
for (size_t i = x1; i <= x2; i++)
{
float px = i / samples - curr->density;
float py = next->density - curr->density;
float part = low( high(px / py, 0.0f), 1.0f );
buffer[i] = - logf( 1.f - (1.f - part) * y1 - part * y2 );
}
curr = next;
iter++;
}
}
// ----------------------------------------------------------------------------
// Generates a diffuse map from the transfer function specification
// ----------------------------------------------------------------------------
void mapSpecular(Image3D<Vector<UInt8,4>> & map, std::list<std::shared_ptr<Node>> transfer)
{
float samples = static_cast<float>(map.width()) - 1;
auto buffer = map.data();
memset(buffer, 0, map.size()*sizeof(Vector<UInt8,4>));
auto iter = transfer.begin();
auto curr = *iter;
iter++;
while (iter != transfer.end())
{
auto next = *iter;
size_t x1 = static_cast<size_t>(curr->density * samples);
if (x1 >= map.width()) x1 = map.width()-1;
size_t x2 = static_cast<size_t>(next->density * samples);
if (x2 >= map.width()) x2 = map.width()-1;
Vector3f s1(curr->material->specular);
Vector3f s2(next->material->specular);
Vector4f y1(s1[0], s1[1], s1[2], curr->material->glossiness*255.0f);
Vector4f y2(s2[0], s2[1], s2[2], next->material->glossiness*255.0f);
for (size_t i = x1; i <= x2; i++)
{
float px = i / samples - curr->density;
float py = next->density - curr->density;
float part = low( high(px / py, 0.0f), 1.0f );
buffer[i] = static_cast<Vector<UInt8,4>>(y1*(1.f - part) + y2*part);
}
curr = next;
iter++;
}
}
// ----------------------------------------------------------------------------
// Generates an emissive map from the tranfer specification
// ----------------------------------------------------------------------------
void mapEmissive(Image3D<Vector4f> & map, std::list<std::shared_ptr<Node>> transfer)
{
float samples = static_cast<float>(map.width()) - 1;
auto buffer = map.data();
memset(buffer, 0, map.size()*sizeof(Vector4f));
auto iter = transfer.begin();
auto curr = *iter;
iter++;
while (iter != transfer.end())
{
auto next = *iter;
size_t x1 = static_cast<size_t>(curr->density * samples);
if (x1 >= map.width()) x1 = map.width()-1;
size_t x2 = static_cast<size_t>(next->density * samples);
if (x2 >= map.width()) x2 = map.width()-1;
Vector3f s1 = Vector3f(curr->material->emissive) / 255.0f * curr->material->emissiveStrength;
Vector3f s2 = Vector3f(next->material->emissive) / 255.0f * curr->material->emissiveStrength;
Vector4f y1(s1[0], s1[1], s1[2], 0.0f);
Vector4f y2(s2[0], s2[1], s2[2], 0.0f);
for (size_t i = x1; i <= x2; i++)
{
float px = i / samples - curr->density;
float py = next->density - curr->density;
float part = low( high(px / py, 0.0f), 1.0f );
buffer[i] = y1*(1.f - part) + y2*part;
}
curr = next;
iter++;
}
}
/*
* Kernel event
*/
KernelEvent::KernelEvent(CommandQueue *parent,
Kernel *kernel,
cl_uint work_dim,
const size_t *global_work_offset,
const size_t *global_work_size,
const size_t *local_work_size,
cl_uint num_events_in_wait_list,
const cl_event *event_wait_list,
cl_int *errcode_ret)
: Event(parent, Queued, num_events_in_wait_list, event_wait_list, errcode_ret),
p_work_dim(work_dim), p_kernel(kernel), p_timeout_ms(0)
{
clRetainKernel(desc(p_kernel));
if (*errcode_ret != CL_SUCCESS) return;
*errcode_ret = CL_SUCCESS;
// Sanity checks
if (!kernel)
{
*errcode_ret = CL_INVALID_KERNEL;
return;
}
// Check that the kernel was built for parent's device.
cl_device_id d_device = 0;
cl_context k_ctx, q_ctx;
size_t max_work_group_size;
cl_uint max_dims = 0;
*errcode_ret = parent->info(CL_QUEUE_DEVICE, sizeof(cl_device_id), &d_device, 0);
if (*errcode_ret != CL_SUCCESS)
return;
auto device = pobj(d_device);
*errcode_ret = parent->info(CL_QUEUE_CONTEXT, sizeof(cl_context), &q_ctx, 0);
*errcode_ret |= kernel->info(CL_KERNEL_CONTEXT, sizeof(cl_context), &k_ctx, 0);
*errcode_ret |= device->info(CL_DEVICE_MAX_WORK_GROUP_SIZE, sizeof(size_t),
&max_work_group_size, 0);
*errcode_ret |= device->info(CL_DEVICE_MAX_WORK_ITEM_DIMENSIONS, sizeof(size_t),
&max_dims, 0);
*errcode_ret |= device->info(CL_DEVICE_MAX_WORK_ITEM_SIZES,
max_dims * sizeof(size_t), p_max_work_item_sizes, 0);
if (*errcode_ret != CL_SUCCESS)
return;
p_dev_kernel = kernel->deviceDependentKernel(device);
if (!p_dev_kernel)
{
*errcode_ret = CL_INVALID_PROGRAM_EXECUTABLE;
return;
}
// Check that contexts match
if (k_ctx != q_ctx)
{
*errcode_ret = CL_INVALID_CONTEXT;
return;
}
// Check args
if (!kernel->argsSpecified())
{
*errcode_ret = CL_INVALID_KERNEL_ARGS;
return;
}
// Check dimension
if ((work_dim == 0 || work_dim > max_dims) ||
(max_dims > MAX_WORK_DIMS))
{
*errcode_ret = CL_INVALID_WORK_DIMENSION;
return;
}
// Populate work_offset, work_size and local_work_size
size_t work_group_size = 1;
cl_uint dims[3];
kernel->reqdWorkGroupSize(kernel->deviceDependentModule(device), dims);
uint32_t reqd_x = dims[0];
uint32_t reqd_y = dims[1];
uint32_t reqd_z = dims[2];
bool reqd_any = reqd_x > 0 || reqd_y > 0 || reqd_z > 0;
if (reqd_any)
{
// if __attribute__((reqd_work_group_size(X, Y, Z))) is set and local size not specified
if (!local_work_size)
{
*errcode_ret = CL_INVALID_WORK_GROUP_SIZE;
return;
}
// if __attribute__((reqd_work_group_size(X, Y, Z))) doesn't match
else
{
if (( local_work_size[0] != reqd_x) ||
(work_dim > 1 && local_work_size[1] != reqd_y) ||
(work_dim > 2 && local_work_size[2] != reqd_z))
{
*errcode_ret = CL_INVALID_WORK_GROUP_SIZE;
return;
}
}
}
// If kernel has zero arguments
if (local_work_size && local_work_size[0] == 0) return;
cl_uint i;
for (i=0; i<work_dim; ++i)
{
if (global_work_offset)
{
p_global_work_offset[i] = global_work_offset[i];
}
else
{
p_global_work_offset[i] = 0;
}
if (!global_work_size || !global_work_size[i])
{
*errcode_ret = CL_INVALID_GLOBAL_WORK_SIZE;
return;
}
p_global_work_size[i] = global_work_size[i];
if (!local_work_size)
{
// Guess the best value according to the device
p_local_work_size[i] =
p_dev_kernel->guessWorkGroupSize(work_dim, i, global_work_size[i]);
}
else
{
// Check divisibility
if ((global_work_size[i] % local_work_size[i]) != 0)
{
*errcode_ret = CL_INVALID_WORK_GROUP_SIZE;
return;
}
// Not too big ?
if (local_work_size[i] > p_max_work_item_sizes[i])
{
*errcode_ret = CL_INVALID_WORK_ITEM_SIZE;
return;
}
p_local_work_size[i] = local_work_size[i];
work_group_size *= local_work_size[i];
}
}
// initialize missing dimensions
for (; i < max_dims; i++)
{
p_global_work_offset[i] = 0;
p_global_work_size[i] = 1;
p_local_work_size[i] = 1;
}
// Check we don't ask too much to the device
if (work_group_size > max_work_group_size)
{
*errcode_ret = CL_INVALID_WORK_GROUP_SIZE;
return;
}
// Check arguments (buffer alignment, image size, ...)
for (unsigned int i=0; i<kernel->numArgs(); ++i)
{
const Kernel::Arg &a = kernel->arg(i);
if (a.kind() == Kernel::Arg::Buffer && a.file() != Kernel::Arg::Local)
{
MemObject *buffer = *(MemObject **)(a.value(0));
if (!BufferEvent::isSubBufferAligned(buffer, device))
{
*errcode_ret = CL_MISALIGNED_SUB_BUFFER_OFFSET;
return;
}
clRetainMemObject(desc(buffer));
p_mem_objects.push_back((MemObject *) buffer);
}
else if (a.kind() == Kernel::Arg::Image2D)
{
Image2D *image = *(Image2D **)(a.value(0));
size_t maxWidth, maxHeight;
*errcode_ret = device->info(CL_DEVICE_IMAGE2D_MAX_WIDTH,
sizeof(size_t), &maxWidth, 0);
*errcode_ret |= device->info(CL_DEVICE_IMAGE2D_MAX_HEIGHT,
sizeof(size_t), &maxHeight, 0);
if (*errcode_ret != CL_SUCCESS)
return;
if (image->width() > maxWidth || image->height() > maxHeight)
{
*errcode_ret = CL_INVALID_IMAGE_SIZE;
return;
}
clRetainMemObject(desc(image));
p_mem_objects.push_back((MemObject *) image);
}
else if (a.kind() == Kernel::Arg::Image3D)
{
Image3D *image = *(Image3D **)a.value(0);
size_t maxWidth, maxHeight, maxDepth;
*errcode_ret = device->info(CL_DEVICE_IMAGE3D_MAX_WIDTH,
sizeof(size_t), &maxWidth, 0);
*errcode_ret |= device->info(CL_DEVICE_IMAGE3D_MAX_HEIGHT,
sizeof(size_t), &maxHeight, 0);
*errcode_ret |= device->info(CL_DEVICE_IMAGE3D_MAX_DEPTH,
sizeof(size_t), &maxDepth, 0);
if (*errcode_ret != CL_SUCCESS)
return;
if (image->width() > maxWidth || image->height() > maxHeight ||
image->depth() > maxDepth)
{
*errcode_ret = CL_INVALID_IMAGE_SIZE;
return;
}
clRetainMemObject(desc(image));
p_mem_objects.push_back((MemObject *) image);
}
}
// Check if kernel has timeout specified and CommandQueue allows it
if (kernel->getTimeout() > 0)
{
cl_command_queue_properties queue_props;
*errcode_ret = parent->info(CL_QUEUE_PROPERTIES,
sizeof(cl_command_queue_properties),
&queue_props, 0);
if (*errcode_ret != CL_SUCCESS) return;
if ((queue_props & CL_QUEUE_KERNEL_TIMEOUT_COMPUTE_UNIT_TI) != 0)
p_timeout_ms = kernel->getTimeout();
}
}
