Page MenuHome
Differential D5968 Diff 18711 intern/cycles/device/device_optix.cpp

Changeset View

Standalone View

View Options

intern/cycles/device/device_optix.cpp

Show First 20 Lines • Show All 163 Lines • ▼ Show 20 Lines# endif
// Need CUDA kernel module for some utility functions // Need CUDA kernel module for some utility functions
CUmodule cuda_module = NULL; CUmodule cuda_module = NULL;
CUmodule cuda_filter_module = NULL; CUmodule cuda_filter_module = NULL;
// All necessary OptiX kernels are in one module // All necessary OptiX kernels are in one module
OptixModule optix_module = NULL; OptixModule optix_module = NULL;
OptixPipeline pipelines[NUM_PIPELINES] = {}; OptixPipeline pipelines[NUM_PIPELINES] = {};
bool motion_blur = false;
bool need_texture_info = false; bool need_texture_info = false;
device_vector<SbtRecord> sbt_data; device_vector<SbtRecord> sbt_data;
device_vector<TextureInfo> texture_info; device_vector<TextureInfo> texture_info;
device_only_memory<KernelParams> launch_params; device_only_memory<KernelParams> launch_params;
vector<device_only_memory<uint8_t>> blas; vector<device_only_memory<uint8_t>> blas;
OptixTraversableHandle tlas_handle = 0; OptixTraversableHandle tlas_handle = 0;
map<device_memory *, CUDAMem> cuda_mem_map; map<device_memory *, CUDAMem> cuda_mem_map;
▲ Show 20 Lines • Show All 152 Lines • ▼ Show 20 Lines
# ifdef WITH_CYCLES_DEBUG # ifdef WITH_CYCLES_DEBUG
pipeline_options.exceptionFlags = OPTIX_EXCEPTION_FLAG_STACK_OVERFLOW | pipeline_options.exceptionFlags = OPTIX_EXCEPTION_FLAG_STACK_OVERFLOW |
OPTIX_EXCEPTION_FLAG_TRACE_DEPTH; OPTIX_EXCEPTION_FLAG_TRACE_DEPTH;
# else # else
pipeline_options.exceptionFlags = OPTIX_EXCEPTION_FLAG_NONE; pipeline_options.exceptionFlags = OPTIX_EXCEPTION_FLAG_NONE;
# endif # endif
pipeline_options.pipelineLaunchParamsVariableName = "__params"; // See kernel_globals.h pipeline_options.pipelineLaunchParamsVariableName = "__params"; // See kernel_globals.h
if (requested_features.use_object_motion) { // Keep track of whether motion blur is enabled, so to enable/disable motion in BVH builds
// This is necessary since objects may be reported to have motion if the Vector pass is
// active, but may still need to be rendered without motion blur if that isn't active as well
motion_blur = requested_features.use_object_motion;
if (motion_blur) {
pipeline_options.usesMotionBlur = true; pipeline_options.usesMotionBlur = true;
// Motion blur can insert motion transforms into the traversal graph // Motion blur can insert motion transforms into the traversal graph
// It is no longer a two-level graph then, so need to set flags to allow any configuration // It is no longer a two-level graph then, so need to set flags to allow any configuration
pipeline_options.traversableGraphFlags = OPTIX_TRAVERSABLE_GRAPH_FLAG_ALLOW_ANY; pipeline_options.traversableGraphFlags = OPTIX_TRAVERSABLE_GRAPH_FLAG_ALLOW_ANY;
} }
{ // Load and compile PTX module with OptiX kernels { // Load and compile PTX module with OptiX kernels
string ptx_data; string ptx_data;
▲ Show 20 Lines • Show All 518 Lines • ▼ Show 20 Lines for (Object *ob : bvh->objects) {
// Build BLAS for curve primitives // Build BLAS for curve primitives
if (bvh->params.primitive_mask & PRIMITIVE_ALL_CURVE && mesh->num_curves() > 0) { if (bvh->params.primitive_mask & PRIMITIVE_ALL_CURVE && mesh->num_curves() > 0) {
const size_t num_curves = mesh->num_curves(); const size_t num_curves = mesh->num_curves();
const size_t num_segments = mesh->num_segments(); const size_t num_segments = mesh->num_segments();
size_t num_motion_steps = 1; size_t num_motion_steps = 1;
Attribute *motion_keys = mesh->curve_attributes.find(ATTR_STD_MOTION_VERTEX_POSITION); Attribute *motion_keys = mesh->curve_attributes.find(ATTR_STD_MOTION_VERTEX_POSITION);
if (mesh->use_motion_blur && motion_keys) { if (motion_blur && mesh->use_motion_blur && motion_keys) {
num_motion_steps = mesh->motion_steps; num_motion_steps = mesh->motion_steps;
} }
device_vector<OptixAabb> aabb_data(this, "temp_aabb_data", MEM_READ_ONLY); device_vector<OptixAabb> aabb_data(this, "temp_aabb_data", MEM_READ_ONLY);
aabb_data.alloc(num_segments * num_motion_steps); aabb_data.alloc(num_segments * num_motion_steps);
// Get AABBs for each motion step // Get AABBs for each motion step
for (size_t step = 0; step < num_motion_steps; ++step) { for (size_t step = 0; step < num_motion_steps; ++step) {
▲ Show 20 Lines • Show All 53 Lines • ▼ Show 20 Lines for (Object *ob : bvh->objects) {
} }
// Build BLAS for triangle primitives // Build BLAS for triangle primitives
if (bvh->params.primitive_mask & PRIMITIVE_ALL_TRIANGLE && mesh->num_triangles() > 0) { if (bvh->params.primitive_mask & PRIMITIVE_ALL_TRIANGLE && mesh->num_triangles() > 0) {
const size_t num_verts = mesh->verts.size(); const size_t num_verts = mesh->verts.size();
size_t num_motion_steps = 1; size_t num_motion_steps = 1;
Attribute *motion_keys = mesh->attributes.find(ATTR_STD_MOTION_VERTEX_POSITION); Attribute *motion_keys = mesh->attributes.find(ATTR_STD_MOTION_VERTEX_POSITION);
if (mesh->use_motion_blur && motion_keys) { if (motion_blur && mesh->use_motion_blur && motion_keys) {
num_motion_steps = mesh->motion_steps; num_motion_steps = mesh->motion_steps;
} }
device_vector<int> index_data(this, "temp_index_data", MEM_READ_ONLY); device_vector<int> index_data(this, "temp_index_data", MEM_READ_ONLY);
index_data.alloc(mesh->triangles.size()); index_data.alloc(mesh->triangles.size());
memcpy(index_data.data(), mesh->triangles.data(), mesh->triangles.size() * sizeof(int)); memcpy(index_data.data(), mesh->triangles.data(), mesh->triangles.size() * sizeof(int));
device_vector<float3> vertex_data(this, "temp_vertex_data", MEM_READ_ONLY); device_vector<float3> vertex_data(this, "temp_vertex_data", MEM_READ_ONLY);
vertex_data.alloc(num_verts * num_motion_steps); vertex_data.alloc(num_verts * num_motion_steps);
▲ Show 20 Lines • Show All 82 Lines • ▼ Show 20 Lines for (Object *ob : bvh->objects) {
// Set user instance ID to object index // Set user instance ID to object index
instance.instanceId = ob->get_device_index(); instance.instanceId = ob->get_device_index();
// Volumes have a special bit set in the visibility mask so a trace can mask only volumes // Volumes have a special bit set in the visibility mask so a trace can mask only volumes
// See 'scene_intersect_volume' in bvh.h // See 'scene_intersect_volume' in bvh.h
instance.visibilityMask = (ob->mesh->has_volume ? 3 : 1); instance.visibilityMask = (ob->mesh->has_volume ? 3 : 1);
// Insert motion traversable if object has motion // Insert motion traversable if object has motion
if (ob->use_motion()) { if (motion_blur && ob->use_motion()) {
blas.emplace_back(this, "motion_transform"); blas.emplace_back(this, "motion_transform");
device_only_memory<uint8_t> &motion_transform_gpu = blas.back(); device_only_memory<uint8_t> &motion_transform_gpu = blas.back();
motion_transform_gpu.alloc_to_device(sizeof(OptixSRTMotionTransform) + motion_transform_gpu.alloc_to_device(sizeof(OptixSRTMotionTransform) +
(max(ob->motion.size(), 2) - 2) * (max(ob->motion.size(), 2) - 2) *
sizeof(OptixSRTData)); sizeof(OptixSRTData));
// Allocate host side memory for motion transform and fill it with transform data // Allocate host side memory for motion transform and fill it with transform data
OptixSRTMotionTransform &motion_transform = *reinterpret_cast<OptixSRTMotionTransform *>( OptixSRTMotionTransform &motion_transform = *reinterpret_cast<OptixSRTMotionTransform *>(
▲ Show 20 Lines • Show All 918 Lines • Show Last 20 Lines