diff options
| author | Brecht Van Lommel <brecht@blender.org> | 2021-09-20 18:59:20 +0300 |
|---|---|---|
| committer | Brecht Van Lommel <brecht@blender.org> | 2021-09-21 15:55:54 +0300 |
| commit | 08031197250aeecbaca3803254e6f25b8c7b7b37 (patch) | |
| tree | 6fe7ab045f0dc0a423d6557c4073f34309ef4740 /intern/cycles/render | |
| parent | fa6b1007bad065440950cd67deb16a04f368856f (diff) | |
Cycles: merge of cycles-x branch, a major update to the renderer
This includes much improved GPU rendering performance, viewport interactivity,
new shadow catcher, revamped sampling settings, subsurface scattering anisotropy,
new GPU volume sampling, improved PMJ sampling pattern, and more.
Some features have also been removed or changed, breaking backwards compatibility.
Including the removal of the OpenCL backend, for which alternatives are under
development.
Release notes and code docs:
https://wiki.blender.org/wiki/Reference/Release_Notes/3.0/Cycles
https://wiki.blender.org/wiki/Source/Render/Cycles
Credits:
* Sergey Sharybin
* Brecht Van Lommel
* Patrick Mours (OptiX backend)
* Christophe Hery (subsurface scattering anisotropy)
* William Leeson (PMJ sampling pattern)
* Alaska (various fixes and tweaks)
* Thomas Dinges (various fixes)
For the full commit history, see the cycles-x branch. This squashes together
all the changes since intermediate changes would often fail building or tests.
Ref T87839, T87837, T87836
Fixes T90734, T89353, T80267, T80267, T77185, T69800
Diffstat (limited to 'intern/cycles/render')
42 files changed, 3472 insertions, 3786 deletions
diff --git a/intern/cycles/render/CMakeLists.txt b/intern/cycles/render/CMakeLists.txt index feead27c5ca..6edb5261b32 100644 --- a/intern/cycles/render/CMakeLists.txt +++ b/intern/cycles/render/CMakeLists.txt @@ -32,10 +32,10 @@ set(SRC camera.cpp colorspace.cpp constant_fold.cpp - coverage.cpp denoising.cpp film.cpp geometry.cpp + gpu_display.cpp graph.cpp hair.cpp image.cpp @@ -54,6 +54,7 @@ set(SRC object.cpp osl.cpp particles.cpp + pass.cpp curves.cpp scene.cpp session.cpp @@ -76,10 +77,10 @@ set(SRC_HEADERS camera.h colorspace.h constant_fold.h - coverage.h denoising.h film.h geometry.h + gpu_display.h graph.h hair.h image.h @@ -95,6 +96,7 @@ set(SRC_HEADERS object.h osl.h particles.h + pass.h procedural.h curves.h scene.h @@ -111,6 +113,7 @@ set(SRC_HEADERS set(LIB cycles_bvh cycles_device + cycles_integrator cycles_subd cycles_util ) diff --git a/intern/cycles/render/background.cpp b/intern/cycles/render/background.cpp index b925e755434..ae6290ac27b 100644 --- a/intern/cycles/render/background.cpp +++ b/intern/cycles/render/background.cpp @@ -34,11 +34,7 @@ NODE_DEFINE(Background) { NodeType *type = NodeType::add("background", create); - SOCKET_FLOAT(ao_factor, "AO Factor", 0.0f); - SOCKET_FLOAT(ao_distance, "AO Distance", FLT_MAX); - SOCKET_BOOLEAN(use_shader, "Use Shader", true); - SOCKET_BOOLEAN(use_ao, "Use AO", false); SOCKET_UINT(visibility, "Visibility", PATH_RAY_ALL_VISIBILITY); SOCKET_BOOLEAN(transparent, "Transparent", false); @@ -80,10 +76,6 @@ void Background::device_update(Device *device, DeviceScene *dscene, Scene *scene /* set shader index and transparent option */ KernelBackground *kbackground = &dscene->data.background; - kbackground->ao_factor = (use_ao) ? ao_factor : 0.0f; - kbackground->ao_bounces_factor = ao_factor; - kbackground->ao_distance = ao_distance; - kbackground->transparent = transparent; kbackground->surface_shader = scene->shader_manager->get_shader_id(bg_shader); @@ -138,10 +130,6 @@ void Background::tag_update(Scene *scene) * and to avoid doing unnecessary updates anywhere else. */ tag_use_shader_modified(); } - - if (ao_factor_is_modified() || use_ao_is_modified()) { - scene->integrator->tag_update(scene, Integrator::BACKGROUND_AO_MODIFIED); - } } Shader *Background::get_shader(const Scene *scene) diff --git a/intern/cycles/render/background.h b/intern/cycles/render/background.h index e89ffbc2445..2f7ef0f7737 100644 --- a/intern/cycles/render/background.h +++ b/intern/cycles/render/background.h @@ -32,11 +32,7 @@ class Background : public Node { public: NODE_DECLARE - NODE_SOCKET_API(float, ao_factor) - NODE_SOCKET_API(float, ao_distance) - NODE_SOCKET_API(bool, use_shader) - NODE_SOCKET_API(bool, use_ao) NODE_SOCKET_API(uint, visibility) NODE_SOCKET_API(Shader *, shader) diff --git a/intern/cycles/render/bake.cpp b/intern/cycles/render/bake.cpp index 317a3937cab..54e496caed6 100644 --- a/intern/cycles/render/bake.cpp +++ b/intern/cycles/render/bake.cpp @@ -26,58 +26,8 @@ CCL_NAMESPACE_BEGIN -static int aa_samples(Scene *scene, Object *object, ShaderEvalType type) -{ - if (type == SHADER_EVAL_UV || type == SHADER_EVAL_ROUGHNESS) { - return 1; - } - else if (type == SHADER_EVAL_NORMAL) { - /* Only antialias normal if mesh has bump mapping. */ - if (object->get_geometry()) { - foreach (Node *node, object->get_geometry()->get_used_shaders()) { - Shader *shader = static_cast<Shader *>(node); - if (shader->has_bump) { - return scene->integrator->get_aa_samples(); - } - } - } - - return 1; - } - else { - return scene->integrator->get_aa_samples(); - } -} - -/* Keep it synced with kernel_bake.h logic */ -static int shader_type_to_pass_filter(ShaderEvalType type, int pass_filter) -{ - const int component_flags = pass_filter & - (BAKE_FILTER_DIRECT | BAKE_FILTER_INDIRECT | BAKE_FILTER_COLOR); - - switch (type) { - case SHADER_EVAL_AO: - return BAKE_FILTER_AO; - case SHADER_EVAL_SHADOW: - return BAKE_FILTER_DIRECT; - case SHADER_EVAL_DIFFUSE: - return BAKE_FILTER_DIFFUSE | component_flags; - case SHADER_EVAL_GLOSSY: - return BAKE_FILTER_GLOSSY | component_flags; - case SHADER_EVAL_TRANSMISSION: - return BAKE_FILTER_TRANSMISSION | component_flags; - case SHADER_EVAL_COMBINED: - return pass_filter; - default: - return 0; - } -} - BakeManager::BakeManager() { - type = SHADER_EVAL_BAKE; - pass_filter = 0; - need_update_ = true; } @@ -85,32 +35,14 @@ BakeManager::~BakeManager() { } -bool BakeManager::get_baking() +bool BakeManager::get_baking() const { return !object_name.empty(); } -void BakeManager::set(Scene *scene, - const std::string &object_name_, - ShaderEvalType type_, - int pass_filter_) +void BakeManager::set(Scene *scene, const std::string &object_name_) { object_name = object_name_; - type = type_; - pass_filter = shader_type_to_pass_filter(type_, pass_filter_); - - Pass::add(PASS_BAKE_PRIMITIVE, scene->passes); - Pass::add(PASS_BAKE_DIFFERENTIAL, scene->passes); - - if (type == SHADER_EVAL_UV) { - /* force UV to be available */ - Pass::add(PASS_UV, scene->passes); - } - - /* force use_light_pass to be true if we bake more than just colors */ - if (pass_filter & ~BAKE_FILTER_COLOR) { - Pass::add(PASS_LIGHT, scene->passes); - } /* create device and update scene */ scene->film->tag_modified(); @@ -127,29 +59,29 @@ void BakeManager::device_update(Device * /*device*/, if (!need_update()) return; - scoped_callback_timer timer([scene](double time) { - if (scene->update_stats) { - scene->update_stats->bake.times.add_entry({"device_update", time}); - } - }); - - KernelIntegrator *kintegrator = &dscene->data.integrator; KernelBake *kbake = &dscene->data.bake; + memset(kbake, 0, sizeof(*kbake)); - kbake->type = type; - kbake->pass_filter = pass_filter; - - int object_index = 0; - foreach (Object *object, scene->objects) { - const Geometry *geom = object->get_geometry(); - if (object->name == object_name && geom->geometry_type == Geometry::MESH) { - kbake->object_index = object_index; - kbake->tri_offset = geom->prim_offset; - kintegrator->aa_samples = aa_samples(scene, object, type); - break; - } + if (!object_name.empty()) { + scoped_callback_timer timer([scene](double time) { + if (scene->update_stats) { + scene->update_stats->bake.times.add_entry({"device_update", time}); + } + }); + + kbake->use = true; - object_index++; + int object_index = 0; + foreach (Object *object, scene->objects) { + const Geometry *geom = object->get_geometry(); + if (object->name == object_name && geom->geometry_type == Geometry::MESH) { + kbake->object_index = object_index; + kbake->tri_offset = geom->prim_offset; + break; + } + + object_index++; + } } need_update_ = false; diff --git a/intern/cycles/render/bake.h b/intern/cycles/render/bake.h index 655b9b1cf7e..39e504490c2 100644 --- a/intern/cycles/render/bake.h +++ b/intern/cycles/render/bake.h @@ -30,8 +30,8 @@ class BakeManager { BakeManager(); ~BakeManager(); - void set(Scene *scene, const std::string &object_name, ShaderEvalType type, int pass_filter); - bool get_baking(); + void set(Scene *scene, const std::string &object_name); + bool get_baking() const; void device_update(Device *device, DeviceScene *dscene, Scene *scene, Progress &progress); void device_free(Device *device, DeviceScene *dscene); @@ -42,8 +42,6 @@ class BakeManager { private: bool need_update_; - ShaderEvalType type; - int pass_filter; std::string object_name; }; diff --git a/intern/cycles/render/buffers.cpp b/intern/cycles/render/buffers.cpp index fcfad58995e..1cdae3af7f5 100644 --- a/intern/cycles/render/buffers.cpp +++ b/intern/cycles/render/buffers.cpp @@ -28,537 +28,334 @@ CCL_NAMESPACE_BEGIN -/* Buffer Params */ +/* -------------------------------------------------------------------- + * Convert part information to an index of `BufferParams::pass_offset_`. + */ -BufferParams::BufferParams() +static int pass_type_mode_to_index(PassType pass_type, PassMode mode) { - width = 0; - height = 0; - - full_x = 0; - full_y = 0; - full_width = 0; - full_height = 0; + int index = static_cast<int>(pass_type) * 2; - denoising_data_pass = false; - denoising_clean_pass = false; - denoising_prefiltered_pass = false; + if (mode == PassMode::DENOISED) { + ++index; + } - Pass::add(PASS_COMBINED, passes); + return index; } -void BufferParams::get_offset_stride(int &offset, int &stride) +static int pass_to_index(const BufferPass &pass) { - offset = -(full_x + full_y * width); - stride = width; + return pass_type_mode_to_index(pass.type, pass.mode); } -bool BufferParams::modified(const BufferParams ¶ms) -{ - return !(full_x == params.full_x && full_y == params.full_y && width == params.width && - height == params.height && full_width == params.full_width && - full_height == params.full_height && Pass::equals(passes, params.passes) && - denoising_data_pass == params.denoising_data_pass && - denoising_clean_pass == params.denoising_clean_pass && - denoising_prefiltered_pass == params.denoising_prefiltered_pass); -} +/* -------------------------------------------------------------------- + * Buffer pass. + */ -int BufferParams::get_passes_size() +NODE_DEFINE(BufferPass) { - int size = 0; + NodeType *type = NodeType::add("buffer_pass", create); - for (size_t i = 0; i < passes.size(); i++) - size += passes[i].components; + const NodeEnum *pass_type_enum = Pass::get_type_enum(); + const NodeEnum *pass_mode_enum = Pass::get_mode_enum(); - if (denoising_data_pass) { - size += DENOISING_PASS_SIZE_BASE; - if (denoising_clean_pass) - size += DENOISING_PASS_SIZE_CLEAN; - if (denoising_prefiltered_pass) - size += DENOISING_PASS_SIZE_PREFILTERED; - } + SOCKET_ENUM(type, "Type", *pass_type_enum, PASS_COMBINED); + SOCKET_ENUM(mode, "Mode", *pass_mode_enum, static_cast<int>(PassMode::DENOISED)); + SOCKET_STRING(name, "Name", ustring()); + SOCKET_BOOLEAN(include_albedo, "Include Albedo", false); - return align_up(size, 4); -} + SOCKET_INT(offset, "Offset", -1); -int BufferParams::get_denoising_offset() -{ - int offset = 0; - - for (size_t i = 0; i < passes.size(); i++) - offset += passes[i].components; - - return offset; + return type; } -int BufferParams::get_denoising_prefiltered_offset() +BufferPass::BufferPass() : Node(get_node_type()) { - assert(denoising_prefiltered_pass); - - int offset = get_denoising_offset(); - - offset += DENOISING_PASS_SIZE_BASE; - if (denoising_clean_pass) { - offset += DENOISING_PASS_SIZE_CLEAN; - } - - return offset; } -/* Render Buffer Task */ - -RenderTile::RenderTile() +BufferPass::BufferPass(const Pass *scene_pass) + : Node(get_node_type()), + type(scene_pass->get_type()), + mode(scene_pass->get_mode()), + name(scene_pass->get_name()), + include_albedo(scene_pass->get_include_albedo()) { - x = 0; - y = 0; - w = 0; - h = 0; - - sample = 0; - start_sample = 0; - num_samples = 0; - resolution = 0; - - offset = 0; - stride = 0; - - buffer = 0; - - buffers = NULL; - stealing_state = NO_STEALING; } -/* Render Buffers */ - -RenderBuffers::RenderBuffers(Device *device) - : buffer(device, "RenderBuffers", MEM_READ_WRITE), - map_neighbor_copied(false), - render_time(0.0f) +PassInfo BufferPass::get_info() const { + return Pass::get_info(type, include_albedo); } -RenderBuffers::~RenderBuffers() -{ - buffer.free(); -} +/* -------------------------------------------------------------------- + * Buffer Params. + */ -void RenderBuffers::reset(BufferParams ¶ms_) +NODE_DEFINE(BufferParams) { - params = params_; - - /* re-allocate buffer */ - buffer.alloc(params.width * params.get_passes_size(), params.height); - buffer.zero_to_device(); + NodeType *type = NodeType::add("buffer_params", create); + + SOCKET_INT(width, "Width", 0); + SOCKET_INT(height, "Height", 0); + + SOCKET_INT(full_x, "Full X", 0); + SOCKET_INT(full_y, "Full Y", 0); + SOCKET_INT(full_width, "Full Width", 0); + SOCKET_INT(full_height, "Full Height", 0); + + SOCKET_STRING(layer, "Layer", ustring()); + SOCKET_STRING(view, "View", ustring()); + SOCKET_FLOAT(exposure, "Exposure", 1.0f); + SOCKET_BOOLEAN(use_approximate_shadow_catcher, "Use Approximate Shadow Catcher", false); + SOCKET_BOOLEAN(use_transparent_background, "Transparent Background", false); + + /* Notes: + * - Skip passes since they do not follow typical container socket definition. + * Might look into covering those as a socket in the future. + * + * - Skip offset, stride, and pass stride since those can be delivered from the passes and + * rest of the sockets. */ + + return type; } -void RenderBuffers::zero() +BufferParams::BufferParams() : Node(get_node_type()) { - buffer.zero_to_device(); + reset_pass_offset(); } -bool RenderBuffers::copy_from_device() +void BufferParams::update_passes() { - if (!buffer.device_pointer) - return false; - - buffer.copy_from_device(0, params.width * params.get_passes_size(), params.height); - - return true; -} - -bool RenderBuffers::get_denoising_pass_rect( - int type, float exposure, int sample, int components, float *pixels) -{ - if (buffer.data() == NULL) { - return false; - } - - float scale = 1.0f; - float alpha_scale = 1.0f / sample; - if (type == DENOISING_PASS_PREFILTERED_COLOR || type == DENOISING_PASS_CLEAN || - type == DENOISING_PASS_PREFILTERED_INTENSITY) { - scale *= exposure; - } - else if (type == DENOISING_PASS_PREFILTERED_VARIANCE) { - scale *= exposure * exposure * (sample - 1); - } + update_offset_stride(); + reset_pass_offset(); + + pass_stride = 0; + for (const BufferPass &pass : passes) { + if (pass.offset != PASS_UNUSED) { + const int index = pass_to_index(pass); + if (pass_offset_[index] == PASS_UNUSED) { + pass_offset_[index] = pass_stride; + } - int offset; - if (type == DENOISING_PASS_CLEAN) { - /* The clean pass isn't changed by prefiltering, so we use the original one there. */ - offset = type + params.get_denoising_offset(); - scale /= sample; - } - else if (params.denoising_prefiltered_pass) { - offset = type + params.get_denoising_prefiltered_offset(); - } - else { - switch (type) { - case DENOISING_PASS_PREFILTERED_DEPTH: - offset = params.get_denoising_offset() + DENOISING_PASS_DEPTH; - break; - case DENOISING_PASS_PREFILTERED_NORMAL: - offset = params.get_denoising_offset() + DENOISING_PASS_NORMAL; - break; - case DENOISING_PASS_PREFILTERED_ALBEDO: - offset = params.get_denoising_offset() + DENOISING_PASS_ALBEDO; - break; - case DENOISING_PASS_PREFILTERED_COLOR: - /* If we're not saving the prefiltering result, return the original noisy pass. */ - offset = params.get_denoising_offset() + DENOISING_PASS_COLOR; - break; - default: - return false; + pass_stride += pass.get_info().num_components; } - scale /= sample; } +} - int pass_stride = params.get_passes_size(); - int size = params.width * params.height; +void BufferParams::update_passes(const vector<Pass *> &scene_passes) +{ + passes.clear(); - float *in = buffer.data() + offset; + pass_stride = 0; + for (const Pass *scene_pass : scene_passes) { + BufferPass buffer_pass(scene_pass); - if (components == 1) { - for (int i = 0; i < size; i++, in += pass_stride, pixels++) { - pixels[0] = in[0] * scale; + if (scene_pass->is_written()) { + buffer_pass.offset = pass_stride; + pass_stride += scene_pass->get_info().num_components; } - } - else if (components == 3) { - for (int i = 0; i < size; i++, in += pass_stride, pixels += 3) { - pixels[0] = in[0] * scale; - pixels[1] = in[1] * scale; - pixels[2] = in[2] * scale; - } - } - else if (components == 4) { - /* Since the alpha channel is not involved in denoising, output the Combined alpha channel. */ - assert(params.passes[0].type == PASS_COMBINED); - float *in_combined = buffer.data(); - - for (int i = 0; i < size; i++, in += pass_stride, in_combined += pass_stride, pixels += 4) { - float3 val = make_float3(in[0], in[1], in[2]); - if (type == DENOISING_PASS_PREFILTERED_COLOR && params.denoising_prefiltered_pass) { - /* Remove highlight compression from the image. */ - val = color_highlight_uncompress(val); - } - pixels[0] = val.x * scale; - pixels[1] = val.y * scale; - pixels[2] = val.z * scale; - pixels[3] = saturate(in_combined[3] * alpha_scale); + else { + buffer_pass.offset = PASS_UNUSED; } - } - else { - return false; + + passes.emplace_back(std::move(buffer_pass)); } - return true; + update_passes(); } -bool RenderBuffers::get_pass_rect( - const string &name, float exposure, int sample, int components, float *pixels) +void BufferParams::reset_pass_offset() { - if (buffer.data() == NULL) { - return false; + for (int i = 0; i < kNumPassOffsets; ++i) { + pass_offset_[i] = PASS_UNUSED; } +} - float *sample_count = NULL; - if (name == "Combined") { - int sample_offset = 0; - for (size_t j = 0; j < params.passes.size(); j++) { - Pass &pass = params.passes[j]; - if (pass.type != PASS_SAMPLE_COUNT) { - sample_offset += pass.components; - continue; - } - else { - sample_count = buffer.data() + sample_offset; - break; - } - } +int BufferParams::get_pass_offset(PassType pass_type, PassMode mode) const +{ + if (pass_type == PASS_NONE || pass_type == PASS_UNUSED) { + return PASS_UNUSED; } - int pass_offset = 0; - - for (size_t j = 0; j < params.passes.size(); j++) { - Pass &pass = params.passes[j]; + const int index = pass_type_mode_to_index(pass_type, mode); + return pass_offset_[index]; +} - /* Pass is identified by both type and name, multiple of the same type - * may exist with a different name. */ - if (pass.name != name) { - pass_offset += pass.components; - continue; +const BufferPass *BufferParams::find_pass(string_view name) const +{ + for (const BufferPass &pass : passes) { + if (pass.name == name) { + return &pass; } + } - PassType type = pass.type; - - float *in = buffer.data() + pass_offset; - int pass_stride = params.get_passes_size(); - - float scale = (pass.filter) ? 1.0f / (float)sample : 1.0f; - float scale_exposure = (pass.exposure) ? scale * exposure : scale; - - int size = params.width * params.height; + return nullptr; +} - if (components == 1 && type == PASS_RENDER_TIME) { - /* Render time is not stored by kernel, but measured per tile. */ - float val = (float)(1000.0 * render_time / (params.width * params.height * sample)); - for (int i = 0; i < size; i++, pixels++) { - pixels[0] = val; - } - } - else if (components == 1) { - assert(pass.components == components); - - /* Scalar */ - if (type == PASS_DEPTH) { - for (int i = 0; i < size; i++, in += pass_stride, pixels++) { - float f = *in; - pixels[0] = (f == 0.0f) ? 1e10f : f * scale_exposure; - } - } - else if (type == PASS_MIST) { - for (int i = 0; i < size; i++, in += pass_stride, pixels++) { - float f = *in; - pixels[0] = saturate(f * scale_exposure); - } - } - else { - for (int i = 0; i < size; i++, in += pass_stride, pixels++) { - float f = *in; - pixels[0] = f * scale_exposure; - } - } - } - else if (components == 3) { - assert(pass.components == 4); - - /* RGBA */ - if (type == PASS_SHADOW) { - for (int i = 0; i < size; i++, in += pass_stride, pixels += 3) { - float4 f = make_float4(in[0], in[1], in[2], in[3]); - float invw = (f.w > 0.0f) ? 1.0f / f.w : 1.0f; - - pixels[0] = f.x * invw; - pixels[1] = f.y * invw; - pixels[2] = f.z * invw; - } - } - else if (pass.divide_type != PASS_NONE) { - /* RGB lighting passes that need to divide out color */ - pass_offset = 0; - for (size_t k = 0; k < params.passes.size(); k++) { - Pass &color_pass = params.passes[k]; - if (color_pass.type == pass.divide_type) - break; - pass_offset += color_pass.components; - } - - float *in_divide = buffer.data() + pass_offset; - - for (int i = 0; i < size; i++, in += pass_stride, in_divide += pass_stride, pixels += 3) { - float3 f = make_float3(in[0], in[1], in[2]); - float3 f_divide = make_float3(in_divide[0], in_divide[1], in_divide[2]); - - f = safe_divide_even_color(f * exposure, f_divide); - - pixels[0] = f.x; - pixels[1] = f.y; - pixels[2] = f.z; - } - } - else { - /* RGB/vector */ - for (int i = 0; i < size; i++, in += pass_stride, pixels += 3) { - float3 f = make_float3(in[0], in[1], in[2]); - - pixels[0] = f.x * scale_exposure; - pixels[1] = f.y * scale_exposure; - pixels[2] = f.z * scale_exposure; - } - } - } - else if (components == 4) { - assert(pass.components == components); - - /* RGBA */ - if (type == PASS_SHADOW) { - for (int i = 0; i < size; i++, in += pass_stride, pixels += 4) { - float4 f = make_float4(in[0], in[1], in[2], in[3]); - float invw = (f.w > 0.0f) ? 1.0f / f.w : 1.0f; - - pixels[0] = f.x * invw; - pixels[1] = f.y * invw; - pixels[2] = f.z * invw; - pixels[3] = 1.0f; - } - } - else if (type == PASS_MOTION) { - /* need to normalize by number of samples accumulated for motion */ - pass_offset = 0; - for (size_t k = 0; k < params.passes.size(); k++) { - Pass &color_pass = params.passes[k]; - if (color_pass.type == PASS_MOTION_WEIGHT) - break; - pass_offset += color_pass.components; - } - - float *in_weight = buffer.data() + pass_offset; - - for (int i = 0; i < size; i++, in += pass_stride, in_weight += pass_stride, pixels += 4) { - float4 f = make_float4(in[0], in[1], in[2], in[3]); - float w = in_weight[0]; - float invw = (w > 0.0f) ? 1.0f / w : 0.0f; - - pixels[0] = f.x * invw; - pixels[1] = f.y * invw; - pixels[2] = f.z * invw; - pixels[3] = f.w * invw; - } - } - else if (type == PASS_CRYPTOMATTE) { - for (int i = 0; i < size; i++, in += pass_stride, pixels += 4) { - float4 f = make_float4(in[0], in[1], in[2], in[3]); - /* x and z contain integer IDs, don't rescale them. - y and w contain matte weights, they get scaled. */ - pixels[0] = f.x; - pixels[1] = f.y * scale; - pixels[2] = f.z; - pixels[3] = f.w * scale; - } - } - else { - for (int i = 0; i < size; i++, in += pass_stride, pixels += 4) { - if (sample_count && sample_count[i * pass_stride] < 0.0f) { - scale = (pass.filter) ? -1.0f / (sample_count[i * pass_stride]) : 1.0f; - scale_exposure = (pass.exposure) ? scale * exposure : scale; - } - - float4 f = make_float4(in[0], in[1], in[2], in[3]); - - pixels[0] = f.x * scale_exposure; - pixels[1] = f.y * scale_exposure; - pixels[2] = f.z * scale_exposure; - - /* Clamp since alpha might be > 1.0 due to Russian roulette. */ - pixels[3] = saturate(f.w * scale); - } - } +const BufferPass *BufferParams::find_pass(PassType type, PassMode mode) const +{ + for (const BufferPass &pass : passes) { + if (pass.type == type && pass.mode == mode) { + return &pass; } - - return true; } - return false; + return nullptr; } -bool RenderBuffers::set_pass_rect(PassType type, int components, float *pixels, int samples) +const BufferPass *BufferParams::get_actual_display_pass(PassType type, PassMode mode) const { - if (buffer.data() == NULL) { - return false; - } - - int pass_offset = 0; + const BufferPass *pass = find_pass(type, mode); + return get_actual_display_pass(pass); +} - for (size_t j = 0; j < params.passes.size(); j++) { - Pass &pass = params.passes[j]; +const BufferPass *BufferParams::get_actual_display_pass(const BufferPass *pass) const +{ + if (!pass) { + return nullptr; + } - if (pass.type != type) { - pass_offset += pass.components; - continue; + if (pass->type == PASS_COMBINED) { + const BufferPass *shadow_catcher_matte_pass = find_pass(PASS_SHADOW_CATCHER_MATTE, pass->mode); + if (shadow_catcher_matte_pass) { + pass = shadow_catcher_matte_pass; } + } - float *out = buffer.data() + pass_offset; - int pass_stride = params.get_passes_size(); - int size = params.width * params.height; - - assert(pass.components == components); + return pass; +} - for (int i = 0; i < size; i++, out += pass_stride, pixels += components) { - if (pass.filter) { - /* Scale by the number of samples, inverse of what we do in get_pass_rect. - * A better solution would be to remove the need for set_pass_rect entirely, - * and change baking to bake multiple objects in a tile at once. */ - for (int j = 0; j < components; j++) { - out[j] = pixels[j] * samples; - } - } - else { - /* For non-filtered passes just straight copy, these may contain non-float data. */ - memcpy(out, pixels, sizeof(float) * components); - } - } +void BufferParams::update_offset_stride() +{ + offset = -(full_x + full_y * width); + stride = width; +} +bool BufferParams::modified(const BufferParams &other) const +{ + if (!(width == other.width && height == other.height && full_x == other.full_x && + full_y == other.full_y && full_width == other.full_width && + full_height == other.full_height && offset == other.offset && stride == other.stride && + pass_stride == other.pass_stride && layer == other.layer && view == other.view && + exposure == other.exposure && + use_approximate_shadow_catcher == other.use_approximate_shadow_catcher && + use_transparent_background == other.use_transparent_background)) { return true; } - return false; + return !(passes == other.passes); } -/* Display Buffer */ +/* -------------------------------------------------------------------- + * Render Buffers. + */ -DisplayBuffer::DisplayBuffer(Device *device, bool linear) - : draw_width(0), - draw_height(0), - transparent(true), /* todo: determine from background */ - half_float(linear), - rgba_byte(device, "display buffer byte"), - rgba_half(device, "display buffer half") +RenderBuffers::RenderBuffers(Device *device) : buffer(device, "RenderBuffers", MEM_READ_WRITE) { } -DisplayBuffer::~DisplayBuffer() +RenderBuffers::~RenderBuffers() { - rgba_byte.free(); - rgba_half.free(); + buffer.free(); } -void DisplayBuffer::reset(BufferParams ¶ms_) +void RenderBuffers::reset(const BufferParams ¶ms_) { - draw_width = 0; - draw_height = 0; + DCHECK(params_.pass_stride != -1); params = params_; - /* allocate display pixels */ - if (half_float) { - rgba_half.alloc_to_device(params.width, params.height); - } - else { - rgba_byte.alloc_to_device(params.width, params.height); - } + /* re-allocate buffer */ + buffer.alloc(params.width * params.pass_stride, params.height); } -void DisplayBuffer::draw_set(int width, int height) +void RenderBuffers::zero() { - assert(width <= params.width && height <= params.height); + buffer.zero_to_device(); +} - draw_width = width; - draw_height = height; +bool RenderBuffers::copy_from_device() +{ + DCHECK(params.pass_stride != -1); + + if (!buffer.device_pointer) + return false; + + buffer.copy_from_device(0, params.width * params.pass_stride, params.height); + + return true; } -void DisplayBuffer::draw(Device *device, const DeviceDrawParams &draw_params) +void RenderBuffers::copy_to_device() { - if (draw_width != 0 && draw_height != 0) { - device_memory &rgba = (half_float) ? (device_memory &)rgba_half : (device_memory &)rgba_byte; - - device->draw_pixels(rgba, - 0, - draw_width, - draw_height, - params.width, - params.height, - params.full_x, - params.full_y, - params.full_width, - params.full_height, - transparent, - draw_params); - } + buffer.copy_to_device(); } -bool DisplayBuffer::draw_ready() +void render_buffers_host_copy_denoised(RenderBuffers *dst, + const BufferParams &dst_params, + const RenderBuffers *src, + const BufferParams &src_params, + const size_t src_offset) { - return (draw_width != 0 && draw_height != 0); + DCHECK_EQ(dst_params.width, src_params.width); + /* TODO(sergey): More sanity checks to avoid buffer overrun. */ + + /* Create a map of pass ofsets to be copied. + * Assume offsets are different to allow copying passes between buffers with different set of + * passes. */ + + struct { + int dst_offset; + int src_offset; + } pass_offsets[PASS_NUM]; + + int num_passes = 0; + + for (int i = 0; i < PASS_NUM; ++i) { + const PassType pass_type = static_cast<PassType>(i); + + const int dst_pass_offset = dst_params.get_pass_offset(pass_type, PassMode::DENOISED); + if (dst_pass_offset == PASS_UNUSED) { + continue; + } + + const int src_pass_offset = src_params.get_pass_offset(pass_type, PassMode::DENOISED); + if (src_pass_offset == PASS_UNUSED) { + continue; + } + + pass_offsets[num_passes].dst_offset = dst_pass_offset; + pass_offsets[num_passes].src_offset = src_pass_offset; + ++num_passes; + } + + /* Copy passes. */ + /* TODO(sergey): Make it more reusable, allowing implement copy of noisy passes. */ + + const int64_t dst_width = dst_params.width; + const int64_t dst_height = dst_params.height; + const int64_t dst_pass_stride = dst_params.pass_stride; + const int64_t dst_num_pixels = dst_width * dst_height; + + const int64_t src_pass_stride = src_params.pass_stride; + const int64_t src_offset_in_floats = src_offset * src_pass_stride; + + const float *src_pixel = src->buffer.data() + src_offset_in_floats; + float *dst_pixel = dst->buffer.data(); + + for (int i = 0; i < dst_num_pixels; + ++i, src_pixel += src_pass_stride, dst_pixel += dst_pass_stride) { + for (int pass_offset_idx = 0; pass_offset_idx < num_passes; ++pass_offset_idx) { + const int dst_pass_offset = pass_offsets[pass_offset_idx].dst_offset; + const int src_pass_offset = pass_offsets[pass_offset_idx].src_offset; + + /* TODO(sergey): Support non-RGBA passes. */ + dst_pixel[dst_pass_offset + 0] = src_pixel[src_pass_offset + 0]; + dst_pixel[dst_pass_offset + 1] = src_pixel[src_pass_offset + 1]; + dst_pixel[dst_pass_offset + 2] = src_pixel[src_pass_offset + 2]; + dst_pixel[dst_pass_offset + 3] = src_pixel[src_pass_offset + 3]; + } + } } CCL_NAMESPACE_END diff --git a/intern/cycles/render/buffers.h b/intern/cycles/render/buffers.h index 4ffc628bb52..c048234167d 100644 --- a/intern/cycles/render/buffers.h +++ b/intern/cycles/render/buffers.h @@ -18,8 +18,8 @@ #define __BUFFERS_H__ #include "device/device_memory.h" - -#include "render/film.h" +#include "graph/node.h" +#include "render/pass.h" #include "kernel/kernel_types.h" @@ -34,170 +34,156 @@ class Device; struct DeviceDrawParams; struct float4; +/* NOTE: Is not a real scene node. Using Node API for ease of (de)serialization. */ +class BufferPass : public Node { + public: + NODE_DECLARE + + PassType type = PASS_NONE; + PassMode mode = PassMode::NOISY; + ustring name; + bool include_albedo = false; + + int offset = -1; + + BufferPass(); + explicit BufferPass(const Pass *scene_pass); + + BufferPass(BufferPass &&other) noexcept = default; + BufferPass(const BufferPass &other) = default; + + BufferPass &operator=(BufferPass &&other) = default; + BufferPass &operator=(const BufferPass &other) = default; + + ~BufferPass() = default; + + PassInfo get_info() const; + + inline bool operator==(const BufferPass &other) const + { + return type == other.type && mode == other.mode && name == other.name && + include_albedo == other.include_albedo && offset == other.offset; + } + inline bool operator!=(const BufferPass &other) const + { + return !(*this == other); + } +}; + /* Buffer Parameters * Size of render buffer and how it fits in the full image (border render). */ -class BufferParams { +/* NOTE: Is not a real scene node. Using Node API for ease of (de)serialization. */ +class BufferParams : public Node { public: - /* width/height of the physical buffer */ - int width; - int height; - - /* offset into and width/height of the full buffer */ - int full_x; - int full_y; - int full_width; - int full_height; - - /* passes */ - vector<Pass> passes; - bool denoising_data_pass; - /* If only some light path types should be target, an additional pass is needed. */ - bool denoising_clean_pass; - /* When we're prefiltering the passes during rendering, we need to keep both the - * original and the prefiltered data around because neighboring tiles might still - * need the original data. */ - bool denoising_prefiltered_pass; - - /* functions */ - BufferParams(); + NODE_DECLARE - void get_offset_stride(int &offset, int &stride); - bool modified(const BufferParams ¶ms); - int get_passes_size(); - int get_denoising_offset(); - int get_denoising_prefiltered_offset(); -}; + /* Width/height of the physical buffer. */ + int width = 0; + int height = 0; -/* Render Buffers */ + /* Offset into and width/height of the full buffer. */ + int full_x = 0; + int full_y = 0; + int full_width = 0; + int full_height = 0; -class RenderBuffers { - public: - /* buffer parameters */ - BufferParams params; + /* Runtime fields, only valid after `update_passes()` or `update_offset_stride()`. */ + int offset = -1, stride = -1; - /* float buffer */ - device_vector<float> buffer; - bool map_neighbor_copied; - double render_time; + /* Runtime fields, only valid after `update_passes()`. */ + int pass_stride = -1; - explicit RenderBuffers(Device *device); - ~RenderBuffers(); + /* Properties which are used for accessing buffer pixels outside of scene graph. */ + vector<BufferPass> passes; + ustring layer; + ustring view; + float exposure = 1.0f; + bool use_approximate_shadow_catcher = false; + bool use_transparent_background = false; - void reset(BufferParams ¶ms); - void zero(); + BufferParams(); - bool copy_from_device(); - bool get_pass_rect( - const string &name, float exposure, int sample, int components, float *pixels); - bool get_denoising_pass_rect( - int offset, float exposure, int sample, int components, float *pixels); - bool set_pass_rect(PassType type, int components, float *pixels, int samples); -}; + BufferParams(BufferParams &&other) noexcept = default; + BufferParams(const BufferParams &other) = default; -/* Display Buffer - * - * The buffer used for drawing during render, filled by converting the render - * buffers to byte of half float storage */ + BufferParams &operator=(BufferParams &&other) = default; + BufferParams &operator=(const BufferParams &other) = default; -class DisplayBuffer { - public: - /* buffer parameters */ - BufferParams params; - /* dimensions for how much of the buffer is actually ready for display. - * with progressive render we can be using only a subset of the buffer. - * if these are zero, it means nothing can be drawn yet */ - int draw_width, draw_height; - /* draw alpha channel? */ - bool transparent; - /* use half float? */ - bool half_float; - /* byte buffer for converted result */ - device_pixels<uchar4> rgba_byte; - device_pixels<half4> rgba_half; - - DisplayBuffer(Device *device, bool linear = false); - ~DisplayBuffer(); - - void reset(BufferParams ¶ms); - - void draw_set(int width, int height); - void draw(Device *device, const DeviceDrawParams &draw_params); - bool draw_ready(); -}; + ~BufferParams() = default; -/* Render Tile - * Rendering task on a buffer */ + /* Pre-calculate all fields which depends on the passes. + * + * When the scene passes are given, the buffer passes will be created from them and stored in + * this params, and then params are updated for those passes. + * The `update_passes()` without parameters updates offsets and stries which are stored outside + * of the passes. */ + void update_passes(); + void update_passes(const vector<Pass *> &scene_passes); -class RenderTile { - public: - typedef enum { PATH_TRACE = (1 << 0), BAKE = (1 << 1), DENOISE = (1 << 2) } Task; + /* Returns PASS_UNUSED if there is no such pass in the buffer. */ + int get_pass_offset(PassType type, PassMode mode = PassMode::NOISY) const; - Task task; - int x, y, w, h; - int start_sample; - int num_samples; - int sample; - int resolution; - int offset; - int stride; - int tile_index; + /* Returns nullptr if pass with given name does not exist. */ + const BufferPass *find_pass(string_view name) const; + const BufferPass *find_pass(PassType type, PassMode mode = PassMode::NOISY) const; - device_ptr buffer; - int device_size; + /* Get display pass from its name. + * Will do special logic to replace combined pass with shadow catcher matte. */ + const BufferPass *get_actual_display_pass(PassType type, PassMode mode = PassMode::NOISY) const; + const BufferPass *get_actual_display_pass(const BufferPass *pass) const; - typedef enum { NO_STEALING = 0, CAN_BE_STOLEN = 1, WAS_STOLEN = 2 } StealingState; - StealingState stealing_state; + void update_offset_stride(); - RenderBuffers *buffers; + bool modified(const BufferParams &other) const; - RenderTile(); + protected: + void reset_pass_offset(); - int4 bounds() const - { - return make_int4(x, /* xmin */ - y, /* ymin */ - x + w, /* xmax */ - y + h); /* ymax */ - } + /* Multipled by 2 to be able to store noisy and denoised pass types. */ + static constexpr int kNumPassOffsets = PASS_NUM * 2; + + /* Indexed by an index derived from pass type and mode, indicates offset of the corresponding + * pass in the buffer. + * If there are multiple passes with same type and mode contains lowest offset of all of them. */ + int pass_offset_[kNumPassOffsets]; }; -/* Render Tile Neighbors - * Set of neighboring tiles used for denoising. Tile order: - * 0 1 2 - * 3 4 5 - * 6 7 8 */ +/* Render Buffers */ -class RenderTileNeighbors { +class RenderBuffers { public: - static const int SIZE = 9; - static const int CENTER = 4; + /* buffer parameters */ + BufferParams params; - RenderTile tiles[SIZE]; - RenderTile target; + /* float buffer */ + device_vector<float> buffer; - RenderTileNeighbors(const RenderTile ¢er) - { - tiles[CENTER] = center; - } + explicit RenderBuffers(Device *device); + ~RenderBuffers(); - int4 bounds() const - { - return make_int4(tiles[3].x, /* xmin */ - tiles[1].y, /* ymin */ - tiles[5].x + tiles[5].w, /* xmax */ - tiles[7].y + tiles[7].h); /* ymax */ - } + void reset(const BufferParams ¶ms); + void zero(); - void set_bounds_from_center() - { - tiles[3].x = tiles[CENTER].x; - tiles[1].y = tiles[CENTER].y; - tiles[5].x = tiles[CENTER].x + tiles[CENTER].w; - tiles[7].y = tiles[CENTER].y + tiles[CENTER].h; - } + bool copy_from_device(); + void copy_to_device(); }; +/* Copy denoised passes form source to destination. + * + * Buffer parameters are provided explicitly, allowing to copy pixelks between render buffers which + * content corresponds to a render result at a non-unit resolution divider. + * + * `src_offset` allows to offset source pixel index which is used when a fraction of the source + * buffer is to be copied. + * + * Copy happens of the number of pixels in the destination. */ +void render_buffers_host_copy_denoised(RenderBuffers *dst, + const BufferParams &dst_params, + const RenderBuffers *src, + const BufferParams &src_params, + const size_t src_offset = 0); + CCL_NAMESPACE_END #endif /* __BUFFERS_H__ */ diff --git a/intern/cycles/render/camera.cpp b/intern/cycles/render/camera.cpp index 327f166f9d8..8b69c971991 100644 --- a/intern/cycles/render/camera.cpp +++ b/intern/cycles/render/camera.cpp @@ -33,9 +33,9 @@ /* needed for calculating differentials */ // clang-format off -#include "kernel/kernel_compat_cpu.h" -#include "kernel/split/kernel_split_data.h" -#include "kernel/kernel_globals.h" +#include "kernel/device/cpu/compat.h" +#include "kernel/device/cpu/globals.h" + #include "kernel/kernel_projection.h" #include "kernel/kernel_differential.h" #include "kernel/kernel_montecarlo.h" @@ -169,7 +169,6 @@ Camera::Camera() : Node(get_node_type()) width = 1024; height = 512; - resolution = 1; use_perspective_motion = false; @@ -455,7 +454,6 @@ void Camera::update(Scene *scene) /* render size */ kcam->width = width; kcam->height = height; - kcam->resolution = resolution; /* store differentials */ kcam->dx = float3_to_float4(dx); @@ -776,9 +774,11 @@ float Camera::world_to_raster_size(float3 P) &ray); #endif - differential_transfer(&ray.dP, ray.dP, ray.D, ray.dD, ray.D, dist); + /* TODO: would it help to use more accurate differentials here? */ + differential3 dP; + differential_transfer_compact(&dP, ray.dP, ray.D, ray.dD, ray.D, dist); - return max(len(ray.dP.dx), len(ray.dP.dy)); + return max(len(dP.dx), len(dP.dy)); } return res; @@ -789,12 +789,11 @@ bool Camera::use_motion() const return motion.size() > 1; } -void Camera::set_screen_size_and_resolution(int width_, int height_, int resolution_) +void Camera::set_screen_size(int width_, int height_) { - if (width_ != width || height_ != height || resolution_ != resolution) { + if (width_ != width || height_ != height) { width = width_; height = height_; - resolution = resolution_; tag_modified(); } } diff --git a/intern/cycles/render/camera.h b/intern/cycles/render/camera.h index 5abb4750764..cb8ecac1a7e 100644 --- a/intern/cycles/render/camera.h +++ b/intern/cycles/render/camera.h @@ -199,7 +199,6 @@ class Camera : public Node { private: int width; int height; - int resolution; public: /* functions */ @@ -225,7 +224,7 @@ class Camera : public Node { int motion_step(float time) const; bool use_motion() const; - void set_screen_size_and_resolution(int width_, int height_, int resolution_); + void set_screen_size(int width_, int height_); private: /* Private utility functions. */ diff --git a/intern/cycles/render/coverage.cpp b/intern/cycles/render/coverage.cpp deleted file mode 100644 index 99d4daa6961..00000000000 --- a/intern/cycles/render/coverage.cpp +++ /dev/null @@ -1,155 +0,0 @@ -/* - * Copyright 2018 Blender Foundation - * - * Licensed under the Apache License, Version 2.0 (the "License"); - * you may not use this file except in compliance with the License. - * You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - * See the License for the specific language governing permissions and - * limitations under the License. - */ - -#include "render/coverage.h" -#include "render/buffers.h" - -#include "kernel/kernel_compat_cpu.h" -#include "kernel/kernel_types.h" -#include "kernel/split/kernel_split_data.h" - -#include "kernel/kernel_globals.h" -#include "kernel/kernel_id_passes.h" - -#include "util/util_map.h" - -CCL_NAMESPACE_BEGIN - -static bool crypomatte_comp(const pair<float, float> &i, const pair<float, float> j) -{ - return i.first > j.first; -} - -void Coverage::finalize() -{ - int pass_offset = 0; - if (kernel_data.film.cryptomatte_passes & CRYPT_OBJECT) { - finalize_buffer(coverage_object, pass_offset); - pass_offset += kernel_data.film.cryptomatte_depth * 4; - } - if (kernel_data.film.cryptomatte_passes & CRYPT_MATERIAL) { - finalize_buffer(coverage_material, pass_offset); - pass_offset += kernel_data.film.cryptomatte_depth * 4; - } - if (kernel_data.film.cryptomatte_passes & CRYPT_ASSET) { - finalize_buffer(coverage_asset, pass_offset); - } -} - -void Coverage::init_path_trace() -{ - kg->coverage_object = kg->coverage_material = kg->coverage_asset = NULL; - - if (kernel_data.film.cryptomatte_passes & CRYPT_ACCURATE) { - if (kernel_data.film.cryptomatte_passes & CRYPT_OBJECT) { - coverage_object.clear(); - coverage_object.resize(tile.w * tile.h); - } - if (kernel_data.film.cryptomatte_passes & CRYPT_MATERIAL) { - coverage_material.clear(); - coverage_material.resize(tile.w * tile.h); - } - if (kernel_data.film.cryptomatte_passes & CRYPT_ASSET) { - coverage_asset.clear(); - coverage_asset.resize(tile.w * tile.h); - } - } -} - -void Coverage::init_pixel(int x, int y) -{ - if (kernel_data.film.cryptomatte_passes & CRYPT_ACCURATE) { - const int pixel_index = tile.w * (y - tile.y) + x - tile.x; - if (kernel_data.film.cryptomatte_passes & CRYPT_OBJECT) { - kg->coverage_object = &coverage_object[pixel_index]; - } - if (kernel_data.film.cryptomatte_passes & CRYPT_MATERIAL) { - kg->coverage_material = &coverage_material[pixel_index]; - } - if (kernel_data.film.cryptomatte_passes & CRYPT_ASSET) { - kg->coverage_asset = &coverage_asset[pixel_index]; - } - } -} - -void Coverage::finalize_buffer(vector<CoverageMap> &coverage, const int pass_offset) -{ - if (kernel_data.film.cryptomatte_passes & CRYPT_ACCURATE) { - flatten_buffer(coverage, pass_offset); - } - else { - sort_buffer(pass_offset); - } -} - -void Coverage::flatten_buffer(vector<CoverageMap> &coverage, const int pass_offset) -{ - /* Sort the coverage map and write it to the output */ - int pixel_index = 0; - int pass_stride = tile.buffers->params.get_passes_size(); - for (int y = 0; y < tile.h; ++y) { - for (int x = 0; x < tile.w; ++x) { - const CoverageMap &pixel = coverage[pixel_index]; - if (!pixel.empty()) { - /* buffer offset */ - int index = x + y * tile.stride; - float *buffer = (float *)tile.buffer + index * pass_stride; - - /* sort the cryptomatte pixel */ - vector<pair<float, float>> sorted_pixel; - for (CoverageMap::const_iterator it = pixel.begin(); it != pixel.end(); ++it) { - sorted_pixel.push_back(std::make_pair(it->second, it->first)); - } - sort(sorted_pixel.begin(), sorted_pixel.end(), crypomatte_comp); - int num_slots = 2 * (kernel_data.film.cryptomatte_depth); - if (sorted_pixel.size() > num_slots) { - float leftover = 0.0f; - for (vector<pair<float, float>>::iterator it = sorted_pixel.begin() + num_slots; - it != sorted_pixel.end(); - ++it) { - leftover += it->first; - } - sorted_pixel[num_slots - 1].first += leftover; - } - int limit = min(num_slots, sorted_pixel.size()); - for (int i = 0; i < limit; ++i) { - kernel_write_id_slots(buffer + kernel_data.film.pass_cryptomatte + pass_offset, - 2 * (kernel_data.film.cryptomatte_depth), - sorted_pixel[i].second, - sorted_pixel[i].first); - } - } - ++pixel_index; - } - } -} - -void Coverage::sort_buffer(const int pass_offset) -{ - /* Sort the coverage map and write it to the output */ - int pass_stride = tile.buffers->params.get_passes_size(); - for (int y = 0; y < tile.h; ++y) { - for (int x = 0; x < tile.w; ++x) { - /* buffer offset */ - int index = x + y * tile.stride; - float *buffer = (float *)tile.buffer + index * pass_stride; - kernel_sort_id_slots(buffer + kernel_data.film.pass_cryptomatte + pass_offset, - 2 * (kernel_data.film.cryptomatte_depth)); - } - } -} - -CCL_NAMESPACE_END diff --git a/intern/cycles/render/coverage.h b/intern/cycles/render/coverage.h deleted file mode 100644 index 12182c614da..00000000000 --- a/intern/cycles/render/coverage.h +++ /dev/null @@ -1,52 +0,0 @@ -/* - * Copyright 2018 Blender Foundation - * - * Licensed under the Apache License, Version 2.0 (the "License"); - * you may not use this file except in compliance with the License. - * You may obtain a copy of the License at - * - * http://www.apache.org/licenses/LICENSE-2.0 - * - * Unless required by applicable law or agreed to in writing, software - * distributed under the License is distributed on an "AS IS" BASIS, - * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. - * See the License for the specific language governing permissions and - * limitations under the License. - */ - -#ifndef __COVERAGE_H__ -#define __COVERAGE_H__ - -#include "util/util_map.h" -#include "util/util_vector.h" - -CCL_NAMESPACE_BEGIN - -struct KernelGlobals; -class RenderTile; - -typedef unordered_map<float, float> CoverageMap; - -class Coverage { - public: - Coverage(KernelGlobals *kg_, RenderTile &tile_) : kg(kg_), tile(tile_) - { - } - void init_path_trace(); - void init_pixel(int x, int y); - void finalize(); - - private: - vector<CoverageMap> coverage_object; - vector<CoverageMap> coverage_material; - vector<CoverageMap> coverage_asset; - KernelGlobals *kg; - RenderTile &tile; - void finalize_buffer(vector<CoverageMap> &coverage, const int pass_offset); - void flatten_buffer(vector<CoverageMap> &coverage, const int pass_offset); - void sort_buffer(const int pass_offset); -}; - -CCL_NAMESPACE_END - -#endif /* __COVERAGE_H__ */ diff --git a/intern/cycles/render/denoising.cpp b/intern/cycles/render/denoising.cpp index ddbe7484800..bcf8d3fa204 100644 --- a/intern/cycles/render/denoising.cpp +++ b/intern/cycles/render/denoising.cpp @@ -16,15 +16,17 @@ #include "render/denoising.h" -#include "kernel/filter/filter_defines.h" +#if 0 -#include "util/util_foreach.h" -#include "util/util_map.h" -#include "util/util_system.h" -#include "util/util_task.h" -#include "util/util_time.h" +# include "kernel/filter/filter_defines.h" -#include <OpenImageIO/filesystem.h> +# include "util/util_foreach.h" +# include "util/util_map.h" +# include "util/util_system.h" +# include "util/util_task.h" +# include "util/util_time.h" + +# include <OpenImageIO/filesystem.h> CCL_NAMESPACE_BEGIN @@ -225,7 +227,7 @@ bool DenoiseImageLayer::match_channels(int neighbor, /* Denoise Task */ DenoiseTask::DenoiseTask(Device *device, - Denoiser *denoiser, + DenoiserPipeline *denoiser, int frame, const vector<int> &neighbor_frames) : denoiser(denoiser), @@ -386,7 +388,6 @@ void DenoiseTask::create_task(DeviceTask &task) task.denoising = denoiser->params; task.denoising.type = DENOISER_NLM; task.denoising.use = true; - task.denoising.store_passes = false; task.denoising_from_render = false; task.denoising_frames.resize(neighbor_frames.size()); @@ -863,7 +864,7 @@ bool DenoiseImage::save_output(const string &out_filepath, string &error) /* File pattern handling and outer loop over frames */ -Denoiser::Denoiser(DeviceInfo &device_info) +DenoiserPipeline::DenoiserPipeline(DeviceInfo &device_info) { samples_override = 0; tile_size = make_int2(64, 64); @@ -876,18 +877,16 @@ Denoiser::Denoiser(DeviceInfo &device_info) /* Initialize device. */ device = Device::create(device_info, stats, profiler, true); - DeviceRequestedFeatures req; - req.use_denoising = true; - device->load_kernels(req); + device->load_kernels(KERNEL_FEATURE_DENOISING); } -Denoiser::~Denoiser() +DenoiserPipeline::~DenoiserPipeline() { delete device; TaskScheduler::exit(); } -bool Denoiser::run() +bool DenoiserPipeline::run() { assert(input.size() == output.size()); @@ -931,3 +930,5 @@ bool Denoiser::run() } CCL_NAMESPACE_END + +#endif diff --git a/intern/cycles/render/denoising.h b/intern/cycles/render/denoising.h index c1b4d0a5596..097cc570d06 100644 --- a/intern/cycles/render/denoising.h +++ b/intern/cycles/render/denoising.h @@ -17,27 +17,31 @@ #ifndef __DENOISING_H__ #define __DENOISING_H__ -#include "device/device.h" -#include "device/device_denoising.h" +#if 0 -#include "render/buffers.h" +/* TODO(sergey): Make it explicit and clear when something is a denoiser, its pipeline or + * parameters. Currently it is an annoying mixture of terms used interchangeably. */ -#include "util/util_string.h" -#include "util/util_unique_ptr.h" -#include "util/util_vector.h" +# include "device/device.h" -#include <OpenImageIO/imageio.h> +# include "render/buffers.h" + +# include "util/util_string.h" +# include "util/util_unique_ptr.h" +# include "util/util_vector.h" + +# include <OpenImageIO/imageio.h> OIIO_NAMESPACE_USING CCL_NAMESPACE_BEGIN -/* Denoiser */ +/* Denoiser pipeline */ -class Denoiser { +class DenoiserPipeline { public: - Denoiser(DeviceInfo &device_info); - ~Denoiser(); + DenoiserPipeline(DeviceInfo &device_info); + ~DenoiserPipeline(); bool run(); @@ -155,7 +159,10 @@ class DenoiseImage { class DenoiseTask { public: - DenoiseTask(Device *device, Denoiser *denoiser, int frame, const vector<int> &neighbor_frames); + DenoiseTask(Device *device, + DenoiserPipeline *denoiser, + int frame, + const vector<int> &neighbor_frames); ~DenoiseTask(); /* Task stages */ @@ -168,7 +175,7 @@ class DenoiseTask { protected: /* Denoiser parameters and device */ - Denoiser *denoiser; + DenoiserPipeline *denoiser; Device *device; /* Frame number to be denoised */ @@ -204,4 +211,6 @@ class DenoiseTask { CCL_NAMESPACE_END +#endif + #endif /* __DENOISING_H__ */ diff --git a/intern/cycles/render/film.cpp b/intern/cycles/render/film.cpp index 5df396394c4..8e14b338bd3 100644 --- a/intern/cycles/render/film.cpp +++ b/intern/cycles/render/film.cpp @@ -16,9 +16,12 @@ #include "render/film.h" #include "device/device.h" +#include "render/background.h" +#include "render/bake.h" #include "render/camera.h" #include "render/integrator.h" #include "render/mesh.h" +#include "render/object.h" #include "render/scene.h" #include "render/stats.h" #include "render/tables.h" @@ -31,261 +34,6 @@ CCL_NAMESPACE_BEGIN -/* Pass */ - -static bool compare_pass_order(const Pass &a, const Pass &b) -{ - if (a.components == b.components) - return (a.type < b.type); - return (a.components > b.components); -} - -static NodeEnum *get_pass_type_enum() -{ - static NodeEnum pass_type_enum; - pass_type_enum.insert("combined", PASS_COMBINED); - pass_type_enum.insert("depth", PASS_DEPTH); - pass_type_enum.insert("normal", PASS_NORMAL); - pass_type_enum.insert("uv", PASS_UV); - pass_type_enum.insert("object_id", PASS_OBJECT_ID); - pass_type_enum.insert("material_id", PASS_MATERIAL_ID); - pass_type_enum.insert("motion", PASS_MOTION); - pass_type_enum.insert("motion_weight", PASS_MOTION_WEIGHT); - pass_type_enum.insert("render_time", PASS_RENDER_TIME); - pass_type_enum.insert("cryptomatte", PASS_CRYPTOMATTE); - pass_type_enum.insert("aov_color", PASS_AOV_COLOR); - pass_type_enum.insert("aov_value", PASS_AOV_VALUE); - pass_type_enum.insert("adaptive_aux_buffer", PASS_ADAPTIVE_AUX_BUFFER); - pass_type_enum.insert("sample_count", PASS_SAMPLE_COUNT); - pass_type_enum.insert("mist", PASS_MIST); - pass_type_enum.insert("emission", PASS_EMISSION); - pass_type_enum.insert("background", PASS_BACKGROUND); - pass_type_enum.insert("ambient_occlusion", PASS_AO); - pass_type_enum.insert("shadow", PASS_SHADOW); - pass_type_enum.insert("diffuse_direct", PASS_DIFFUSE_DIRECT); - pass_type_enum.insert("diffuse_indirect", PASS_DIFFUSE_INDIRECT); - pass_type_enum.insert("diffuse_color", PASS_DIFFUSE_COLOR); - pass_type_enum.insert("glossy_direct", PASS_GLOSSY_DIRECT); - pass_type_enum.insert("glossy_indirect", PASS_GLOSSY_INDIRECT); - pass_type_enum.insert("glossy_color", PASS_GLOSSY_COLOR); - pass_type_enum.insert("transmission_direct", PASS_TRANSMISSION_DIRECT); - pass_type_enum.insert("transmission_indirect", PASS_TRANSMISSION_INDIRECT); - pass_type_enum.insert("transmission_color", PASS_TRANSMISSION_COLOR); - pass_type_enum.insert("volume_direct", PASS_VOLUME_DIRECT); - pass_type_enum.insert("volume_indirect", PASS_VOLUME_INDIRECT); - pass_type_enum.insert("bake_primitive", PASS_BAKE_PRIMITIVE); - pass_type_enum.insert("bake_differential", PASS_BAKE_DIFFERENTIAL); - - return &pass_type_enum; -} - -NODE_DEFINE(Pass) -{ - NodeType *type = NodeType::add("pass", create); - - NodeEnum *pass_type_enum = get_pass_type_enum(); - SOCKET_ENUM(type, "Type", *pass_type_enum, PASS_COMBINED); - SOCKET_STRING(name, "Name", ustring()); - - return type; -} - -Pass::Pass() : Node(get_node_type()) -{ -} - -void Pass::add(PassType type, vector<Pass> &passes, const char *name) -{ - for (size_t i = 0; i < passes.size(); i++) { - if (passes[i].type != type) { - continue; - } - - /* An empty name is used as a placeholder to signal that any pass of - * that type is fine (because the content always is the same). - * This is important to support divide_type: If the pass that has a - * divide_type is added first, a pass for divide_type with an empty - * name will be added. Then, if a matching pass with a name is later - * requested, the existing placeholder will be renamed to that. - * If the divide_type is explicitly allocated with a name first and - * then again as part of another pass, the second one will just be - * skipped because that type already exists. */ - - /* If no name is specified, any pass of the correct type will match. */ - if (name == NULL) { - return; - } - - /* If we already have a placeholder pass, rename that one. */ - if (passes[i].name.empty()) { - passes[i].name = name; - return; - } - - /* If neither existing nor requested pass have placeholder name, they - * must match. */ - if (name == passes[i].name) { - return; - } - } - - Pass pass; - - pass.type = type; - pass.filter = true; - pass.exposure = false; - pass.divide_type = PASS_NONE; - if (name) { - pass.name = name; - } - - switch (type) { - case PASS_NONE: - pass.components = 0; - break; - case PASS_COMBINED: - pass.components = 4; - pass.exposure = true; - break; - case PASS_DEPTH: - pass.components = 1; - pass.filter = false; - break; - case PASS_MIST: - pass.components = 1; - break; - case PASS_NORMAL: - pass.components = 4; - break; - case PASS_UV: - pass.components = 4; - break; - case PASS_MOTION: - pass.components = 4; - pass.divide_type = PASS_MOTION_WEIGHT; - break; - case PASS_MOTION_WEIGHT: - pass.components = 1; - break; - case PASS_OBJECT_ID: - case PASS_MATERIAL_ID: - pass.components = 1; - pass.filter = false; - break; - - case PASS_EMISSION: - case PASS_BACKGROUND: - pass.components = 4; - pass.exposure = true; - break; - case PASS_AO: - pass.components = 4; - break; - case PASS_SHADOW: - pass.components = 4; - pass.exposure = false; - break; - case PASS_LIGHT: - /* This isn't a real pass, used by baking to see whether - * light data is needed or not. - * - * Set components to 0 so pass sort below happens in a - * determined way. - */ - pass.components = 0; - break; - case PASS_RENDER_TIME: - /* This pass is handled entirely on the host side. */ - pass.components = 0; - break; - - case PASS_DIFFUSE_COLOR: - case PASS_GLOSSY_COLOR: - case PASS_TRANSMISSION_COLOR: - pass.components = 4; - break; - case PASS_DIFFUSE_DIRECT: - case PASS_DIFFUSE_INDIRECT: - pass.components = 4; - pass.exposure = true; - pass.divide_type = PASS_DIFFUSE_COLOR; - break; - case PASS_GLOSSY_DIRECT: - case PASS_GLOSSY_INDIRECT: - pass.components = 4; - pass.exposure = true; - pass.divide_type = PASS_GLOSSY_COLOR; - break; - case PASS_TRANSMISSION_DIRECT: - case PASS_TRANSMISSION_INDIRECT: - pass.components = 4; - pass.exposure = true; - pass.divide_type = PASS_TRANSMISSION_COLOR; - break; - case PASS_VOLUME_DIRECT: - case PASS_VOLUME_INDIRECT: - pass.components = 4; - pass.exposure = true; - break; - case PASS_CRYPTOMATTE: - pass.components = 4; - break; - case PASS_ADAPTIVE_AUX_BUFFER: - pass.components = 4; - break; - case PASS_SAMPLE_COUNT: - pass.components = 1; - pass.exposure = false; - break; - case PASS_AOV_COLOR: - pass.components = 4; - break; - case PASS_AOV_VALUE: - pass.components = 1; - break; - case PASS_BAKE_PRIMITIVE: - case PASS_BAKE_DIFFERENTIAL: - pass.components = 4; - pass.exposure = false; - pass.filter = false; - break; - default: - assert(false); - break; - } - - passes.push_back(pass); - - /* Order from by components, to ensure alignment so passes with size 4 - * come first and then passes with size 1. Note this must use stable sort - * so cryptomatte passes remain in the right order. */ - stable_sort(&passes[0], &passes[0] + passes.size(), compare_pass_order); - - if (pass.divide_type != PASS_NONE) - Pass::add(pass.divide_type, passes); -} - -bool Pass::equals(const vector<Pass> &A, const vector<Pass> &B) -{ - if (A.size() != B.size()) - return false; - - for (int i = 0; i < A.size(); i++) - if (A[i].type != B[i].type || A[i].name != B[i].name) - return false; - - return true; -} - -bool Pass::contains(const vector<Pass> &passes, PassType type) -{ - for (size_t i = 0; i < passes.size(); i++) - if (passes[i].type == type) - return true; - - return false; -} - /* Pixel Filter */ static float filter_func_box(float /*v*/, float /*width*/) @@ -368,17 +116,11 @@ NODE_DEFINE(Film) SOCKET_FLOAT(mist_depth, "Mist Depth", 100.0f); SOCKET_FLOAT(mist_falloff, "Mist Falloff", 1.0f); - SOCKET_BOOLEAN(denoising_data_pass, "Generate Denoising Data Pass", false); - SOCKET_BOOLEAN(denoising_clean_pass, "Generate Denoising Clean Pass", false); - SOCKET_BOOLEAN(denoising_prefiltered_pass, "Generate Denoising Prefiltered Pass", false); - SOCKET_INT(denoising_flags, "Denoising Flags", 0); - SOCKET_BOOLEAN(use_adaptive_sampling, "Use Adaptive Sampling", false); - - SOCKET_BOOLEAN(use_light_visibility, "Use Light Visibility", false); - - NodeEnum *pass_type_enum = get_pass_type_enum(); + const NodeEnum *pass_type_enum = Pass::get_type_enum(); SOCKET_ENUM(display_pass, "Display Pass", *pass_type_enum, PASS_COMBINED); + SOCKET_BOOLEAN(show_active_pixels, "Show Active Pixels", false); + static NodeEnum cryptomatte_passes_enum; cryptomatte_passes_enum.insert("none", CRYPT_NONE); cryptomatte_passes_enum.insert("object", CRYPT_OBJECT); @@ -389,15 +131,13 @@ NODE_DEFINE(Film) SOCKET_INT(cryptomatte_depth, "Cryptomatte Depth", 0); + SOCKET_BOOLEAN(use_approximate_shadow_catcher, "Use Approximate Shadow Catcher", false); + return type; } -Film::Film() : Node(get_node_type()) +Film::Film() : Node(get_node_type()), filter_table_offset_(TABLE_OFFSET_INVALID) { - use_light_visibility = false; - filter_table_offset = TABLE_OFFSET_INVALID; - cryptomatte_passes = CRYPT_NONE; - display_pass = PASS_COMBINED; } Film::~Film() @@ -406,7 +146,8 @@ Film::~Film() void Film::add_default(Scene *scene) { - Pass::add(PASS_COMBINED, scene->passes); + Pass *pass = scene->create_node<Pass>(); + pass->set_type(PASS_COMBINED); } void Film::device_update(Device *device, DeviceScene *dscene, Scene *scene) @@ -426,50 +167,77 @@ void Film::device_update(Device *device, DeviceScene *dscene, Scene *scene) /* update __data */ kfilm->exposure = exposure; + kfilm->pass_alpha_threshold = pass_alpha_threshold; kfilm->pass_flag = 0; - kfilm->display_pass_stride = -1; - kfilm->display_pass_components = 0; - kfilm->display_divide_pass_stride = -1; - kfilm->use_display_exposure = false; - kfilm->use_display_pass_alpha = (display_pass == PASS_COMBINED); + kfilm->use_approximate_shadow_catcher = get_use_approximate_shadow_catcher(); kfilm->light_pass_flag = 0; kfilm->pass_stride = 0; - kfilm->use_light_pass = use_light_visibility; - kfilm->pass_aov_value_num = 0; - kfilm->pass_aov_color_num = 0; + + /* Mark with PASS_UNUSED to avoid mask test in the kernel. */ + kfilm->pass_background = PASS_UNUSED; + kfilm->pass_emission = PASS_UNUSED; + kfilm->pass_ao = PASS_UNUSED; + kfilm->pass_diffuse_direct = PASS_UNUSED; + kfilm->pass_diffuse_indirect = PASS_UNUSED; + kfilm->pass_glossy_direct = PASS_UNUSED; + kfilm->pass_glossy_indirect = PASS_UNUSED; + kfilm->pass_transmission_direct = PASS_UNUSED; + kfilm->pass_transmission_indirect = PASS_UNUSED; + kfilm->pass_volume_direct = PASS_UNUSED; + kfilm->pass_volume_indirect = PASS_UNUSED; + kfilm->pass_volume_direct = PASS_UNUSED; + kfilm->pass_volume_indirect = PASS_UNUSED; + kfilm->pass_shadow = PASS_UNUSED; + + /* Mark passes as unused so that the kernel knows the pass is inaccessible. */ + kfilm->pass_denoising_normal = PASS_UNUSED; + kfilm->pass_denoising_albedo = PASS_UNUSED; + kfilm->pass_sample_count = PASS_UNUSED; + kfilm->pass_adaptive_aux_buffer = PASS_UNUSED; + kfilm->pass_shadow_catcher = PASS_UNUSED; + kfilm->pass_shadow_catcher_sample_count = PASS_UNUSED; + kfilm->pass_shadow_catcher_matte = PASS_UNUSED; bool have_cryptomatte = false; + bool have_aov_color = false; + bool have_aov_value = false; for (size_t i = 0; i < scene->passes.size(); i++) { - Pass &pass = scene->passes[i]; + const Pass *pass = scene->passes[i]; - if (pass.type == PASS_NONE) { + if (pass->get_type() == PASS_NONE || !pass->is_written()) { + continue; + } + + if (pass->get_mode() == PassMode::DENOISED) { + /* Generally we only storing offsets of the noisy passes. The display pass is an exception + * since it is a read operation and not a write. */ + kfilm->pass_stride += pass->get_info().num_components; continue; } /* Can't do motion pass if no motion vectors are available. */ - if (pass.type == PASS_MOTION || pass.type == PASS_MOTION_WEIGHT) { + if (pass->get_type() == PASS_MOTION || pass->get_type() == PASS_MOTION_WEIGHT) { if (scene->need_motion() != Scene::MOTION_PASS) { - kfilm->pass_stride += pass.components; + kfilm->pass_stride += pass->get_info().num_components; continue; } } - int pass_flag = (1 << (pass.type % 32)); - if (pass.type <= PASS_CATEGORY_MAIN_END) { - kfilm->pass_flag |= pass_flag; - } - else if (pass.type <= PASS_CATEGORY_LIGHT_END) { - kfilm->use_light_pass = 1; + const int pass_flag = (1 << (pass->get_type() % 32)); + if (pass->get_type() <= PASS_CATEGORY_LIGHT_END) { kfilm->light_pass_flag |= pass_flag; } + else if (pass->get_type() <= PASS_CATEGORY_DATA_END) { + kfilm->pass_flag |= pass_flag; + } else { - assert(pass.type <= PASS_CATEGORY_BAKE_END); + assert(pass->get_type() <= PASS_CATEGORY_BAKE_END); } - switch (pass.type) { + switch (pass->get_type()) { case PASS_COMBINED: kfilm->pass_combined = kfilm->pass_stride; break; @@ -479,6 +247,12 @@ void Film::device_update(Device *device, DeviceScene *dscene, Scene *scene) case PASS_NORMAL: kfilm->pass_normal = kfilm->pass_stride; break; + case PASS_POSITION: + kfilm->pass_position = kfilm->pass_stride; + break; + case PASS_ROUGHNESS: + kfilm->pass_roughness = kfilm->pass_stride; + break; case PASS_UV: kfilm->pass_uv = kfilm->pass_stride; break; @@ -511,9 +285,6 @@ void Film::device_update(Device *device, DeviceScene *dscene, Scene *scene) kfilm->pass_shadow = kfilm->pass_stride; break; - case PASS_LIGHT: - break; - case PASS_DIFFUSE_COLOR: kfilm->pass_diffuse_color = kfilm->pass_stride; break; @@ -563,78 +334,56 @@ void Film::device_update(Device *device, DeviceScene *dscene, Scene *scene) kfilm->pass_stride; have_cryptomatte = true; break; + + case PASS_DENOISING_NORMAL: + kfilm->pass_denoising_normal = kfilm->pass_stride; + break; + case PASS_DENOISING_ALBEDO: + kfilm->pass_denoising_albedo = kfilm->pass_stride; + break; + + case PASS_SHADOW_CATCHER: + kfilm->pass_shadow_catcher = kfilm->pass_stride; + break; + case PASS_SHADOW_CATCHER_SAMPLE_COUNT: + kfilm->pass_shadow_catcher_sample_count = kfilm->pass_stride; + break; + case PASS_SHADOW_CATCHER_MATTE: + kfilm->pass_shadow_catcher_matte = kfilm->pass_stride; + break; + case PASS_ADAPTIVE_AUX_BUFFER: kfilm->pass_adaptive_aux_buffer = kfilm->pass_stride; break; case PASS_SAMPLE_COUNT: kfilm->pass_sample_count = kfilm->pass_stride; break; + case PASS_AOV_COLOR: - if (kfilm->pass_aov_color_num == 0) { + if (!have_aov_color) { kfilm->pass_aov_color = kfilm->pass_stride; + have_aov_color = true; } - kfilm->pass_aov_color_num++; break; case PASS_AOV_VALUE: - if (kfilm->pass_aov_value_num == 0) { + if (!have_aov_value) { kfilm->pass_aov_value = kfilm->pass_stride; + have_aov_value = true; } - kfilm->pass_aov_value_num++; break; default: assert(false); break; } - if (pass.type == display_pass) { - kfilm->display_pass_stride = kfilm->pass_stride; - kfilm->display_pass_components = pass.components; - kfilm->use_display_exposure = pass.exposure && (kfilm->exposure != 1.0f); - } - else if (pass.type == PASS_DIFFUSE_COLOR || pass.type == PASS_TRANSMISSION_COLOR || - pass.type == PASS_GLOSSY_COLOR) { - kfilm->display_divide_pass_stride = kfilm->pass_stride; - } - - kfilm->pass_stride += pass.components; - } - - kfilm->pass_denoising_data = 0; - kfilm->pass_denoising_clean = 0; - kfilm->denoising_flags = 0; - if (denoising_data_pass) { - kfilm->pass_denoising_data = kfilm->pass_stride; - kfilm->pass_stride += DENOISING_PASS_SIZE_BASE; - kfilm->denoising_flags = denoising_flags; - if (denoising_clean_pass) { - kfilm->pass_denoising_clean = kfilm->pass_stride; - kfilm->pass_stride += DENOISING_PASS_SIZE_CLEAN; - kfilm->use_light_pass = 1; - } - if (denoising_prefiltered_pass) { - kfilm->pass_stride += DENOISING_PASS_SIZE_PREFILTERED; - } - } - - kfilm->pass_stride = align_up(kfilm->pass_stride, 4); - - /* When displaying the normal/uv pass in the viewport we need to disable - * transparency. - * - * We also don't need to perform light accumulations. Later we want to optimize this to suppress - * light calculations. */ - if (display_pass == PASS_NORMAL || display_pass == PASS_UV) { - kfilm->use_light_pass = 0; - } - else { - kfilm->pass_alpha_threshold = pass_alpha_threshold; + kfilm->pass_stride += pass->get_info().num_components; } /* update filter table */ vector<float> table = filter_table(filter_type, filter_width); - scene->lookup_tables->remove_table(&filter_table_offset); - filter_table_offset = scene->lookup_tables->add_table(dscene, table); - kfilm->filter_table_offset = (int)filter_table_offset; + scene->lookup_tables->remove_table(&filter_table_offset_); + filter_table_offset_ = scene->lookup_tables->add_table(dscene, table); + kfilm->filter_table_offset = (int)filter_table_offset_; /* mist pass parameters */ kfilm->mist_start = mist_start; @@ -644,79 +393,298 @@ void Film::device_update(Device *device, DeviceScene *dscene, Scene *scene) kfilm->cryptomatte_passes = cryptomatte_passes; kfilm->cryptomatte_depth = cryptomatte_depth; - pass_stride = kfilm->pass_stride; - denoising_data_offset = kfilm->pass_denoising_data; - denoising_clean_offset = kfilm->pass_denoising_clean; - clear_modified(); } void Film::device_free(Device * /*device*/, DeviceScene * /*dscene*/, Scene *scene) { - scene->lookup_tables->remove_table(&filter_table_offset); + scene->lookup_tables->remove_table(&filter_table_offset_); } -void Film::tag_passes_update(Scene *scene, const vector<Pass> &passes_, bool update_passes) +int Film::get_aov_offset(Scene *scene, string name, bool &is_color) { - if (Pass::contains(scene->passes, PASS_UV) != Pass::contains(passes_, PASS_UV)) { - scene->geometry_manager->tag_update(scene, GeometryManager::UV_PASS_NEEDED); + int offset_color = 0, offset_value = 0; + foreach (const Pass *pass, scene->passes) { + if (pass->get_name() == name) { + if (pass->get_type() == PASS_AOV_VALUE) { + is_color = false; + return offset_value; + } + else if (pass->get_type() == PASS_AOV_COLOR) { + is_color = true; + return offset_color; + } + } + + if (pass->get_type() == PASS_AOV_VALUE) { + offset_value += pass->get_info().num_components; + } + else if (pass->get_type() == PASS_AOV_COLOR) { + offset_color += pass->get_info().num_components; + } + } + + return -1; +} + +void Film::update_passes(Scene *scene, bool add_sample_count_pass) +{ + const Background *background = scene->background; + const BakeManager *bake_manager = scene->bake_manager; + const ObjectManager *object_manager = scene->object_manager; + Integrator *integrator = scene->integrator; + + if (!is_modified() && !object_manager->need_update() && !integrator->is_modified()) { + return; + } + + /* Remove auto generated passes and recreate them. */ + remove_auto_passes(scene); + + /* Display pass for viewport. */ + const PassType display_pass = get_display_pass(); + add_auto_pass(scene, display_pass); + + /* Assumption is that a combined pass always exists for now, for example + * adaptive sampling is always based on a combined pass. But we should + * try to lift this limitation in the future for faster rendering of + * individual passes. */ + if (display_pass != PASS_COMBINED) { + add_auto_pass(scene, PASS_COMBINED); + } + + /* Create passes needed for adaptive sampling. */ + const AdaptiveSampling adaptive_sampling = integrator->get_adaptive_sampling(); + if (adaptive_sampling.use) { + add_auto_pass(scene, PASS_SAMPLE_COUNT); + add_auto_pass(scene, PASS_ADAPTIVE_AUX_BUFFER); + } + + /* Create passes needed for denoising. */ + const bool use_denoise = integrator->get_use_denoise(); + if (use_denoise) { + if (integrator->get_use_denoise_pass_normal()) { + add_auto_pass(scene, PASS_DENOISING_NORMAL); + } + if (integrator->get_use_denoise_pass_albedo()) { + add_auto_pass(scene, PASS_DENOISING_ALBEDO); + } + } + + /* Create passes for shadow catcher. */ + if (scene->has_shadow_catcher()) { + const bool need_background = get_use_approximate_shadow_catcher() && + !background->get_transparent(); + + add_auto_pass(scene, PASS_SHADOW_CATCHER); + add_auto_pass(scene, PASS_SHADOW_CATCHER_SAMPLE_COUNT); + add_auto_pass(scene, PASS_SHADOW_CATCHER_MATTE); + + if (need_background) { + add_auto_pass(scene, PASS_BACKGROUND); + } + } + else if (Pass::contains(scene->passes, PASS_SHADOW_CATCHER)) { + add_auto_pass(scene, PASS_SHADOW_CATCHER); + add_auto_pass(scene, PASS_SHADOW_CATCHER_SAMPLE_COUNT); + } + + const vector<Pass *> passes_immutable = scene->passes; + for (const Pass *pass : passes_immutable) { + const PassInfo info = pass->get_info(); + /* Add utility passes needed to generate some light passes. */ + if (info.divide_type != PASS_NONE) { + add_auto_pass(scene, info.divide_type); + } + if (info.direct_type != PASS_NONE) { + add_auto_pass(scene, info.direct_type); + } + if (info.indirect_type != PASS_NONE) { + add_auto_pass(scene, info.indirect_type); + } + + /* NOTE: Enable all denoised passes when storage is requested. + * This way it is possible to tweak denoiser parameters later on. */ + if (info.support_denoise && use_denoise) { + add_auto_pass(scene, pass->get_type(), PassMode::DENOISED); + } + } + + if (bake_manager->get_baking()) { + add_auto_pass(scene, PASS_BAKE_PRIMITIVE, "BakePrimitive"); + add_auto_pass(scene, PASS_BAKE_DIFFERENTIAL, "BakeDifferential"); + } + + if (add_sample_count_pass) { + if (!Pass::contains(scene->passes, PASS_SAMPLE_COUNT)) { + add_auto_pass(scene, PASS_SAMPLE_COUNT); + } + } + + /* Remove duplicates and initialize internal pass info. */ + finalize_passes(scene, use_denoise); + /* Flush scene updates. */ + const bool have_uv_pass = Pass::contains(scene->passes, PASS_UV); + const bool have_motion_pass = Pass::contains(scene->passes, PASS_MOTION); + const bool have_ao_pass = Pass::contains(scene->passes, PASS_AO); + + if (have_uv_pass != prev_have_uv_pass) { + scene->geometry_manager->tag_update(scene, GeometryManager::UV_PASS_NEEDED); foreach (Shader *shader, scene->shaders) shader->need_update_uvs = true; } - else if (Pass::contains(scene->passes, PASS_MOTION) != Pass::contains(passes_, PASS_MOTION)) { + if (have_motion_pass != prev_have_motion_pass) { scene->geometry_manager->tag_update(scene, GeometryManager::MOTION_PASS_NEEDED); } - else if (Pass::contains(scene->passes, PASS_AO) != Pass::contains(passes_, PASS_AO)) { + if (have_ao_pass != prev_have_ao_pass) { scene->integrator->tag_update(scene, Integrator::AO_PASS_MODIFIED); } - if (update_passes) { - scene->passes = passes_; + prev_have_uv_pass = have_uv_pass; + prev_have_motion_pass = have_motion_pass; + prev_have_ao_pass = have_ao_pass; + + tag_modified(); + + /* Debug logging. */ + if (VLOG_IS_ON(2)) { + VLOG(2) << "Effective scene passes:"; + for (const Pass *pass : scene->passes) { + VLOG(2) << "- " << *pass; + } } } -int Film::get_aov_offset(Scene *scene, string name, bool &is_color) +void Film::add_auto_pass(Scene *scene, PassType type, const char *name) { - int num_color = 0, num_value = 0; - foreach (const Pass &pass, scene->passes) { - if (pass.type == PASS_AOV_COLOR) { - num_color++; - } - else if (pass.type == PASS_AOV_VALUE) { - num_value++; + add_auto_pass(scene, type, PassMode::NOISY, name); +} + +void Film::add_auto_pass(Scene *scene, PassType type, PassMode mode, const char *name) +{ + Pass *pass = new Pass(); + pass->set_type(type); + pass->set_mode(mode); + pass->set_name(ustring((name) ? name : "")); + pass->is_auto_ = true; + + pass->set_owner(scene); + scene->passes.push_back(pass); +} + +void Film::remove_auto_passes(Scene *scene) +{ + /* Remove all passes which were automatically created. */ + vector<Pass *> new_passes; + + for (Pass *pass : scene->passes) { + if (!pass->is_auto_) { + new_passes.push_back(pass); } else { - continue; - } - - if (pass.name == name) { - is_color = (pass.type == PASS_AOV_COLOR); - return (is_color ? num_color : num_value) - 1; + delete pass; } } - return -1; + scene->passes = new_passes; } -int Film::get_pass_stride() const +static bool compare_pass_order(const Pass *a, const Pass *b) { - return pass_stride; -} + const int num_components_a = a->get_info().num_components; + const int num_components_b = b->get_info().num_components; -int Film::get_denoising_data_offset() const -{ - return denoising_data_offset; + if (num_components_a == num_components_b) { + return (a->get_type() < b->get_type()); + } + + return num_components_a > num_components_b; } -int Film::get_denoising_clean_offset() const +void Film::finalize_passes(Scene *scene, const bool use_denoise) { - return denoising_clean_offset; + /* Remove duplicate passes. */ + vector<Pass *> new_passes; + + for (Pass *pass : scene->passes) { + /* Disable denoising on passes if denoising is disabled, or if the + * pass does not support it. */ + pass->set_mode((use_denoise && pass->get_info().support_denoise) ? pass->get_mode() : + PassMode::NOISY); + + /* Merge duplicate passes. */ + bool duplicate_found = false; + for (Pass *new_pass : new_passes) { + /* If different type or denoising, don't merge. */ + if (new_pass->get_type() != pass->get_type() || new_pass->get_mode() != pass->get_mode()) { + continue; + } + + /* If both passes have a name and the names are different, don't merge. + * If either pass has a name, we'll use that name. */ + if (!pass->get_name().empty() && !new_pass->get_name().empty() && + pass->get_name() != new_pass->get_name()) { + continue; + } + + if (!pass->get_name().empty() && new_pass->get_name().empty()) { + new_pass->set_name(pass->get_name()); + } + + new_pass->is_auto_ &= pass->is_auto_; + duplicate_found = true; + break; + } + + if (!duplicate_found) { + new_passes.push_back(pass); + } + else { + delete pass; + } + } + + /* Order from by components and type, This is required to for AOVs and cryptomatte passes, + * which the kernel assumes to be in order. Note this must use stable sort so cryptomatte + * passes remain in the right order. */ + stable_sort(new_passes.begin(), new_passes.end(), compare_pass_order); + + scene->passes = new_passes; } -size_t Film::get_filter_table_offset() const +uint Film::get_kernel_features(const Scene *scene) const { - return filter_table_offset; + uint kernel_features = 0; + + for (const Pass *pass : scene->passes) { + if (!pass->is_written()) { + continue; + } + + const PassType pass_type = pass->get_type(); + const PassMode pass_mode = pass->get_mode(); + + if (pass_mode == PassMode::DENOISED || pass_type == PASS_DENOISING_NORMAL || + pass_type == PASS_DENOISING_ALBEDO) { + kernel_features |= KERNEL_FEATURE_DENOISING; + } + + if (pass_type != PASS_NONE && pass_type != PASS_COMBINED && + pass_type <= PASS_CATEGORY_LIGHT_END) { + kernel_features |= KERNEL_FEATURE_LIGHT_PASSES; + + if (pass_type == PASS_SHADOW) { + kernel_features |= KERNEL_FEATURE_SHADOW_PASS; + } + } + + if (pass_type == PASS_AO) { + kernel_features |= KERNEL_FEATURE_NODE_RAYTRACE; + } + } + + return kernel_features; } CCL_NAMESPACE_END diff --git a/intern/cycles/render/film.h b/intern/cycles/render/film.h index 462a7275491..5d327353361 100644 --- a/intern/cycles/render/film.h +++ b/intern/cycles/render/film.h @@ -17,6 +17,7 @@ #ifndef __FILM_H__ #define __FILM_H__ +#include "render/pass.h" #include "util/util_string.h" #include "util/util_vector.h" @@ -38,36 +39,15 @@ typedef enum FilterType { FILTER_NUM_TYPES, } FilterType; -class Pass : public Node { - public: - NODE_DECLARE - - Pass(); - - PassType type; - int components; - bool filter; - bool exposure; - PassType divide_type; - ustring name; - - static void add(PassType type, vector<Pass> &passes, const char *name = NULL); - static bool equals(const vector<Pass> &A, const vector<Pass> &B); - static bool contains(const vector<Pass> &passes, PassType); -}; - class Film : public Node { public: NODE_DECLARE NODE_SOCKET_API(float, exposure) - NODE_SOCKET_API(bool, denoising_data_pass) - NODE_SOCKET_API(bool, denoising_clean_pass) - NODE_SOCKET_API(bool, denoising_prefiltered_pass) - NODE_SOCKET_API(int, denoising_flags) NODE_SOCKET_API(float, pass_alpha_threshold) NODE_SOCKET_API(PassType, display_pass) + NODE_SOCKET_API(bool, show_active_pixels) NODE_SOCKET_API(FilterType, filter_type) NODE_SOCKET_API(float, filter_width) @@ -76,17 +56,18 @@ class Film : public Node { NODE_SOCKET_API(float, mist_depth) NODE_SOCKET_API(float, mist_falloff) - NODE_SOCKET_API(bool, use_light_visibility) NODE_SOCKET_API(CryptomatteType, cryptomatte_passes) NODE_SOCKET_API(int, cryptomatte_depth) - NODE_SOCKET_API(bool, use_adaptive_sampling) + /* Approximate shadow catcher pass into its matte pass, so that both artificial objects and + * shadows can be alpha-overed onto a backdrop. */ + NODE_SOCKET_API(bool, use_approximate_shadow_catcher) private: - int pass_stride; - int denoising_data_offset; - int denoising_clean_offset; - size_t filter_table_offset; + size_t filter_table_offset_; + bool prev_have_uv_pass = false; + bool prev_have_motion_pass = false; + bool prev_have_ao_pass = false; public: Film(); @@ -98,14 +79,20 @@ class Film : public Node { void device_update(Device *device, DeviceScene *dscene, Scene *scene); void device_free(Device *device, DeviceScene *dscene, Scene *scene); - void tag_passes_update(Scene *scene, const vector<Pass> &passes_, bool update_passes = true); - int get_aov_offset(Scene *scene, string name, bool &is_color); - int get_pass_stride() const; - int get_denoising_data_offset() const; - int get_denoising_clean_offset() const; - size_t get_filter_table_offset() const; + /* Update passes so that they contain all passes required for the configured functionality. + * + * If `add_sample_count_pass` is true then the SAMPLE_COUNT pass is ensured to be added. */ + void update_passes(Scene *scene, bool add_sample_count_pass); + + uint get_kernel_features(const Scene *scene) const; + + private: + void add_auto_pass(Scene *scene, PassType type, const char *name = nullptr); + void add_auto_pass(Scene *scene, PassType type, PassMode mode, const char *name = nullptr); + void remove_auto_passes(Scene *scene); + void finalize_passes(Scene *scene, const bool use_denoise); }; CCL_NAMESPACE_END diff --git a/intern/cycles/render/geometry.cpp b/intern/cycles/render/geometry.cpp index 7ec1d2d9abb..6804a006fe6 100644 --- a/intern/cycles/render/geometry.cpp +++ b/intern/cycles/render/geometry.cpp @@ -215,6 +215,12 @@ void Geometry::compute_bvh( msg += string_printf("%s %u/%u", name.c_str(), (uint)(n + 1), (uint)total); Object object; + + /* Ensure all visibility bits are set at the geometry level BVH. In + * the object level BVH is where actual visibility is tested. */ + object.set_is_shadow_catcher(true); + object.set_visibility(~0); + object.set_geometry(this); vector<Geometry *> geometry; @@ -315,7 +321,7 @@ void GeometryManager::update_osl_attributes(Device *device, { #ifdef WITH_OSL /* for OSL, a hash map is used to lookup the attribute by name. */ - OSLGlobals *og = (OSLGlobals *)device->osl_memory(); + OSLGlobals *og = (OSLGlobals *)device->get_cpu_osl_memory(); og->object_name_map.clear(); og->attribute_map.clear(); @@ -1855,8 +1861,8 @@ void GeometryManager::device_update(Device *device, }); Camera *dicing_camera = scene->dicing_camera; - dicing_camera->set_screen_size_and_resolution( - dicing_camera->get_full_width(), dicing_camera->get_full_height(), 1); + dicing_camera->set_screen_size(dicing_camera->get_full_width(), + dicing_camera->get_full_height()); dicing_camera->update(scene); size_t i = 0; @@ -2157,7 +2163,7 @@ void GeometryManager::device_free(Device *device, DeviceScene *dscene, bool forc dscene->data.bvh.bvh_layout = BVH_LAYOUT_NONE; #ifdef WITH_OSL - OSLGlobals *og = (OSLGlobals *)device->osl_memory(); + OSLGlobals *og = (OSLGlobals *)device->get_cpu_osl_memory(); if (og) { og->object_name_map.clear(); diff --git a/intern/cycles/render/gpu_display.cpp b/intern/cycles/render/gpu_display.cpp new file mode 100644 index 00000000000..a8f0cc50583 --- /dev/null +++ b/intern/cycles/render/gpu_display.cpp @@ -0,0 +1,227 @@ +/* + * Copyright 2021 Blender Foundation + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include "render/gpu_display.h" + +#include "render/buffers.h" +#include "util/util_logging.h" + +CCL_NAMESPACE_BEGIN + +void GPUDisplay::reset(const BufferParams &buffer_params) +{ + thread_scoped_lock lock(mutex_); + + const GPUDisplayParams old_params = params_; + + params_.offset = make_int2(buffer_params.full_x, buffer_params.full_y); + params_.full_size = make_int2(buffer_params.full_width, buffer_params.full_height); + params_.size = make_int2(buffer_params.width, buffer_params.height); + + /* If the parameters did change tag texture as unusable. This avoids drawing old texture content + * in an updated configuration of the viewport. For example, avoids drawing old frame when render + * border did change. + * If the parameters did not change, allow drawing the current state of the texture, which will + * not count as an up-to-date redraw. This will avoid flickering when doping camera navigation by + * showing a previously rendered frame for until the new one is ready. */ + if (old_params.modified(params_)) { + texture_state_.is_usable = false; + } + + texture_state_.is_outdated = true; +} + +void GPUDisplay::mark_texture_updated() +{ + texture_state_.is_outdated = false; + texture_state_.is_usable = true; +} + +/* -------------------------------------------------------------------- + * Update procedure. + */ + +bool GPUDisplay::update_begin(int texture_width, int texture_height) +{ + DCHECK(!update_state_.is_active); + + if (update_state_.is_active) { + LOG(ERROR) << "Attempt to re-activate update process."; + return false; + } + + /* Get parameters within a mutex lock, to avoid reset() modifying them at the same time. + * The update itself is non-blocking however, for better performance and to avoid + * potential deadlocks due to locks held by the subclass. */ + GPUDisplayParams params; + { + thread_scoped_lock lock(mutex_); + params = params_; + texture_state_.size = make_int2(texture_width, texture_height); + } + + if (!do_update_begin(params, texture_width, texture_height)) { + LOG(ERROR) << "GPUDisplay implementation could not begin update."; + return false; + } + + update_state_.is_active = true; + + return true; +} + +void GPUDisplay::update_end() +{ + DCHECK(update_state_.is_active); + + if (!update_state_.is_active) { + LOG(ERROR) << "Attempt to deactivate inactive update process."; + return; + } + + do_update_end(); + + update_state_.is_active = false; +} + +int2 GPUDisplay::get_texture_size() const +{ + return texture_state_.size; +} + +/* -------------------------------------------------------------------- + * Texture update from CPU buffer. + */ + +void GPUDisplay::copy_pixels_to_texture( + const half4 *rgba_pixels, int texture_x, int texture_y, int pixels_width, int pixels_height) +{ + DCHECK(update_state_.is_active); + + if (!update_state_.is_active) { + LOG(ERROR) << "Attempt to copy pixels data outside of GPUDisplay update."; + return; + } + + mark_texture_updated(); + do_copy_pixels_to_texture(rgba_pixels, texture_x, texture_y, pixels_width, pixels_height); +} + +/* -------------------------------------------------------------------- + * Texture buffer mapping. + */ + +half4 *GPUDisplay::map_texture_buffer() +{ + DCHECK(!texture_buffer_state_.is_mapped); + DCHECK(update_state_.is_active); + + if (texture_buffer_state_.is_mapped) { + LOG(ERROR) << "Attempt to re-map an already mapped texture buffer."; + return nullptr; + } + + if (!update_state_.is_active) { + LOG(ERROR) << "Attempt to copy pixels data outside of GPUDisplay update."; + return nullptr; + } + + half4 *mapped_rgba_pixels = do_map_texture_buffer(); + + if (mapped_rgba_pixels) { + texture_buffer_state_.is_mapped = true; + } + + return mapped_rgba_pixels; +} + +void GPUDisplay::unmap_texture_buffer() +{ + DCHECK(texture_buffer_state_.is_mapped); + + if (!texture_buffer_state_.is_mapped) { + LOG(ERROR) << "Attempt to unmap non-mapped texture buffer."; + return; + } + + texture_buffer_state_.is_mapped = false; + + mark_texture_updated(); + do_unmap_texture_buffer(); +} + +/* -------------------------------------------------------------------- + * Graphics interoperability. + */ + +DeviceGraphicsInteropDestination GPUDisplay::graphics_interop_get() +{ + DCHECK(!texture_buffer_state_.is_mapped); + DCHECK(update_state_.is_active); + + if (texture_buffer_state_.is_mapped) { + LOG(ERROR) + << "Attempt to use graphics interoperability mode while the texture buffer is mapped."; + return DeviceGraphicsInteropDestination(); + } + + if (!update_state_.is_active) { + LOG(ERROR) << "Attempt to use graphics interoperability outside of GPUDisplay update."; + return DeviceGraphicsInteropDestination(); + } + + /* Assume that interop will write new values to the texture. */ + mark_texture_updated(); + + return do_graphics_interop_get(); +} + +void GPUDisplay::graphics_interop_activate() +{ +} + +void GPUDisplay::graphics_interop_deactivate() +{ +} + +/* -------------------------------------------------------------------- + * Drawing. + */ + +bool GPUDisplay::draw() +{ + /* Get parameters within a mutex lock, to avoid reset() modifying them at the same time. + * The drawing itself is non-blocking however, for better performance and to avoid + * potential deadlocks due to locks held by the subclass. */ + GPUDisplayParams params; + bool is_usable; + bool is_outdated; + + { + thread_scoped_lock lock(mutex_); + params = params_; + is_usable = texture_state_.is_usable; + is_outdated = texture_state_.is_outdated; + } + + if (is_usable) { + do_draw(params); + } + + return !is_outdated; +} + +CCL_NAMESPACE_END diff --git a/intern/cycles/render/gpu_display.h b/intern/cycles/render/gpu_display.h new file mode 100644 index 00000000000..cbe347895a1 --- /dev/null +++ b/intern/cycles/render/gpu_display.h @@ -0,0 +1,247 @@ +/* + * Copyright 2021 Blender Foundation + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#pragma once + +#include "device/device_graphics_interop.h" +#include "util/util_half.h" +#include "util/util_thread.h" +#include "util/util_types.h" + +CCL_NAMESPACE_BEGIN + +class BufferParams; + +/* GPUDisplay class takes care of drawing render result in a viewport. The render result is stored + * in a GPU-side texture, which is updated from a path tracer and drawn by an application. + * + * The base GPUDisplay does some special texture state tracking, which allows render Session to + * make decisions on whether reset for an updated state is possible or not. This state should only + * be tracked in a base class and a particular implementation should not worry about it. + * + * The subclasses should only implement the pure virtual methods, which allows them to not worry + * about parent method calls, which helps them to be as small and reliable as possible. */ + +class GPUDisplayParams { + public: + /* Offset of the display within a viewport. + * For example, set to a lower-bottom corner of border render in Blender's viewport. */ + int2 offset = make_int2(0, 0); + + /* Full viewport size. + * + * NOTE: Is not affected by the resolution divider. */ + int2 full_size = make_int2(0, 0); + + /* Effective vieport size. + * In the case of border render, size of the border rectangle. + * + * NOTE: Is not affected by the resolution divider. */ + int2 size = make_int2(0, 0); + + bool modified(const GPUDisplayParams &other) const + { + return !(offset == other.offset && full_size == other.full_size && size == other.size); + } +}; + +class GPUDisplay { + public: + GPUDisplay() = default; + virtual ~GPUDisplay() = default; + + /* Reset the display for the new state of render session. Is called whenever session is reset, + * which happens on changes like viewport navigation or viewport dimension change. + * + * This call will configure parameters for a changed buffer and reset the texture state. */ + void reset(const BufferParams &buffer_params); + + const GPUDisplayParams &get_params() const + { + return params_; + } + + /* -------------------------------------------------------------------- + * Update procedure. + * + * These calls indicates a desire of the caller to update content of the displayed texture. */ + + /* Returns true when update is ready. Update should be finished with update_end(). + * + * If false is returned then no update is possible, and no update_end() call is needed. + * + * The texture width and height denotes an actual resolution of the underlying render result. */ + bool update_begin(int texture_width, int texture_height); + + void update_end(); + + /* Get currently configured texture size of the display (as configured by `update_begin()`. */ + int2 get_texture_size() const; + + /* -------------------------------------------------------------------- + * Texture update from CPU buffer. + * + * NOTE: The GPUDisplay should be marked for an update being in process with `update_begin()`. + * + * Most portable implementation, which must be supported by all platforms. Might not be the most + * efficient one. + */ + + /* Copy buffer of rendered pixels of a given size into a given position of the texture. + * + * This function does not acquire a lock. The reason for this is is to allow use of this function + * for partial updates from different devices. In this case the caller will acquire the lock + * once, update all the slices and release + * the lock once. This will ensure that draw() will never use partially updated texture. */ + void copy_pixels_to_texture( + const half4 *rgba_pixels, int texture_x, int texture_y, int pixels_width, int pixels_height); + + /* -------------------------------------------------------------------- + * Texture buffer mapping. + * + * This functionality is used to update GPU-side texture content without need to maintain CPU + * side buffer on the caller. + * + * NOTE: The GPUDisplay should be marked for an update being in process with `update_begin()`. + * + * NOTE: Texture buffer can not be mapped while graphics interopeability is active. This means + * that `map_texture_buffer()` is not allowed between `graphics_interop_begin()` and + * `graphics_interop_end()` calls. + */ + + /* Map pixels memory form texture to a buffer available for write from CPU. Width and height will + * define a requested size of the texture to write to. + * Upon success a non-null pointer is returned and the texture buffer is to be unmapped. + * If an error happens during mapping, or if mapoping is not supported by this GPU display a + * null pointer is returned and the buffer is NOT to be unmapped. + * + * NOTE: Usually the implementation will rely on a GPU context of some sort, and the GPU context + * is often can not be bound to two threads simultaneously, and can not be released from a + * different thread. This means that the mapping API should be used from the single thread only, + */ + half4 *map_texture_buffer(); + void unmap_texture_buffer(); + + /* -------------------------------------------------------------------- + * Graphics interoperability. + * + * A special code path which allows to update texture content directly from the GPU compute + * device. Complementary part of DeviceGraphicsInterop. + * + * NOTE: Graphics interoperability can not be used while the texture buffer is mapped. This means + * that `graphics_interop_get()` is not allowed between `map_texture_buffer()` and + * `unmap_texture_buffer()` calls. */ + + /* Get GPUDisplay graphics interoperability information which acts as a destination for the + * device API. */ + DeviceGraphicsInteropDestination graphics_interop_get(); + + /* (De)activate GPU display for graphics interoperability outside of regular display udpate + * routines. */ + virtual void graphics_interop_activate(); + virtual void graphics_interop_deactivate(); + + /* -------------------------------------------------------------------- + * Drawing. + */ + + /* Clear the texture by filling it with all zeroes. + * + * This call might happen in parallel with draw, but can never happen in parallel with the + * update. + * + * The actual zero-ing can be deferred to a later moment. What is important is that after clear + * and before pixels update the drawing texture will be fully empty, and that partial update + * after clear will write new pixel values for an updating area, leaving everything else zeroed. + * + * If the GPU display supports graphics interoperability then the zeroing the display is to be + * delegated to the device via the `DeviceGraphicsInteropDestination`. */ + virtual void clear() = 0; + + /* Draw the current state of the texture. + * + * Returns true if this call did draw an updated state of the texture. */ + bool draw(); + + protected: + /* Implementation-specific calls which subclasses are to implement. + * These `do_foo()` method corresponds to their `foo()` calls, but they are purely virtual to + * simplify their particular implementation. */ + virtual bool do_update_begin(const GPUDisplayParams ¶ms, + int texture_width, + int texture_height) = 0; + virtual void do_update_end() = 0; + + virtual void do_copy_pixels_to_texture(const half4 *rgba_pixels, + int texture_x, + int texture_y, + int pixels_width, + int pixels_height) = 0; + + virtual half4 *do_map_texture_buffer() = 0; + virtual void do_unmap_texture_buffer() = 0; + + /* Note that this might be called in parallel to do_update_begin() and do_update_end(), + * the subclass is responsible for appropriate mutex locks to avoid multiple threads + * editing and drawing the texture at the same time. */ + virtual void do_draw(const GPUDisplayParams ¶ms) = 0; + + virtual DeviceGraphicsInteropDestination do_graphics_interop_get() = 0; + + private: + thread_mutex mutex_; + GPUDisplayParams params_; + + /* Mark texture as its content has been updated. + * Used from places which knows that the texture content has been brough up-to-date, so that the + * drawing knows whether it can be performed, and whether drawing happenned with an up-to-date + * texture state. */ + void mark_texture_updated(); + + /* State of the update process. */ + struct { + /* True when update is in process, indicated by `update_begin()` / `update_end()`. */ + bool is_active = false; + } update_state_; + + /* State of the texture, which is needed for an integration with render session and interactive + * updates and navigation. */ + struct { + /* Denotes whether possibly existing state of GPU side texture is still usable. + * It will not be usable in cases like render border did change (in this case we don't want + * previous texture to be rendered at all). + * + * However, if only navigation or object in scene did change, then the outdated state of the + * texture is still usable for draw, preventing display viewport flickering on navigation and + * object modifications. */ + bool is_usable = false; + + /* Texture is considered outdated after `reset()` until the next call of + * `copy_pixels_to_texture()`. */ + bool is_outdated = true; + + /* Texture size in pixels. */ + int2 size = make_int2(0, 0); + } texture_state_; + + /* State of the texture buffer. Is tracked to perform sanity checks. */ + struct { + /* True when the texture buffer is mapped with `map_texture_buffer()`. */ + bool is_mapped = false; + } texture_buffer_state_; +}; + +CCL_NAMESPACE_END diff --git a/intern/cycles/render/graph.h b/intern/cycles/render/graph.h index 5102b182593..3584754fad1 100644 --- a/intern/cycles/render/graph.h +++ b/intern/cycles/render/graph.h @@ -224,10 +224,6 @@ class ShaderNode : public Node { { return false; } - virtual bool has_raytrace() - { - return false; - } vector<ShaderInput *> inputs; vector<ShaderOutput *> outputs; @@ -242,22 +238,13 @@ class ShaderNode : public Node { * that those functions are for selective compilation only? */ - /* Nodes are split into several groups, group of level 0 contains - * nodes which are most commonly used, further levels are extension - * of previous one and includes less commonly used nodes. - */ - virtual int get_group() - { - return NODE_GROUP_LEVEL_0; - } - /* Node feature are used to disable huge nodes inside the group, * so it's possible to disable huge nodes inside of the required * nodes group. */ virtual int get_feature() { - return bump == SHADER_BUMP_NONE ? 0 : NODE_FEATURE_BUMP; + return bump == SHADER_BUMP_NONE ? 0 : KERNEL_FEATURE_NODE_BUMP; } /* Get closure ID to which the node compiles into. */ diff --git a/intern/cycles/render/integrator.cpp b/intern/cycles/render/integrator.cpp index d8749cec9fa..d74d14242bb 100644 --- a/intern/cycles/render/integrator.cpp +++ b/intern/cycles/render/integrator.cpp @@ -53,6 +53,8 @@ NODE_DEFINE(Integrator) SOCKET_INT(transparent_max_bounce, "Transparent Max Bounce", 7); SOCKET_INT(ao_bounces, "AO Bounces", 0); + SOCKET_FLOAT(ao_factor, "AO Factor", 0.0f); + SOCKET_FLOAT(ao_distance, "AO Distance", FLT_MAX); SOCKET_INT(volume_max_steps, "Volume Max Steps", 1024); SOCKET_FLOAT(volume_step_rate, "Volume Step Rate", 1.0f); @@ -66,33 +68,39 @@ NODE_DEFINE(Integrator) SOCKET_BOOLEAN(motion_blur, "Motion Blur", false); SOCKET_INT(aa_samples, "AA Samples", 0); - SOCKET_INT(diffuse_samples, "Diffuse Samples", 1); - SOCKET_INT(glossy_samples, "Glossy Samples", 1); - SOCKET_INT(transmission_samples, "Transmission Samples", 1); - SOCKET_INT(ao_samples, "AO Samples", 1); - SOCKET_INT(mesh_light_samples, "Mesh Light Samples", 1); - SOCKET_INT(subsurface_samples, "Subsurface Samples", 1); - SOCKET_INT(volume_samples, "Volume Samples", 1); SOCKET_INT(start_sample, "Start Sample", 0); + SOCKET_BOOLEAN(use_adaptive_sampling, "Use Adaptive Sampling", false); SOCKET_FLOAT(adaptive_threshold, "Adaptive Threshold", 0.0f); SOCKET_INT(adaptive_min_samples, "Adaptive Min Samples", 0); - SOCKET_BOOLEAN(sample_all_lights_direct, "Sample All Lights Direct", true); - SOCKET_BOOLEAN(sample_all_lights_indirect, "Sample All Lights Indirect", true); SOCKET_FLOAT(light_sampling_threshold, "Light Sampling Threshold", 0.05f); - static NodeEnum method_enum; - method_enum.insert("path", PATH); - method_enum.insert("branched_path", BRANCHED_PATH); - SOCKET_ENUM(method, "Method", method_enum, PATH); - static NodeEnum sampling_pattern_enum; sampling_pattern_enum.insert("sobol", SAMPLING_PATTERN_SOBOL); - sampling_pattern_enum.insert("cmj", SAMPLING_PATTERN_CMJ); sampling_pattern_enum.insert("pmj", SAMPLING_PATTERN_PMJ); SOCKET_ENUM(sampling_pattern, "Sampling Pattern", sampling_pattern_enum, SAMPLING_PATTERN_SOBOL); + static NodeEnum denoiser_type_enum; + denoiser_type_enum.insert("optix", DENOISER_OPTIX); + denoiser_type_enum.insert("openimagedenoise", DENOISER_OPENIMAGEDENOISE); + + static NodeEnum denoiser_prefilter_enum; + denoiser_prefilter_enum.insert("none", DENOISER_PREFILTER_NONE); + denoiser_prefilter_enum.insert("fast", DENOISER_PREFILTER_FAST); + denoiser_prefilter_enum.insert("accurate", DENOISER_PREFILTER_ACCURATE); + + /* Default to accurate denoising with OpenImageDenoise. For interactive viewport + * it's best use OptiX and disable the normal pass since it does not always have + * the desired effect for that denoiser. */ + SOCKET_BOOLEAN(use_denoise, "Use Denoiser", false); + SOCKET_ENUM(denoiser_type, "Denoiser Type", denoiser_type_enum, DENOISER_OPENIMAGEDENOISE); + SOCKET_INT(denoise_start_sample, "Start Sample to Denoise", 0); + SOCKET_BOOLEAN(use_denoise_pass_albedo, "Use Albedo Pass for Denoiser", true); + SOCKET_BOOLEAN(use_denoise_pass_normal, "Use Normal Pass for Denoiser", true); + SOCKET_ENUM( + denoiser_prefilter, "Denoiser Type", denoiser_prefilter_enum, DENOISER_PREFILTER_ACCURATE); + return type; } @@ -115,13 +123,20 @@ void Integrator::device_update(Device *device, DeviceScene *dscene, Scene *scene } }); - const bool need_update_lut = ao_samples_is_modified() || diffuse_samples_is_modified() || - glossy_samples_is_modified() || max_bounce_is_modified() || - max_transmission_bounce_is_modified() || - mesh_light_samples_is_modified() || method_is_modified() || - sampling_pattern_is_modified() || - subsurface_samples_is_modified() || - transmission_samples_is_modified() || volume_samples_is_modified(); + KernelIntegrator *kintegrator = &dscene->data.integrator; + + /* Adaptive sampling requires PMJ samples. + * + * This also makes detection of sampling pattern a bit more involved: can not rely on the changed + * state of socket, since its value might be different from the effective value used here. So + * instead compare with previous value in the KernelIntegrator. Only do it if the device was + * updated once (in which case the `sample_pattern_lut` will be allocated to a non-zero size). */ + const SamplingPattern new_sampling_pattern = (use_adaptive_sampling) ? SAMPLING_PATTERN_PMJ : + sampling_pattern; + + const bool need_update_lut = max_bounce_is_modified() || max_transmission_bounce_is_modified() || + dscene->sample_pattern_lut.size() == 0 || + kintegrator->sampling_pattern != new_sampling_pattern; if (need_update_lut) { dscene->sample_pattern_lut.tag_realloc(); @@ -129,8 +144,6 @@ void Integrator::device_update(Device *device, DeviceScene *dscene, Scene *scene device_free(device, dscene); - KernelIntegrator *kintegrator = &dscene->data.integrator; - /* integrator parameters */ kintegrator->min_bounce = min_bounce + 1; kintegrator->max_bounce = max_bounce + 1; @@ -143,12 +156,9 @@ void Integrator::device_update(Device *device, DeviceScene *dscene, Scene *scene kintegrator->transparent_min_bounce = transparent_min_bounce + 1; kintegrator->transparent_max_bounce = transparent_max_bounce + 1; - if (ao_bounces == 0) { - kintegrator->ao_bounces = INT_MAX; - } - else { - kintegrator->ao_bounces = ao_bounces - 1; - } + kintegrator->ao_bounces = ao_bounces; + kintegrator->ao_bounces_distance = ao_distance; + kintegrator->ao_bounces_factor = ao_factor; /* Transparent Shadows * We only need to enable transparent shadows, if we actually have @@ -171,10 +181,7 @@ void Integrator::device_update(Device *device, DeviceScene *dscene, Scene *scene kintegrator->caustics_refractive = caustics_refractive; kintegrator->filter_glossy = (filter_glossy == 0.0f) ? FLT_MAX : 1.0f / filter_glossy; - kintegrator->seed = hash_uint2(seed, 0); - - kintegrator->use_ambient_occlusion = ((Pass::contains(scene->passes, PASS_AO)) || - dscene->data.background.ao_factor != 0.0f); + kintegrator->seed = seed; kintegrator->sample_clamp_direct = (sample_clamp_direct == 0.0f) ? FLT_MAX : sample_clamp_direct * 3.0f; @@ -182,51 +189,7 @@ void Integrator::device_update(Device *device, DeviceScene *dscene, Scene *scene FLT_MAX : sample_clamp_indirect * 3.0f; - kintegrator->branched = (method == BRANCHED_PATH) && device->info.has_branched_path; - kintegrator->volume_decoupled = device->info.has_volume_decoupled; - kintegrator->diffuse_samples = diffuse_samples; - kintegrator->glossy_samples = glossy_samples; - kintegrator->transmission_samples = transmission_samples; - kintegrator->ao_samples = ao_samples; - kintegrator->mesh_light_samples = mesh_light_samples; - kintegrator->subsurface_samples = subsurface_samples; - kintegrator->volume_samples = volume_samples; - kintegrator->start_sample = start_sample; - - if (kintegrator->branched) { - kintegrator->sample_all_lights_direct = sample_all_lights_direct; - kintegrator->sample_all_lights_indirect = sample_all_lights_indirect; - } - else { - kintegrator->sample_all_lights_direct = false; - kintegrator->sample_all_lights_indirect = false; - } - - kintegrator->sampling_pattern = sampling_pattern; - kintegrator->aa_samples = aa_samples; - if (aa_samples > 0 && adaptive_min_samples == 0) { - kintegrator->adaptive_min_samples = max(4, (int)sqrtf(aa_samples)); - VLOG(1) << "Cycles adaptive sampling: automatic min samples = " - << kintegrator->adaptive_min_samples; - } - else { - kintegrator->adaptive_min_samples = max(4, adaptive_min_samples); - } - - kintegrator->adaptive_step = 4; - kintegrator->adaptive_stop_per_sample = device->info.has_adaptive_stop_per_sample; - - /* Adaptive step must be a power of two for bitwise operations to work. */ - assert((kintegrator->adaptive_step & (kintegrator->adaptive_step - 1)) == 0); - - if (aa_samples > 0 && adaptive_threshold == 0.0f) { - kintegrator->adaptive_threshold = max(0.001f, 1.0f / (float)aa_samples); - VLOG(1) << "Cycles adaptive sampling: automatic threshold = " - << kintegrator->adaptive_threshold; - } - else { - kintegrator->adaptive_threshold = adaptive_threshold; - } + kintegrator->sampling_pattern = new_sampling_pattern; if (light_sampling_threshold > 0.0f) { kintegrator->light_inv_rr_threshold = 1.0f / light_sampling_threshold; @@ -236,29 +199,15 @@ void Integrator::device_update(Device *device, DeviceScene *dscene, Scene *scene } /* sobol directions table */ - int max_samples = 1; - - if (kintegrator->branched) { - foreach (Light *light, scene->lights) - max_samples = max(max_samples, light->get_samples()); - - max_samples = max(max_samples, - max(diffuse_samples, max(glossy_samples, transmission_samples))); - max_samples = max(max_samples, max(ao_samples, max(mesh_light_samples, subsurface_samples))); - max_samples = max(max_samples, volume_samples); - } - - uint total_bounces = max_bounce + transparent_max_bounce + 3 + VOLUME_BOUNDS_MAX + - max(BSSRDF_MAX_HITS, BSSRDF_MAX_BOUNCES); - - max_samples *= total_bounces; + int max_samples = max_bounce + transparent_max_bounce + 3 + VOLUME_BOUNDS_MAX + + max(BSSRDF_MAX_HITS, BSSRDF_MAX_BOUNCES); int dimensions = PRNG_BASE_NUM + max_samples * PRNG_BOUNCE_NUM; dimensions = min(dimensions, SOBOL_MAX_DIMENSIONS); if (need_update_lut) { - if (sampling_pattern == SAMPLING_PATTERN_SOBOL) { - uint *directions = dscene->sample_pattern_lut.alloc(SOBOL_BITS * dimensions); + if (kintegrator->sampling_pattern == SAMPLING_PATTERN_SOBOL) { + uint *directions = (uint *)dscene->sample_pattern_lut.alloc(SOBOL_BITS * dimensions); sobol_generate_direction_vectors((uint(*)[SOBOL_BITS])directions, dimensions); @@ -276,10 +225,13 @@ void Integrator::device_update(Device *device, DeviceScene *dscene, Scene *scene function_bind(&progressive_multi_jitter_02_generate_2D, sequence, sequence_size, j)); } pool.wait_work(); + dscene->sample_pattern_lut.copy_to_device(); } } + kintegrator->has_shadow_catcher = scene->has_shadow_catcher(); + dscene->sample_pattern_lut.clear_modified(); clear_modified(); } @@ -295,17 +247,12 @@ void Integrator::tag_update(Scene *scene, uint32_t flag) tag_modified(); } - if (flag & (AO_PASS_MODIFIED | BACKGROUND_AO_MODIFIED)) { + if (flag & AO_PASS_MODIFIED) { /* tag only the ao_bounces socket as modified so we avoid updating sample_pattern_lut * unnecessarily */ tag_ao_bounces_modified(); } - if ((flag & LIGHT_SAMPLES_MODIFIED) && (method == BRANCHED_PATH)) { - /* the number of light samples may affect the size of the sample_pattern_lut */ - tag_sampling_pattern_modified(); - } - if (filter_glossy_is_modified()) { foreach (Shader *shader, scene->shaders) { if (shader->has_integrator_dependency) { @@ -321,4 +268,65 @@ void Integrator::tag_update(Scene *scene, uint32_t flag) } } +AdaptiveSampling Integrator::get_adaptive_sampling() const +{ + AdaptiveSampling adaptive_sampling; + + adaptive_sampling.use = use_adaptive_sampling; + + if (!adaptive_sampling.use) { + return adaptive_sampling; + } + + if (aa_samples > 0 && adaptive_threshold == 0.0f) { + adaptive_sampling.threshold = max(0.001f, 1.0f / (float)aa_samples); + VLOG(1) << "Cycles adaptive sampling: automatic threshold = " << adaptive_sampling.threshold; + } + else { + adaptive_sampling.threshold = adaptive_threshold; + } + + if (adaptive_sampling.threshold > 0 && adaptive_min_samples == 0) { + /* Threshold 0.1 -> 32, 0.01 -> 64, 0.001 -> 128. + * This is highly scene dependent, we make a guess that seemed to work well + * in various test scenes. */ + const int min_samples = (int)ceilf(16.0f / powf(adaptive_sampling.threshold, 0.3f)); + adaptive_sampling.min_samples = max(4, min_samples); + VLOG(1) << "Cycles adaptive sampling: automatic min samples = " + << adaptive_sampling.min_samples; + } + else { + adaptive_sampling.min_samples = max(4, adaptive_min_samples); + } + + /* Arbitrary factor that makes the threshold more similar to what is was before, + * and gives arguably more intuitive values. */ + adaptive_sampling.threshold *= 5.0f; + + adaptive_sampling.adaptive_step = 16; + + DCHECK(is_power_of_two(adaptive_sampling.adaptive_step)) + << "Adaptive step must be a power of two for bitwise operations to work"; + + return adaptive_sampling; +} + +DenoiseParams Integrator::get_denoise_params() const +{ + DenoiseParams denoise_params; + + denoise_params.use = use_denoise; + + denoise_params.type = denoiser_type; + + denoise_params.start_sample = denoise_start_sample; + + denoise_params.use_pass_albedo = use_denoise_pass_albedo; + denoise_params.use_pass_normal = use_denoise_pass_normal; + + denoise_params.prefilter = denoiser_prefilter; + + return denoise_params; +} + CCL_NAMESPACE_END diff --git a/intern/cycles/render/integrator.h b/intern/cycles/render/integrator.h index 4eeeda92d41..32e108d62ca 100644 --- a/intern/cycles/render/integrator.h +++ b/intern/cycles/render/integrator.h @@ -19,7 +19,9 @@ #include "kernel/kernel_types.h" +#include "device/device_denoise.h" /* For the paramaters and type enum. */ #include "graph/node.h" +#include "integrator/adaptive_sampling.h" CCL_NAMESPACE_BEGIN @@ -43,6 +45,8 @@ class Integrator : public Node { NODE_SOCKET_API(int, transparent_max_bounce) NODE_SOCKET_API(int, ao_bounces) + NODE_SOCKET_API(float, ao_factor) + NODE_SOCKET_API(float, ao_distance) NODE_SOCKET_API(int, volume_max_steps) NODE_SOCKET_API(float, volume_step_rate) @@ -62,37 +66,26 @@ class Integrator : public Node { static const int MAX_SAMPLES = (1 << 24); NODE_SOCKET_API(int, aa_samples) - NODE_SOCKET_API(int, diffuse_samples) - NODE_SOCKET_API(int, glossy_samples) - NODE_SOCKET_API(int, transmission_samples) - NODE_SOCKET_API(int, ao_samples) - NODE_SOCKET_API(int, mesh_light_samples) - NODE_SOCKET_API(int, subsurface_samples) - NODE_SOCKET_API(int, volume_samples) NODE_SOCKET_API(int, start_sample) - NODE_SOCKET_API(bool, sample_all_lights_direct) - NODE_SOCKET_API(bool, sample_all_lights_indirect) NODE_SOCKET_API(float, light_sampling_threshold) + NODE_SOCKET_API(bool, use_adaptive_sampling) NODE_SOCKET_API(int, adaptive_min_samples) NODE_SOCKET_API(float, adaptive_threshold) - enum Method { - BRANCHED_PATH = 0, - PATH = 1, - - NUM_METHODS, - }; - - NODE_SOCKET_API(Method, method) - NODE_SOCKET_API(SamplingPattern, sampling_pattern) + NODE_SOCKET_API(bool, use_denoise); + NODE_SOCKET_API(DenoiserType, denoiser_type); + NODE_SOCKET_API(int, denoise_start_sample); + NODE_SOCKET_API(bool, use_denoise_pass_albedo); + NODE_SOCKET_API(bool, use_denoise_pass_normal); + NODE_SOCKET_API(DenoiserPrefilter, denoiser_prefilter); + enum : uint32_t { AO_PASS_MODIFIED = (1 << 0), - BACKGROUND_AO_MODIFIED = (1 << 1), - LIGHT_SAMPLES_MODIFIED = (1 << 2), + OBJECT_MANAGER = (1 << 1), /* tag everything in the manager for an update */ UPDATE_ALL = ~0u, @@ -107,6 +100,9 @@ class Integrator : public Node { void device_free(Device *device, DeviceScene *dscene, bool force_free = false); void tag_update(Scene *scene, uint32_t flag); + + AdaptiveSampling get_adaptive_sampling() const; + DenoiseParams get_denoise_params() const; }; CCL_NAMESPACE_END diff --git a/intern/cycles/render/jitter.cpp b/intern/cycles/render/jitter.cpp index fc47b0e8f0a..e31f8abd446 100644 --- a/intern/cycles/render/jitter.cpp +++ b/intern/cycles/render/jitter.cpp @@ -242,12 +242,6 @@ class PMJ02_Generator : public PMJ_Generator { static void shuffle(float2 points[], int size, int rng_seed) { - /* Offset samples by 1.0 for faster scrambling in kernel_random.h */ - for (int i = 0; i < size; ++i) { - points[i].x += 1.0f; - points[i].y += 1.0f; - } - if (rng_seed == 0) { return; } diff --git a/intern/cycles/render/light.cpp b/intern/cycles/render/light.cpp index 15aa4e047b5..ae1150fc07b 100644 --- a/intern/cycles/render/light.cpp +++ b/intern/cycles/render/light.cpp @@ -14,12 +14,13 @@ * limitations under the License. */ -#include "render/light.h" #include "device/device.h" + #include "render/background.h" #include "render/film.h" #include "render/graph.h" #include "render/integrator.h" +#include "render/light.h" #include "render/mesh.h" #include "render/nodes.h" #include "render/object.h" @@ -27,6 +28,8 @@ #include "render/shader.h" #include "render/stats.h" +#include "integrator/shader_eval.h" + #include "util/util_foreach.h" #include "util/util_hash.h" #include "util/util_logging.h" @@ -43,63 +46,49 @@ static void shade_background_pixels(Device *device, vector<float3> &pixels, Progress &progress) { - /* create input */ - device_vector<uint4> d_input(device, "background_input", MEM_READ_ONLY); - device_vector<float4> d_output(device, "background_output", MEM_READ_WRITE); - - uint4 *d_input_data = d_input.alloc(width * height); - - for (int y = 0; y < height; y++) { - for (int x = 0; x < width; x++) { - float u = (x + 0.5f) / width; - float v = (y + 0.5f) / height; - - uint4 in = make_uint4(__float_as_int(u), __float_as_int(v), 0, 0); - d_input_data[x + y * width] = in; - } - } - - /* compute on device */ - d_output.alloc(width * height); - d_output.zero_to_device(); - d_input.copy_to_device(); - + /* Needs to be up to data for attribute access. */ device->const_copy_to("__data", &dscene->data, sizeof(dscene->data)); - DeviceTask main_task(DeviceTask::SHADER); - main_task.shader_input = d_input.device_pointer; - main_task.shader_output = d_output.device_pointer; - main_task.shader_eval_type = SHADER_EVAL_BACKGROUND; - main_task.shader_x = 0; - main_task.shader_w = width * height; - main_task.num_samples = 1; - main_task.get_cancel = function_bind(&Progress::get_cancel, &progress); - - /* disabled splitting for now, there's an issue with multi-GPU mem_copy_from */ - list<DeviceTask> split_tasks; - main_task.split(split_tasks, 1, 128 * 128); - - foreach (DeviceTask &task, split_tasks) { - device->task_add(task); - device->task_wait(); - d_output.copy_from_device(task.shader_x, 1, task.shader_w); - } - - d_input.free(); - - float4 *d_output_data = d_output.data(); - - pixels.resize(width * height); - - for (int y = 0; y < height; y++) { - for (int x = 0; x < width; x++) { - pixels[y * width + x].x = d_output_data[y * width + x].x; - pixels[y * width + x].y = d_output_data[y * width + x].y; - pixels[y * width + x].z = d_output_data[y * width + x].z; - } - } + const int size = width * height; + pixels.resize(size); + + /* Evaluate shader on device. */ + ShaderEval shader_eval(device, progress); + shader_eval.eval( + SHADER_EVAL_BACKGROUND, + size, + [&](device_vector<KernelShaderEvalInput> &d_input) { + /* Fill coordinates for shading. */ + KernelShaderEvalInput *d_input_data = d_input.data(); + + for (int y = 0; y < height; y++) { + for (int x = 0; x < width; x++) { + float u = (x + 0.5f) / width; + float v = (y + 0.5f) / height; + + KernelShaderEvalInput in; + in.object = OBJECT_NONE; + in.prim = PRIM_NONE; + in.u = u; + in.v = v; + d_input_data[x + y * width] = in; + } + } - d_output.free(); + return size; + }, + [&](device_vector<float4> &d_output) { + /* Copy output to pixel buffer. */ + float4 *d_output_data = d_output.data(); + + for (int y = 0; y < height; y++) { + for (int x = 0; x < width; x++) { + pixels[y * width + x].x = d_output_data[y * width + x].x; + pixels[y * width + x].y = d_output_data[y * width + x].y; + pixels[y * width + x].z = d_output_data[y * width + x].z; + } + } + }); } /* Light */ @@ -140,15 +129,16 @@ NODE_DEFINE(Light) SOCKET_BOOLEAN(cast_shadow, "Cast Shadow", true); SOCKET_BOOLEAN(use_mis, "Use Mis", false); + SOCKET_BOOLEAN(use_camera, "Use Camera", true); SOCKET_BOOLEAN(use_diffuse, "Use Diffuse", true); SOCKET_BOOLEAN(use_glossy, "Use Glossy", true); SOCKET_BOOLEAN(use_transmission, "Use Transmission", true); SOCKET_BOOLEAN(use_scatter, "Use Scatter", true); - SOCKET_INT(samples, "Samples", 1); SOCKET_INT(max_bounces, "Max Bounces", 1024); SOCKET_UINT(random_id, "Random ID", 0); + SOCKET_BOOLEAN(is_shadow_catcher, "Shadow Catcher", true); SOCKET_BOOLEAN(is_portal, "Is Portal", false); SOCKET_BOOLEAN(is_enabled, "Is Enabled", true); @@ -166,10 +156,6 @@ void Light::tag_update(Scene *scene) { if (is_modified()) { scene->light_manager->tag_update(scene, LightManager::LIGHT_MODIFIED); - - if (samples_is_modified()) { - scene->integrator->tag_update(scene, Integrator::LIGHT_SAMPLES_MODIFIED); - } } } @@ -193,7 +179,6 @@ LightManager::LightManager() { update_flags = UPDATE_ALL; need_update_background = true; - use_light_visibility = false; last_background_enabled = false; last_background_resolution = 0; } @@ -357,21 +342,23 @@ void LightManager::device_update_distribution(Device *, int object_id = j; int shader_flag = 0; + if (!(object->get_visibility() & PATH_RAY_CAMERA)) { + shader_flag |= SHADER_EXCLUDE_CAMERA; + } if (!(object->get_visibility() & PATH_RAY_DIFFUSE)) { shader_flag |= SHADER_EXCLUDE_DIFFUSE; - use_light_visibility = true; } if (!(object->get_visibility() & PATH_RAY_GLOSSY)) { shader_flag |= SHADER_EXCLUDE_GLOSSY; - use_light_visibility = true; } if (!(object->get_visibility() & PATH_RAY_TRANSMIT)) { shader_flag |= SHADER_EXCLUDE_TRANSMIT; - use_light_visibility = true; } if (!(object->get_visibility() & PATH_RAY_VOLUME_SCATTER)) { shader_flag |= SHADER_EXCLUDE_SCATTER; - use_light_visibility = true; + } + if (!(object->get_is_shadow_catcher())) { + shader_flag |= SHADER_EXCLUDE_SHADOW_CATCHER; } size_t mesh_num_triangles = mesh->num_triangles(); @@ -496,10 +483,10 @@ void LightManager::device_update_distribution(Device *, kfilm->pass_shadow_scale = 1.0f; if (kintegrator->pdf_triangles != 0.0f) - kfilm->pass_shadow_scale *= 0.5f; + kfilm->pass_shadow_scale /= 0.5f; if (num_background_lights < num_lights) - kfilm->pass_shadow_scale *= (float)(num_lights - num_background_lights) / (float)num_lights; + kfilm->pass_shadow_scale /= (float)(num_lights - num_background_lights) / (float)num_lights; /* CDF */ dscene->light_distribution.copy_to_device(); @@ -766,25 +753,26 @@ void LightManager::device_update_points(Device *, DeviceScene *dscene, Scene *sc if (!light->cast_shadow) shader_id &= ~SHADER_CAST_SHADOW; + if (!light->use_camera) { + shader_id |= SHADER_EXCLUDE_CAMERA; + } if (!light->use_diffuse) { shader_id |= SHADER_EXCLUDE_DIFFUSE; - use_light_visibility = true; } if (!light->use_glossy) { shader_id |= SHADER_EXCLUDE_GLOSSY; - use_light_visibility = true; } if (!light->use_transmission) { shader_id |= SHADER_EXCLUDE_TRANSMIT; - use_light_visibility = true; } if (!light->use_scatter) { shader_id |= SHADER_EXCLUDE_SCATTER; - use_light_visibility = true; + } + if (!light->is_shadow_catcher) { + shader_id |= SHADER_EXCLUDE_SHADOW_CATCHER; } klights[light_index].type = light->light_type; - klights[light_index].samples = light->samples; klights[light_index].strength[0] = light->strength.x; klights[light_index].strength[1] = light->strength.y; klights[light_index].strength[2] = light->strength.z; @@ -836,19 +824,15 @@ void LightManager::device_update_points(Device *, DeviceScene *dscene, Scene *sc if (!(visibility & PATH_RAY_DIFFUSE)) { shader_id |= SHADER_EXCLUDE_DIFFUSE; - use_light_visibility = true; } if (!(visibility & PATH_RAY_GLOSSY)) { shader_id |= SHADER_EXCLUDE_GLOSSY; - use_light_visibility = true; } if (!(visibility & PATH_RAY_TRANSMIT)) { shader_id |= SHADER_EXCLUDE_TRANSMIT; - use_light_visibility = true; } if (!(visibility & PATH_RAY_VOLUME_SCATTER)) { shader_id |= SHADER_EXCLUDE_SCATTER; - use_light_visibility = true; } } else if (light->light_type == LIGHT_AREA) { @@ -998,8 +982,6 @@ void LightManager::device_update(Device *device, device_free(device, dscene, need_update_background); - use_light_visibility = false; - device_update_points(device, dscene, scene); if (progress.get_cancel()) return; @@ -1018,8 +1000,6 @@ void LightManager::device_update(Device *device, if (progress.get_cancel()) return; - scene->film->set_use_light_visibility(use_light_visibility); - update_flags = UPDATE_NONE; need_update_background = false; } diff --git a/intern/cycles/render/light.h b/intern/cycles/render/light.h index fbd709125ff..7f86237c8b3 100644 --- a/intern/cycles/render/light.h +++ b/intern/cycles/render/light.h @@ -69,16 +69,17 @@ class Light : public Node { NODE_SOCKET_API(bool, cast_shadow) NODE_SOCKET_API(bool, use_mis) + NODE_SOCKET_API(bool, use_camera) NODE_SOCKET_API(bool, use_diffuse) NODE_SOCKET_API(bool, use_glossy) NODE_SOCKET_API(bool, use_transmission) NODE_SOCKET_API(bool, use_scatter) + NODE_SOCKET_API(bool, is_shadow_catcher) NODE_SOCKET_API(bool, is_portal) NODE_SOCKET_API(bool, is_enabled) NODE_SOCKET_API(Shader *, shader) - NODE_SOCKET_API(int, samples) NODE_SOCKET_API(int, max_bounces) NODE_SOCKET_API(uint, random_id) @@ -108,8 +109,6 @@ class LightManager { UPDATE_NONE = 0u, }; - bool use_light_visibility; - /* Need to update background (including multiple importance map) */ bool need_update_background; diff --git a/intern/cycles/render/mesh_displace.cpp b/intern/cycles/render/mesh_displace.cpp index b39d81023d9..c00c4c24211 100644 --- a/intern/cycles/render/mesh_displace.cpp +++ b/intern/cycles/render/mesh_displace.cpp @@ -16,6 +16,8 @@ #include "device/device.h" +#include "integrator/shader_eval.h" + #include "render/mesh.h" #include "render/object.h" #include "render/scene.h" @@ -43,40 +45,28 @@ static float3 compute_face_normal(const Mesh::Triangle &t, float3 *verts) return norm / normlen; } -bool GeometryManager::displace( - Device *device, DeviceScene *dscene, Scene *scene, Mesh *mesh, Progress &progress) +/* Fill in coordinates for mesh displacement shader evaluation on device. */ +static int fill_shader_input(const Scene *scene, + const Mesh *mesh, + const int object_index, + device_vector<KernelShaderEvalInput> &d_input) { - /* verify if we have a displacement shader */ - if (!mesh->has_true_displacement()) { - return false; - } - - string msg = string_printf("Computing Displacement %s", mesh->name.c_str()); - progress.set_status("Updating Mesh", msg); + int d_input_size = 0; + KernelShaderEvalInput *d_input_data = d_input.data(); - /* find object index. todo: is arbitrary */ - size_t object_index = OBJECT_NONE; + const array<int> &mesh_shaders = mesh->get_shader(); + const array<Node *> &mesh_used_shaders = mesh->get_used_shaders(); + const array<float3> &mesh_verts = mesh->get_verts(); - for (size_t i = 0; i < scene->objects.size(); i++) { - if (scene->objects[i]->get_geometry() == mesh) { - object_index = i; - break; - } - } - - /* setup input for device task */ - const size_t num_verts = mesh->verts.size(); + const int num_verts = mesh_verts.size(); vector<bool> done(num_verts, false); - device_vector<uint4> d_input(device, "displace_input", MEM_READ_ONLY); - uint4 *d_input_data = d_input.alloc(num_verts); - size_t d_input_size = 0; - size_t num_triangles = mesh->num_triangles(); - for (size_t i = 0; i < num_triangles; i++) { + int num_triangles = mesh->num_triangles(); + for (int i = 0; i < num_triangles; i++) { Mesh::Triangle t = mesh->get_triangle(i); - int shader_index = mesh->shader[i]; - Shader *shader = (shader_index < mesh->used_shaders.size()) ? - static_cast<Shader *>(mesh->used_shaders[shader_index]) : + int shader_index = mesh_shaders[i]; + Shader *shader = (shader_index < mesh_used_shaders.size()) ? + static_cast<Shader *>(mesh_used_shaders[shader_index]) : scene->default_surface; if (!shader->has_displacement || shader->get_displacement_method() == DISPLACE_BUMP) { @@ -110,57 +100,41 @@ bool GeometryManager::displace( } /* back */ - uint4 in = make_uint4(object, prim, __float_as_int(u), __float_as_int(v)); + KernelShaderEvalInput in; + in.object = object; + in.prim = prim; + in.u = u; + in.v = v; d_input_data[d_input_size++] = in; } } - if (d_input_size == 0) - return false; - - /* run device task */ - device_vector<float4> d_output(device, "displace_output", MEM_READ_WRITE); - d_output.alloc(d_input_size); - d_output.zero_to_device(); - d_input.copy_to_device(); - - /* needs to be up to data for attribute access */ - device->const_copy_to("__data", &dscene->data, sizeof(dscene->data)); - - DeviceTask task(DeviceTask::SHADER); - task.shader_input = d_input.device_pointer; - task.shader_output = d_output.device_pointer; - task.shader_eval_type = SHADER_EVAL_DISPLACE; - task.shader_x = 0; - task.shader_w = d_output.size(); - task.num_samples = 1; - task.get_cancel = function_bind(&Progress::get_cancel, &progress); - - device->task_add(task); - device->task_wait(); - - if (progress.get_cancel()) { - d_input.free(); - d_output.free(); - return false; - } + return d_input_size; +} - d_output.copy_from_device(0, 1, d_output.size()); - d_input.free(); +/* Read back mesh displacement shader output. */ +static void read_shader_output(const Scene *scene, + Mesh *mesh, + const device_vector<float4> &d_output) +{ + const array<int> &mesh_shaders = mesh->get_shader(); + const array<Node *> &mesh_used_shaders = mesh->get_used_shaders(); + array<float3> &mesh_verts = mesh->get_verts(); - /* read result */ - done.clear(); - done.resize(num_verts, false); - int k = 0; + const int num_verts = mesh_verts.size(); + const int num_motion_steps = mesh->get_motion_steps(); + vector<bool> done(num_verts, false); - float4 *offset = d_output.data(); + const float4 *d_output_data = d_output.data(); + int d_output_index = 0; Attribute *attr_mP = mesh->attributes.find(ATTR_STD_MOTION_VERTEX_POSITION); - for (size_t i = 0; i < num_triangles; i++) { + int num_triangles = mesh->num_triangles(); + for (int i = 0; i < num_triangles; i++) { Mesh::Triangle t = mesh->get_triangle(i); - int shader_index = mesh->shader[i]; - Shader *shader = (shader_index < mesh->used_shaders.size()) ? - static_cast<Shader *>(mesh->used_shaders[shader_index]) : + int shader_index = mesh_shaders[i]; + Shader *shader = (shader_index < mesh_used_shaders.size()) ? + static_cast<Shader *>(mesh_used_shaders[shader_index]) : scene->default_surface; if (!shader->has_displacement || shader->get_displacement_method() == DISPLACE_BUMP) { @@ -170,12 +144,12 @@ bool GeometryManager::displace( for (int j = 0; j < 3; j++) { if (!done[t.v[j]]) { done[t.v[j]] = true; - float3 off = float4_to_float3(offset[k++]); + float3 off = float4_to_float3(d_output_data[d_output_index++]); /* Avoid illegal vertex coordinates. */ off = ensure_finite3(off); - mesh->verts[t.v[j]] += off; + mesh_verts[t.v[j]] += off; if (attr_mP != NULL) { - for (int step = 0; step < mesh->motion_steps - 1; step++) { + for (int step = 0; step < num_motion_steps - 1; step++) { float3 *mP = attr_mP->data_float3() + step * num_verts; mP[t.v[j]] += off; } @@ -183,8 +157,47 @@ bool GeometryManager::displace( } } } +} - d_output.free(); +bool GeometryManager::displace( + Device *device, DeviceScene *dscene, Scene *scene, Mesh *mesh, Progress &progress) +{ + /* verify if we have a displacement shader */ + if (!mesh->has_true_displacement()) { + return false; + } + + const size_t num_verts = mesh->verts.size(); + const size_t num_triangles = mesh->num_triangles(); + + if (num_triangles == 0) { + return false; + } + + string msg = string_printf("Computing Displacement %s", mesh->name.c_str()); + progress.set_status("Updating Mesh", msg); + + /* find object index. todo: is arbitrary */ + size_t object_index = OBJECT_NONE; + + for (size_t i = 0; i < scene->objects.size(); i++) { + if (scene->objects[i]->get_geometry() == mesh) { + object_index = i; + break; + } + } + + /* Needs to be up to data for attribute access. */ + device->const_copy_to("__data", &dscene->data, sizeof(dscene->data)); + + /* Evaluate shader on device. */ + ShaderEval shader_eval(device, progress); + if (!shader_eval.eval(SHADER_EVAL_DISPLACE, + num_verts, + function_bind(&fill_shader_input, scene, mesh, object_index, _1), + function_bind(&read_shader_output, scene, mesh, _1))) { + return false; + } /* stitch */ unordered_set<int> stitch_keys; @@ -297,8 +310,7 @@ bool GeometryManager::displace( } /* normalize vertex normals */ - done.clear(); - done.resize(num_verts, false); + vector<bool> done(num_verts, false); for (size_t i = 0; i < num_triangles; i++) { if (tri_has_true_disp[i]) { @@ -368,8 +380,7 @@ bool GeometryManager::displace( } /* normalize vertex normals */ - done.clear(); - done.resize(num_verts, false); + vector<bool> done(num_verts, false); for (size_t i = 0; i < num_triangles; i++) { if (tri_has_true_disp[i]) { diff --git a/intern/cycles/render/nodes.cpp b/intern/cycles/render/nodes.cpp index 795166bcf4c..5303d55242e 100644 --- a/intern/cycles/render/nodes.cpp +++ b/intern/cycles/render/nodes.cpp @@ -2736,18 +2736,21 @@ NODE_DEFINE(PrincipledBsdfNode) distribution, "Distribution", distribution_enum, CLOSURE_BSDF_MICROFACET_MULTI_GGX_GLASS_ID); static NodeEnum subsurface_method_enum; - subsurface_method_enum.insert("burley", CLOSURE_BSSRDF_PRINCIPLED_ID); - subsurface_method_enum.insert("random_walk", CLOSURE_BSSRDF_PRINCIPLED_RANDOM_WALK_ID); + subsurface_method_enum.insert("random_walk_fixed_radius", + CLOSURE_BSSRDF_RANDOM_WALK_FIXED_RADIUS_ID); + subsurface_method_enum.insert("random_walk", CLOSURE_BSSRDF_RANDOM_WALK_ID); SOCKET_ENUM(subsurface_method, "Subsurface Method", subsurface_method_enum, - CLOSURE_BSSRDF_PRINCIPLED_ID); + CLOSURE_BSSRDF_RANDOM_WALK_ID); SOCKET_IN_COLOR(base_color, "Base Color", make_float3(0.8f, 0.8f, 0.8f)); SOCKET_IN_COLOR(subsurface_color, "Subsurface Color", make_float3(0.8f, 0.8f, 0.8f)); SOCKET_IN_FLOAT(metallic, "Metallic", 0.0f); SOCKET_IN_FLOAT(subsurface, "Subsurface", 0.0f); SOCKET_IN_VECTOR(subsurface_radius, "Subsurface Radius", make_float3(0.1f, 0.1f, 0.1f)); + SOCKET_IN_FLOAT(subsurface_ior, "Subsurface IOR", 1.4f); + SOCKET_IN_FLOAT(subsurface_anisotropy, "Subsurface Anisotropy", 0.0f); SOCKET_IN_FLOAT(specular, "Specular", 0.0f); SOCKET_IN_FLOAT(roughness, "Roughness", 0.5f); SOCKET_IN_FLOAT(specular_tint, "Specular Tint", 0.0f); @@ -2857,6 +2860,8 @@ void PrincipledBsdfNode::compile(SVMCompiler &compiler, ShaderInput *p_metallic, ShaderInput *p_subsurface, ShaderInput *p_subsurface_radius, + ShaderInput *p_subsurface_ior, + ShaderInput *p_subsurface_anisotropy, ShaderInput *p_specular, ShaderInput *p_roughness, ShaderInput *p_specular_tint, @@ -2896,6 +2901,8 @@ void PrincipledBsdfNode::compile(SVMCompiler &compiler, int transmission_roughness_offset = compiler.stack_assign(p_transmission_roughness); int anisotropic_rotation_offset = compiler.stack_assign(p_anisotropic_rotation); int subsurface_radius_offset = compiler.stack_assign(p_subsurface_radius); + int subsurface_ior_offset = compiler.stack_assign(p_subsurface_ior); + int subsurface_anisotropy_offset = compiler.stack_assign(p_subsurface_anisotropy); compiler.add_node(NODE_CLOSURE_BSDF, compiler.encode_uchar4(closure, @@ -2929,8 +2936,10 @@ void PrincipledBsdfNode::compile(SVMCompiler &compiler, __float_as_int(bc_default.y), __float_as_int(bc_default.z)); - compiler.add_node( - clearcoat_normal_offset, subsurface_radius_offset, SVM_STACK_INVALID, SVM_STACK_INVALID); + compiler.add_node(clearcoat_normal_offset, + subsurface_radius_offset, + subsurface_ior_offset, + subsurface_anisotropy_offset); float3 ss_default = get_float3(subsurface_color_in->socket_type); @@ -2953,6 +2962,8 @@ void PrincipledBsdfNode::compile(SVMCompiler &compiler) input("Metallic"), input("Subsurface"), input("Subsurface Radius"), + input("Subsurface IOR"), + input("Subsurface Anisotropy"), input("Specular"), input("Roughness"), input("Specular Tint"), @@ -3048,16 +3059,16 @@ NODE_DEFINE(SubsurfaceScatteringNode) SOCKET_IN_NORMAL(normal, "Normal", zero_float3(), SocketType::LINK_NORMAL); SOCKET_IN_FLOAT(surface_mix_weight, "SurfaceMixWeight", 0.0f, SocketType::SVM_INTERNAL); - static NodeEnum falloff_enum; - falloff_enum.insert("cubic", CLOSURE_BSSRDF_CUBIC_ID); - falloff_enum.insert("gaussian", CLOSURE_BSSRDF_GAUSSIAN_ID); - falloff_enum.insert("burley", CLOSURE_BSSRDF_BURLEY_ID); - falloff_enum.insert("random_walk", CLOSURE_BSSRDF_RANDOM_WALK_ID); - SOCKET_ENUM(falloff, "Falloff", falloff_enum, CLOSURE_BSSRDF_BURLEY_ID); + static NodeEnum method_enum; + method_enum.insert("random_walk_fixed_radius", CLOSURE_BSSRDF_RANDOM_WALK_FIXED_RADIUS_ID); + method_enum.insert("random_walk", CLOSURE_BSSRDF_RANDOM_WALK_ID); + SOCKET_ENUM(method, "Method", method_enum, CLOSURE_BSSRDF_RANDOM_WALK_ID); + SOCKET_IN_FLOAT(scale, "Scale", 0.01f); SOCKET_IN_VECTOR(radius, "Radius", make_float3(0.1f, 0.1f, 0.1f)); - SOCKET_IN_FLOAT(sharpness, "Sharpness", 0.0f); - SOCKET_IN_FLOAT(texture_blur, "Texture Blur", 1.0f); + + SOCKET_IN_FLOAT(subsurface_ior, "IOR", 1.4f); + SOCKET_IN_FLOAT(subsurface_anisotropy, "Anisotropy", 0.0f); SOCKET_OUT_CLOSURE(BSSRDF, "BSSRDF"); @@ -3066,20 +3077,19 @@ NODE_DEFINE(SubsurfaceScatteringNode) SubsurfaceScatteringNode::SubsurfaceScatteringNode() : BsdfNode(get_node_type()) { - closure = falloff; + closure = method; } void SubsurfaceScatteringNode::compile(SVMCompiler &compiler) { - closure = falloff; - BsdfNode::compile( - compiler, input("Scale"), input("Texture Blur"), input("Radius"), input("Sharpness")); + closure = method; + BsdfNode::compile(compiler, input("Scale"), input("IOR"), input("Radius"), input("Anisotropy")); } void SubsurfaceScatteringNode::compile(OSLCompiler &compiler) { - closure = falloff; - compiler.parameter(this, "falloff"); + closure = method; + compiler.parameter(this, "method"); compiler.add(this, "node_subsurface_scattering"); } @@ -3786,20 +3796,6 @@ void GeometryNode::compile(OSLCompiler &compiler) compiler.add(this, "node_geometry"); } -int GeometryNode::get_group() -{ - ShaderOutput *out; - int result = ShaderNode::get_group(); - - /* Backfacing uses NODE_LIGHT_PATH */ - out = output("Backfacing"); - if (!out->links.empty()) { - result = max(result, NODE_GROUP_LEVEL_1); - } - - return result; -} - /* TextureCoordinate */ NODE_DEFINE(TextureCoordinateNode) @@ -5926,33 +5922,33 @@ NODE_DEFINE(OutputAOVNode) OutputAOVNode::OutputAOVNode() : ShaderNode(get_node_type()) { special_type = SHADER_SPECIAL_TYPE_OUTPUT_AOV; - slot = -1; + offset = -1; } void OutputAOVNode::simplify_settings(Scene *scene) { - slot = scene->film->get_aov_offset(scene, name.string(), is_color); - if (slot == -1) { - slot = scene->film->get_aov_offset(scene, name.string(), is_color); + offset = scene->film->get_aov_offset(scene, name.string(), is_color); + if (offset == -1) { + offset = scene->film->get_aov_offset(scene, name.string(), is_color); } - if (slot == -1 || is_color) { + if (offset == -1 || is_color) { input("Value")->disconnect(); } - if (slot == -1 || !is_color) { + if (offset == -1 || !is_color) { input("Color")->disconnect(); } } void OutputAOVNode::compile(SVMCompiler &compiler) { - assert(slot >= 0); + assert(offset >= 0); if (is_color) { - compiler.add_node(NODE_AOV_COLOR, compiler.stack_assign(input("Color")), slot); + compiler.add_node(NODE_AOV_COLOR, compiler.stack_assign(input("Color")), offset); } else { - compiler.add_node(NODE_AOV_VALUE, compiler.stack_assign(input("Value")), slot); + compiler.add_node(NODE_AOV_VALUE, compiler.stack_assign(input("Value")), offset); } } diff --git a/intern/cycles/render/nodes.h b/intern/cycles/render/nodes.h index 3013e9b1866..22bdb06b059 100644 --- a/intern/cycles/render/nodes.h +++ b/intern/cycles/render/nodes.h @@ -143,10 +143,6 @@ class EnvironmentTextureNode : public ImageSlotTextureNode { { return true; } - virtual int get_group() - { - return NODE_GROUP_LEVEL_2; - } virtual bool equals(const ShaderNode &other) { @@ -170,11 +166,6 @@ class SkyTextureNode : public TextureNode { public: SHADER_NODE_CLASS(SkyTextureNode) - virtual int get_group() - { - return NODE_GROUP_LEVEL_2; - } - NODE_SOCKET_API(NodeSkyType, sky_type) NODE_SOCKET_API(float3, sun_direction) NODE_SOCKET_API(float, turbidity) @@ -224,18 +215,13 @@ class OutputAOVNode : public ShaderNode { NODE_SOCKET_API(ustring, name) - virtual int get_group() - { - return NODE_GROUP_LEVEL_4; - } - /* Don't allow output node de-duplication. */ virtual bool equals(const ShaderNode & /*other*/) { return false; } - int slot; + int offset; bool is_color; }; @@ -243,11 +229,6 @@ class GradientTextureNode : public TextureNode { public: SHADER_NODE_CLASS(GradientTextureNode) - virtual int get_group() - { - return NODE_GROUP_LEVEL_2; - } - NODE_SOCKET_API(NodeGradientType, gradient_type) NODE_SOCKET_API(float3, vector) }; @@ -269,19 +250,14 @@ class VoronoiTextureNode : public TextureNode { public: SHADER_NODE_CLASS(VoronoiTextureNode) - virtual int get_group() - { - return NODE_GROUP_LEVEL_2; - } - virtual int get_feature() { int result = ShaderNode::get_feature(); if (dimensions == 4) { - result |= NODE_FEATURE_VORONOI_EXTRA; + result |= KERNEL_FEATURE_NODE_VORONOI_EXTRA; } else if (dimensions >= 2 && feature == NODE_VORONOI_SMOOTH_F1) { - result |= NODE_FEATURE_VORONOI_EXTRA; + result |= KERNEL_FEATURE_NODE_VORONOI_EXTRA; } return result; } @@ -301,11 +277,6 @@ class MusgraveTextureNode : public TextureNode { public: SHADER_NODE_CLASS(MusgraveTextureNode) - virtual int get_group() - { - return NODE_GROUP_LEVEL_2; - } - NODE_SOCKET_API(int, dimensions) NODE_SOCKET_API(NodeMusgraveType, musgrave_type) NODE_SOCKET_API(float, w) @@ -322,11 +293,6 @@ class WaveTextureNode : public TextureNode { public: SHADER_NODE_CLASS(WaveTextureNode) - virtual int get_group() - { - return NODE_GROUP_LEVEL_2; - } - NODE_SOCKET_API(NodeWaveType, wave_type) NODE_SOCKET_API(NodeWaveBandsDirection, bands_direction) NODE_SOCKET_API(NodeWaveRingsDirection, rings_direction) @@ -345,11 +311,6 @@ class MagicTextureNode : public TextureNode { public: SHADER_NODE_CLASS(MagicTextureNode) - virtual int get_group() - { - return NODE_GROUP_LEVEL_2; - } - NODE_SOCKET_API(int, depth) NODE_SOCKET_API(float3, vector) NODE_SOCKET_API(float, scale) @@ -364,11 +325,6 @@ class CheckerTextureNode : public TextureNode { NODE_SOCKET_API(float3, color1) NODE_SOCKET_API(float3, color2) NODE_SOCKET_API(float, scale) - - virtual int get_group() - { - return NODE_GROUP_LEVEL_2; - } }; class BrickTextureNode : public TextureNode { @@ -390,20 +346,11 @@ class BrickTextureNode : public TextureNode { NODE_SOCKET_API(float, brick_width) NODE_SOCKET_API(float, row_height) NODE_SOCKET_API(float3, vector) - - virtual int get_group() - { - return NODE_GROUP_LEVEL_2; - } }; class PointDensityTextureNode : public ShaderNode { public: SHADER_NODE_NO_CLONE_CLASS(PointDensityTextureNode) - virtual int get_group() - { - return NODE_GROUP_LEVEL_4; - } ~PointDensityTextureNode(); ShaderNode *clone(ShaderGraph *graph) const; @@ -443,10 +390,6 @@ class IESLightNode : public TextureNode { ~IESLightNode(); ShaderNode *clone(ShaderGraph *graph) const; - virtual int get_group() - { - return NODE_GROUP_LEVEL_2; - } NODE_SOCKET_API(ustring, filename) NODE_SOCKET_API(ustring, ies) @@ -464,10 +407,6 @@ class IESLightNode : public TextureNode { class WhiteNoiseTextureNode : public ShaderNode { public: SHADER_NODE_CLASS(WhiteNoiseTextureNode) - virtual int get_group() - { - return NODE_GROUP_LEVEL_2; - } NODE_SOCKET_API(int, dimensions) NODE_SOCKET_API(float3, vector) @@ -477,10 +416,6 @@ class WhiteNoiseTextureNode : public ShaderNode { class MappingNode : public ShaderNode { public: SHADER_NODE_CLASS(MappingNode) - virtual int get_group() - { - return NODE_GROUP_LEVEL_2; - } void constant_fold(const ConstantFolder &folder); NODE_SOCKET_API(float3, vector) @@ -546,6 +481,11 @@ class BsdfBaseNode : public ShaderNode { return false; } + virtual int get_feature() + { + return ShaderNode::get_feature() | KERNEL_FEATURE_NODE_BSDF; + } + protected: ClosureType closure; }; @@ -606,6 +546,8 @@ class PrincipledBsdfNode : public BsdfBaseNode { ShaderInput *metallic, ShaderInput *subsurface, ShaderInput *subsurface_radius, + ShaderInput *subsurface_ior, + ShaderInput *subsurface_anisotropy, ShaderInput *specular, ShaderInput *roughness, ShaderInput *specular_tint, @@ -622,6 +564,8 @@ class PrincipledBsdfNode : public BsdfBaseNode { NODE_SOCKET_API(float3, base_color) NODE_SOCKET_API(float3, subsurface_color) NODE_SOCKET_API(float3, subsurface_radius) + NODE_SOCKET_API(float, subsurface_ior) + NODE_SOCKET_API(float, subsurface_anisotropy) NODE_SOCKET_API(float, metallic) NODE_SOCKET_API(float, subsurface) NODE_SOCKET_API(float, specular) @@ -758,14 +702,14 @@ class SubsurfaceScatteringNode : public BsdfNode { bool has_bssrdf_bump(); ClosureType get_closure_type() { - return falloff; + return method; } NODE_SOCKET_API(float, scale) NODE_SOCKET_API(float3, radius) - NODE_SOCKET_API(float, sharpness) - NODE_SOCKET_API(float, texture_blur) - NODE_SOCKET_API(ClosureType, falloff) + NODE_SOCKET_API(float, subsurface_ior) + NODE_SOCKET_API(float, subsurface_anisotropy) + NODE_SOCKET_API(ClosureType, method) }; class EmissionNode : public ShaderNode { @@ -782,6 +726,11 @@ class EmissionNode : public ShaderNode { return true; } + virtual int get_feature() + { + return ShaderNode::get_feature() | KERNEL_FEATURE_NODE_EMISSION; + } + NODE_SOCKET_API(float3, color) NODE_SOCKET_API(float, strength) NODE_SOCKET_API(float, surface_mix_weight) @@ -792,6 +741,11 @@ class BackgroundNode : public ShaderNode { SHADER_NODE_CLASS(BackgroundNode) void constant_fold(const ConstantFolder &folder); + virtual int get_feature() + { + return ShaderNode::get_feature() | KERNEL_FEATURE_NODE_EMISSION; + } + NODE_SOCKET_API(float3, color) NODE_SOCKET_API(float, strength) NODE_SOCKET_API(float, surface_mix_weight) @@ -800,10 +754,6 @@ class BackgroundNode : public ShaderNode { class HoldoutNode : public ShaderNode { public: SHADER_NODE_CLASS(HoldoutNode) - virtual int get_group() - { - return NODE_GROUP_LEVEL_1; - } virtual ClosureType get_closure_type() { return CLOSURE_HOLDOUT_ID; @@ -821,13 +771,9 @@ class AmbientOcclusionNode : public ShaderNode { { return true; } - virtual int get_group() - { - return NODE_GROUP_LEVEL_3; - } - virtual bool has_raytrace() + virtual int get_feature() { - return true; + return KERNEL_FEATURE_NODE_RAYTRACE; } NODE_SOCKET_API(float3, color) @@ -845,13 +791,9 @@ class VolumeNode : public ShaderNode { SHADER_NODE_BASE_CLASS(VolumeNode) void compile(SVMCompiler &compiler, ShaderInput *param1, ShaderInput *param2); - virtual int get_group() - { - return NODE_GROUP_LEVEL_1; - } virtual int get_feature() { - return ShaderNode::get_feature() | NODE_FEATURE_VOLUME; + return ShaderNode::get_feature() | KERNEL_FEATURE_NODE_VOLUME; } virtual ClosureType get_closure_type() { @@ -1013,10 +955,6 @@ class UVMapNode : public ShaderNode { { return true; } - virtual int get_group() - { - return NODE_GROUP_LEVEL_1; - } NODE_SOCKET_API(ustring, attribute) NODE_SOCKET_API(bool, from_dupli) @@ -1025,10 +963,6 @@ class UVMapNode : public ShaderNode { class LightPathNode : public ShaderNode { public: SHADER_NODE_CLASS(LightPathNode) - virtual int get_group() - { - return NODE_GROUP_LEVEL_1; - } }; class LightFalloffNode : public ShaderNode { @@ -1038,10 +972,6 @@ class LightFalloffNode : public ShaderNode { { return true; } - virtual int get_group() - { - return NODE_GROUP_LEVEL_2; - } NODE_SOCKET_API(float, strength) NODE_SOCKET_API(float, smooth) @@ -1050,10 +980,6 @@ class LightFalloffNode : public ShaderNode { class ObjectInfoNode : public ShaderNode { public: SHADER_NODE_CLASS(ObjectInfoNode) - virtual int get_group() - { - return NODE_GROUP_LEVEL_1; - } }; class ParticleInfoNode : public ShaderNode { @@ -1064,10 +990,6 @@ class ParticleInfoNode : public ShaderNode { { return true; } - virtual int get_group() - { - return NODE_GROUP_LEVEL_1; - } }; class HairInfoNode : public ShaderNode { @@ -1083,13 +1005,9 @@ class HairInfoNode : public ShaderNode { { return true; } - virtual int get_group() - { - return NODE_GROUP_LEVEL_1; - } virtual int get_feature() { - return ShaderNode::get_feature() | NODE_FEATURE_HAIR; + return ShaderNode::get_feature() | KERNEL_FEATURE_NODE_HAIR; } }; @@ -1168,10 +1086,6 @@ class InvertNode : public ShaderNode { public: SHADER_NODE_CLASS(InvertNode) void constant_fold(const ConstantFolder &folder); - virtual int get_group() - { - return NODE_GROUP_LEVEL_3; - } NODE_SOCKET_API(float, fac) NODE_SOCKET_API(float3, color) @@ -1182,11 +1096,6 @@ class MixNode : public ShaderNode { SHADER_NODE_CLASS(MixNode) void constant_fold(const ConstantFolder &folder); - virtual int get_group() - { - return NODE_GROUP_LEVEL_3; - } - NODE_SOCKET_API(NodeMix, mix_type) NODE_SOCKET_API(bool, use_clamp) NODE_SOCKET_API(float3, color1) @@ -1198,10 +1107,6 @@ class CombineRGBNode : public ShaderNode { public: SHADER_NODE_CLASS(CombineRGBNode) void constant_fold(const ConstantFolder &folder); - virtual int get_group() - { - return NODE_GROUP_LEVEL_3; - } NODE_SOCKET_API(float, r) NODE_SOCKET_API(float, g) @@ -1212,10 +1117,6 @@ class CombineHSVNode : public ShaderNode { public: SHADER_NODE_CLASS(CombineHSVNode) void constant_fold(const ConstantFolder &folder); - virtual int get_group() - { - return NODE_GROUP_LEVEL_3; - } NODE_SOCKET_API(float, h) NODE_SOCKET_API(float, s) @@ -1226,10 +1127,6 @@ class CombineXYZNode : public ShaderNode { public: SHADER_NODE_CLASS(CombineXYZNode) void constant_fold(const ConstantFolder &folder); - virtual int get_group() - { - return NODE_GROUP_LEVEL_3; - } NODE_SOCKET_API(float, x) NODE_SOCKET_API(float, y) @@ -1240,10 +1137,6 @@ class GammaNode : public ShaderNode { public: SHADER_NODE_CLASS(GammaNode) void constant_fold(const ConstantFolder &folder); - virtual int get_group() - { - return NODE_GROUP_LEVEL_1; - } NODE_SOCKET_API(float3, color) NODE_SOCKET_API(float, gamma) @@ -1253,10 +1146,6 @@ class BrightContrastNode : public ShaderNode { public: SHADER_NODE_CLASS(BrightContrastNode) void constant_fold(const ConstantFolder &folder); - virtual int get_group() - { - return NODE_GROUP_LEVEL_1; - } NODE_SOCKET_API(float3, color) NODE_SOCKET_API(float, bright) @@ -1267,10 +1156,6 @@ class SeparateRGBNode : public ShaderNode { public: SHADER_NODE_CLASS(SeparateRGBNode) void constant_fold(const ConstantFolder &folder); - virtual int get_group() - { - return NODE_GROUP_LEVEL_3; - } NODE_SOCKET_API(float3, color) }; @@ -1279,10 +1164,6 @@ class SeparateHSVNode : public ShaderNode { public: SHADER_NODE_CLASS(SeparateHSVNode) void constant_fold(const ConstantFolder &folder); - virtual int get_group() - { - return NODE_GROUP_LEVEL_3; - } NODE_SOCKET_API(float3, color) }; @@ -1291,10 +1172,6 @@ class SeparateXYZNode : public ShaderNode { public: SHADER_NODE_CLASS(SeparateXYZNode) void constant_fold(const ConstantFolder &folder); - virtual int get_group() - { - return NODE_GROUP_LEVEL_3; - } NODE_SOCKET_API(float3, vector) }; @@ -1333,10 +1210,6 @@ class CameraNode : public ShaderNode { { return true; } - virtual int get_group() - { - return NODE_GROUP_LEVEL_2; - } }; class FresnelNode : public ShaderNode { @@ -1346,10 +1219,6 @@ class FresnelNode : public ShaderNode { { return true; } - virtual int get_group() - { - return NODE_GROUP_LEVEL_1; - } NODE_SOCKET_API(float3, normal) NODE_SOCKET_API(float, IOR) @@ -1362,10 +1231,6 @@ class LayerWeightNode : public ShaderNode { { return true; } - virtual int get_group() - { - return NODE_GROUP_LEVEL_1; - } NODE_SOCKET_API(float3, normal) NODE_SOCKET_API(float, blend) @@ -1378,10 +1243,6 @@ class WireframeNode : public ShaderNode { { return true; } - virtual int get_group() - { - return NODE_GROUP_LEVEL_3; - } NODE_SOCKET_API(float, size) NODE_SOCKET_API(bool, use_pixel_size) @@ -1390,10 +1251,6 @@ class WireframeNode : public ShaderNode { class WavelengthNode : public ShaderNode { public: SHADER_NODE_CLASS(WavelengthNode) - virtual int get_group() - { - return NODE_GROUP_LEVEL_3; - } NODE_SOCKET_API(float, wavelength) }; @@ -1402,10 +1259,6 @@ class BlackbodyNode : public ShaderNode { public: SHADER_NODE_CLASS(BlackbodyNode) void constant_fold(const ConstantFolder &folder); - virtual int get_group() - { - return NODE_GROUP_LEVEL_3; - } NODE_SOCKET_API(float, temperature) }; @@ -1413,10 +1266,6 @@ class BlackbodyNode : public ShaderNode { class MapRangeNode : public ShaderNode { public: SHADER_NODE_CLASS(MapRangeNode) - virtual int get_group() - { - return NODE_GROUP_LEVEL_3; - } void expand(ShaderGraph *graph); NODE_SOCKET_API(float, value) @@ -1433,10 +1282,6 @@ class ClampNode : public ShaderNode { public: SHADER_NODE_CLASS(ClampNode) void constant_fold(const ConstantFolder &folder); - virtual int get_group() - { - return NODE_GROUP_LEVEL_3; - } NODE_SOCKET_API(float, value) NODE_SOCKET_API(float, min) NODE_SOCKET_API(float, max) @@ -1446,10 +1291,6 @@ class ClampNode : public ShaderNode { class MathNode : public ShaderNode { public: SHADER_NODE_CLASS(MathNode) - virtual int get_group() - { - return NODE_GROUP_LEVEL_1; - } void expand(ShaderGraph *graph); void constant_fold(const ConstantFolder &folder); @@ -1463,10 +1304,6 @@ class MathNode : public ShaderNode { class NormalNode : public ShaderNode { public: SHADER_NODE_CLASS(NormalNode) - virtual int get_group() - { - return NODE_GROUP_LEVEL_2; - } NODE_SOCKET_API(float3, direction) NODE_SOCKET_API(float3, normal) @@ -1475,10 +1312,6 @@ class NormalNode : public ShaderNode { class VectorMathNode : public ShaderNode { public: SHADER_NODE_CLASS(VectorMathNode) - virtual int get_group() - { - return NODE_GROUP_LEVEL_1; - } void constant_fold(const ConstantFolder &folder); NODE_SOCKET_API(float3, vector1) @@ -1492,10 +1325,6 @@ class VectorRotateNode : public ShaderNode { public: SHADER_NODE_CLASS(VectorRotateNode) - virtual int get_group() - { - return NODE_GROUP_LEVEL_3; - } NODE_SOCKET_API(NodeVectorRotateType, rotate_type) NODE_SOCKET_API(bool, invert) NODE_SOCKET_API(float3, vector) @@ -1509,11 +1338,6 @@ class VectorTransformNode : public ShaderNode { public: SHADER_NODE_CLASS(VectorTransformNode) - virtual int get_group() - { - return NODE_GROUP_LEVEL_3; - } - NODE_SOCKET_API(NodeVectorTransformType, transform_type) NODE_SOCKET_API(NodeVectorTransformConvertSpace, convert_from) NODE_SOCKET_API(NodeVectorTransformConvertSpace, convert_to) @@ -1530,7 +1354,7 @@ class BumpNode : public ShaderNode { } virtual int get_feature() { - return NODE_FEATURE_BUMP; + return KERNEL_FEATURE_NODE_BUMP; } NODE_SOCKET_API(bool, invert) @@ -1549,11 +1373,6 @@ class CurvesNode : public ShaderNode { explicit CurvesNode(const NodeType *node_type); SHADER_NODE_BASE_CLASS(CurvesNode) - virtual int get_group() - { - return NODE_GROUP_LEVEL_3; - } - NODE_SOCKET_API_ARRAY(array<float3>, curves) NODE_SOCKET_API(float, min_x) NODE_SOCKET_API(float, max_x) @@ -1583,10 +1402,6 @@ class RGBRampNode : public ShaderNode { public: SHADER_NODE_CLASS(RGBRampNode) void constant_fold(const ConstantFolder &folder); - virtual int get_group() - { - return NODE_GROUP_LEVEL_1; - } NODE_SOCKET_API_ARRAY(array<float3>, ramp) NODE_SOCKET_API_ARRAY(array<float>, ramp_alpha) @@ -1656,10 +1471,6 @@ class NormalMapNode : public ShaderNode { { return true; } - virtual int get_group() - { - return NODE_GROUP_LEVEL_3; - } NODE_SOCKET_API(NodeNormalMapSpace, space) NODE_SOCKET_API(ustring, attribute) @@ -1680,10 +1491,6 @@ class TangentNode : public ShaderNode { { return true; } - virtual int get_group() - { - return NODE_GROUP_LEVEL_3; - } NODE_SOCKET_API(NodeTangentDirectionType, direction_type) NODE_SOCKET_API(NodeTangentAxis, axis) @@ -1698,13 +1505,9 @@ class BevelNode : public ShaderNode { { return true; } - virtual int get_group() - { - return NODE_GROUP_LEVEL_3; - } - virtual bool has_raytrace() + virtual int get_feature() { - return true; + return KERNEL_FEATURE_NODE_RAYTRACE; } NODE_SOCKET_API(float, radius) @@ -1718,7 +1521,7 @@ class DisplacementNode : public ShaderNode { void constant_fold(const ConstantFolder &folder); virtual int get_feature() { - return NODE_FEATURE_BUMP; + return KERNEL_FEATURE_NODE_BUMP; } NODE_SOCKET_API(NodeNormalMapSpace, space) @@ -1739,7 +1542,7 @@ class VectorDisplacementNode : public ShaderNode { void constant_fold(const ConstantFolder &folder); virtual int get_feature() { - return NODE_FEATURE_BUMP; + return KERNEL_FEATURE_NODE_BUMP; } NODE_SOCKET_API(NodeNormalMapSpace, space) diff --git a/intern/cycles/render/object.cpp b/intern/cycles/render/object.cpp index c88d94fe4c2..4637f8fe989 100644 --- a/intern/cycles/render/object.cpp +++ b/intern/cycles/render/object.cpp @@ -216,6 +216,10 @@ void Object::tag_update(Scene *scene) if (use_holdout_is_modified()) { flag |= ObjectManager::HOLDOUT_MODIFIED; } + + if (is_shadow_catcher_is_modified()) { + scene->tag_shadow_catcher_modified(); + } } if (geometry) { @@ -273,14 +277,7 @@ bool Object::is_traceable() const uint Object::visibility_for_tracing() const { - uint trace_visibility = visibility; - if (is_shadow_catcher) { - trace_visibility &= ~PATH_RAY_SHADOW_NON_CATCHER; - } - else { - trace_visibility &= ~PATH_RAY_SHADOW_CATCHER; - } - return trace_visibility; + return SHADOW_CATCHER_OBJECT_VISIBILITY(is_shadow_catcher, visibility & PATH_RAY_ALL_VISIBILITY); } float Object::compute_volume_step_size() const @@ -680,7 +677,7 @@ void ObjectManager::device_update(Device *device, /* prepare for static BVH building */ /* todo: do before to support getting object level coords? */ - if (scene->params.bvh_type == SceneParams::BVH_STATIC) { + if (scene->params.bvh_type == BVH_TYPE_STATIC) { scoped_callback_timer timer([scene](double time) { if (scene->update_stats) { scene->update_stats->object.times.add_entry( @@ -932,6 +929,11 @@ void ObjectManager::tag_update(Scene *scene, uint32_t flag) } scene->light_manager->tag_update(scene, LightManager::OBJECT_MANAGER); + + /* Integrator's shadow catcher settings depends on object visibility settings. */ + if (flag & (OBJECT_ADDED | OBJECT_REMOVED | OBJECT_MODIFIED)) { + scene->integrator->tag_update(scene, Integrator::OBJECT_MANAGER); + } } bool ObjectManager::need_update() const diff --git a/intern/cycles/render/osl.cpp b/intern/cycles/render/osl.cpp index 7dc79f48145..d28b222c10e 100644 --- a/intern/cycles/render/osl.cpp +++ b/intern/cycles/render/osl.cpp @@ -113,7 +113,7 @@ void OSLShaderManager::device_update_specific(Device *device, scene->image_manager->set_osl_texture_system((void *)ts); /* create shaders */ - OSLGlobals *og = (OSLGlobals *)device->osl_memory(); + OSLGlobals *og = (OSLGlobals *)device->get_cpu_osl_memory(); Shader *background_shader = scene->background->get_shader(scene); foreach (Shader *shader, scene->shaders) { @@ -174,7 +174,7 @@ void OSLShaderManager::device_update_specific(Device *device, void OSLShaderManager::device_free(Device *device, DeviceScene *dscene, Scene *scene) { - OSLGlobals *og = (OSLGlobals *)device->osl_memory(); + OSLGlobals *og = (OSLGlobals *)device->get_cpu_osl_memory(); device_free_common(device, dscene, scene); @@ -257,25 +257,36 @@ void OSLShaderManager::shading_system_init() /* our own ray types */ static const char *raytypes[] = { - "camera", /* PATH_RAY_CAMERA */ - "reflection", /* PATH_RAY_REFLECT */ - "refraction", /* PATH_RAY_TRANSMIT */ - "diffuse", /* PATH_RAY_DIFFUSE */ - "glossy", /* PATH_RAY_GLOSSY */ - "singular", /* PATH_RAY_SINGULAR */ - "transparent", /* PATH_RAY_TRANSPARENT */ - - "shadow", /* PATH_RAY_SHADOW_OPAQUE_NON_CATCHER */ - "shadow", /* PATH_RAY_SHADOW_OPAQUE_CATCHER */ - "shadow", /* PATH_RAY_SHADOW_TRANSPARENT_NON_CATCHER */ - "shadow", /* PATH_RAY_SHADOW_TRANSPARENT_CATCHER */ - - "__unused__", "volume_scatter", /* PATH_RAY_VOLUME_SCATTER */ - "__unused__", - - "__unused__", "diffuse_ancestor", /* PATH_RAY_DIFFUSE_ANCESTOR */ - "__unused__", "__unused__", "__unused__", "__unused__", - "__unused__", "__unused__", "__unused__", + "camera", /* PATH_RAY_CAMERA */ + "reflection", /* PATH_RAY_REFLECT */ + "refraction", /* PATH_RAY_TRANSMIT */ + "diffuse", /* PATH_RAY_DIFFUSE */ + "glossy", /* PATH_RAY_GLOSSY */ + "singular", /* PATH_RAY_SINGULAR */ + "transparent", /* PATH_RAY_TRANSPARENT */ + "volume_scatter", /* PATH_RAY_VOLUME_SCATTER */ + + "shadow", /* PATH_RAY_SHADOW_OPAQUE */ + "shadow", /* PATH_RAY_SHADOW_TRANSPARENT */ + + "__unused__", /* PATH_RAY_NODE_UNALIGNED */ + "__unused__", /* PATH_RAY_MIS_SKIP */ + + "diffuse_ancestor", /* PATH_RAY_DIFFUSE_ANCESTOR */ + + "__unused__", /* PATH_RAY_SINGLE_PASS_DONE */ + "__unused__", /* PATH_RAY_TRANSPARENT_BACKGROUND */ + "__unused__", /* PATH_RAY_TERMINATE_IMMEDIATE */ + "__unused__", /* PATH_RAY_TERMINATE_AFTER_TRANSPARENT */ + "__unused__", /* PATH_RAY_EMISSION */ + "__unused__", /* PATH_RAY_SUBSURFACE */ + "__unused__", /* PATH_RAY_DENOISING_FEATURES */ + "__unused__", /* PATH_RAY_REFLECT_PASS */ + "__unused__", /* PATH_RAY_TRANSMISSION_PASS */ + "__unused__", /* PATH_RAY_VOLUME_PASS */ + "__unused__", /* PATH_RAY_SHADOW_FOR_LIGHT */ + "__unused__", /* PATH_RAY_SHADOW_CATCHER_HIT */ + "__unused__", /* PATH_RAY_SHADOW_CATCHER_PASS */ }; const int nraytypes = sizeof(raytypes) / sizeof(raytypes[0]); @@ -758,7 +769,8 @@ void OSLCompiler::add(ShaderNode *node, const char *name, bool isfilepath) current_shader->has_surface_bssrdf = true; current_shader->has_bssrdf_bump = true; /* can't detect yet */ } - current_shader->has_bump = true; /* can't detect yet */ + current_shader->has_bump = true; /* can't detect yet */ + current_shader->has_surface_raytrace = true; /* can't detect yet */ } if (node->has_spatial_varying()) { @@ -1054,6 +1066,8 @@ void OSLCompiler::generate_nodes(const ShaderNodeSet &nodes) current_shader->has_surface_emission = true; if (node->has_surface_transparent()) current_shader->has_surface_transparent = true; + if (node->get_feature() & KERNEL_FEATURE_NODE_RAYTRACE) + current_shader->has_surface_raytrace = true; if (node->has_spatial_varying()) current_shader->has_surface_spatial_varying = true; if (node->has_surface_bssrdf()) { diff --git a/intern/cycles/render/pass.cpp b/intern/cycles/render/pass.cpp new file mode 100644 index 00000000000..27ad7c0db97 --- /dev/null +++ b/intern/cycles/render/pass.cpp @@ -0,0 +1,427 @@ +/* + * Copyright 2011-2021 Blender Foundation + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include "render/pass.h" + +#include "util/util_algorithm.h" +#include "util/util_logging.h" + +CCL_NAMESPACE_BEGIN + +const char *pass_type_as_string(const PassType type) +{ + const int type_int = static_cast<int>(type); + + const NodeEnum *type_enum = Pass::get_type_enum(); + + if (!type_enum->exists(type_int)) { + LOG(DFATAL) << "Unhandled pass type " << static_cast<int>(type) << ", not supposed to happen."; + return "UNKNOWN"; + } + + return (*type_enum)[type_int].c_str(); +} + +const char *pass_mode_as_string(PassMode mode) +{ + switch (mode) { + case PassMode::NOISY: + return "NOISY"; + case PassMode::DENOISED: + return "DENOISED"; + } + + LOG(DFATAL) << "Unhandled pass mode " << static_cast<int>(mode) << ", should never happen."; + return "UNKNOWN"; +} + +std::ostream &operator<<(std::ostream &os, PassMode mode) +{ + os << pass_mode_as_string(mode); + return os; +} + +const NodeEnum *Pass::get_type_enum() +{ + static NodeEnum pass_type_enum; + + if (pass_type_enum.empty()) { + + /* Light Passes. */ + pass_type_enum.insert("combined", PASS_COMBINED); + pass_type_enum.insert("emission", PASS_EMISSION); + pass_type_enum.insert("background", PASS_BACKGROUND); + pass_type_enum.insert("ao", PASS_AO); + pass_type_enum.insert("shadow", PASS_SHADOW); + pass_type_enum.insert("diffuse", PASS_DIFFUSE); + pass_type_enum.insert("diffuse_direct", PASS_DIFFUSE_DIRECT); + pass_type_enum.insert("diffuse_indirect", PASS_DIFFUSE_INDIRECT); + pass_type_enum.insert("glossy", PASS_GLOSSY); + pass_type_enum.insert("glossy_direct", PASS_GLOSSY_DIRECT); + pass_type_enum.insert("glossy_indirect", PASS_GLOSSY_INDIRECT); + pass_type_enum.insert("transmission", PASS_TRANSMISSION); + pass_type_enum.insert("transmission_direct", PASS_TRANSMISSION_DIRECT); + pass_type_enum.insert("transmission_indirect", PASS_TRANSMISSION_INDIRECT); + pass_type_enum.insert("volume", PASS_VOLUME); + pass_type_enum.insert("volume_direct", PASS_VOLUME_DIRECT); + pass_type_enum.insert("volume_indirect", PASS_VOLUME_INDIRECT); + + /* Data passes. */ + pass_type_enum.insert("depth", PASS_DEPTH); + pass_type_enum.insert("position", PASS_POSITION); + pass_type_enum.insert("normal", PASS_NORMAL); + pass_type_enum.insert("roughness", PASS_ROUGHNESS); + pass_type_enum.insert("uv", PASS_UV); + pass_type_enum.insert("object_id", PASS_OBJECT_ID); + pass_type_enum.insert("material_id", PASS_MATERIAL_ID); + pass_type_enum.insert("motion", PASS_MOTION); + pass_type_enum.insert("motion_weight", PASS_MOTION_WEIGHT); + pass_type_enum.insert("render_time", PASS_RENDER_TIME); + pass_type_enum.insert("cryptomatte", PASS_CRYPTOMATTE); + pass_type_enum.insert("aov_color", PASS_AOV_COLOR); + pass_type_enum.insert("aov_value", PASS_AOV_VALUE); + pass_type_enum.insert("adaptive_aux_buffer", PASS_ADAPTIVE_AUX_BUFFER); + pass_type_enum.insert("sample_count", PASS_SAMPLE_COUNT); + pass_type_enum.insert("diffuse_color", PASS_DIFFUSE_COLOR); + pass_type_enum.insert("glossy_color", PASS_GLOSSY_COLOR); + pass_type_enum.insert("transmission_color", PASS_TRANSMISSION_COLOR); + pass_type_enum.insert("mist", PASS_MIST); + pass_type_enum.insert("denoising_normal", PASS_DENOISING_NORMAL); + pass_type_enum.insert("denoising_albedo", PASS_DENOISING_ALBEDO); + + pass_type_enum.insert("shadow_catcher", PASS_SHADOW_CATCHER); + pass_type_enum.insert("shadow_catcher_sample_count", PASS_SHADOW_CATCHER_SAMPLE_COUNT); + pass_type_enum.insert("shadow_catcher_matte", PASS_SHADOW_CATCHER_MATTE); + + pass_type_enum.insert("bake_primitive", PASS_BAKE_PRIMITIVE); + pass_type_enum.insert("bake_differential", PASS_BAKE_DIFFERENTIAL); + } + + return &pass_type_enum; +} + +const NodeEnum *Pass::get_mode_enum() +{ + static NodeEnum pass_mode_enum; + + if (pass_mode_enum.empty()) { + pass_mode_enum.insert("noisy", static_cast<int>(PassMode::NOISY)); + pass_mode_enum.insert("denoised", static_cast<int>(PassMode::DENOISED)); + } + + return &pass_mode_enum; +} + +NODE_DEFINE(Pass) +{ + NodeType *type = NodeType::add("pass", create); + + const NodeEnum *pass_type_enum = get_type_enum(); + const NodeEnum *pass_mode_enum = get_mode_enum(); + + SOCKET_ENUM(type, "Type", *pass_type_enum, PASS_COMBINED); + SOCKET_ENUM(mode, "Mode", *pass_mode_enum, static_cast<int>(PassMode::DENOISED)); + SOCKET_STRING(name, "Name", ustring()); + SOCKET_BOOLEAN(include_albedo, "Include Albedo", false); + + return type; +} + +Pass::Pass() : Node(get_node_type()), is_auto_(false) +{ +} + +PassInfo Pass::get_info() const +{ + return get_info(type, include_albedo); +} + +bool Pass::is_written() const +{ + return get_info().is_written; +} + +PassInfo Pass::get_info(const PassType type, const bool include_albedo) +{ + PassInfo pass_info; + + pass_info.use_filter = true; + pass_info.use_exposure = false; + pass_info.divide_type = PASS_NONE; + pass_info.use_compositing = false; + pass_info.use_denoising_albedo = true; + + switch (type) { + case PASS_NONE: + pass_info.num_components = 0; + break; + case PASS_COMBINED: + pass_info.num_components = 4; + pass_info.use_exposure = true; + pass_info.support_denoise = true; + break; + case PASS_DEPTH: + pass_info.num_components = 1; + pass_info.use_filter = false; + break; + case PASS_MIST: + pass_info.num_components = 1; + break; + case PASS_POSITION: + pass_info.num_components = 3; + break; + case PASS_NORMAL: + pass_info.num_components = 3; + break; + case PASS_ROUGHNESS: + pass_info.num_components = 1; + break; + case PASS_UV: + pass_info.num_components = 3; + break; + case PASS_MOTION: + pass_info.num_components = 4; + pass_info.divide_type = PASS_MOTION_WEIGHT; + break; + case PASS_MOTION_WEIGHT: + pass_info.num_components = 1; + break; + case PASS_OBJECT_ID: + case PASS_MATERIAL_ID: + pass_info.num_components = 1; + pass_info.use_filter = false; + break; + + case PASS_EMISSION: + case PASS_BACKGROUND: + pass_info.num_components = 3; + pass_info.use_exposure = true; + break; + case PASS_AO: + pass_info.num_components = 3; + break; + case PASS_SHADOW: + pass_info.num_components = 3; + pass_info.use_exposure = false; + break; + case PASS_RENDER_TIME: + /* This pass is handled entirely on the host side. */ + pass_info.num_components = 0; + break; + + case PASS_DIFFUSE_COLOR: + case PASS_GLOSSY_COLOR: + case PASS_TRANSMISSION_COLOR: + pass_info.num_components = 3; + break; + case PASS_DIFFUSE: + pass_info.num_components = 3; + pass_info.use_exposure = true; + pass_info.direct_type = PASS_DIFFUSE_DIRECT; + pass_info.indirect_type = PASS_DIFFUSE_INDIRECT; + pass_info.divide_type = (!include_albedo) ? PASS_DIFFUSE_COLOR : PASS_NONE; + pass_info.use_compositing = true; + pass_info.is_written = false; + break; + case PASS_DIFFUSE_DIRECT: + case PASS_DIFFUSE_INDIRECT: + pass_info.num_components = 3; + pass_info.use_exposure = true; + pass_info.divide_type = (!include_albedo) ? PASS_DIFFUSE_COLOR : PASS_NONE; + pass_info.use_compositing = true; + break; + case PASS_GLOSSY: + pass_info.num_components = 3; + pass_info.use_exposure = true; + pass_info.direct_type = PASS_GLOSSY_DIRECT; + pass_info.indirect_type = PASS_GLOSSY_INDIRECT; + pass_info.divide_type = (!include_albedo) ? PASS_GLOSSY_COLOR : PASS_NONE; + pass_info.use_compositing = true; + pass_info.is_written = false; + break; + case PASS_GLOSSY_DIRECT: + case PASS_GLOSSY_INDIRECT: + pass_info.num_components = 3; + pass_info.use_exposure = true; + pass_info.divide_type = (!include_albedo) ? PASS_GLOSSY_COLOR : PASS_NONE; + pass_info.use_compositing = true; + break; + case PASS_TRANSMISSION: + pass_info.num_components = 3; + pass_info.use_exposure = true; + pass_info.direct_type = PASS_TRANSMISSION_DIRECT; + pass_info.indirect_type = PASS_TRANSMISSION_INDIRECT; + pass_info.divide_type = (!include_albedo) ? PASS_TRANSMISSION_COLOR : PASS_NONE; + pass_info.use_compositing = true; + pass_info.is_written = false; + break; + case PASS_TRANSMISSION_DIRECT: + case PASS_TRANSMISSION_INDIRECT: + pass_info.num_components = 3; + pass_info.use_exposure = true; + pass_info.divide_type = (!include_albedo) ? PASS_TRANSMISSION_COLOR : PASS_NONE; + pass_info.use_compositing = true; + break; + case PASS_VOLUME: + pass_info.num_components = 3; + pass_info.use_exposure = true; + pass_info.direct_type = PASS_VOLUME_DIRECT; + pass_info.indirect_type = PASS_VOLUME_INDIRECT; + pass_info.use_compositing = true; + pass_info.is_written = false; + break; + case PASS_VOLUME_DIRECT: + case PASS_VOLUME_INDIRECT: + pass_info.num_components = 3; + pass_info.use_exposure = true; + break; + + case PASS_CRYPTOMATTE: + pass_info.num_components = 4; + break; + + case PASS_DENOISING_NORMAL: + pass_info.num_components = 3; + break; + case PASS_DENOISING_ALBEDO: + pass_info.num_components = 3; + break; + + case PASS_SHADOW_CATCHER: + pass_info.num_components = 3; + pass_info.use_exposure = true; + pass_info.use_compositing = true; + pass_info.use_denoising_albedo = false; + pass_info.support_denoise = true; + break; + case PASS_SHADOW_CATCHER_SAMPLE_COUNT: + pass_info.num_components = 1; + break; + case PASS_SHADOW_CATCHER_MATTE: + pass_info.num_components = 4; + pass_info.use_exposure = true; + pass_info.support_denoise = true; + /* Without shadow catcher approximation compositing is not needed. + * Since we don't know here whether approximation is used or not, leave the decision up to + * the caller which will know that. */ + break; + + case PASS_ADAPTIVE_AUX_BUFFER: + pass_info.num_components = 4; + break; + case PASS_SAMPLE_COUNT: + pass_info.num_components = 1; + pass_info.use_exposure = false; + break; + + case PASS_AOV_COLOR: + pass_info.num_components = 3; + break; + case PASS_AOV_VALUE: + pass_info.num_components = 1; + break; + + case PASS_BAKE_PRIMITIVE: + case PASS_BAKE_DIFFERENTIAL: + pass_info.num_components = 4; + pass_info.use_exposure = false; + pass_info.use_filter = false; + break; + + case PASS_CATEGORY_LIGHT_END: + case PASS_CATEGORY_DATA_END: + case PASS_CATEGORY_BAKE_END: + case PASS_NUM: + LOG(DFATAL) << "Unexpected pass type is used " << type; + pass_info.num_components = 0; + break; + } + + return pass_info; +} + +bool Pass::contains(const vector<Pass *> &passes, PassType type) +{ + for (const Pass *pass : passes) { + if (pass->get_type() != type) { + continue; + } + + return true; + } + + return false; +} + +const Pass *Pass::find(const vector<Pass *> &passes, const string &name) +{ + for (const Pass *pass : passes) { + if (pass->get_name() == name) { + return pass; + } + } + + return nullptr; +} + +const Pass *Pass::find(const vector<Pass *> &passes, PassType type, PassMode mode) +{ + for (const Pass *pass : passes) { + if (pass->get_type() != type || pass->get_mode() != mode) { + continue; + } + + return pass; + } + + return nullptr; +} + +int Pass::get_offset(const vector<Pass *> &passes, const Pass *pass) +{ + int pass_offset = 0; + + for (const Pass *current_pass : passes) { + /* Note that pass name is allowed to be empty. This is why we check for type and mode. */ + if (current_pass->get_type() == pass->get_type() && + current_pass->get_mode() == pass->get_mode() && + current_pass->get_name() == pass->get_name()) { + if (current_pass->is_written()) { + return pass_offset; + } + else { + return PASS_UNUSED; + } + } + if (current_pass->is_written()) { + pass_offset += current_pass->get_info().num_components; + } + } + + return PASS_UNUSED; +} + +std::ostream &operator<<(std::ostream &os, const Pass &pass) +{ + os << "type: " << pass_type_as_string(pass.get_type()); + os << ", name: \"" << pass.get_name() << "\""; + os << ", mode: " << pass.get_mode(); + os << ", is_written: " << string_from_bool(pass.is_written()); + + return os; +} + +CCL_NAMESPACE_END diff --git a/intern/cycles/render/pass.h b/intern/cycles/render/pass.h new file mode 100644 index 00000000000..82230c62cb0 --- /dev/null +++ b/intern/cycles/render/pass.h @@ -0,0 +1,106 @@ +/* + * Copyright 2011-2021 Blender Foundation + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#pragma once + +#include <ostream> // NOLINT + +#include "util/util_string.h" +#include "util/util_vector.h" + +#include "kernel/kernel_types.h" + +#include "graph/node.h" + +CCL_NAMESPACE_BEGIN + +const char *pass_type_as_string(const PassType type); + +enum class PassMode { + NOISY, + DENOISED, +}; +const char *pass_mode_as_string(PassMode mode); +std::ostream &operator<<(std::ostream &os, PassMode mode); + +struct PassInfo { + int num_components = -1; + bool use_filter = false; + bool use_exposure = false; + bool is_written = true; + PassType divide_type = PASS_NONE; + PassType direct_type = PASS_NONE; + PassType indirect_type = PASS_NONE; + + /* Pass access for read can not happen directly and needs some sort of compositing (for example, + * light passes due to divide_type, or shadow catcher pass. */ + bool use_compositing = false; + + /* Used to disable albedo pass for denoising. + * Light and shadow catcher passes should not have discontinuity in the denoised result based on + * the underlying albedo. */ + bool use_denoising_albedo = true; + + /* Pass supports denoising. */ + bool support_denoise = false; +}; + +class Pass : public Node { + public: + NODE_DECLARE + + NODE_SOCKET_API(PassType, type) + NODE_SOCKET_API(PassMode, mode) + NODE_SOCKET_API(ustring, name) + NODE_SOCKET_API(bool, include_albedo) + + Pass(); + + PassInfo get_info() const; + + /* The pass is written by the render pipeline (kernel or denoiser). If the pass is written it + * will have pixels allocated in a RenderBuffer. Passes which are not written do not have their + * pixels allocated to save memory. */ + bool is_written() const; + + protected: + /* The has been created automatically as a requirement to various rendering functionality (such + * as adaptive sampling). */ + bool is_auto_; + + public: + static const NodeEnum *get_type_enum(); + static const NodeEnum *get_mode_enum(); + + static PassInfo get_info(PassType type, const bool include_albedo = false); + + static bool contains(const vector<Pass *> &passes, PassType type); + + /* Returns nullptr if there is no pass with the given name or type+mode. */ + static const Pass *find(const vector<Pass *> &passes, const string &name); + static const Pass *find(const vector<Pass *> &passes, + PassType type, + PassMode mode = PassMode::NOISY); + + /* Returns PASS_UNUSED if there is no corresponding pass. */ + static int get_offset(const vector<Pass *> &passes, const Pass *pass); + + friend class Film; +}; + +std::ostream &operator<<(std::ostream &os, const Pass &pass); + +CCL_NAMESPACE_END diff --git a/intern/cycles/render/scene.cpp b/intern/cycles/render/scene.cpp index c4e7d2c79d6..a4b030190dc 100644 --- a/intern/cycles/render/scene.cpp +++ b/intern/cycles/render/scene.cpp @@ -163,12 +163,15 @@ void Scene::free_memory(bool final) delete p; foreach (Light *l, lights) delete l; + foreach (Pass *p, passes) + delete p; geometry.clear(); objects.clear(); lights.clear(); particle_systems.clear(); procedurals.clear(); + passes.clear(); if (device) { camera->device_free(device, &dscene, this); @@ -253,7 +256,6 @@ void Scene::device_update(Device *device_, Progress &progress) * - Camera may be used for adaptive subdivision. * - Displacement shader must have all shader data available. * - Light manager needs lookup tables and final mesh data to compute emission CDF. - * - Film needs light manager to run for use_light_visibility * - Lookup tables are done a second time to handle film tables */ @@ -469,88 +471,110 @@ void Scene::enable_update_stats() } } -DeviceRequestedFeatures Scene::get_requested_device_features() +void Scene::update_kernel_features() { - DeviceRequestedFeatures requested_features; + if (!need_update()) { + return; + } - shader_manager->get_requested_features(this, &requested_features); + /* These features are not being tweaked as often as shaders, + * so could be done selective magic for the viewport as well. */ + uint kernel_features = shader_manager->get_kernel_features(this); - /* This features are not being tweaked as often as shaders, - * so could be done selective magic for the viewport as well. - */ bool use_motion = need_motion() == Scene::MotionType::MOTION_BLUR; - requested_features.use_hair = false; - requested_features.use_hair_thick = (params.hair_shape == CURVE_THICK); - requested_features.use_object_motion = false; - requested_features.use_camera_motion = use_motion && camera->use_motion(); + kernel_features |= KERNEL_FEATURE_PATH_TRACING; + if (params.hair_shape == CURVE_THICK) { + kernel_features |= KERNEL_FEATURE_HAIR_THICK; + } + if (use_motion && camera->use_motion()) { + kernel_features |= KERNEL_FEATURE_CAMERA_MOTION; + } foreach (Object *object, objects) { Geometry *geom = object->get_geometry(); if (use_motion) { - requested_features.use_object_motion |= object->use_motion() | geom->get_use_motion_blur(); - requested_features.use_camera_motion |= geom->get_use_motion_blur(); + if (object->use_motion() || geom->get_use_motion_blur()) { + kernel_features |= KERNEL_FEATURE_OBJECT_MOTION; + } + if (geom->get_use_motion_blur()) { + kernel_features |= KERNEL_FEATURE_CAMERA_MOTION; + } } if (object->get_is_shadow_catcher()) { - requested_features.use_shadow_tricks = true; + kernel_features |= KERNEL_FEATURE_SHADOW_CATCHER; } if (geom->is_mesh()) { Mesh *mesh = static_cast<Mesh *>(geom); #ifdef WITH_OPENSUBDIV if (mesh->get_subdivision_type() != Mesh::SUBDIVISION_NONE) { - requested_features.use_patch_evaluation = true; + kernel_features |= KERNEL_FEATURE_PATCH_EVALUATION; } #endif - requested_features.use_true_displacement |= mesh->has_true_displacement(); } else if (geom->is_hair()) { - requested_features.use_hair = true; + kernel_features |= KERNEL_FEATURE_HAIR; } } - requested_features.use_background_light = light_manager->has_background_light(this); - - requested_features.use_baking = bake_manager->get_baking(); - requested_features.use_integrator_branched = (integrator->get_method() == - Integrator::BRANCHED_PATH); - if (film->get_denoising_data_pass()) { - requested_features.use_denoising = true; - requested_features.use_shadow_tricks = true; + if (bake_manager->get_baking()) { + kernel_features |= KERNEL_FEATURE_BAKING; } - return requested_features; -} + kernel_features |= film->get_kernel_features(this); -bool Scene::update(Progress &progress, bool &kernel_switch_needed) -{ - /* update scene */ - if (need_update()) { - /* Update max_closures. */ - KernelIntegrator *kintegrator = &dscene.data.integrator; - if (params.background) { - kintegrator->max_closures = get_max_closure_count(); - } - else { - /* Currently viewport render is faster with higher max_closures, needs investigating. */ - kintegrator->max_closures = MAX_CLOSURE; - } - - /* Load render kernels, before device update where we upload data to the GPU. */ - bool new_kernels_needed = load_kernels(progress, false); - - progress.set_status("Updating Scene"); - MEM_GUARDED_CALL(&progress, device_update, device, progress); + dscene.data.kernel_features = kernel_features; - DeviceKernelStatus kernel_switch_status = device->get_active_kernel_switch_state(); - kernel_switch_needed = kernel_switch_status == DEVICE_KERNEL_FEATURE_KERNEL_AVAILABLE || - kernel_switch_status == DEVICE_KERNEL_FEATURE_KERNEL_INVALID; - if (new_kernels_needed || kernel_switch_needed) { - progress.set_kernel_status("Compiling render kernels"); - device->wait_for_availability(loaded_kernel_features); - progress.set_kernel_status(""); - } + /* Currently viewport render is faster with higher max_closures, needs investigating. */ + const uint max_closures = (params.background) ? get_max_closure_count() : MAX_CLOSURE; + dscene.data.max_closures = max_closures; + dscene.data.max_shaders = shaders.size(); +} - return true; +bool Scene::update(Progress &progress) +{ + if (!need_update()) { + return false; } - return false; + + /* Load render kernels, before device update where we upload data to the GPU. */ + load_kernels(progress, false); + + /* Upload scene data to the GPU. */ + progress.set_status("Updating Scene"); + MEM_GUARDED_CALL(&progress, device_update, device, progress); + + return true; +} + +static void log_kernel_features(const uint features) +{ + VLOG(2) << "Requested features:\n"; + VLOG(2) << "Use BSDF " << string_from_bool(features & KERNEL_FEATURE_NODE_BSDF) << "\n"; + VLOG(2) << "Use Principled BSDF " << string_from_bool(features & KERNEL_FEATURE_PRINCIPLED) + << "\n"; + VLOG(2) << "Use Emission " << string_from_bool(features & KERNEL_FEATURE_NODE_EMISSION) << "\n"; + VLOG(2) << "Use Volume " << string_from_bool(features & KERNEL_FEATURE_NODE_VOLUME) << "\n"; + VLOG(2) << "Use Hair " << string_from_bool(features & KERNEL_FEATURE_NODE_HAIR) << "\n"; + VLOG(2) << "Use Bump " << string_from_bool(features & KERNEL_FEATURE_NODE_BUMP) << "\n"; + VLOG(2) << "Use Voronoi " << string_from_bool(features & KERNEL_FEATURE_NODE_VORONOI_EXTRA) + << "\n"; + VLOG(2) << "Use Shader Raytrace " << string_from_bool(features & KERNEL_FEATURE_NODE_RAYTRACE) + << "\n"; + VLOG(2) << "Use Transparent " << string_from_bool(features & KERNEL_FEATURE_TRANSPARENT) << "\n"; + VLOG(2) << "Use Denoising " << string_from_bool(features & KERNEL_FEATURE_DENOISING) << "\n"; + VLOG(2) << "Use Path Tracing " << string_from_bool(features & KERNEL_FEATURE_PATH_TRACING) + << "\n"; + VLOG(2) << "Use Hair " << string_from_bool(features & KERNEL_FEATURE_HAIR) << "\n"; + VLOG(2) << "Use Object Motion " << string_from_bool(features & KERNEL_FEATURE_OBJECT_MOTION) + << "\n"; + VLOG(2) << "Use Camera Motion " << string_from_bool(features & KERNEL_FEATURE_CAMERA_MOTION) + << "\n"; + VLOG(2) << "Use Baking " << string_from_bool(features & KERNEL_FEATURE_BAKING) << "\n"; + VLOG(2) << "Use Subsurface " << string_from_bool(features & KERNEL_FEATURE_SUBSURFACE) << "\n"; + VLOG(2) << "Use Volume " << string_from_bool(features & KERNEL_FEATURE_VOLUME) << "\n"; + VLOG(2) << "Use Patch Evaluation " + << string_from_bool(features & KERNEL_FEATURE_PATCH_EVALUATION) << "\n"; + VLOG(2) << "Use Shadow Catcher " << string_from_bool(features & KERNEL_FEATURE_SHADOW_CATCHER) + << "\n"; } bool Scene::load_kernels(Progress &progress, bool lock_scene) @@ -560,15 +584,15 @@ bool Scene::load_kernels(Progress &progress, bool lock_scene) scene_lock = thread_scoped_lock(mutex); } - DeviceRequestedFeatures requested_features = get_requested_device_features(); + const uint kernel_features = dscene.data.kernel_features; - if (!kernels_loaded || loaded_kernel_features.modified(requested_features)) { + if (!kernels_loaded || loaded_kernel_features != kernel_features) { progress.set_status("Loading render kernels (may take a few minutes the first time)"); scoped_timer timer; - VLOG(2) << "Requested features:\n" << requested_features; - if (!device->load_kernels(requested_features)) { + log_kernel_features(kernel_features); + if (!device->load_kernels(kernel_features)) { string message = device->error_message(); if (message.empty()) message = "Failed loading render kernel, see console for errors"; @@ -580,7 +604,7 @@ bool Scene::load_kernels(Progress &progress, bool lock_scene) } kernels_loaded = true; - loaded_kernel_features = requested_features; + loaded_kernel_features = kernel_features; return true; } return false; @@ -618,6 +642,28 @@ int Scene::get_max_closure_count() return max_closure_global; } +bool Scene::has_shadow_catcher() +{ + if (shadow_catcher_modified_) { + has_shadow_catcher_ = false; + for (Object *object : objects) { + if (object->get_is_shadow_catcher()) { + has_shadow_catcher_ = true; + break; + } + } + + shadow_catcher_modified_ = false; + } + + return has_shadow_catcher_; +} + +void Scene::tag_shadow_catcher_modified() +{ + shadow_catcher_modified_ = true; +} + template<> Light *Scene::create_node<Light>() { Light *node = new Light(); @@ -694,6 +740,15 @@ template<> AlembicProcedural *Scene::create_node<AlembicProcedural>() #endif } +template<> Pass *Scene::create_node<Pass>() +{ + Pass *node = new Pass(); + node->set_owner(this); + passes.push_back(node); + film->tag_modified(); + return node; +} + template<typename T> void delete_node_from_array(vector<T> &nodes, T node) { for (size_t i = 0; i < nodes.size(); ++i) { @@ -779,6 +834,12 @@ template<> void Scene::delete_node_impl(AlembicProcedural *node) #endif } +template<> void Scene::delete_node_impl(Pass *node) +{ + delete_node_from_array(passes, node); + film->tag_modified(); +} + template<typename T> static void remove_nodes_in_set(const set<T *> &nodes_set, vector<T *> &nodes_array, @@ -842,4 +903,10 @@ template<> void Scene::delete_nodes(const set<Procedural *> &nodes, const NodeOw procedural_manager->tag_update(); } +template<> void Scene::delete_nodes(const set<Pass *> &nodes, const NodeOwner *owner) +{ + remove_nodes_in_set(nodes, passes, owner); + film->tag_modified(); +} + CCL_NAMESPACE_END diff --git a/intern/cycles/render/scene.h b/intern/cycles/render/scene.h index 7d8a6774381..cf4a3ba6b12 100644 --- a/intern/cycles/render/scene.h +++ b/intern/cycles/render/scene.h @@ -128,7 +128,7 @@ class DeviceScene { device_vector<float> lookup_table; /* integrator */ - device_vector<uint> sample_pattern_lut; + device_vector<float> sample_pattern_lut; /* ies lights */ device_vector<float> ies_lights; @@ -142,27 +142,6 @@ class DeviceScene { class SceneParams { public: - /* Type of BVH, in terms whether it is supported dynamic updates of meshes - * or whether modifying geometry requires full BVH rebuild. - */ - enum BVHType { - /* BVH supports dynamic updates of geometry. - * - * Faster for updating BVH tree when doing modifications in viewport, - * but slower for rendering. - */ - BVH_DYNAMIC = 0, - /* BVH tree is calculated for specific scene, updates in geometry - * requires full tree rebuild. - * - * Slower to update BVH tree when modifying objects in viewport, also - * slower to build final BVH tree but gives best possible render speed. - */ - BVH_STATIC = 1, - - BVH_NUM_TYPES, - }; - ShadingSystem shadingsystem; /* Requested BVH layout. @@ -186,7 +165,7 @@ class SceneParams { { shadingsystem = SHADINGSYSTEM_SVM; bvh_layout = BVH_LAYOUT_BVH2; - bvh_type = BVH_DYNAMIC; + bvh_type = BVH_TYPE_DYNAMIC; use_bvh_spatial_split = false; use_bvh_unaligned_nodes = true; num_bvh_time_steps = 0; @@ -196,7 +175,7 @@ class SceneParams { background = true; } - bool modified(const SceneParams ¶ms) + bool modified(const SceneParams ¶ms) const { return !(shadingsystem == params.shadingsystem && bvh_layout == params.bvh_layout && bvh_type == params.bvh_type && @@ -236,7 +215,7 @@ class Scene : public NodeOwner { vector<Shader *> shaders; vector<Light *> lights; vector<ParticleSystem *> particle_systems; - vector<Pass> passes; + vector<Pass *> passes; vector<Procedural *> procedurals; /* data managers */ @@ -291,7 +270,11 @@ class Scene : public NodeOwner { void enable_update_stats(); - bool update(Progress &progress, bool &kernel_switch_needed); + void update_kernel_features(); + bool update(Progress &progress); + + bool has_shadow_catcher(); + void tag_shadow_catcher_modified(); /* This function is used to create a node of a specified type instead of * calling 'new', and sets the scene as the owner of the node. @@ -348,13 +331,12 @@ class Scene : public NodeOwner { void free_memory(bool final); bool kernels_loaded; - DeviceRequestedFeatures loaded_kernel_features; + uint loaded_kernel_features; bool load_kernels(Progress &progress, bool lock_scene = true); - /* ** Split kernel routines ** */ - - DeviceRequestedFeatures get_requested_device_features(); + bool has_shadow_catcher_ = false; + bool shadow_catcher_modified_ = true; /* Maximum number of closure during session lifetime. */ int max_closure_global; @@ -384,6 +366,8 @@ template<> Shader *Scene::create_node<Shader>(); template<> AlembicProcedural *Scene::create_node<AlembicProcedural>(); +template<> Pass *Scene::create_node<Pass>(); + template<> void Scene::delete_node_impl(Light *node); template<> void Scene::delete_node_impl(Mesh *node); @@ -404,6 +388,8 @@ template<> void Scene::delete_node_impl(Procedural *node); template<> void Scene::delete_node_impl(AlembicProcedural *node); +template<> void Scene::delete_node_impl(Pass *node); + template<> void Scene::delete_nodes(const set<Light *> &nodes, const NodeOwner *owner); template<> void Scene::delete_nodes(const set<Geometry *> &nodes, const NodeOwner *owner); @@ -416,6 +402,8 @@ template<> void Scene::delete_nodes(const set<Shader *> &nodes, const NodeOwner template<> void Scene::delete_nodes(const set<Procedural *> &nodes, const NodeOwner *owner); +template<> void Scene::delete_nodes(const set<Pass *> &nodes, const NodeOwner *owner); + CCL_NAMESPACE_END #endif /* __SCENE_H__ */ diff --git a/intern/cycles/render/session.cpp b/intern/cycles/render/session.cpp index 1b91c49f0ea..84407f8e6dd 100644 --- a/intern/cycles/render/session.cpp +++ b/intern/cycles/render/session.cpp @@ -17,10 +17,15 @@ #include <limits.h> #include <string.h> +#include "device/cpu/device.h" #include "device/device.h" +#include "integrator/pass_accessor_cpu.h" +#include "integrator/path_trace.h" +#include "render/background.h" #include "render/bake.h" #include "render/buffers.h" #include "render/camera.h" +#include "render/gpu_display.h" #include "render/graph.h" #include "render/integrator.h" #include "render/light.h" @@ -39,70 +44,63 @@ CCL_NAMESPACE_BEGIN -/* Note about preserve_tile_device option for tile manager: - * progressive refine and viewport rendering does requires tiles to - * always be allocated for the same device - */ -Session::Session(const SessionParams ¶ms_) - : params(params_), - tile_manager(params.progressive, - params.samples, - params.tile_size, - params.start_resolution, - params.background == false || params.progressive_refine, - params.background, - params.tile_order, - max(params.device.multi_devices.size(), 1), - params.pixel_size), - stats(), - profiler() +Session::Session(const SessionParams ¶ms_, const SceneParams &scene_params) + : params(params_), render_scheduler_(tile_manager_, params) { - device_use_gl_ = ((params.device.type != DEVICE_CPU) && !params.background); - TaskScheduler::init(params.threads); - session_thread_ = NULL; - scene = NULL; - - reset_time_ = 0.0; - last_update_time_ = 0.0; + session_thread_ = nullptr; delayed_reset_.do_reset = false; - delayed_reset_.samples = 0; - - display_outdated_ = false; - gpu_draw_ready_ = false; - gpu_need_display_buffer_update_ = false; pause_ = false; cancel_ = false; new_work_added_ = false; - buffers = NULL; - display = NULL; + device = Device::create(params.device, stats, profiler); - /* Validate denoising parameters. */ - set_denoising(params.denoising); + scene = new Scene(scene_params, device); - /* Create CPU/GPU devices. */ - device = Device::create(params.device, stats, profiler, params.background); - - if (!device->error_message().empty()) { - progress.set_error(device->error_message()); - return; - } + /* Configure path tracer. */ + path_trace_ = make_unique<PathTrace>( + device, scene->film, &scene->dscene, render_scheduler_, tile_manager_); + path_trace_->set_progress(&progress); + path_trace_->tile_buffer_update_cb = [&]() { + if (!update_render_tile_cb) { + return; + } + update_render_tile_cb(); + }; + path_trace_->tile_buffer_write_cb = [&]() { + if (!write_render_tile_cb) { + return; + } + write_render_tile_cb(); + }; + path_trace_->tile_buffer_read_cb = [&]() -> bool { + if (!read_render_tile_cb) { + return false; + } + read_render_tile_cb(); + return true; + }; + path_trace_->progress_update_cb = [&]() { update_status_time(); }; - /* Create buffers for interactive rendering. */ - if (!(params.background && !params.write_render_cb)) { - buffers = new RenderBuffers(device); - display = new DisplayBuffer(device, params.display_buffer_linear); - } + tile_manager_.full_buffer_written_cb = [&](string_view filename) { + if (!full_buffer_written_cb) { + return; + } + full_buffer_written_cb(filename); + }; } Session::~Session() { cancel(); + /* TODO(sergey): Bring the passes in viewport back. + * It is unclear why there is such an exception needed though. */ +#if 0 if (buffers && params.write_render_cb) { /* Copy to display buffer and write out image if requested */ delete display; @@ -116,12 +114,14 @@ Session::~Session() uchar4 *pixels = display->rgba_byte.copy_from_device(0, w, h); params.write_render_cb((uchar *)pixels, w, h, 4); } +#endif - /* clean up */ - tile_manager.device_free(); + /* Make sure path tracer is destroyed before the deviec. This is needed because destruction might + * need to access device for device memory free. */ + /* TODO(sergey): Convert device to be unique_ptr, and rely on C++ to destruct objects in the + * pre-defined order. */ + path_trace_.reset(); - delete buffers; - delete display; delete scene; delete device; @@ -135,15 +135,16 @@ void Session::start() } } -void Session::cancel() +void Session::cancel(bool quick) { + if (quick && path_trace_) { + path_trace_->cancel(); + } + if (session_thread_) { /* wait for session thread to end */ progress.set_cancel("Exiting"); - gpu_need_display_buffer_update_ = false; - gpu_need_display_buffer_update_cond_.notify_all(); - { thread_scoped_lock pause_lock(pause_mutex_); pause_ = false; @@ -157,570 +158,43 @@ void Session::cancel() bool Session::ready_to_reset() { - double dt = time_dt() - reset_time_; - - if (!display_outdated_) - return (dt > params.reset_timeout); - else - return (dt > params.cancel_timeout); + return path_trace_->ready_to_reset(); } -/* GPU Session */ - -void Session::reset_gpu(BufferParams &buffer_params, int samples) +void Session::run_main_render_loop() { - thread_scoped_lock pause_lock(pause_mutex_); - - /* block for buffer access and reset immediately. we can't do this - * in the thread, because we need to allocate an OpenGL buffer, and - * that only works in the main thread */ - thread_scoped_lock display_lock(display_mutex_); - thread_scoped_lock buffers_lock(buffers_mutex_); + path_trace_->clear_gpu_display(); - display_outdated_ = true; - reset_time_ = time_dt(); + while (true) { + RenderWork render_work = run_update_for_next_iteration(); - reset_(buffer_params, samples); - - gpu_need_display_buffer_update_ = false; - gpu_need_display_buffer_update_cond_.notify_all(); - - new_work_added_ = true; - - pause_cond_.notify_all(); -} - -bool Session::draw_gpu(BufferParams &buffer_params, DeviceDrawParams &draw_params) -{ - /* block for buffer access */ - thread_scoped_lock display_lock(display_mutex_); - - /* first check we already rendered something */ - if (gpu_draw_ready_) { - /* then verify the buffers have the expected size, so we don't - * draw previous results in a resized window */ - if (buffer_params.width == display->params.width && - buffer_params.height == display->params.height) { - /* for CUDA we need to do tone-mapping still, since we can - * only access GL buffers from the main thread. */ - if (gpu_need_display_buffer_update_) { - thread_scoped_lock buffers_lock(buffers_mutex_); - copy_to_display_buffer(tile_manager.state.sample); - gpu_need_display_buffer_update_ = false; - gpu_need_display_buffer_update_cond_.notify_all(); + if (!render_work) { + if (VLOG_IS_ON(2)) { + double total_time, render_time; + progress.get_time(total_time, render_time); + VLOG(2) << "Rendering in main loop is done in " << render_time << " seconds."; + VLOG(2) << path_trace_->full_report(); } - display->draw(device, draw_params); - - if (display_outdated_ && (time_dt() - reset_time_) > params.text_timeout) - return false; - - return true; - } - } - - return false; -} - -void Session::run_gpu() -{ - bool tiles_written = false; - - reset_time_ = time_dt(); - last_update_time_ = time_dt(); - last_display_time_ = last_update_time_; - - progress.set_render_start_time(); - - while (!progress.get_cancel()) { - const bool no_tiles = !run_update_for_next_iteration(); - - if (no_tiles) { if (params.background) { - /* if no work left and in background mode, we can stop immediately */ + /* if no work left and in background mode, we can stop immediately. */ progress.set_status("Finished"); break; } } - if (run_wait_for_work(no_tiles)) { - continue; - } - - if (progress.get_cancel()) { - break; - } - - if (!no_tiles) { - if (!device->error_message().empty()) - progress.set_error(device->error_message()); - - if (progress.get_cancel()) - break; - - /* buffers mutex is locked entirely while rendering each - * sample, and released/reacquired on each iteration to allow - * reset and draw in between */ - thread_scoped_lock buffers_lock(buffers_mutex_); - - /* update status and timing */ - update_status_time(); - - /* render */ - bool delayed_denoise = false; - const bool need_denoise = render_need_denoise(delayed_denoise); - render(need_denoise); - - device->task_wait(); - - if (!device->error_message().empty()) - progress.set_cancel(device->error_message()); - - /* update status and timing */ - update_status_time(); - - gpu_need_display_buffer_update_ = !delayed_denoise; - gpu_draw_ready_ = true; - progress.set_update(); - - /* wait for until display buffer is updated */ - if (!params.background) { - while (gpu_need_display_buffer_update_) { - if (progress.get_cancel()) - break; - - gpu_need_display_buffer_update_cond_.wait(buffers_lock); - } - } - - if (!device->error_message().empty()) - progress.set_error(device->error_message()); - - tiles_written = update_progressive_refine(progress.get_cancel()); - - if (progress.get_cancel()) - break; - } - } - - if (!tiles_written) - update_progressive_refine(true); -} - -/* CPU Session */ - -void Session::reset_cpu(BufferParams &buffer_params, int samples) -{ - thread_scoped_lock reset_lock(delayed_reset_.mutex); - thread_scoped_lock pause_lock(pause_mutex_); - - display_outdated_ = true; - reset_time_ = time_dt(); - - delayed_reset_.params = buffer_params; - delayed_reset_.samples = samples; - delayed_reset_.do_reset = true; - device->task_cancel(); - - pause_cond_.notify_all(); -} - -bool Session::draw_cpu(BufferParams &buffer_params, DeviceDrawParams &draw_params) -{ - thread_scoped_lock display_lock(display_mutex_); - - /* first check we already rendered something */ - if (display->draw_ready()) { - /* then verify the buffers have the expected size, so we don't - * draw previous results in a resized window */ - if (buffer_params.width == display->params.width && - buffer_params.height == display->params.height) { - display->draw(device, draw_params); - - if (display_outdated_ && (time_dt() - reset_time_) > params.text_timeout) - return false; - - return true; - } - } - - return false; -} - -bool Session::steal_tile(RenderTile &rtile, Device *tile_device, thread_scoped_lock &tile_lock) -{ - /* Devices that can get their tiles stolen don't steal tiles themselves. - * Additionally, if there are no stealable tiles in flight, give up here. */ - if (tile_device->info.type == DEVICE_CPU || stealable_tiles_ == 0) { - return false; - } - - /* Wait until no other thread is trying to steal a tile. */ - while (tile_stealing_state_ != NOT_STEALING && stealable_tiles_ > 0) { - /* Someone else is currently trying to get a tile. - * Wait on the condition variable and try later. */ - tile_steal_cond_.wait(tile_lock); - } - /* If another thread stole the last stealable tile in the meantime, give up. */ - if (stealable_tiles_ == 0) { - return false; - } - - /* There are stealable tiles in flight, so signal that one should be released. */ - tile_stealing_state_ = WAITING_FOR_TILE; - - /* Wait until a device notices the signal and releases its tile. */ - while (tile_stealing_state_ != GOT_TILE && stealable_tiles_ > 0) { - tile_steal_cond_.wait(tile_lock); - } - /* If the last stealable tile finished on its own, give up. */ - if (tile_stealing_state_ != GOT_TILE) { - tile_stealing_state_ = NOT_STEALING; - return false; - } - - /* Successfully stole a tile, now move it to the new device. */ - rtile = stolen_tile_; - rtile.buffers->buffer.move_device(tile_device); - rtile.buffer = rtile.buffers->buffer.device_pointer; - rtile.stealing_state = RenderTile::NO_STEALING; - rtile.num_samples -= (rtile.sample - rtile.start_sample); - rtile.start_sample = rtile.sample; - - tile_stealing_state_ = NOT_STEALING; - - /* Poke any threads which might be waiting for NOT_STEALING above. */ - tile_steal_cond_.notify_one(); - - return true; -} - -bool Session::get_tile_stolen() -{ - /* If tile_stealing_state is WAITING_FOR_TILE, atomically set it to RELEASING_TILE - * and return true. */ - TileStealingState expected = WAITING_FOR_TILE; - return tile_stealing_state_.compare_exchange_weak(expected, RELEASING_TILE); -} - -bool Session::acquire_tile(RenderTile &rtile, Device *tile_device, uint tile_types) -{ - if (progress.get_cancel()) { - if (params.progressive_refine == false) { - /* for progressive refine current sample should be finished for all tiles */ - return false; - } - } - - thread_scoped_lock tile_lock(tile_mutex_); - - /* get next tile from manager */ - Tile *tile; - int device_num = device->device_number(tile_device); - - while (!tile_manager.next_tile(tile, device_num, tile_types)) { - /* Can only steal tiles on devices that support rendering - * This is because denoising tiles cannot be stolen (see below) - */ - if ((tile_types & (RenderTile::PATH_TRACE | RenderTile::BAKE)) && - steal_tile(rtile, tile_device, tile_lock)) { - return true; - } - - /* Wait for denoising tiles to become available */ - if ((tile_types & RenderTile::DENOISE) && !progress.get_cancel() && tile_manager.has_tiles()) { - denoising_cond_.wait(tile_lock); - continue; - } - - return false; - } - - /* fill render tile */ - rtile.x = tile_manager.state.buffer.full_x + tile->x; - rtile.y = tile_manager.state.buffer.full_y + tile->y; - rtile.w = tile->w; - rtile.h = tile->h; - rtile.start_sample = tile_manager.state.sample; - rtile.num_samples = tile_manager.state.num_samples; - rtile.resolution = tile_manager.state.resolution_divider; - rtile.tile_index = tile->index; - rtile.stealing_state = RenderTile::NO_STEALING; - - if (tile->state == Tile::DENOISE) { - rtile.task = RenderTile::DENOISE; - } - else { - if (tile_device->info.type == DEVICE_CPU) { - stealable_tiles_++; - rtile.stealing_state = RenderTile::CAN_BE_STOLEN; - } - - if (read_bake_tile_cb) { - rtile.task = RenderTile::BAKE; - } - else { - rtile.task = RenderTile::PATH_TRACE; - } - } - - tile_lock.unlock(); - - /* in case of a permanent buffer, return it, otherwise we will allocate - * a new temporary buffer */ - if (buffers) { - tile_manager.state.buffer.get_offset_stride(rtile.offset, rtile.stride); - - rtile.buffer = buffers->buffer.device_pointer; - rtile.buffers = buffers; - - device->map_tile(tile_device, rtile); - - /* Reset copy state, since buffer contents change after the tile was acquired */ - buffers->map_neighbor_copied = false; - - /* This hack ensures that the copy in 'MultiDevice::map_neighbor_tiles' accounts - * for the buffer resolution divider. */ - buffers->buffer.data_width = (buffers->params.width * buffers->params.get_passes_size()) / - tile_manager.state.resolution_divider; - buffers->buffer.data_height = buffers->params.height / tile_manager.state.resolution_divider; - - return true; - } - - if (tile->buffers == NULL) { - /* fill buffer parameters */ - BufferParams buffer_params = tile_manager.params; - buffer_params.full_x = rtile.x; - buffer_params.full_y = rtile.y; - buffer_params.width = rtile.w; - buffer_params.height = rtile.h; - - /* allocate buffers */ - tile->buffers = new RenderBuffers(tile_device); - tile->buffers->reset(buffer_params); - } - else if (tile->buffers->buffer.device != tile_device) { - /* Move buffer to current tile device again in case it was stolen before. - * Not needed for denoising since that already handles mapping of tiles and - * neighbors to its own device. */ - if (rtile.task != RenderTile::DENOISE) { - tile->buffers->buffer.move_device(tile_device); - } - } - - tile->buffers->map_neighbor_copied = false; - - tile->buffers->params.get_offset_stride(rtile.offset, rtile.stride); - - rtile.buffer = tile->buffers->buffer.device_pointer; - rtile.buffers = tile->buffers; - rtile.sample = tile_manager.state.sample; - - if (read_bake_tile_cb) { - /* This will read any passes needed as input for baking. */ - if (tile_manager.state.sample == tile_manager.range_start_sample) { - { - thread_scoped_lock tile_lock(tile_mutex_); - read_bake_tile_cb(rtile); - } - rtile.buffers->buffer.copy_to_device(); - } - } - else { - /* This will tag tile as IN PROGRESS in blender-side render pipeline, - * which is needed to highlight currently rendering tile before first - * sample was processed for it. */ - update_tile_sample(rtile); - } - - return true; -} - -void Session::update_tile_sample(RenderTile &rtile) -{ - thread_scoped_lock tile_lock(tile_mutex_); - - if (update_render_tile_cb) { - if (params.progressive_refine == false) { - /* todo: optimize this by making it thread safe and removing lock */ - - update_render_tile_cb(rtile, true); - } - } - - update_status_time(); -} - -void Session::release_tile(RenderTile &rtile, const bool need_denoise) -{ - thread_scoped_lock tile_lock(tile_mutex_); - - if (rtile.stealing_state != RenderTile::NO_STEALING) { - stealable_tiles_--; - if (rtile.stealing_state == RenderTile::WAS_STOLEN) { - /* If the tile is being stolen, don't release it here - the new device will pick up where - * the old one left off. */ - - assert(tile_stealing_state_ == RELEASING_TILE); - assert(rtile.sample < rtile.start_sample + rtile.num_samples); - - tile_stealing_state_ = GOT_TILE; - stolen_tile_ = rtile; - tile_steal_cond_.notify_all(); - return; - } - else if (stealable_tiles_ == 0) { - /* If this was the last stealable tile, wake up any threads still waiting for one. */ - tile_steal_cond_.notify_all(); - } - } - - progress.add_finished_tile(rtile.task == RenderTile::DENOISE); - - bool delete_tile; - - if (tile_manager.finish_tile(rtile.tile_index, need_denoise, delete_tile)) { - /* Finished tile pixels write. */ - if (write_render_tile_cb && params.progressive_refine == false) { - write_render_tile_cb(rtile); - } - - if (delete_tile) { - delete rtile.buffers; - tile_manager.state.tiles[rtile.tile_index].buffers = NULL; - } - } - else { - /* In progress tile pixels update. */ - if (update_render_tile_cb && params.progressive_refine == false) { - update_render_tile_cb(rtile, false); - } - } - - update_status_time(); - - /* Notify denoising thread that a tile was finished. */ - denoising_cond_.notify_all(); -} - -void Session::map_neighbor_tiles(RenderTileNeighbors &neighbors, Device *tile_device) -{ - thread_scoped_lock tile_lock(tile_mutex_); - - const int4 image_region = make_int4( - tile_manager.state.buffer.full_x, - tile_manager.state.buffer.full_y, - tile_manager.state.buffer.full_x + tile_manager.state.buffer.width, - tile_manager.state.buffer.full_y + tile_manager.state.buffer.height); - - RenderTile ¢er_tile = neighbors.tiles[RenderTileNeighbors::CENTER]; - - if (!tile_manager.schedule_denoising) { - /* Fix up tile slices with overlap. */ - if (tile_manager.slice_overlap != 0) { - int y = max(center_tile.y - tile_manager.slice_overlap, image_region.y); - center_tile.h = min(center_tile.y + center_tile.h + tile_manager.slice_overlap, - image_region.w) - - y; - center_tile.y = y; - } - - /* Tiles are not being denoised individually, which means the entire image is processed. */ - neighbors.set_bounds_from_center(); - } - else { - int center_idx = center_tile.tile_index; - assert(tile_manager.state.tiles[center_idx].state == Tile::DENOISE); - - for (int dy = -1, i = 0; dy <= 1; dy++) { - for (int dx = -1; dx <= 1; dx++, i++) { - RenderTile &rtile = neighbors.tiles[i]; - int nindex = tile_manager.get_neighbor_index(center_idx, i); - if (nindex >= 0) { - Tile *tile = &tile_manager.state.tiles[nindex]; - - rtile.x = image_region.x + tile->x; - rtile.y = image_region.y + tile->y; - rtile.w = tile->w; - rtile.h = tile->h; - - if (buffers) { - tile_manager.state.buffer.get_offset_stride(rtile.offset, rtile.stride); - - rtile.buffer = buffers->buffer.device_pointer; - rtile.buffers = buffers; - } - else { - assert(tile->buffers); - tile->buffers->params.get_offset_stride(rtile.offset, rtile.stride); - - rtile.buffer = tile->buffers->buffer.device_pointer; - rtile.buffers = tile->buffers; - } - } - else { - int px = center_tile.x + dx * params.tile_size.x; - int py = center_tile.y + dy * params.tile_size.y; - - rtile.x = clamp(px, image_region.x, image_region.z); - rtile.y = clamp(py, image_region.y, image_region.w); - rtile.w = rtile.h = 0; - - rtile.buffer = (device_ptr)NULL; - rtile.buffers = NULL; - } - } - } - } - - assert(center_tile.buffers); - device->map_neighbor_tiles(tile_device, neighbors); - - /* The denoised result is written back to the original tile. */ - neighbors.target = center_tile; -} - -void Session::unmap_neighbor_tiles(RenderTileNeighbors &neighbors, Device *tile_device) -{ - thread_scoped_lock tile_lock(tile_mutex_); - device->unmap_neighbor_tiles(tile_device, neighbors); -} - -void Session::run_cpu() -{ - bool tiles_written = false; - - last_update_time_ = time_dt(); - last_display_time_ = last_update_time_; - - while (!progress.get_cancel()) { - const bool no_tiles = !run_update_for_next_iteration(); - bool need_copy_to_display_buffer = false; - - if (no_tiles) { - if (params.background) { - /* if no work left and in background mode, we can stop immediately */ - progress.set_status("Finished"); + const bool did_cancel = progress.get_cancel(); + if (did_cancel) { + render_scheduler_.render_work_reschedule_on_cancel(render_work); + if (!render_work) { break; } } - - if (run_wait_for_work(no_tiles)) { + else if (run_wait_for_work(render_work)) { continue; } - if (progress.get_cancel()) { - break; - } - - if (!no_tiles) { - if (!device->error_message().empty()) - progress.set_error(device->error_message()); - - if (progress.get_cancel()) - break; - + { /* buffers mutex is locked entirely while rendering each * sample, and released/reacquired on each iteration to allow * reset and draw in between */ @@ -730,49 +204,25 @@ void Session::run_cpu() update_status_time(); /* render */ - bool delayed_denoise = false; - const bool need_denoise = render_need_denoise(delayed_denoise); - render(need_denoise); + path_trace_->render(render_work); /* update status and timing */ update_status_time(); - if (!params.background) - need_copy_to_display_buffer = !delayed_denoise; - - if (!device->error_message().empty()) - progress.set_error(device->error_message()); - } - - device->task_wait(); - - { - thread_scoped_lock reset_lock(delayed_reset_.mutex); - thread_scoped_lock buffers_lock(buffers_mutex_); - thread_scoped_lock display_lock(display_mutex_); - - if (delayed_reset_.do_reset) { - /* reset rendering if request from main thread */ - delayed_reset_.do_reset = false; - reset_(delayed_reset_.params, delayed_reset_.samples); - } - else if (need_copy_to_display_buffer) { - /* Only copy to display_buffer if we do not reset, we don't - * want to show the result of an incomplete sample */ - copy_to_display_buffer(tile_manager.state.sample); + if (device->have_error()) { + const string &error_message = device->error_message(); + progress.set_error(error_message); + progress.set_cancel(error_message); + break; } - - if (!device->error_message().empty()) - progress.set_error(device->error_message()); - - tiles_written = update_progressive_refine(progress.get_cancel()); } progress.set_update(); - } - if (!tiles_written) - update_progressive_refine(true); + if (did_cancel) { + break; + } + } } void Session::run() @@ -789,10 +239,7 @@ void Session::run() /* reset number of rendered samples */ progress.reset_sample(); - if (device_use_gl_) - run_gpu(); - else - run_cpu(); + run_main_render_loop(); } profiler.stop(); @@ -804,31 +251,92 @@ void Session::run() progress.set_update(); } -bool Session::run_update_for_next_iteration() +RenderWork Session::run_update_for_next_iteration() { + RenderWork render_work; + thread_scoped_lock scene_lock(scene->mutex); thread_scoped_lock reset_lock(delayed_reset_.mutex); + bool have_tiles = true; + bool switched_to_new_tile = false; + if (delayed_reset_.do_reset) { thread_scoped_lock buffers_lock(buffers_mutex_); - reset_(delayed_reset_.params, delayed_reset_.samples); - delayed_reset_.do_reset = false; + do_delayed_reset(); + + /* After reset make sure the tile manager is at the first big tile. */ + have_tiles = tile_manager_.next(); + switched_to_new_tile = true; + } + + /* Update number of samples in the integrator. + * Ideally this would need to happen once in `Session::set_samples()`, but the issue there is + * the initial configuration when Session is created where the `set_samples()` is not used. */ + scene->integrator->set_aa_samples(params.samples); + + /* Update denoiser settings. */ + { + const DenoiseParams denoise_params = scene->integrator->get_denoise_params(); + path_trace_->set_denoiser_params(denoise_params); + } + + /* Update adaptive sampling. */ + { + const AdaptiveSampling adaptive_sampling = scene->integrator->get_adaptive_sampling(); + path_trace_->set_adaptive_sampling(adaptive_sampling); } - const bool have_tiles = tile_manager.next(); + render_scheduler_.set_num_samples(params.samples); + render_scheduler_.set_time_limit(params.time_limit); + + while (have_tiles) { + render_work = render_scheduler_.get_render_work(); + if (render_work) { + break; + } - if (have_tiles) { + progress.add_finished_tile(false); + + have_tiles = tile_manager_.next(); + if (have_tiles) { + render_scheduler_.reset_for_next_tile(); + switched_to_new_tile = true; + } + } + + if (render_work) { scoped_timer update_timer; - if (update_scene()) { + + if (switched_to_new_tile) { + BufferParams tile_params = buffer_params_; + + const Tile &tile = tile_manager_.get_current_tile(); + tile_params.width = tile.width; + tile_params.height = tile.height; + tile_params.full_x = tile.x + buffer_params_.full_x; + tile_params.full_y = tile.y + buffer_params_.full_y; + tile_params.full_width = buffer_params_.full_width; + tile_params.full_height = buffer_params_.full_height; + tile_params.update_offset_stride(); + + path_trace_->reset(buffer_params_, tile_params); + } + + const int resolution = render_work.resolution_divider; + const int width = max(1, buffer_params_.full_width / resolution); + const int height = max(1, buffer_params_.full_height / resolution); + + if (update_scene(width, height)) { profiler.reset(scene->shaders.size(), scene->objects.size()); } progress.add_skip_time(update_timer, params.background); } - return have_tiles; + return render_work; } -bool Session::run_wait_for_work(bool no_tiles) +bool Session::run_wait_for_work(const RenderWork &render_work) { /* In an offline rendering there is no pause, and no tiles will mean the job is fully done. */ if (params.background) { @@ -837,19 +345,20 @@ bool Session::run_wait_for_work(bool no_tiles) thread_scoped_lock pause_lock(pause_mutex_); - if (!pause_ && !no_tiles) { + if (!pause_ && render_work) { /* Rendering is not paused and there is work to be done. No need to wait for anything. */ return false; } - update_status_time(pause_, no_tiles); + const bool no_work = !render_work; + update_status_time(pause_, no_work); /* Only leave the loop when rendering is not paused. But even if the current render is un-paused * but there is nothing to render keep waiting until new work is added. */ while (!cancel_) { scoped_timer pause_timer; - if (!pause_ && (!no_tiles || new_work_added_ || delayed_reset_.do_reset)) { + if (!pause_ && (render_work || new_work_added_ || delayed_reset_.do_reset)) { break; } @@ -860,52 +369,88 @@ bool Session::run_wait_for_work(bool no_tiles) progress.add_skip_time(pause_timer, params.background); } - update_status_time(pause_, no_tiles); + update_status_time(pause_, no_work); progress.set_update(); } new_work_added_ = false; - return no_tiles; + return no_work; } -bool Session::draw(BufferParams &buffer_params, DeviceDrawParams &draw_params) +void Session::draw() { - if (device_use_gl_) - return draw_gpu(buffer_params, draw_params); - else - return draw_cpu(buffer_params, draw_params); + path_trace_->draw(); } -void Session::reset_(BufferParams &buffer_params, int samples) +int2 Session::get_effective_tile_size() const { - if (buffers && buffer_params.modified(tile_manager.params)) { - gpu_draw_ready_ = false; - buffers->reset(buffer_params); - if (display) { - display->reset(buffer_params); - } + /* No support yet for baking with tiles. */ + if (!params.use_auto_tile || scene->bake_manager->get_baking()) { + return make_int2(buffer_params_.width, buffer_params_.height); } - tile_manager.reset(buffer_params, samples); - stealable_tiles_ = 0; - tile_stealing_state_ = NOT_STEALING; - progress.reset_sample(); + /* TODO(sergey): Take available memory into account, and if there is enough memory do not tile + * and prefer optimal performance. */ + + return make_int2(params.tile_size, params.tile_size); +} + +void Session::do_delayed_reset() +{ + if (!delayed_reset_.do_reset) { + return; + } + delayed_reset_.do_reset = false; + + params = delayed_reset_.session_params; + buffer_params_ = delayed_reset_.buffer_params; + + /* Store parameters used for buffers access outside of scene graph. */ + buffer_params_.exposure = scene->film->get_exposure(); + buffer_params_.use_approximate_shadow_catcher = + scene->film->get_use_approximate_shadow_catcher(); + buffer_params_.use_transparent_background = scene->background->get_transparent(); - bool show_progress = params.background || tile_manager.get_num_effective_samples() != INT_MAX; - progress.set_total_pixel_samples(show_progress ? tile_manager.state.total_pixel_samples : 0); + /* Tile and work scheduling. */ + tile_manager_.reset_scheduling(buffer_params_, get_effective_tile_size()); + render_scheduler_.reset(buffer_params_, params.samples); - if (!params.background) + /* Passes. */ + /* When multiple tiles are used SAMPLE_COUNT pass is used to keep track of possible partial + * tile results. It is safe to use generic update function here which checks for changes since + * changes in tile settings re-creates session, which ensures film is fully updated on tile + * changes. */ + scene->film->update_passes(scene, tile_manager_.has_multiple_tiles()); + + /* Update for new state of scene and passes. */ + buffer_params_.update_passes(scene->passes); + tile_manager_.update(buffer_params_, scene); + + /* Progress. */ + progress.reset_sample(); + progress.set_total_pixel_samples(buffer_params_.width * buffer_params_.height * params.samples); + + if (!params.background) { progress.set_start_time(); + } progress.set_render_start_time(); } -void Session::reset(BufferParams &buffer_params, int samples) +void Session::reset(const SessionParams &session_params, const BufferParams &buffer_params) { - if (device_use_gl_) - reset_gpu(buffer_params, samples); - else - reset_cpu(buffer_params, samples); + { + thread_scoped_lock reset_lock(delayed_reset_.mutex); + thread_scoped_lock pause_lock(pause_mutex_); + + delayed_reset_.do_reset = true; + delayed_reset_.session_params = session_params; + delayed_reset_.buffer_params = buffer_params; + + path_trace_->cancel(); + } + + pause_cond_.notify_all(); } void Session::set_samples(int samples) @@ -915,7 +460,22 @@ void Session::set_samples(int samples) } params.samples = samples; - tile_manager.set_samples(samples); + + { + thread_scoped_lock pause_lock(pause_mutex_); + new_work_added_ = true; + } + + pause_cond_.notify_all(); +} + +void Session::set_time_limit(double time_limit) +{ + if (time_limit == params.time_limit) { + return; + } + + params.time_limit = time_limit; { thread_scoped_lock pause_lock(pause_mutex_); @@ -948,38 +508,9 @@ void Session::set_pause(bool pause) } } -void Session::set_denoising(const DenoiseParams &denoising) +void Session::set_gpu_display(unique_ptr<GPUDisplay> gpu_display) { - bool need_denoise = denoising.need_denoising_task(); - - /* Lock buffers so no denoising operation is triggered while the settings are changed here. */ - thread_scoped_lock buffers_lock(buffers_mutex_); - params.denoising = denoising; - - if (!(params.device.denoisers & denoising.type)) { - if (need_denoise) { - progress.set_error("Denoiser type not supported by compute device"); - } - - params.denoising.use = false; - need_denoise = false; - } - - // TODO(pmours): Query the required overlap value for denoising from the device? - tile_manager.slice_overlap = need_denoise && !params.background ? 64 : 0; - - /* Schedule per tile denoising for final renders if we are either denoising or - * need prefiltered passes for the native denoiser. */ - tile_manager.schedule_denoising = need_denoise && !buffers; -} - -void Session::set_denoising_start_sample(int sample) -{ - if (sample != params.denoising.start_sample) { - params.denoising.start_sample = sample; - - pause_cond_.notify_all(); - } + path_trace_->set_gpu_display(move(gpu_display)); } void Session::wait() @@ -989,81 +520,67 @@ void Session::wait() delete session_thread_; } - session_thread_ = NULL; + session_thread_ = nullptr; } -bool Session::update_scene() +bool Session::update_scene(int width, int height) { - /* update camera if dimensions changed for progressive render. the camera + /* Update camera if dimensions changed for progressive render. the camera * knows nothing about progressive or cropped rendering, it just gets the - * image dimensions passed in */ + * image dimensions passed in. */ Camera *cam = scene->camera; - int width = tile_manager.state.buffer.full_width; - int height = tile_manager.state.buffer.full_height; - int resolution = tile_manager.state.resolution_divider; - - cam->set_screen_size_and_resolution(width, height, resolution); + cam->set_screen_size(width, height); - /* number of samples is needed by multi jittered - * sampling pattern and by baking */ - Integrator *integrator = scene->integrator; - BakeManager *bake_manager = scene->bake_manager; + /* First detect which kernel features are used and allocate working memory. + * This helps estimate how may device memory is available for the scene and + * how much we need to allocate on the host instead. */ + scene->update_kernel_features(); - if (integrator->get_sampling_pattern() != SAMPLING_PATTERN_SOBOL || bake_manager->get_baking()) { - integrator->set_aa_samples(tile_manager.num_samples); - } + path_trace_->load_kernels(); + path_trace_->alloc_work_memory(); - bool kernel_switch_needed = false; - if (scene->update(progress, kernel_switch_needed)) { - if (kernel_switch_needed) { - reset(tile_manager.params, params.samples); - } + if (scene->update(progress)) { return true; } + return false; } +static string status_append(const string &status, const string &suffix) +{ + string prefix = status; + if (!prefix.empty()) { + prefix += ", "; + } + return prefix + suffix; +} + void Session::update_status_time(bool show_pause, bool show_done) { - int progressive_sample = tile_manager.state.sample; - int num_samples = tile_manager.get_num_effective_samples(); + string status, substatus; - int tile = progress.get_rendered_tiles(); - int num_tiles = tile_manager.state.num_tiles; + const int current_tile = progress.get_rendered_tiles(); + const int num_tiles = tile_manager_.get_num_tiles(); - /* update status */ - string status, substatus; + const int current_sample = progress.get_current_sample(); + const int num_samples = render_scheduler_.get_num_samples(); - if (!params.progressive) { - const bool is_cpu = params.device.type == DEVICE_CPU; - const bool rendering_finished = (tile == num_tiles); - const bool is_last_tile = (tile + 1) == num_tiles; - - substatus = string_printf("Rendered %d/%d Tiles", tile, num_tiles); - - if (!rendering_finished && (device->show_samples() || (is_cpu && is_last_tile))) { - /* Some devices automatically support showing the sample number: - * - CUDADevice - * - OpenCLDevice when using the megakernel (the split kernel renders multiple - * samples at the same time, so the current sample isn't really defined) - * - CPUDevice when using one thread - * For these devices, the current sample is always shown. - * - * The other option is when the last tile is currently being rendered by the CPU. - */ - substatus += string_printf(", Sample %d/%d", progress.get_current_sample(), num_samples); - } - if (params.denoising.use && params.denoising.type != DENOISER_OPENIMAGEDENOISE) { - substatus += string_printf(", Denoised %d tiles", progress.get_denoised_tiles()); - } - else if (params.denoising.store_passes && params.denoising.type == DENOISER_NLM) { - substatus += string_printf(", Prefiltered %d tiles", progress.get_denoised_tiles()); - } + /* TIle. */ + if (tile_manager_.has_multiple_tiles()) { + substatus = status_append(substatus, + string_printf("Rendered %d/%d Tiles", current_tile, num_tiles)); } - else if (tile_manager.num_samples == Integrator::MAX_SAMPLES) - substatus = string_printf("Path Tracing Sample %d", progressive_sample + 1); - else - substatus = string_printf("Path Tracing Sample %d/%d", progressive_sample + 1, num_samples); + + /* Sample. */ + if (num_samples == Integrator::MAX_SAMPLES) { + substatus = status_append(substatus, string_printf("Sample %d", current_sample)); + } + else { + substatus = status_append(substatus, + string_printf("Sample %d/%d", current_sample, num_samples)); + } + + /* TODO(sergey): Denoising status from the path trace. */ if (show_pause) { status = "Rendering Paused"; @@ -1080,210 +597,122 @@ void Session::update_status_time(bool show_pause, bool show_done) progress.set_status(status, substatus); } -bool Session::render_need_denoise(bool &delayed) +void Session::device_free() { - delayed = false; - - /* Not supported yet for baking. */ - if (read_bake_tile_cb) { - return false; - } - - /* Denoising enabled? */ - if (!params.denoising.need_denoising_task()) { - return false; - } - - if (params.background) { - /* Background render, only denoise when rendering the last sample. */ - return tile_manager.done(); - } - - /* Viewport render. */ - - /* It can happen that denoising was already enabled, but the scene still needs an update. */ - if (scene->film->is_modified() || !scene->film->get_denoising_data_offset()) { - return false; - } + scene->device_free(); + path_trace_->device_free(); +} - /* Immediately denoise when we reach the start sample or last sample. */ - const int num_samples_finished = tile_manager.state.sample + 1; - if (num_samples_finished == params.denoising.start_sample || - num_samples_finished == params.samples) { - return true; +void Session::collect_statistics(RenderStats *render_stats) +{ + scene->collect_statistics(render_stats); + if (params.use_profiling && (params.device.type == DEVICE_CPU)) { + render_stats->collect_profiling(scene, profiler); } +} - /* Do not denoise until the sample at which denoising should start is reached. */ - if (num_samples_finished < params.denoising.start_sample) { - return false; - } +/* -------------------------------------------------------------------- + * Tile and tile pixels aceess. + */ - /* Avoid excessive denoising in viewport after reaching a certain amount of samples. */ - delayed = (tile_manager.state.sample >= 20 && - (time_dt() - last_display_time_) < params.progressive_update_timeout); - return !delayed; +bool Session::has_multiple_render_tiles() const +{ + return tile_manager_.has_multiple_tiles(); } -void Session::render(bool need_denoise) +int2 Session::get_render_tile_size() const { - if (buffers && tile_manager.state.sample == tile_manager.range_start_sample) { - /* Clear buffers. */ - buffers->zero(); - } - - if (tile_manager.state.buffer.width == 0 || tile_manager.state.buffer.height == 0) { - return; /* Avoid empty launches. */ - } + return path_trace_->get_render_tile_size(); +} - /* Add path trace task. */ - DeviceTask task(DeviceTask::RENDER); - - task.acquire_tile = function_bind(&Session::acquire_tile, this, _2, _1, _3); - task.release_tile = function_bind(&Session::release_tile, this, _1, need_denoise); - task.map_neighbor_tiles = function_bind(&Session::map_neighbor_tiles, this, _1, _2); - task.unmap_neighbor_tiles = function_bind(&Session::unmap_neighbor_tiles, this, _1, _2); - task.get_cancel = function_bind(&Progress::get_cancel, &this->progress); - task.update_tile_sample = function_bind(&Session::update_tile_sample, this, _1); - task.update_progress_sample = function_bind(&Progress::add_samples, &this->progress, _1, _2); - task.get_tile_stolen = function_bind(&Session::get_tile_stolen, this); - task.need_finish_queue = params.progressive_refine; - task.integrator_branched = scene->integrator->get_method() == Integrator::BRANCHED_PATH; - - task.adaptive_sampling.use = (scene->integrator->get_sampling_pattern() == - SAMPLING_PATTERN_PMJ) && - scene->dscene.data.film.pass_adaptive_aux_buffer; - task.adaptive_sampling.min_samples = scene->dscene.data.integrator.adaptive_min_samples; - task.adaptive_sampling.adaptive_step = scene->dscene.data.integrator.adaptive_step; - - /* Acquire render tiles by default. */ - task.tile_types = RenderTile::PATH_TRACE; - - if (need_denoise) { - task.denoising = params.denoising; - - task.pass_stride = scene->film->get_pass_stride(); - task.target_pass_stride = task.pass_stride; - task.pass_denoising_data = scene->film->get_denoising_data_offset(); - task.pass_denoising_clean = scene->film->get_denoising_clean_offset(); - - task.denoising_from_render = true; - - if (tile_manager.schedule_denoising) { - /* Acquire denoising tiles during rendering. */ - task.tile_types |= RenderTile::DENOISE; - } - else { - assert(buffers); - - /* Schedule rendering and wait for it to finish. */ - device->task_add(task); - device->task_wait(); - - /* Then run denoising on the whole image at once. */ - task.type = DeviceTask::DENOISE_BUFFER; - task.x = tile_manager.state.buffer.full_x; - task.y = tile_manager.state.buffer.full_y; - task.w = tile_manager.state.buffer.width; - task.h = tile_manager.state.buffer.height; - task.buffer = buffers->buffer.device_pointer; - task.sample = tile_manager.state.sample; - task.num_samples = tile_manager.state.num_samples; - tile_manager.state.buffer.get_offset_stride(task.offset, task.stride); - task.buffers = buffers; - } - } +int2 Session::get_render_tile_offset() const +{ + return path_trace_->get_render_tile_offset(); +} - device->task_add(task); +string_view Session::get_render_tile_layer() const +{ + const BufferParams &buffer_params = path_trace_->get_render_tile_params(); + return buffer_params.layer; } -void Session::copy_to_display_buffer(int sample) +string_view Session::get_render_tile_view() const { - /* add film conversion task */ - DeviceTask task(DeviceTask::FILM_CONVERT); - - task.x = tile_manager.state.buffer.full_x; - task.y = tile_manager.state.buffer.full_y; - task.w = tile_manager.state.buffer.width; - task.h = tile_manager.state.buffer.height; - task.rgba_byte = display->rgba_byte.device_pointer; - task.rgba_half = display->rgba_half.device_pointer; - task.buffer = buffers->buffer.device_pointer; - task.sample = sample; - tile_manager.state.buffer.get_offset_stride(task.offset, task.stride); - - if (task.w > 0 && task.h > 0) { - device->task_add(task); - device->task_wait(); - - /* set display to new size */ - display->draw_set(task.w, task.h); - - last_display_time_ = time_dt(); - } + const BufferParams &buffer_params = path_trace_->get_render_tile_params(); + return buffer_params.view; +} - display_outdated_ = false; +bool Session::copy_render_tile_from_device() +{ + return path_trace_->copy_render_tile_from_device(); } -bool Session::update_progressive_refine(bool cancel) +bool Session::get_render_tile_pixels(const string &pass_name, int num_components, float *pixels) { - int sample = tile_manager.state.sample + 1; - bool write = sample == tile_manager.num_samples || cancel; + /* NOTE: The code relies on a fact that session is fully update and no scene/buffer modification + * is happenning while this function runs. */ - double current_time = time_dt(); + const BufferParams &buffer_params = path_trace_->get_render_tile_params(); - if (current_time - last_update_time_ < params.progressive_update_timeout) { - /* If last sample was processed, we need to write buffers anyway. */ - if (!write && sample != 1) - return false; + const BufferPass *pass = buffer_params.find_pass(pass_name); + if (pass == nullptr) { + return false; } - if (params.progressive_refine) { - foreach (Tile &tile, tile_manager.state.tiles) { - if (!tile.buffers) { - continue; - } - - RenderTile rtile; - rtile.x = tile_manager.state.buffer.full_x + tile.x; - rtile.y = tile_manager.state.buffer.full_y + tile.y; - rtile.w = tile.w; - rtile.h = tile.h; - rtile.sample = sample; - rtile.buffers = tile.buffers; - - if (write) { - if (write_render_tile_cb) - write_render_tile_cb(rtile); - } - else { - if (update_render_tile_cb) - update_render_tile_cb(rtile, true); - } + const bool has_denoised_result = path_trace_->has_denoised_result(); + if (pass->mode == PassMode::DENOISED && !has_denoised_result) { + pass = buffer_params.find_pass(pass->type); + if (pass == nullptr) { + /* Happens when denoised result pass is requested but is never written by the kernel. */ + return false; } } - last_update_time_ = current_time; + pass = buffer_params.get_actual_display_pass(pass); + + const float exposure = buffer_params.exposure; + const int num_samples = path_trace_->get_num_render_tile_samples(); - return write; + PassAccessor::PassAccessInfo pass_access_info(*pass); + pass_access_info.use_approximate_shadow_catcher = buffer_params.use_approximate_shadow_catcher; + pass_access_info.use_approximate_shadow_catcher_background = + pass_access_info.use_approximate_shadow_catcher && !buffer_params.use_transparent_background; + + const PassAccessorCPU pass_accessor(pass_access_info, exposure, num_samples); + const PassAccessor::Destination destination(pixels, num_components); + + return path_trace_->get_render_tile_pixels(pass_accessor, destination); } -void Session::device_free() +bool Session::set_render_tile_pixels(const string &pass_name, + int num_components, + const float *pixels) { - scene->device_free(); + /* NOTE: The code relies on a fact that session is fully update and no scene/buffer modification + * is happenning while this function runs. */ + + const BufferPass *pass = buffer_params_.find_pass(pass_name); + if (!pass) { + return false; + } + + const float exposure = scene->film->get_exposure(); + const int num_samples = render_scheduler_.get_num_rendered_samples(); - tile_manager.device_free(); + const PassAccessor::PassAccessInfo pass_access_info(*pass); + PassAccessorCPU pass_accessor(pass_access_info, exposure, num_samples); + PassAccessor::Source source(pixels, num_components); - /* used from background render only, so no need to - * re-create render/display buffers here - */ + return path_trace_->set_render_tile_pixels(pass_accessor, source); } -void Session::collect_statistics(RenderStats *render_stats) +/* -------------------------------------------------------------------- + * Full-frame on-disk storage. + */ + +void Session::process_full_buffer_from_disk(string_view filename) { - scene->collect_statistics(render_stats); - if (params.use_profiling && (params.device.type == DEVICE_CPU)) { - render_stats->collect_profiling(scene, profiler); - } + path_trace_->process_full_buffer_from_disk(filename); } CCL_NAMESPACE_END diff --git a/intern/cycles/render/session.h b/intern/cycles/render/session.h index 05025c10f9c..492cfdd1c09 100644 --- a/intern/cycles/render/session.h +++ b/intern/cycles/render/session.h @@ -18,6 +18,7 @@ #define __SESSION_H__ #include "device/device.h" +#include "integrator/render_scheduler.h" #include "render/buffers.h" #include "render/shader.h" #include "render/stats.h" @@ -26,6 +27,7 @@ #include "util/util_progress.h" #include "util/util_stats.h" #include "util/util_thread.h" +#include "util/util_unique_ptr.h" #include "util/util_vector.h" CCL_NAMESPACE_BEGIN @@ -33,41 +35,35 @@ CCL_NAMESPACE_BEGIN class BufferParams; class Device; class DeviceScene; -class DeviceRequestedFeatures; -class DisplayBuffer; +class PathTrace; class Progress; +class GPUDisplay; class RenderBuffers; class Scene; +class SceneParams; /* Session Parameters */ class SessionParams { public: DeviceInfo device; + + bool headless; bool background; - bool progressive_refine; - bool progressive; bool experimental; int samples; - int2 tile_size; - TileOrder tile_order; - int start_resolution; - int denoising_start_sample; int pixel_size; int threads; - bool adaptive_sampling; - - bool use_profiling; - bool display_buffer_linear; + /* Limit in seconds for how long path tracing is allowed to happen. + * Zero means no limit is applied. */ + double time_limit; - DenoiseParams denoising; + bool use_profiling; - double cancel_timeout; - double reset_timeout; - double text_timeout; - double progressive_update_timeout; + bool use_auto_tile; + int tile_size; ShadingSystem shadingsystem; @@ -75,50 +71,32 @@ class SessionParams { SessionParams() { + headless = false; background = false; - progressive_refine = false; - progressive = false; experimental = false; samples = 1024; - tile_size = make_int2(64, 64); - start_resolution = INT_MAX; - denoising_start_sample = 0; pixel_size = 1; threads = 0; - adaptive_sampling = false; + time_limit = 0.0; use_profiling = false; - display_buffer_linear = false; - - cancel_timeout = 0.1; - reset_timeout = 0.1; - text_timeout = 1.0; - progressive_update_timeout = 1.0; + use_auto_tile = true; + tile_size = 2048; shadingsystem = SHADINGSYSTEM_SVM; - tile_order = TILE_CENTER; } - bool modified(const SessionParams ¶ms) + bool modified(const SessionParams ¶ms) const { /* Modified means we have to recreate the session, any parameter changes * that can be handled by an existing Session are omitted. */ - return !(device == params.device && background == params.background && - progressive_refine == params.progressive_refine && - progressive == params.progressive && experimental == params.experimental && - tile_size == params.tile_size && start_resolution == params.start_resolution && + return !(device == params.device && headless == params.headless && + background == params.background && experimental == params.experimental && pixel_size == params.pixel_size && threads == params.threads && - adaptive_sampling == params.adaptive_sampling && - use_profiling == params.use_profiling && - display_buffer_linear == params.display_buffer_linear && - cancel_timeout == params.cancel_timeout && reset_timeout == params.reset_timeout && - text_timeout == params.text_timeout && - progressive_update_timeout == params.progressive_update_timeout && - tile_order == params.tile_order && shadingsystem == params.shadingsystem && - denoising.type == params.denoising.type && - (denoising.use == params.denoising.use || (device.denoisers & denoising.type))); + use_profiling == params.use_profiling && shadingsystem == params.shadingsystem && + use_auto_tile == params.use_auto_tile && tile_size == params.tile_size); } }; @@ -131,34 +109,41 @@ class Session { public: Device *device; Scene *scene; - RenderBuffers *buffers; - DisplayBuffer *display; Progress progress; SessionParams params; - TileManager tile_manager; Stats stats; Profiler profiler; - function<void(RenderTile &)> write_render_tile_cb; - function<void(RenderTile &, bool)> update_render_tile_cb; - function<void(RenderTile &)> read_bake_tile_cb; + function<void(void)> write_render_tile_cb; + function<void(void)> update_render_tile_cb; + function<void(void)> read_render_tile_cb; + + /* Callback is invoked by tile manager whenever on-dist tiles storage file is closed after + * writing. Allows an engine integration to keep track of those files without worry about + * transfering the information when it needs to re-create session during rendering. */ + function<void(string_view)> full_buffer_written_cb; - explicit Session(const SessionParams ¶ms); + explicit Session(const SessionParams ¶ms, const SceneParams &scene_params); ~Session(); void start(); - void cancel(); - bool draw(BufferParams ¶ms, DeviceDrawParams &draw_params); + + /* When quick cancel is requested path tracing is cancelles as soon as possible, without waiting + * for the buffer to be uniformly sampled. */ + void cancel(bool quick = false); + + void draw(); void wait(); bool ready_to_reset(); - void reset(BufferParams ¶ms, int samples); + void reset(const SessionParams &session_params, const BufferParams &buffer_params); + void set_pause(bool pause); + void set_samples(int samples); - void set_denoising(const DenoiseParams &denoising); - void set_denoising_start_sample(int sample); + void set_time_limit(double time_limit); - bool update_scene(); + void set_gpu_display(unique_ptr<GPUDisplay> gpu_display); void device_free(); @@ -168,83 +153,95 @@ class Session { void collect_statistics(RenderStats *stats); - protected: - struct DelayedReset { - thread_mutex mutex; - bool do_reset; - BufferParams params; - int samples; - } delayed_reset_; + /* -------------------------------------------------------------------- + * Tile and tile pixels aceess. + */ - void run(); + bool has_multiple_render_tiles() const; - bool run_update_for_next_iteration(); - bool run_wait_for_work(bool no_tiles); + /* Get size and offset (relative to the buffer's full x/y) of the currently rendering tile. */ + int2 get_render_tile_size() const; + int2 get_render_tile_offset() const; - void update_status_time(bool show_pause = false, bool show_done = false); + string_view get_render_tile_layer() const; + string_view get_render_tile_view() const; - void render(bool use_denoise); - void copy_to_display_buffer(int sample); + bool copy_render_tile_from_device(); - void reset_(BufferParams ¶ms, int samples); + bool get_render_tile_pixels(const string &pass_name, int num_components, float *pixels); + bool set_render_tile_pixels(const string &pass_name, int num_components, const float *pixels); - void run_cpu(); - bool draw_cpu(BufferParams ¶ms, DeviceDrawParams &draw_params); - void reset_cpu(BufferParams ¶ms, int samples); + /* -------------------------------------------------------------------- + * Full-frame on-disk storage. + */ - void run_gpu(); - bool draw_gpu(BufferParams ¶ms, DeviceDrawParams &draw_params); - void reset_gpu(BufferParams ¶ms, int samples); + /* Read given full-frame file from disk, perform needed processing and write it to the software + * via the write callback. */ + void process_full_buffer_from_disk(string_view filename); - bool render_need_denoise(bool &delayed); + protected: + struct DelayedReset { + thread_mutex mutex; + bool do_reset; + SessionParams session_params; + BufferParams buffer_params; + } delayed_reset_; - bool steal_tile(RenderTile &tile, Device *tile_device, thread_scoped_lock &tile_lock); - bool get_tile_stolen(); - bool acquire_tile(RenderTile &tile, Device *tile_device, uint tile_types); - void update_tile_sample(RenderTile &tile); - void release_tile(RenderTile &tile, const bool need_denoise); + void run(); - void map_neighbor_tiles(RenderTileNeighbors &neighbors, Device *tile_device); - void unmap_neighbor_tiles(RenderTileNeighbors &neighbors, Device *tile_device); + /* Update for the new iteration of the main loop in run implementation (run_cpu and run_gpu). + * + * Will take care of the following things: + * - Delayed reset + * - Scene update + * - Tile manager advance + * - Render scheduler work request + * + * The updates are done in a proper order with proper locking around them, which guarantees + * that the device side of scene and render buffers are always in a consistent state. + * + * Returns render work which is to be rendered next. */ + RenderWork run_update_for_next_iteration(); + + /* Wait for rendering to be unpaused, or for new tiles for render to arrive. + * Returns true if new main render loop iteration is required after this function call. + * + * The `render_work` is the work which was scheduled by the render scheduler right before + * checking the pause. */ + bool run_wait_for_work(const RenderWork &render_work); + + void run_main_render_loop(); + + bool update_scene(int width, int height); - bool device_use_gl_; + void update_status_time(bool show_pause = false, bool show_done = false); - thread *session_thread_; + void do_delayed_reset(); - volatile bool display_outdated_; + int2 get_effective_tile_size() const; - volatile bool gpu_draw_ready_; - volatile bool gpu_need_display_buffer_update_; - thread_condition_variable gpu_need_display_buffer_update_cond_; + thread *session_thread_; - bool pause_; - bool cancel_; - bool new_work_added_; + bool pause_ = false; + bool cancel_ = false; + bool new_work_added_ = false; thread_condition_variable pause_cond_; thread_mutex pause_mutex_; thread_mutex tile_mutex_; thread_mutex buffers_mutex_; - thread_mutex display_mutex_; - thread_condition_variable denoising_cond_; - thread_condition_variable tile_steal_cond_; - - double reset_time_; - double last_update_time_; - double last_display_time_; - - RenderTile stolen_tile_; - typedef enum { - NOT_STEALING, /* There currently is no tile stealing in progress. */ - WAITING_FOR_TILE, /* A device is waiting for another device to release a tile. */ - RELEASING_TILE, /* A device has releasing a stealable tile. */ - GOT_TILE /* A device has released a stealable tile, which is now stored in stolen_tile. */ - } TileStealingState; - std::atomic<TileStealingState> tile_stealing_state_; - int stealable_tiles_; - - /* progressive refine */ - bool update_progressive_refine(bool cancel); + + TileManager tile_manager_; + BufferParams buffer_params_; + + /* Render scheduler is used to get work to be rendered with the current big tile. */ + RenderScheduler render_scheduler_; + + /* Path tracer object. + * + * Is a single full-frame path tracer for interactive viewport rendering. + * A path tracer for the current big-tile for an offline rendering. */ + unique_ptr<PathTrace> path_trace_; }; CCL_NAMESPACE_END diff --git a/intern/cycles/render/shader.cpp b/intern/cycles/render/shader.cpp index 59b60904746..f6b23606e58 100644 --- a/intern/cycles/render/shader.cpp +++ b/intern/cycles/render/shader.cpp @@ -203,6 +203,7 @@ Shader::Shader() : Node(get_node_type()) has_surface = false; has_surface_transparent = false; has_surface_emission = false; + has_surface_raytrace = false; has_surface_bssrdf = false; has_volume = false; has_displacement = false; @@ -485,7 +486,7 @@ void ShaderManager::device_update(Device *device, device_update_specific(device, dscene, scene, progress); } -void ShaderManager::device_update_common(Device *device, +void ShaderManager::device_update_common(Device * /*device*/, DeviceScene *dscene, Scene *scene, Progress & /*progress*/) @@ -508,6 +509,8 @@ void ShaderManager::device_update_common(Device *device, flag |= SD_HAS_EMISSION; if (shader->has_surface_transparent && shader->get_use_transparent_shadow()) flag |= SD_HAS_TRANSPARENT_SHADOW; + if (shader->has_surface_raytrace) + flag |= SD_HAS_RAYTRACE; if (shader->has_volume) { flag |= SD_HAS_VOLUME; has_volumes = true; @@ -528,12 +531,10 @@ void ShaderManager::device_update_common(Device *device, flag |= SD_NEED_VOLUME_ATTRIBUTES; if (shader->has_bssrdf_bump) flag |= SD_HAS_BSSRDF_BUMP; - if (device->info.has_volume_decoupled) { - if (shader->get_volume_sampling_method() == VOLUME_SAMPLING_EQUIANGULAR) - flag |= SD_VOLUME_EQUIANGULAR; - if (shader->get_volume_sampling_method() == VOLUME_SAMPLING_MULTIPLE_IMPORTANCE) - flag |= SD_VOLUME_MIS; - } + if (shader->get_volume_sampling_method() == VOLUME_SAMPLING_EQUIANGULAR) + flag |= SD_VOLUME_EQUIANGULAR; + if (shader->get_volume_sampling_method() == VOLUME_SAMPLING_MULTIPLE_IMPORTANCE) + flag |= SD_VOLUME_MIS; if (shader->get_volume_interpolation_method() == VOLUME_INTERPOLATION_CUBIC) flag |= SD_VOLUME_CUBIC; if (shader->has_bump) @@ -682,39 +683,35 @@ void ShaderManager::add_default(Scene *scene) } } -void ShaderManager::get_requested_graph_features(ShaderGraph *graph, - DeviceRequestedFeatures *requested_features) +uint ShaderManager::get_graph_kernel_features(ShaderGraph *graph) { + uint kernel_features = 0; + foreach (ShaderNode *node, graph->nodes) { - requested_features->max_nodes_group = max(requested_features->max_nodes_group, - node->get_group()); - requested_features->nodes_features |= node->get_feature(); + kernel_features |= node->get_feature(); if (node->special_type == SHADER_SPECIAL_TYPE_CLOSURE) { BsdfBaseNode *bsdf_node = static_cast<BsdfBaseNode *>(node); if (CLOSURE_IS_VOLUME(bsdf_node->get_closure_type())) { - requested_features->nodes_features |= NODE_FEATURE_VOLUME; + kernel_features |= KERNEL_FEATURE_NODE_VOLUME; } else if (CLOSURE_IS_PRINCIPLED(bsdf_node->get_closure_type())) { - requested_features->use_principled = true; + kernel_features |= KERNEL_FEATURE_PRINCIPLED; } } if (node->has_surface_bssrdf()) { - requested_features->use_subsurface = true; + kernel_features |= KERNEL_FEATURE_SUBSURFACE; } if (node->has_surface_transparent()) { - requested_features->use_transparent = true; - } - if (node->has_raytrace()) { - requested_features->use_shader_raytrace = true; + kernel_features |= KERNEL_FEATURE_TRANSPARENT; } } + + return kernel_features; } -void ShaderManager::get_requested_features(Scene *scene, - DeviceRequestedFeatures *requested_features) +uint ShaderManager::get_kernel_features(Scene *scene) { - requested_features->max_nodes_group = NODE_GROUP_LEVEL_0; - requested_features->nodes_features = 0; + uint kernel_features = KERNEL_FEATURE_NODE_BSDF | KERNEL_FEATURE_NODE_EMISSION; for (int i = 0; i < scene->shaders.size(); i++) { Shader *shader = scene->shaders[i]; if (!shader->reference_count()) { @@ -722,21 +719,22 @@ void ShaderManager::get_requested_features(Scene *scene, } /* Gather requested features from all the nodes from the graph nodes. */ - get_requested_graph_features(shader->graph, requested_features); + kernel_features |= get_graph_kernel_features(shader->graph); ShaderNode *output_node = shader->graph->output(); if (output_node->input("Displacement")->link != NULL) { - requested_features->nodes_features |= NODE_FEATURE_BUMP; + kernel_features |= KERNEL_FEATURE_NODE_BUMP; if (shader->get_displacement_method() == DISPLACE_BOTH) { - requested_features->nodes_features |= NODE_FEATURE_BUMP_STATE; - requested_features->max_nodes_group = max(requested_features->max_nodes_group, - NODE_GROUP_LEVEL_1); + kernel_features |= KERNEL_FEATURE_NODE_BUMP_STATE; } } /* On top of volume nodes, also check if we need volume sampling because - * e.g. an Emission node would slip through the NODE_FEATURE_VOLUME check */ - if (shader->has_volume) - requested_features->use_volume |= true; + * e.g. an Emission node would slip through the KERNEL_FEATURE_NODE_VOLUME check */ + if (shader->has_volume) { + kernel_features |= KERNEL_FEATURE_VOLUME; + } } + + return kernel_features; } void ShaderManager::free_memory() diff --git a/intern/cycles/render/shader.h b/intern/cycles/render/shader.h index c65cac351a4..5f9adea3949 100644 --- a/intern/cycles/render/shader.h +++ b/intern/cycles/render/shader.h @@ -38,7 +38,6 @@ CCL_NAMESPACE_BEGIN class Device; class DeviceScene; -class DeviceRequestedFeatures; class Mesh; class Progress; class Scene; @@ -117,6 +116,7 @@ class Shader : public Node { bool has_surface; bool has_surface_emission; bool has_surface_transparent; + bool has_surface_raytrace; bool has_volume; bool has_displacement; bool has_surface_bssrdf; @@ -216,7 +216,7 @@ class ShaderManager { static void add_default(Scene *scene); /* Selective nodes compilation. */ - void get_requested_features(Scene *scene, DeviceRequestedFeatures *requested_features); + uint get_kernel_features(Scene *scene); static void free_memory(); @@ -244,8 +244,7 @@ class ShaderManager { size_t beckmann_table_offset; - void get_requested_graph_features(ShaderGraph *graph, - DeviceRequestedFeatures *requested_features); + uint get_graph_kernel_features(ShaderGraph *graph); thread_spin_lock attribute_lock_; diff --git a/intern/cycles/render/stats.cpp b/intern/cycles/render/stats.cpp index 2c6273842e2..73eb7e21ff9 100644 --- a/intern/cycles/render/stats.cpp +++ b/intern/cycles/render/stats.cpp @@ -264,53 +264,34 @@ void RenderStats::collect_profiling(Scene *scene, Profiler &prof) has_profiling = true; kernel = NamedNestedSampleStats("Total render time", prof.get_event(PROFILING_UNKNOWN)); - kernel.add_entry("Ray setup", prof.get_event(PROFILING_RAY_SETUP)); - kernel.add_entry("Result writing", prof.get_event(PROFILING_WRITE_RESULT)); - - NamedNestedSampleStats &integrator = kernel.add_entry("Path integration", - prof.get_event(PROFILING_PATH_INTEGRATE)); - integrator.add_entry("Scene intersection", prof.get_event(PROFILING_SCENE_INTERSECT)); - integrator.add_entry("Indirect emission", prof.get_event(PROFILING_INDIRECT_EMISSION)); - integrator.add_entry("Volumes", prof.get_event(PROFILING_VOLUME)); - - NamedNestedSampleStats &shading = integrator.add_entry("Shading", 0); - shading.add_entry("Shader Setup", prof.get_event(PROFILING_SHADER_SETUP)); - shading.add_entry("Shader Eval", prof.get_event(PROFILING_SHADER_EVAL)); - shading.add_entry("Shader Apply", prof.get_event(PROFILING_SHADER_APPLY)); - shading.add_entry("Ambient Occlusion", prof.get_event(PROFILING_AO)); - shading.add_entry("Subsurface", prof.get_event(PROFILING_SUBSURFACE)); - - integrator.add_entry("Connect Light", prof.get_event(PROFILING_CONNECT_LIGHT)); - integrator.add_entry("Surface Bounce", prof.get_event(PROFILING_SURFACE_BOUNCE)); - - NamedNestedSampleStats &intersection = kernel.add_entry("Intersection", 0); - intersection.add_entry("Full Intersection", prof.get_event(PROFILING_INTERSECT)); - intersection.add_entry("Local Intersection", prof.get_event(PROFILING_INTERSECT_LOCAL)); - intersection.add_entry("Shadow All Intersection", - prof.get_event(PROFILING_INTERSECT_SHADOW_ALL)); - intersection.add_entry("Volume Intersection", prof.get_event(PROFILING_INTERSECT_VOLUME)); - intersection.add_entry("Volume All Intersection", - prof.get_event(PROFILING_INTERSECT_VOLUME_ALL)); - - NamedNestedSampleStats &closure = kernel.add_entry("Closures", 0); - closure.add_entry("Surface Closure Evaluation", prof.get_event(PROFILING_CLOSURE_EVAL)); - closure.add_entry("Surface Closure Sampling", prof.get_event(PROFILING_CLOSURE_SAMPLE)); - closure.add_entry("Volume Closure Evaluation", prof.get_event(PROFILING_CLOSURE_VOLUME_EVAL)); - closure.add_entry("Volume Closure Sampling", prof.get_event(PROFILING_CLOSURE_VOLUME_SAMPLE)); - - NamedNestedSampleStats &denoising = kernel.add_entry("Denoising", - prof.get_event(PROFILING_DENOISING)); - denoising.add_entry("Construct Transform", - prof.get_event(PROFILING_DENOISING_CONSTRUCT_TRANSFORM)); - denoising.add_entry("Reconstruct", prof.get_event(PROFILING_DENOISING_RECONSTRUCT)); - - NamedNestedSampleStats &prefilter = denoising.add_entry("Prefiltering", 0); - prefilter.add_entry("Divide Shadow", prof.get_event(PROFILING_DENOISING_DIVIDE_SHADOW)); - prefilter.add_entry("Non-Local means", prof.get_event(PROFILING_DENOISING_NON_LOCAL_MEANS)); - prefilter.add_entry("Get Feature", prof.get_event(PROFILING_DENOISING_GET_FEATURE)); - prefilter.add_entry("Detect Outliers", prof.get_event(PROFILING_DENOISING_DETECT_OUTLIERS)); - prefilter.add_entry("Combine Halves", prof.get_event(PROFILING_DENOISING_COMBINE_HALVES)); + kernel.add_entry("Intersect Closest", prof.get_event(PROFILING_INTERSECT_CLOSEST)); + kernel.add_entry("Intersect Shadow", prof.get_event(PROFILING_INTERSECT_SHADOW)); + kernel.add_entry("Intersect Subsurface", prof.get_event(PROFILING_INTERSECT_SUBSURFACE)); + kernel.add_entry("Intersect Volume Stack", prof.get_event(PROFILING_INTERSECT_VOLUME_STACK)); + + NamedNestedSampleStats &surface = kernel.add_entry("Shade Surface", 0); + surface.add_entry("Setup", prof.get_event(PROFILING_SHADE_SURFACE_SETUP)); + surface.add_entry("Shader Evaluation", prof.get_event(PROFILING_SHADE_SURFACE_EVAL)); + surface.add_entry("Render Passes", prof.get_event(PROFILING_SHADE_SURFACE_PASSES)); + surface.add_entry("Direct Light", prof.get_event(PROFILING_SHADE_SURFACE_DIRECT_LIGHT)); + surface.add_entry("Indirect Light", prof.get_event(PROFILING_SHADE_SURFACE_INDIRECT_LIGHT)); + surface.add_entry("Ambient Occlusion", prof.get_event(PROFILING_SHADE_SURFACE_AO)); + + NamedNestedSampleStats &volume = kernel.add_entry("Shade Volume", 0); + volume.add_entry("Setup", prof.get_event(PROFILING_SHADE_VOLUME_SETUP)); + volume.add_entry("Integrate", prof.get_event(PROFILING_SHADE_VOLUME_INTEGRATE)); + volume.add_entry("Direct Light", prof.get_event(PROFILING_SHADE_VOLUME_DIRECT_LIGHT)); + volume.add_entry("Indirect Light", prof.get_event(PROFILING_SHADE_VOLUME_INDIRECT_LIGHT)); + + NamedNestedSampleStats &shadow = kernel.add_entry("Shade Shadow", 0); + shadow.add_entry("Setup", prof.get_event(PROFILING_SHADE_SHADOW_SETUP)); + shadow.add_entry("Surface", prof.get_event(PROFILING_SHADE_SHADOW_SURFACE)); + shadow.add_entry("Volume", prof.get_event(PROFILING_SHADE_SHADOW_VOLUME)); + + NamedNestedSampleStats &light = kernel.add_entry("Shade Light", 0); + light.add_entry("Setup", prof.get_event(PROFILING_SHADE_LIGHT_SETUP)); + light.add_entry("Shader Evaluation", prof.get_event(PROFILING_SHADE_LIGHT_EVAL)); shaders.entries.clear(); foreach (Shader *shader, scene->shaders) { diff --git a/intern/cycles/render/svm.cpp b/intern/cycles/render/svm.cpp index dcb3976e15c..2379eb775a0 100644 --- a/intern/cycles/render/svm.cpp +++ b/intern/cycles/render/svm.cpp @@ -446,6 +446,8 @@ void SVMCompiler::generate_node(ShaderNode *node, ShaderNodeSet &done) if (current_type == SHADER_TYPE_SURFACE) { if (node->has_spatial_varying()) current_shader->has_surface_spatial_varying = true; + if (node->get_feature() & KERNEL_FEATURE_NODE_RAYTRACE) + current_shader->has_surface_raytrace = true; } else if (current_type == SHADER_TYPE_VOLUME) { if (node->has_spatial_varying()) @@ -492,6 +494,13 @@ void SVMCompiler::generate_svm_nodes(const ShaderNodeSet &nodes, CompilerState * void SVMCompiler::generate_closure_node(ShaderNode *node, CompilerState *state) { + /* Skip generating closure that are not supported or needed for a particular + * type of shader. For example a BSDF in a volume shader. */ + const int node_feature = node->get_feature(); + if ((state->node_feature_mask & node_feature) != node_feature) { + return; + } + /* execute dependencies for closure */ foreach (ShaderInput *in, node->inputs) { if (in->link != NULL) { @@ -555,7 +564,7 @@ void SVMCompiler::find_aov_nodes_and_dependencies(ShaderNodeSet &aov_nodes, foreach (ShaderNode *node, graph->nodes) { if (node->special_type == SHADER_SPECIAL_TYPE_OUTPUT_AOV) { OutputAOVNode *aov_node = static_cast<OutputAOVNode *>(node); - if (aov_node->slot >= 0) { + if (aov_node->offset >= 0) { aov_nodes.insert(aov_node); foreach (ShaderInput *in, node->inputs) { if (in->link != NULL) { @@ -785,17 +794,21 @@ void SVMCompiler::compile_type(Shader *shader, ShaderGraph *graph, ShaderType ty case SHADER_TYPE_SURFACE: /* generate surface shader */ generate = true; shader->has_surface = true; + state.node_feature_mask = KERNEL_FEATURE_NODE_MASK_SURFACE; break; case SHADER_TYPE_VOLUME: /* generate volume shader */ generate = true; shader->has_volume = true; + state.node_feature_mask = KERNEL_FEATURE_NODE_MASK_VOLUME; break; case SHADER_TYPE_DISPLACEMENT: /* generate displacement shader */ generate = true; shader->has_displacement = true; + state.node_feature_mask = KERNEL_FEATURE_NODE_MASK_DISPLACEMENT; break; case SHADER_TYPE_BUMP: /* generate bump shader */ generate = true; + state.node_feature_mask = KERNEL_FEATURE_NODE_MASK_BUMP; break; default: break; @@ -867,6 +880,7 @@ void SVMCompiler::compile(Shader *shader, array<int4> &svm_nodes, int index, Sum shader->has_surface = false; shader->has_surface_emission = false; shader->has_surface_transparent = false; + shader->has_surface_raytrace = false; shader->has_surface_bssrdf = false; shader->has_bump = has_bump; shader->has_bssrdf_bump = has_bump; @@ -964,6 +978,7 @@ SVMCompiler::CompilerState::CompilerState(ShaderGraph *graph) max_id = max(node->id, max_id); } nodes_done_flag.resize(max_id + 1, false); + node_feature_mask = 0; } CCL_NAMESPACE_END diff --git a/intern/cycles/render/svm.h b/intern/cycles/render/svm.h index d23ff3e2a47..0353c393ae4 100644 --- a/intern/cycles/render/svm.h +++ b/intern/cycles/render/svm.h @@ -192,6 +192,9 @@ class SVMCompiler { * all areas to use this flags array. */ vector<bool> nodes_done_flag; + + /* Node features that can be compiled. */ + uint node_feature_mask; }; void stack_clear_temporary(ShaderNode *node); diff --git a/intern/cycles/render/tile.cpp b/intern/cycles/render/tile.cpp index 375c9fd8e09..eed75cc2372 100644 --- a/intern/cycles/render/tile.cpp +++ b/intern/cycles/render/tile.cpp @@ -16,601 +16,559 @@ #include "render/tile.h" +#include <atomic> + +#include "graph/node.h" +#include "render/background.h" +#include "render/film.h" +#include "render/integrator.h" +#include "render/scene.h" #include "util/util_algorithm.h" #include "util/util_foreach.h" +#include "util/util_logging.h" +#include "util/util_path.h" +#include "util/util_string.h" +#include "util/util_system.h" #include "util/util_types.h" CCL_NAMESPACE_BEGIN -namespace { +/* -------------------------------------------------------------------- + * Internal functions. + */ -class TileComparator { - public: - TileComparator(TileOrder order_, int2 center_, Tile *tiles_) - : order(order_), center(center_), tiles(tiles_) - { - } +static const char *ATTR_PASSES_COUNT = "cycles.passes.count"; +static const char *ATTR_PASS_SOCKET_PREFIX_FORMAT = "cycles.passes.%d."; +static const char *ATTR_BUFFER_SOCKET_PREFIX = "cycles.buffer."; +static const char *ATTR_DENOISE_SOCKET_PREFIX = "cycles.denoise."; - bool operator()(int a, int b) - { - switch (order) { - case TILE_CENTER: { - float2 dist_a = make_float2(center.x - (tiles[a].x + tiles[a].w / 2), - center.y - (tiles[a].y + tiles[a].h / 2)); - float2 dist_b = make_float2(center.x - (tiles[b].x + tiles[b].w / 2), - center.y - (tiles[b].y + tiles[b].h / 2)); - return dot(dist_a, dist_a) < dot(dist_b, dist_b); - } - case TILE_LEFT_TO_RIGHT: - return (tiles[a].x == tiles[b].x) ? (tiles[a].y < tiles[b].y) : (tiles[a].x < tiles[b].x); - case TILE_RIGHT_TO_LEFT: - return (tiles[a].x == tiles[b].x) ? (tiles[a].y < tiles[b].y) : (tiles[a].x > tiles[b].x); - case TILE_TOP_TO_BOTTOM: - return (tiles[a].y == tiles[b].y) ? (tiles[a].x < tiles[b].x) : (tiles[a].y > tiles[b].y); - case TILE_BOTTOM_TO_TOP: - default: - return (tiles[a].y == tiles[b].y) ? (tiles[a].x < tiles[b].x) : (tiles[a].y < tiles[b].y); +/* Global counter of ToleManager object instances. */ +static std::atomic<uint64_t> g_instance_index = 0; + +/* Construct names of EXR channels which will ensure order of all channels to match exact offsets + * in render buffers corresponding to the given passes. + * + * Returns `std` datatypes so that it can be assigned directly to the OIIO's `ImageSpec`. */ +static std::vector<std::string> exr_channel_names_for_passes(const BufferParams &buffer_params) +{ + static const char *component_suffixes[] = {"R", "G", "B", "A"}; + + int pass_index = 0; + int num_channels = 0; + std::vector<std::string> channel_names; + for (const BufferPass &pass : buffer_params.passes) { + if (pass.offset == PASS_UNUSED) { + continue; } - } - protected: - TileOrder order; - int2 center; - Tile *tiles; -}; + const PassInfo pass_info = pass.get_info(); + num_channels += pass_info.num_components; -inline int2 hilbert_index_to_pos(int n, int d) -{ - int2 r, xy = make_int2(0, 0); - for (int s = 1; s < n; s *= 2) { - r.x = (d >> 1) & 1; - r.y = (d ^ r.x) & 1; - if (!r.y) { - if (r.x) { - xy = make_int2(s - 1, s - 1) - xy; - } - swap(xy.x, xy.y); + /* EXR canonically expects first part of channel names to be sorted alphabetically, which is + * not guaranteed to be the case with passes names. Assign a prefix based on the pass index + * with a fixed width to ensure ordering. This makes it possible to dump existing render + * buffers memory to disk and read it back without doing extra mapping. */ + const string prefix = string_printf("%08d", pass_index); + + const string channel_name_prefix = prefix + string(pass.name) + "."; + + for (int i = 0; i < pass_info.num_components; ++i) { + channel_names.push_back(channel_name_prefix + component_suffixes[i]); } - xy += r * make_int2(s, s); - d >>= 2; + + ++pass_index; } - return xy; + + return channel_names; } -enum SpiralDirection { - DIRECTION_UP, - DIRECTION_LEFT, - DIRECTION_DOWN, - DIRECTION_RIGHT, -}; - -} /* namespace */ - -TileManager::TileManager(bool progressive_, - int num_samples_, - int2 tile_size_, - int start_resolution_, - bool preserve_tile_device_, - bool background_, - TileOrder tile_order_, - int num_devices_, - int pixel_size_) +inline string node_socket_attribute_name(const SocketType &socket, const string &attr_name_prefix) { - progressive = progressive_; - tile_size = tile_size_; - tile_order = tile_order_; - start_resolution = start_resolution_; - pixel_size = pixel_size_; - slice_overlap = 0; - num_samples = num_samples_; - num_devices = num_devices_; - preserve_tile_device = preserve_tile_device_; - background = background_; - schedule_denoising = false; - - range_start_sample = 0; - range_num_samples = -1; - - BufferParams buffer_params; - reset(buffer_params, 0); + return attr_name_prefix + string(socket.name); } -TileManager::~TileManager() +template<typename ValidateValueFunc, typename GetValueFunc> +static bool node_socket_generic_to_image_spec_atttributes( + ImageSpec *image_spec, + const Node *node, + const SocketType &socket, + const string &attr_name_prefix, + const ValidateValueFunc &validate_value_func, + const GetValueFunc &get_value_func) { + if (!validate_value_func(node, socket)) { + return false; + } + + image_spec->attribute(node_socket_attribute_name(socket, attr_name_prefix), + get_value_func(node, socket)); + + return true; } -void TileManager::device_free() +static bool node_socket_to_image_spec_atttributes(ImageSpec *image_spec, + const Node *node, + const SocketType &socket, + const string &attr_name_prefix) { - if (schedule_denoising || progressive) { - for (int i = 0; i < state.tiles.size(); i++) { - delete state.tiles[i].buffers; - state.tiles[i].buffers = NULL; + const string attr_name = node_socket_attribute_name(socket, attr_name_prefix); + + switch (socket.type) { + case SocketType::ENUM: { + const ustring value = node->get_string(socket); + + /* Validate that the node is consistent with the node type definition. */ + const NodeEnum &enum_values = *socket.enum_values; + if (!enum_values.exists(value)) { + LOG(DFATAL) << "Node enum contains invalid value " << value; + return false; + } + + image_spec->attribute(attr_name, value); + + return true; } - } - state.tiles.clear(); + case SocketType::STRING: + image_spec->attribute(attr_name, node->get_string(socket)); + return true; + + case SocketType::INT: + image_spec->attribute(attr_name, node->get_int(socket)); + return true; + + case SocketType::FLOAT: + image_spec->attribute(attr_name, node->get_float(socket)); + return true; + + case SocketType::BOOLEAN: + image_spec->attribute(attr_name, node->get_bool(socket)); + return true; + + default: + LOG(DFATAL) << "Unhandled socket type " << socket.type << ", should never happen."; + return false; + } } -static int get_divider(int w, int h, int start_resolution) +static bool node_socket_from_image_spec_atttributes(Node *node, + const SocketType &socket, + const ImageSpec &image_spec, + const string &attr_name_prefix) { - int divider = 1; - if (start_resolution != INT_MAX) { - while (w * h > start_resolution * start_resolution) { - w = max(1, w / 2); - h = max(1, h / 2); + const string attr_name = node_socket_attribute_name(socket, attr_name_prefix); + + switch (socket.type) { + case SocketType::ENUM: { + /* TODO(sergey): Avoid construction of `ustring` by using `string_view` in the Node API. */ + const ustring value(image_spec.get_string_attribute(attr_name, "")); + + /* Validate that the node is consistent with the node type definition. */ + const NodeEnum &enum_values = *socket.enum_values; + if (!enum_values.exists(value)) { + LOG(ERROR) << "Invalid enumerator value " << value; + return false; + } - divider <<= 1; + node->set(socket, enum_values[value]); + + return true; } + + case SocketType::STRING: + /* TODO(sergey): Avoid construction of `ustring` by using `string_view` in the Node API. */ + node->set(socket, ustring(image_spec.get_string_attribute(attr_name, ""))); + return true; + + case SocketType::INT: + node->set(socket, image_spec.get_int_attribute(attr_name, 0)); + return true; + + case SocketType::FLOAT: + node->set(socket, image_spec.get_float_attribute(attr_name, 0)); + return true; + + case SocketType::BOOLEAN: + node->set(socket, static_cast<bool>(image_spec.get_int_attribute(attr_name, 0))); + return true; + + default: + LOG(DFATAL) << "Unhandled socket type " << socket.type << ", should never happen."; + return false; } - return divider; } -void TileManager::reset(BufferParams ¶ms_, int num_samples_) +static bool node_to_image_spec_atttributes(ImageSpec *image_spec, + const Node *node, + const string &attr_name_prefix) { - params = params_; - - set_samples(num_samples_); - - state.buffer = BufferParams(); - state.sample = range_start_sample - 1; - state.num_tiles = 0; - state.num_samples = 0; - state.resolution_divider = get_divider(params.width, params.height, start_resolution); - state.render_tiles.clear(); - state.denoising_tiles.clear(); - device_free(); + for (const SocketType &socket : node->type->inputs) { + if (!node_socket_to_image_spec_atttributes(image_spec, node, socket, attr_name_prefix)) { + return false; + } + } + + return true; } -void TileManager::set_samples(int num_samples_) +static bool node_from_image_spec_atttributes(Node *node, + const ImageSpec &image_spec, + const string &attr_name_prefix) { - num_samples = num_samples_; + for (const SocketType &socket : node->type->inputs) { + if (!node_socket_from_image_spec_atttributes(node, socket, image_spec, attr_name_prefix)) { + return false; + } + } + + return true; +} - /* No real progress indication is possible when using unlimited samples. */ - if (num_samples == INT_MAX) { - state.total_pixel_samples = 0; +static bool buffer_params_to_image_spec_atttributes(ImageSpec *image_spec, + const BufferParams &buffer_params) +{ + if (!node_to_image_spec_atttributes(image_spec, &buffer_params, ATTR_BUFFER_SOCKET_PREFIX)) { + return false; } - else { - uint64_t pixel_samples = 0; - /* While rendering in the viewport, the initial preview resolution is increased to the native - * resolution before the actual rendering begins. Therefore, additional pixel samples will be - * rendered. */ - int divider = max(get_divider(params.width, params.height, start_resolution) / 2, pixel_size); - while (divider > pixel_size) { - int image_w = max(1, params.width / divider); - int image_h = max(1, params.height / divider); - pixel_samples += image_w * image_h; - divider >>= 1; - } - int image_w = max(1, params.width / divider); - int image_h = max(1, params.height / divider); - state.total_pixel_samples = pixel_samples + - (uint64_t)get_num_effective_samples() * image_w * image_h; - if (schedule_denoising) { - state.total_pixel_samples += params.width * params.height; + /* Passes storage is not covered by the node socket. so "expand" the loop manually. */ + + const int num_passes = buffer_params.passes.size(); + image_spec->attribute(ATTR_PASSES_COUNT, num_passes); + + for (int pass_index = 0; pass_index < num_passes; ++pass_index) { + const string attr_name_prefix = string_printf(ATTR_PASS_SOCKET_PREFIX_FORMAT, pass_index); + + const BufferPass *pass = &buffer_params.passes[pass_index]; + if (!node_to_image_spec_atttributes(image_spec, pass, attr_name_prefix)) { + return false; } } + + return true; } -/* If sliced is false, splits image into tiles and assigns equal amount of tiles to every render - * device. If sliced is true, slice image into as much pieces as how many devices are rendering - * this image. */ -int TileManager::gen_tiles(bool sliced) +static bool buffer_params_from_image_spec_atttributes(BufferParams *buffer_params, + const ImageSpec &image_spec) { - int resolution = state.resolution_divider; - int image_w = max(1, params.width / resolution); - int image_h = max(1, params.height / resolution); - int2 center = make_int2(image_w / 2, image_h / 2); - - int num = preserve_tile_device || sliced ? min(image_h, num_devices) : 1; - int slice_num = sliced ? num : 1; - int tile_w = (tile_size.x >= image_w) ? 1 : divide_up(image_w, tile_size.x); - - device_free(); - state.render_tiles.clear(); - state.denoising_tiles.clear(); - state.render_tiles.resize(num); - state.denoising_tiles.resize(num); - state.tile_stride = tile_w; - vector<list<int>>::iterator tile_list; - tile_list = state.render_tiles.begin(); - - if (tile_order == TILE_HILBERT_SPIRAL) { - assert(!sliced && slice_overlap == 0); - - int tile_h = (tile_size.y >= image_h) ? 1 : divide_up(image_h, tile_size.y); - state.tiles.resize(tile_w * tile_h); - - /* Size of blocks in tiles, must be a power of 2 */ - const int hilbert_size = (max(tile_size.x, tile_size.y) <= 12) ? 8 : 4; - - int tiles_per_device = divide_up(tile_w * tile_h, num); - int cur_device = 0, cur_tiles = 0; - - int2 block_size = tile_size * make_int2(hilbert_size, hilbert_size); - /* Number of blocks to fill the image */ - int blocks_x = (block_size.x >= image_w) ? 1 : divide_up(image_w, block_size.x); - int blocks_y = (block_size.y >= image_h) ? 1 : divide_up(image_h, block_size.y); - int n = max(blocks_x, blocks_y) | 0x1; /* Side length of the spiral (must be odd) */ - /* Offset of spiral (to keep it centered) */ - int2 offset = make_int2((image_w - n * block_size.x) / 2, (image_h - n * block_size.y) / 2); - offset = (offset / tile_size) * tile_size; /* Round to tile border. */ - - int2 block = make_int2(0, 0); /* Current block */ - SpiralDirection prev_dir = DIRECTION_UP, dir = DIRECTION_UP; - for (int i = 0;;) { - /* Generate the tiles in the current block. */ - for (int hilbert_index = 0; hilbert_index < hilbert_size * hilbert_size; hilbert_index++) { - int2 tile, hilbert_pos = hilbert_index_to_pos(hilbert_size, hilbert_index); - /* Rotate block according to spiral direction. */ - if (prev_dir == DIRECTION_UP && dir == DIRECTION_UP) { - tile = make_int2(hilbert_pos.y, hilbert_pos.x); - } - else if (dir == DIRECTION_LEFT || prev_dir == DIRECTION_LEFT) { - tile = hilbert_pos; - } - else if (dir == DIRECTION_DOWN) { - tile = make_int2(hilbert_size - 1 - hilbert_pos.y, hilbert_size - 1 - hilbert_pos.x); - } - else { - tile = make_int2(hilbert_size - 1 - hilbert_pos.x, hilbert_size - 1 - hilbert_pos.y); - } - - int2 pos = block * block_size + tile * tile_size + offset; - /* Only add tiles which are in the image (tiles outside of the image can be generated since - * the spiral is always square). */ - if (pos.x >= 0 && pos.y >= 0 && pos.x < image_w && pos.y < image_h) { - int w = min(tile_size.x, image_w - pos.x); - int h = min(tile_size.y, image_h - pos.y); - int2 ipos = pos / tile_size; - int idx = ipos.y * tile_w + ipos.x; - state.tiles[idx] = Tile(idx, pos.x, pos.y, w, h, cur_device, Tile::RENDER); - tile_list->push_front(idx); - cur_tiles++; - - if (cur_tiles == tiles_per_device) { - tile_list++; - cur_tiles = 0; - cur_device++; - } - } - } + if (!node_from_image_spec_atttributes(buffer_params, image_spec, ATTR_BUFFER_SOCKET_PREFIX)) { + return false; + } - /* Stop as soon as the spiral has reached the center block. */ - if (block.x == (n - 1) / 2 && block.y == (n - 1) / 2) - break; - - /* Advance to next block. */ - prev_dir = dir; - switch (dir) { - case DIRECTION_UP: - block.y++; - if (block.y == (n - i - 1)) { - dir = DIRECTION_LEFT; - } - break; - case DIRECTION_LEFT: - block.x++; - if (block.x == (n - i - 1)) { - dir = DIRECTION_DOWN; - } - break; - case DIRECTION_DOWN: - block.y--; - if (block.y == i) { - dir = DIRECTION_RIGHT; - } - break; - case DIRECTION_RIGHT: - block.x--; - if (block.x == i + 1) { - dir = DIRECTION_UP; - i++; - } - break; - } - } - return tile_w * tile_h; + /* Passes storage is not covered by the node socket. so "expand" the loop manually. */ + + const int num_passes = image_spec.get_int_attribute(ATTR_PASSES_COUNT, 0); + if (num_passes == 0) { + LOG(ERROR) << "Missing passes count attribute."; + return false; } - int idx = 0; - for (int slice = 0; slice < slice_num; slice++) { - int slice_y = (image_h / slice_num) * slice; - int slice_h = (slice == slice_num - 1) ? image_h - slice * (image_h / slice_num) : - image_h / slice_num; + for (int pass_index = 0; pass_index < num_passes; ++pass_index) { + const string attr_name_prefix = string_printf(ATTR_PASS_SOCKET_PREFIX_FORMAT, pass_index); - if (slice_overlap != 0) { - int slice_y_offset = max(slice_y - slice_overlap, 0); - slice_h = min(slice_y + slice_h + slice_overlap, image_h) - slice_y_offset; - slice_y = slice_y_offset; - } + BufferPass pass; - int tile_h = (tile_size.y >= slice_h) ? 1 : divide_up(slice_h, tile_size.y); - - int tiles_per_device = divide_up(tile_w * tile_h, num); - int cur_device = 0, cur_tiles = 0; - - for (int tile_y = 0; tile_y < tile_h; tile_y++) { - for (int tile_x = 0; tile_x < tile_w; tile_x++, idx++) { - int x = tile_x * tile_size.x; - int y = tile_y * tile_size.y; - int w = (tile_x == tile_w - 1) ? image_w - x : tile_size.x; - int h = (tile_y == tile_h - 1) ? slice_h - y : tile_size.y; - - state.tiles.push_back( - Tile(idx, x, y + slice_y, w, h, sliced ? slice : cur_device, Tile::RENDER)); - tile_list->push_back(idx); - - if (!sliced) { - cur_tiles++; - - if (cur_tiles == tiles_per_device) { - /* Tiles are already generated in Bottom-to-Top order, so no sort is necessary in that - * case. */ - if (tile_order != TILE_BOTTOM_TO_TOP) { - tile_list->sort(TileComparator(tile_order, center, &state.tiles[0])); - } - tile_list++; - cur_tiles = 0; - cur_device++; - } - } - } - } - if (sliced) { - tile_list++; + if (!node_from_image_spec_atttributes(&pass, image_spec, attr_name_prefix)) { + return false; } + + buffer_params->passes.emplace_back(std::move(pass)); } - return idx; + buffer_params->update_passes(); + + return true; } -void TileManager::gen_render_tiles() +/* Configure image specification for the given buffer parameters and passes. + * + * Image channels will ber strictly ordered to match content of corresponding buffer, and the + * metadata will be set so that the render buffers and passes can be reconstructed from it. + * + * If the tile size different from (0, 0) the image specification will be configured to use the + * given tile size for tiled IO. */ +static bool configure_image_spec_from_buffer(ImageSpec *image_spec, + const BufferParams &buffer_params, + const int2 tile_size = make_int2(0, 0)) { - /* Regenerate just the render tiles for progressive render. */ - foreach (Tile &tile, state.tiles) { - tile.state = Tile::RENDER; - state.render_tiles[tile.device].push_back(tile.index); + const std::vector<std::string> channel_names = exr_channel_names_for_passes(buffer_params); + const int num_channels = channel_names.size(); + + *image_spec = ImageSpec( + buffer_params.width, buffer_params.height, num_channels, TypeDesc::FLOAT); + + image_spec->channelnames = move(channel_names); + + if (!buffer_params_to_image_spec_atttributes(image_spec, buffer_params)) { + return false; + } + + if (tile_size.x != 0 || tile_size.y != 0) { + DCHECK_GT(tile_size.x, 0); + DCHECK_GT(tile_size.y, 0); + + image_spec->tile_width = tile_size.x; + image_spec->tile_height = tile_size.y; } + + return true; } -void TileManager::set_tiles() +/* -------------------------------------------------------------------- + * Tile Manager. + */ + +TileManager::TileManager() { - int resolution = state.resolution_divider; - int image_w = max(1, params.width / resolution); - int image_h = max(1, params.height / resolution); + /* Use process ID to separate different processes. + * To ensure uniqueness from within a process use combination of object address and instance + * index. This solves problem of possible object re-allocation at the same time, and solves + * possible conflict when the counter overflows while there are still active instances of the + * class. */ + const int tile_manager_id = g_instance_index.fetch_add(1, std::memory_order_relaxed); + tile_file_unique_part_ = to_string(system_self_process_id()) + "-" + + to_string(reinterpret_cast<uintptr_t>(this)) + "-" + + to_string(tile_manager_id); +} - state.num_tiles = gen_tiles(!background); +TileManager::~TileManager() +{ +} + +void TileManager::reset_scheduling(const BufferParams ¶ms, int2 tile_size) +{ + VLOG(3) << "Using tile size of " << tile_size; + + close_tile_output(); + + tile_size_ = tile_size; + + tile_state_.num_tiles_x = divide_up(params.width, tile_size_.x); + tile_state_.num_tiles_y = divide_up(params.height, tile_size_.y); + tile_state_.num_tiles = tile_state_.num_tiles_x * tile_state_.num_tiles_y; + + tile_state_.next_tile_index = 0; + + tile_state_.current_tile = Tile(); +} + +void TileManager::update(const BufferParams ¶ms, const Scene *scene) +{ + DCHECK_NE(params.pass_stride, -1); + + buffer_params_ = params; - state.buffer.width = image_w; - state.buffer.height = image_h; + /* TODO(sergey): Proper Error handling, so that if configuration has failed we dont' attempt to + * write to a partially configured file. */ + configure_image_spec_from_buffer(&write_state_.image_spec, buffer_params_, tile_size_); - state.buffer.full_x = params.full_x / resolution; - state.buffer.full_y = params.full_y / resolution; - state.buffer.full_width = max(1, params.full_width / resolution); - state.buffer.full_height = max(1, params.full_height / resolution); + const DenoiseParams denoise_params = scene->integrator->get_denoise_params(); + node_to_image_spec_atttributes( + &write_state_.image_spec, &denoise_params, ATTR_DENOISE_SOCKET_PREFIX); } -int TileManager::get_neighbor_index(int index, int neighbor) +bool TileManager::done() { - /* Neighbor indices: - * 0 1 2 - * 3 4 5 - * 6 7 8 - */ - static const int dx[] = {-1, 0, 1, -1, 0, 1, -1, 0, 1}; - static const int dy[] = {-1, -1, -1, 0, 0, 0, 1, 1, 1}; - - int resolution = state.resolution_divider; - int image_w = max(1, params.width / resolution); - int image_h = max(1, params.height / resolution); - - int num = min(image_h, num_devices); - int slice_num = !background ? num : 1; - int slice_h = image_h / slice_num; - - int tile_w = (tile_size.x >= image_w) ? 1 : divide_up(image_w, tile_size.x); - int tile_h = (tile_size.y >= slice_h) ? 1 : divide_up(slice_h, tile_size.y); - - /* Tiles in the state tile list are always indexed from left to right, top to bottom. */ - int nx = (index % tile_w) + dx[neighbor]; - int ny = (index / tile_w) + dy[neighbor]; - if (nx < 0 || ny < 0 || nx >= tile_w || ny >= tile_h * slice_num) - return -1; - - return ny * state.tile_stride + nx; + return tile_state_.next_tile_index == tile_state_.num_tiles; } -/* Checks whether all neighbors of a tile (as well as the tile itself) are at least at state - * min_state. */ -bool TileManager::check_neighbor_state(int index, Tile::State min_state) +bool TileManager::next() { - if (index < 0 || state.tiles[index].state < min_state) { + if (done()) { return false; } - for (int neighbor = 0; neighbor < 9; neighbor++) { - int nindex = get_neighbor_index(index, neighbor); - /* Out-of-bounds tiles don't matter. */ - if (nindex >= 0 && state.tiles[nindex].state < min_state) { - return false; - } - } + + tile_state_.current_tile = get_tile_for_index(tile_state_.next_tile_index); + + ++tile_state_.next_tile_index; return true; } -/* Returns whether the tile should be written (and freed if no denoising is used) instead of - * updating. */ -bool TileManager::finish_tile(const int index, const bool need_denoise, bool &delete_tile) +Tile TileManager::get_tile_for_index(int index) const { - delete_tile = false; - - switch (state.tiles[index].state) { - case Tile::RENDER: { - if (!(schedule_denoising && need_denoise)) { - state.tiles[index].state = Tile::DONE; - delete_tile = !progressive; - return true; - } - state.tiles[index].state = Tile::RENDERED; - /* For each neighbor and the tile itself, check whether all of its neighbors have been - * rendered. If yes, it can be denoised. */ - for (int neighbor = 0; neighbor < 9; neighbor++) { - int nindex = get_neighbor_index(index, neighbor); - if (check_neighbor_state(nindex, Tile::RENDERED)) { - state.tiles[nindex].state = Tile::DENOISE; - state.denoising_tiles[state.tiles[nindex].device].push_back(nindex); - } - } - return false; - } - case Tile::DENOISE: { - state.tiles[index].state = Tile::DENOISED; - /* For each neighbor and the tile itself, check whether all of its neighbors have been - * denoised. If yes, it can be freed. */ - for (int neighbor = 0; neighbor < 9; neighbor++) { - int nindex = get_neighbor_index(index, neighbor); - if (check_neighbor_state(nindex, Tile::DENOISED)) { - state.tiles[nindex].state = Tile::DONE; - /* Do not delete finished tiles in progressive mode. */ - if (!progressive) { - /* It can happen that the tile just finished denoising and already can be freed here. - * However, in that case it still has to be written before deleting, so we can't delete - * it yet. */ - if (neighbor == 4) { - delete_tile = true; - } - else { - delete state.tiles[nindex].buffers; - state.tiles[nindex].buffers = NULL; - } - } - } - } - return true; - } - default: - assert(false); - return true; + /* TODO(sergey): Consider using hilbert spiral, or. maybe, even configurable. Not sure this + * brings a lot of value since this is only applicable to BIG tiles. */ + + const int tile_y = index / tile_state_.num_tiles_x; + const int tile_x = index - tile_y * tile_state_.num_tiles_x; + + Tile tile; + + tile.x = tile_x * tile_size_.x; + tile.y = tile_y * tile_size_.y; + tile.width = tile_size_.x; + tile.height = tile_size_.y; + + tile.width = min(tile.width, buffer_params_.width - tile.x); + tile.height = min(tile.height, buffer_params_.height - tile.y); + + return tile; +} + +const Tile &TileManager::get_current_tile() const +{ + return tile_state_.current_tile; +} + +bool TileManager::open_tile_output() +{ + write_state_.filename = path_temp_get("cycles-tile-buffer-" + tile_file_unique_part_ + "-" + + to_string(write_state_.tile_file_index) + ".exr"); + + write_state_.tile_out = ImageOutput::create(write_state_.filename); + if (!write_state_.tile_out) { + LOG(ERROR) << "Error creating image output for " << write_state_.filename; + return false; + } + + if (!write_state_.tile_out->supports("tiles")) { + LOG(ERROR) << "Progress tile file format does not support tiling."; + return false; } + + write_state_.tile_out->open(write_state_.filename, write_state_.image_spec); + write_state_.num_tiles_written = 0; + + VLOG(3) << "Opened tile file " << write_state_.filename; + + return true; } -bool TileManager::next_tile(Tile *&tile, int device, uint tile_types) +bool TileManager::close_tile_output() { - /* Preserve device if requested, unless this is a separate denoising device that just wants to - * grab any available tile. */ - const bool preserve_device = preserve_tile_device && device < num_devices; - - if (tile_types & RenderTile::DENOISE) { - int tile_index = -1; - int logical_device = preserve_device ? device : 0; - - while (logical_device < state.denoising_tiles.size()) { - if (state.denoising_tiles[logical_device].empty()) { - if (preserve_device) { - break; - } - else { - logical_device++; - continue; - } - } + if (!write_state_.tile_out) { + return true; + } - tile_index = state.denoising_tiles[logical_device].front(); - state.denoising_tiles[logical_device].pop_front(); - break; - } + const bool success = write_state_.tile_out->close(); + write_state_.tile_out = nullptr; - if (tile_index >= 0) { - tile = &state.tiles[tile_index]; - return true; - } + if (!success) { + LOG(ERROR) << "Error closing tile file."; + return false; } - if (tile_types & RenderTile::PATH_TRACE) { - int tile_index = -1; - int logical_device = preserve_device ? device : 0; - - while (logical_device < state.render_tiles.size()) { - if (state.render_tiles[logical_device].empty()) { - if (preserve_device) { - break; - } - else { - logical_device++; - continue; - } - } + VLOG(3) << "Tile output is closed."; - tile_index = state.render_tiles[logical_device].front(); - state.render_tiles[logical_device].pop_front(); - break; + return true; +} + +bool TileManager::write_tile(const RenderBuffers &tile_buffers) +{ + if (!write_state_.tile_out) { + if (!open_tile_output()) { + return false; } + } - if (tile_index >= 0) { - tile = &state.tiles[tile_index]; - return true; + DCHECK_EQ(tile_buffers.params.pass_stride, buffer_params_.pass_stride); + + const BufferParams &tile_params = tile_buffers.params; + + vector<float> pixel_storage; + const float *pixels = tile_buffers.buffer.data(); + + /* Tiled writing expects pixels to contain data for an entire tile. Pad the render buffers with + * empty pixels for tiles which are on the image boundary. */ + if (tile_params.width != tile_size_.x || tile_params.height != tile_size_.y) { + const int64_t pass_stride = tile_params.pass_stride; + const int64_t src_row_stride = tile_params.width * pass_stride; + + const int64_t dst_row_stride = tile_size_.x * pass_stride; + pixel_storage.resize(dst_row_stride * tile_size_.y); + + const float *src = tile_buffers.buffer.data(); + float *dst = pixel_storage.data(); + pixels = dst; + + for (int y = 0; y < tile_params.height; ++y, src += src_row_stride, dst += dst_row_stride) { + memcpy(dst, src, src_row_stride * sizeof(float)); } } - return false; -} + const int tile_x = tile_params.full_x - buffer_params_.full_x; + const int tile_y = tile_params.full_y - buffer_params_.full_y; -bool TileManager::done() -{ - int end_sample = (range_num_samples == -1) ? num_samples : - range_start_sample + range_num_samples; - return (state.resolution_divider == pixel_size) && - (state.sample + state.num_samples >= end_sample); + VLOG(3) << "Write tile at " << tile_x << ", " << tile_y; + if (!write_state_.tile_out->write_tile(tile_x, tile_y, 0, TypeDesc::FLOAT, pixels)) { + LOG(ERROR) << "Error writing tile " << write_state_.tile_out->geterror(); + } + + ++write_state_.num_tiles_written; + + return true; } -bool TileManager::has_tiles() +void TileManager::finish_write_tiles() { - foreach (Tile &tile, state.tiles) { - if (tile.state != Tile::DONE) { - return true; + if (!write_state_.tile_out) { + /* None of the tiles were written hence the file was not created. + * Avoid creation of fully empty file since it is redundant. */ + return; + } + + /* EXR expects all tiles to present in file. So explicitly write missing tiles as all-zero. */ + if (write_state_.num_tiles_written < tile_state_.num_tiles) { + vector<float> pixel_storage(tile_size_.x * tile_size_.y * buffer_params_.pass_stride); + + for (int tile_index = write_state_.num_tiles_written; tile_index < tile_state_.num_tiles; + ++tile_index) { + const Tile tile = get_tile_for_index(tile_index); + + VLOG(3) << "Write dummy tile at " << tile.x << ", " << tile.y; + + write_state_.tile_out->write_tile(tile.x, tile.y, 0, TypeDesc::FLOAT, pixel_storage.data()); } } - return false; + + close_tile_output(); + + if (full_buffer_written_cb) { + full_buffer_written_cb(write_state_.filename); + } + + /* Advance the counter upon explicit finish of the file. + * Makes it possible to re-use tile manager for another scene, and avoids unnecessary increments + * of the tile-file-within-session index. */ + ++write_state_.tile_file_index; + + write_state_.filename = ""; } -bool TileManager::next() +bool TileManager::read_full_buffer_from_disk(const string_view filename, + RenderBuffers *buffers, + DenoiseParams *denoise_params) { - if (done()) + unique_ptr<ImageInput> in(ImageInput::open(filename)); + if (!in) { + LOG(ERROR) << "Error opening tile file " << filename; return false; + } + + const ImageSpec &image_spec = in->spec(); - if (progressive && state.resolution_divider > pixel_size) { - state.sample = 0; - state.resolution_divider = max(state.resolution_divider / 2, pixel_size); - state.num_samples = 1; - set_tiles(); + BufferParams buffer_params; + if (!buffer_params_from_image_spec_atttributes(&buffer_params, image_spec)) { + return false; } - else { - state.sample++; + buffers->reset(buffer_params); - if (progressive) - state.num_samples = 1; - else if (range_num_samples == -1) - state.num_samples = num_samples; - else - state.num_samples = range_num_samples; + if (!node_from_image_spec_atttributes(denoise_params, image_spec, ATTR_DENOISE_SOCKET_PREFIX)) { + return false; + } - state.resolution_divider = pixel_size; + if (!in->read_image(TypeDesc::FLOAT, buffers->buffer.data())) { + LOG(ERROR) << "Error reading pixels from the tile file " << in->geterror(); + return false; + } - if (state.sample == range_start_sample) { - set_tiles(); - } - else { - gen_render_tiles(); - } + if (!in->close()) { + LOG(ERROR) << "Error closing tile file " << in->geterror(); + return false; } return true; } -int TileManager::get_num_effective_samples() -{ - return (range_num_samples == -1) ? num_samples : range_num_samples; -} - CCL_NAMESPACE_END diff --git a/intern/cycles/render/tile.h b/intern/cycles/render/tile.h index 790a56f9445..124d0b3652c 100644 --- a/intern/cycles/render/tile.h +++ b/intern/cycles/render/tile.h @@ -14,159 +14,151 @@ * limitations under the License. */ -#ifndef __TILE_H__ -#define __TILE_H__ - -#include <limits.h> +#pragma once #include "render/buffers.h" -#include "util/util_list.h" +#include "util/util_image.h" +#include "util/util_string.h" +#include "util/util_unique_ptr.h" CCL_NAMESPACE_BEGIN -/* Tile */ +class DenoiseParams; +class Scene; + +/* -------------------------------------------------------------------- + * Tile. + */ class Tile { public: - int index; - int x, y, w, h; - int device; - /* RENDER: The tile has to be rendered. - * RENDERED: The tile has been rendered, but can't be denoised yet (waiting for neighbors). - * DENOISE: The tile can be denoised now. - * DENOISED: The tile has been denoised, but can't be freed yet (waiting for neighbors). - * DONE: The tile is finished and has been freed. */ - typedef enum { RENDER = 0, RENDERED, DENOISE, DENOISED, DONE } State; - State state; - RenderBuffers *buffers; + int x = 0, y = 0; + int width = 0, height = 0; Tile() { } - - Tile(int index_, int x_, int y_, int w_, int h_, int device_, State state_ = RENDER) - : index(index_), x(x_), y(y_), w(w_), h(h_), device(device_), state(state_), buffers(NULL) - { - } }; -/* Tile order */ - -/* Note: this should match enum_tile_order in properties.py */ -enum TileOrder { - TILE_CENTER = 0, - TILE_RIGHT_TO_LEFT = 1, - TILE_LEFT_TO_RIGHT = 2, - TILE_TOP_TO_BOTTOM = 3, - TILE_BOTTOM_TO_TOP = 4, - TILE_HILBERT_SPIRAL = 5, -}; - -/* Tile Manager */ +/* -------------------------------------------------------------------- + * Tile Manager. + */ class TileManager { public: - BufferParams params; - - struct State { - vector<Tile> tiles; - int tile_stride; - BufferParams buffer; - int sample; - int num_samples; - int resolution_divider; - int num_tiles; - - /* Total samples over all pixels: Generally num_samples*num_pixels, - * but can be higher due to the initial resolution division for previews. */ - uint64_t total_pixel_samples; - - /* These lists contain the indices of the tiles to be rendered/denoised and are used - * when acquiring a new tile for the device. - * Each list in each vector is for one logical device. */ - vector<list<int>> render_tiles; - vector<list<int>> denoising_tiles; - } state; - - int num_samples; - int slice_overlap; - - TileManager(bool progressive, - int num_samples, - int2 tile_size, - int start_resolution, - bool preserve_tile_device, - bool background, - TileOrder tile_order, - int num_devices = 1, - int pixel_size = 1); + /* This callback is invoked by whenever on-dist tiles storage file is closed after writing. */ + function<void(string_view)> full_buffer_written_cb; + + TileManager(); ~TileManager(); - void device_free(); - void reset(BufferParams ¶ms, int num_samples); - void set_samples(int num_samples); + TileManager(const TileManager &other) = delete; + TileManager(TileManager &&other) noexcept = delete; + TileManager &operator=(const TileManager &other) = delete; + TileManager &operator=(TileManager &&other) = delete; + + /* Reset current progress and start new rendering of the full-frame parameters in tiles of the + * given size. + * Only touches scheduling-related state of the tile manager. */ + /* TODO(sergey): Consider using tile area instead of exact size to help dealing with extreme + * cases of stretched renders. */ + void reset_scheduling(const BufferParams ¶ms, int2 tile_size); + + /* Update for the known buffer passes and scene parameters. + * Will store all parameters needed for buffers access outside of the scene graph. */ + void update(const BufferParams ¶ms, const Scene *scene); + + inline int get_num_tiles() const + { + return tile_state_.num_tiles; + } + + inline bool has_multiple_tiles() const + { + return tile_state_.num_tiles > 1; + } + bool next(); - bool next_tile(Tile *&tile, int device, uint tile_types); - bool finish_tile(const int index, const bool need_denoise, bool &delete_tile); bool done(); - bool has_tiles(); - void set_tile_order(TileOrder tile_order_) + const Tile &get_current_tile() const; + + /* Write render buffer of a tile to a file on disk. + * + * Opens file for write when first tile is written. + * + * Returns true on success. */ + bool write_tile(const RenderBuffers &tile_buffers); + + /* Inform the tile manager that no more tiles will be written to disk. + * The file will be considered final, all handles to it will be closed. */ + void finish_write_tiles(); + + /* Check whether any tile ahs been written to disk. */ + inline bool has_written_tiles() const { - tile_order = tile_order_; + return write_state_.num_tiles_written != 0; } - int get_neighbor_index(int index, int neighbor); - bool check_neighbor_state(int index, Tile::State state); + /* Read full frame render buffer from tiles file on disk. + * + * Returns true on success. */ + bool read_full_buffer_from_disk(string_view filename, + RenderBuffers *buffers, + DenoiseParams *denoise_params); - /* ** Sample range rendering. ** */ + protected: + /* Get tile configuration for its index. + * The tile index must be within [0, state_.tile_state_). */ + Tile get_tile_for_index(int index) const; - /* Start sample in the range. */ - int range_start_sample; + bool open_tile_output(); + bool close_tile_output(); - /* Number to samples in the rendering range. */ - int range_num_samples; + /* Part of an on-disk tile file name which avoids conflicts between several Cycles instances or + * several sessions. */ + string tile_file_unique_part_; - /* Get number of actual samples to render. */ - int get_num_effective_samples(); + int2 tile_size_ = make_int2(0, 0); - /* Schedule tiles for denoising after they've been rendered. */ - bool schedule_denoising; + BufferParams buffer_params_; - protected: - void set_tiles(); - - bool progressive; - int2 tile_size; - TileOrder tile_order; - int start_resolution; - int pixel_size; - int num_devices; - - /* in some cases it is important that the same tile will be returned for the same - * device it was originally generated for (i.e. viewport rendering when buffer is - * allocating once for tile and then always used by it) - * - * in other cases any tile could be handled by any device (i.e. final rendering - * without progressive refine) - */ - bool preserve_tile_device; - - /* for background render tiles should exactly match render parts generated from - * blender side, which means image first gets split into tiles and then tiles are - * assigning to render devices - * - * however viewport rendering expects tiles to be allocated in a special way, - * meaning image is being sliced horizontally first and every device handles - * its own slice - */ - bool background; - - /* Generate tile list, return number of tiles. */ - int gen_tiles(bool sliced); - void gen_render_tiles(); + /* Tile scheduling state. */ + struct { + int num_tiles_x = 0; + int num_tiles_y = 0; + int num_tiles = 0; + + int next_tile_index; + + Tile current_tile; + } tile_state_; + + /* State of tiles writing to a file on disk. */ + struct { + /* Index of a tile file used during the current session. + * This number is used for the file name construction, making it possible to render several + * scenes throughout duration of the session and keep all results available for later read + * access. */ + int tile_file_index = 0; + + string filename; + + /* Specification of the tile image which corresponds to the buffer parameters. + * Contains channels configured according to the passes configuration in the path traces. + * + * Output images are saved using this specification, input images are expected to have matched + * specification. */ + ImageSpec image_spec; + + /* Output handle for the tile file. + * + * This file can not be closed until all tiles has been provided, so the handle is stored in + * the state and is created whenever writing is requested. */ + unique_ptr<ImageOutput> tile_out; + + int num_tiles_written = 0; + } write_state_; }; CCL_NAMESPACE_END - -#endif /* __TILE_H__ */ |