diff options
Diffstat (limited to 'servers/rendering/renderer_rd')
6 files changed, 698 insertions, 29 deletions
diff --git a/servers/rendering/renderer_rd/renderer_scene_render_forward.cpp b/servers/rendering/renderer_rd/renderer_scene_render_forward.cpp index 5412688e3f..9de7af3c22 100644 --- a/servers/rendering/renderer_rd/renderer_scene_render_forward.cpp +++ b/servers/rendering/renderer_rd/renderer_scene_render_forward.cpp @@ -1079,7 +1079,11 @@ void RendererSceneRenderForward::_render_list(RenderingDevice::DrawListID p_draw switch (e->instance->base_type) { case RS::INSTANCE_MESH: { - storage->mesh_surface_get_arrays_and_format(e->instance->base, e->surface_index, pipeline->get_vertex_input_mask(), vertex_array_rd, index_array_rd, vertex_format); + if (e->instance->mesh_instance.is_valid()) { //skeleton and blend shape + storage->mesh_instance_surface_get_arrays_and_format(e->instance->mesh_instance, e->surface_index, pipeline->get_vertex_input_mask(), vertex_array_rd, index_array_rd, vertex_format); + } else { + storage->mesh_surface_get_arrays_and_format(e->instance->base, e->surface_index, pipeline->get_vertex_input_mask(), vertex_array_rd, index_array_rd, vertex_format); + } } break; case RS::INSTANCE_MULTIMESH: { RID mesh = storage->multimesh_get_mesh(e->instance->base); @@ -2500,7 +2504,7 @@ RID RendererSceneRenderForward::_setup_render_pass_uniform_set(RID p_render_buff RD::Uniform u; u.binding = 4; u.uniform_type = RD::UNIFORM_TYPE_TEXTURE; - RID texture = false && rb && rb->depth.is_valid() ? rb->depth : storage->texture_rd_get_default(RendererStorageRD::DEFAULT_RD_TEXTURE_WHITE); + RID texture = rb && rb->depth.is_valid() ? rb->depth : storage->texture_rd_get_default(RendererStorageRD::DEFAULT_RD_TEXTURE_WHITE); u.ids.push_back(texture); uniforms.push_back(u); } @@ -2578,7 +2582,7 @@ RID RendererSceneRenderForward::_setup_render_pass_uniform_set(RID p_render_buff RD::Uniform u; u.binding = 12; u.uniform_type = RD::UNIFORM_TYPE_UNIFORM_BUFFER; - u.ids.push_back(rb ? render_buffers_get_default_gi_probe_buffer() : render_buffers_get_gi_probe_buffer(p_render_buffers)); + u.ids.push_back(rb ? render_buffers_get_gi_probe_buffer(p_render_buffers) : render_buffers_get_default_gi_probe_buffer()); uniforms.push_back(u); } { diff --git a/servers/rendering/renderer_rd/renderer_storage_rd.cpp b/servers/rendering/renderer_rd/renderer_storage_rd.cpp index 564d61f9fb..286281c83d 100644 --- a/servers/rendering/renderer_rd/renderer_storage_rd.cpp +++ b/servers/rendering/renderer_rd/renderer_storage_rd.cpp @@ -2392,13 +2392,23 @@ RID RendererStorageRD::mesh_create() { return mesh_owner.make_rid(Mesh()); } +void RendererStorageRD::mesh_set_blend_shape_count(RID p_mesh, int p_blend_shape_count) { + ERR_FAIL_COND(p_blend_shape_count < 0); + + Mesh *mesh = mesh_owner.getornull(p_mesh); + ERR_FAIL_COND(!mesh); + + ERR_FAIL_COND(mesh->surface_count > 0); //surfaces already exist + + mesh->blend_shape_count = p_blend_shape_count; +} + /// Returns stride void RendererStorageRD::mesh_add_surface(RID p_mesh, const RS::SurfaceData &p_surface) { Mesh *mesh = mesh_owner.getornull(p_mesh); ERR_FAIL_COND(!mesh); //ensure blend shape consistency - ERR_FAIL_COND(mesh->blend_shape_count && p_surface.blend_shape_count != mesh->blend_shape_count); ERR_FAIL_COND(mesh->blend_shape_count && p_surface.bone_aabbs.size() != mesh->bone_aabbs.size()); #ifdef DEBUG_ENABLED @@ -2453,7 +2463,7 @@ void RendererStorageRD::mesh_add_surface(RID p_mesh, const RS::SurfaceData &p_su case RS::ARRAY_BONES: { //uses a separate array bool use_8 = p_surface.format & RS::ARRAY_FLAG_USE_8_BONE_WEIGHTS; - skin_stride += sizeof(int16_t) * (use_8 ? 8 : 4); + skin_stride += sizeof(int16_t) * (use_8 ? 16 : 8); } break; } } @@ -2461,6 +2471,11 @@ void RendererStorageRD::mesh_add_surface(RID p_mesh, const RS::SurfaceData &p_su int expected_size = stride * p_surface.vertex_count; ERR_FAIL_COND_MSG(expected_size != p_surface.vertex_data.size(), "Size of vertex data provided (" + itos(p_surface.vertex_data.size()) + ") does not match expected (" + itos(expected_size) + ")"); + + int bs_expected_size = expected_size * mesh->blend_shape_count; + + ERR_FAIL_COND_MSG(bs_expected_size != p_surface.blend_shape_data.size(), "Size of blend shape data provided (" + itos(p_surface.blend_shape_data.size()) + ") does not match expected (" + itos(bs_expected_size) + ")"); + int expected_attrib_size = attrib_stride * p_surface.vertex_count; ERR_FAIL_COND_MSG(expected_attrib_size != p_surface.attribute_data.size(), "Size of attribute data provided (" + itos(p_surface.attribute_data.size()) + ") does not match expected (" + itos(expected_attrib_size) + ")"); @@ -2477,15 +2492,25 @@ void RendererStorageRD::mesh_add_surface(RID p_mesh, const RS::SurfaceData &p_su s->format = p_surface.format; s->primitive = p_surface.primitive; - s->vertex_buffer = RD::get_singleton()->vertex_buffer_create(p_surface.vertex_data.size(), p_surface.vertex_data); + bool use_as_storage = (p_surface.skin_data.size() || mesh->blend_shape_count > 0); + + s->vertex_buffer = RD::get_singleton()->vertex_buffer_create(p_surface.vertex_data.size(), p_surface.vertex_data, use_as_storage); + s->vertex_buffer_size = p_surface.vertex_data.size(); + if (p_surface.attribute_data.size()) { s->attribute_buffer = RD::get_singleton()->vertex_buffer_create(p_surface.attribute_data.size(), p_surface.attribute_data); } if (p_surface.skin_data.size()) { - s->skin_buffer = RD::get_singleton()->vertex_buffer_create(p_surface.skin_data.size(), p_surface.skin_data); + s->skin_buffer = RD::get_singleton()->vertex_buffer_create(p_surface.skin_data.size(), p_surface.skin_data, use_as_storage); + s->skin_buffer_size = p_surface.skin_data.size(); } + s->vertex_count = p_surface.vertex_count; + if (p_surface.format & RS::ARRAY_FORMAT_BONES) { + mesh->has_bone_weights = true; + } + if (p_surface.index_count) { bool is_index_16 = p_surface.vertex_count <= 65536; @@ -2507,17 +2532,45 @@ void RendererStorageRD::mesh_add_surface(RID p_mesh, const RS::SurfaceData &p_su s->aabb = p_surface.aabb; s->bone_aabbs = p_surface.bone_aabbs; //only really useful for returning them. -#if 0 - for (int i = 0; i < p_surface.blend_shapes.size(); i++) { - if (p_surface.blend_shapes[i].size() != p_surface.vertex_data.size()) { - memdelete(s); - ERR_FAIL_COND(p_surface.blend_shapes[i].size() != p_surface.vertex_data.size()); + + if (mesh->blend_shape_count > 0) { + s->blend_shape_buffer = RD::get_singleton()->storage_buffer_create(p_surface.blend_shape_data.size(), p_surface.blend_shape_data); + } + + if (use_as_storage) { + Vector<RD::Uniform> uniforms; + { + RD::Uniform u; + u.binding = 0; + u.uniform_type = RD::UNIFORM_TYPE_STORAGE_BUFFER; + u.ids.push_back(s->vertex_buffer); + uniforms.push_back(u); + } + { + RD::Uniform u; + u.binding = 1; + u.uniform_type = RD::UNIFORM_TYPE_STORAGE_BUFFER; + if (s->skin_buffer.is_valid()) { + u.ids.push_back(s->skin_buffer); + } else { + u.ids.push_back(default_rd_storage_buffer); + } + uniforms.push_back(u); } - RID vertex_buffer = RD::get_singleton()->vertex_buffer_create(p_surface.blend_shapes[i].size(), p_surface.blend_shapes[i]); - s->blend_shapes.push_back(vertex_buffer); + { + RD::Uniform u; + u.binding = 2; + u.uniform_type = RD::UNIFORM_TYPE_STORAGE_BUFFER; + if (s->blend_shape_buffer.is_valid()) { + u.ids.push_back(s->blend_shape_buffer); + } else { + u.ids.push_back(default_rd_storage_buffer); + } + uniforms.push_back(u); + } + + s->uniform_set = RD::get_singleton()->uniform_set_create(uniforms, skeleton_shader.version_shader[0], SkeletonShader::UNIFORM_SET_SURFACE); } -#endif - mesh->blend_shape_count = p_surface.blend_shape_count; if (mesh->surface_count == 0) { mesh->bone_aabbs = p_surface.bone_aabbs; @@ -2535,6 +2588,12 @@ void RendererStorageRD::mesh_add_surface(RID p_mesh, const RS::SurfaceData &p_su mesh->surfaces[mesh->surface_count] = s; mesh->surface_count++; + for (List<MeshInstance *>::Element *E = mesh->instances.front(); E; E = E->next()) { + //update instances + MeshInstance *mi = E->get(); + _mesh_instance_add_surface(mi, mesh, mesh->surface_count - 1); + } + mesh->instance_dependency.instance_notify_changed(true, true); mesh->material_cache.clear(); @@ -2792,16 +2851,223 @@ void RendererStorageRD::mesh_clear(RID p_mesh) { mesh->surfaces = nullptr; mesh->surface_count = 0; mesh->material_cache.clear(); + //clear instance data + for (List<MeshInstance *>::Element *E = mesh->instances.front(); E; E = E->next()) { + MeshInstance *mi = E->get(); + _mesh_instance_clear(mi); + } mesh->instance_dependency.instance_notify_changed(true, true); + mesh->has_bone_weights = false; +} + +bool RendererStorageRD::mesh_needs_instance(RID p_mesh, bool p_has_skeleton) { + Mesh *mesh = mesh_owner.getornull(p_mesh); + ERR_FAIL_COND_V(!mesh, false); + + return mesh->blend_shape_count > 0 || (mesh->has_bone_weights && p_has_skeleton); +} + +/* MESH INSTANCE */ + +RID RendererStorageRD::mesh_instance_create(RID p_base) { + Mesh *mesh = mesh_owner.getornull(p_base); + ERR_FAIL_COND_V(!mesh, RID()); + + MeshInstance *mi = memnew(MeshInstance); + + mi->mesh = mesh; + + for (uint32_t i = 0; i < mesh->surface_count; i++) { + _mesh_instance_add_surface(mi, mesh, i); + } + + mi->I = mesh->instances.push_back(mi); + + mi->dirty = true; + + return mesh_instance_owner.make_rid(mi); +} +void RendererStorageRD::mesh_instance_set_skeleton(RID p_mesh_instance, RID p_skeleton) { + MeshInstance *mi = mesh_instance_owner.getornull(p_mesh_instance); + if (mi->skeleton == p_skeleton) { + return; + } + mi->skeleton = p_skeleton; + mi->skeleton_version = 0; + mi->dirty = true; } -void RendererStorageRD::_mesh_surface_generate_version_for_input_mask(Mesh::Surface *s, uint32_t p_input_mask) { - uint32_t version = s->version_count; - s->version_count++; - s->versions = (Mesh::Surface::Version *)memrealloc(s->versions, sizeof(Mesh::Surface::Version) * s->version_count); +void RendererStorageRD::mesh_instance_set_blend_shape_weight(RID p_mesh_instance, int p_shape, float p_weight) { + MeshInstance *mi = mesh_instance_owner.getornull(p_mesh_instance); + ERR_FAIL_COND(!mi); + ERR_FAIL_INDEX(p_shape, (int)mi->blend_weights.size()); + mi->blend_weights[p_shape] = p_weight; + mi->weights_dirty = true; + //will be eventually updated +} + +void RendererStorageRD::_mesh_instance_clear(MeshInstance *mi) { + for (uint32_t i = 0; i < mi->surfaces.size(); i++) { + if (mi->surfaces[i].vertex_buffer.is_valid()) { + RD::get_singleton()->free(mi->surfaces[i].vertex_buffer); + } + if (mi->surfaces[i].versions) { + for (uint32_t j = 0; j < mi->surfaces[i].version_count; j++) { + RD::get_singleton()->free(mi->surfaces[i].versions[j].vertex_array); + } + memfree(mi->surfaces[i].versions); + } + } + mi->surfaces.clear(); + + if (mi->blend_weights_buffer.is_valid()) { + RD::get_singleton()->free(mi->blend_weights_buffer); + } + mi->blend_weights.clear(); + mi->weights_dirty = false; + mi->skeleton_version = 0; +} + +void RendererStorageRD::_mesh_instance_add_surface(MeshInstance *mi, Mesh *mesh, uint32_t p_surface) { + if (mesh->blend_shape_count > 0 && mi->blend_weights_buffer.is_null()) { + mi->blend_weights.resize(mesh->blend_shape_count); + for (uint32_t i = 0; i < mi->blend_weights.size(); i++) { + mi->blend_weights[i] = 0; + } + mi->blend_weights_buffer = RD::get_singleton()->storage_buffer_create(sizeof(float) * mi->blend_weights.size(), mi->blend_weights.to_byte_array()); + mi->weights_dirty = true; + } - Mesh::Surface::Version &v = s->versions[version]; + MeshInstance::Surface s; + if (mesh->blend_shape_count > 0 || (mesh->surfaces[p_surface]->format & RS::ARRAY_FORMAT_BONES)) { + //surface warrants transform + s.vertex_buffer = RD::get_singleton()->vertex_buffer_create(mesh->surfaces[p_surface]->vertex_buffer_size, Vector<uint8_t>(), true); + Vector<RD::Uniform> uniforms; + { + RD::Uniform u; + u.binding = 1; + u.uniform_type = RD::UNIFORM_TYPE_STORAGE_BUFFER; + u.ids.push_back(s.vertex_buffer); + uniforms.push_back(u); + } + { + RD::Uniform u; + u.binding = 2; + u.uniform_type = RD::UNIFORM_TYPE_STORAGE_BUFFER; + if (mi->blend_weights_buffer.is_valid()) { + u.ids.push_back(mi->blend_weights_buffer); + } else { + u.ids.push_back(default_rd_storage_buffer); + } + uniforms.push_back(u); + } + s.uniform_set = RD::get_singleton()->uniform_set_create(uniforms, skeleton_shader.version_shader[0], SkeletonShader::UNIFORM_SET_INSTANCE); + } + + mi->surfaces.push_back(s); + mi->dirty = true; +} + +void RendererStorageRD::mesh_instance_check_for_update(RID p_mesh_instance) { + MeshInstance *mi = mesh_instance_owner.getornull(p_mesh_instance); + + bool needs_update = mi->dirty; + + if (mi->weights_dirty && !mi->weight_update_list.in_list()) { + dirty_mesh_instance_weights.add(&mi->weight_update_list); + needs_update = true; + } + + if (mi->array_update_list.in_list()) { + return; + } + + if (!needs_update && mi->skeleton.is_valid()) { + Skeleton *sk = skeleton_owner.getornull(mi->skeleton); + if (sk && sk->version != mi->skeleton_version) { + needs_update = true; + } + } + + if (needs_update) { + dirty_mesh_instance_arrays.add(&mi->array_update_list); + } +} + +void RendererStorageRD::update_mesh_instances() { + while (dirty_mesh_instance_weights.first()) { + MeshInstance *mi = dirty_mesh_instance_weights.first()->self(); + + if (mi->blend_weights_buffer.is_valid()) { + RD::get_singleton()->buffer_update(mi->blend_weights_buffer, 0, mi->blend_weights.size() * sizeof(float), mi->blend_weights.ptr(), true); + } + dirty_mesh_instance_weights.remove(&mi->weight_update_list); + mi->weights_dirty = false; + } + if (dirty_mesh_instance_arrays.first() == nullptr) { + return; //nothing to do + } + + //process skeletons and blend shapes + RD::ComputeListID compute_list = RD::get_singleton()->compute_list_begin(); + + while (dirty_mesh_instance_arrays.first()) { + MeshInstance *mi = dirty_mesh_instance_arrays.first()->self(); + + Skeleton *sk = skeleton_owner.getornull(mi->skeleton); + + for (uint32_t i = 0; i < mi->surfaces.size(); i++) { + if (mi->surfaces[i].uniform_set == RID() || mi->mesh->surfaces[i]->uniform_set == RID()) { + continue; + } + + bool array_is_2d = mi->mesh->surfaces[i]->format & RS::ARRAY_FLAG_USE_2D_VERTICES; + + RD::get_singleton()->compute_list_bind_compute_pipeline(compute_list, skeleton_shader.pipeline[array_is_2d ? SkeletonShader::SHADER_MODE_2D : SkeletonShader::SHADER_MODE_3D]); + + RD::get_singleton()->compute_list_bind_uniform_set(compute_list, mi->surfaces[i].uniform_set, SkeletonShader::UNIFORM_SET_INSTANCE); + RD::get_singleton()->compute_list_bind_uniform_set(compute_list, mi->mesh->surfaces[i]->uniform_set, SkeletonShader::UNIFORM_SET_SURFACE); + if (sk && sk->uniform_set_mi.is_valid()) { + RD::get_singleton()->compute_list_bind_uniform_set(compute_list, sk->uniform_set_mi, SkeletonShader::UNIFORM_SET_SKELETON); + } else { + RD::get_singleton()->compute_list_bind_uniform_set(compute_list, skeleton_shader.default_skeleton_uniform_set, SkeletonShader::UNIFORM_SET_SKELETON); + } + + SkeletonShader::PushConstant push_constant; + + push_constant.has_normal = mi->mesh->surfaces[i]->format & RS::ARRAY_FORMAT_NORMAL; + push_constant.has_tangent = mi->mesh->surfaces[i]->format & RS::ARRAY_FORMAT_TANGENT; + push_constant.has_skeleton = sk != nullptr && sk->use_2d == array_is_2d && (mi->mesh->surfaces[i]->format & RS::ARRAY_FORMAT_BONES); + push_constant.has_blend_shape = mi->mesh->blend_shape_count > 0; + + push_constant.vertex_count = mi->mesh->surfaces[i]->vertex_count; + push_constant.vertex_stride = (mi->mesh->surfaces[i]->vertex_buffer_size / mi->mesh->surfaces[i]->vertex_count) / 4; + push_constant.skin_stride = (mi->mesh->surfaces[i]->skin_buffer_size / mi->mesh->surfaces[i]->vertex_count) / 4; + push_constant.skin_weight_offset = (mi->mesh->surfaces[i]->format & RS::ARRAY_FLAG_USE_8_BONE_WEIGHTS) ? 4 : 2; + + push_constant.blend_shape_count = mi->mesh->blend_shape_count; + push_constant.normalized_blend_shapes = mi->mesh->blend_shape_mode == RS::BLEND_SHAPE_MODE_NORMALIZED; + push_constant.pad0 = 0; + push_constant.pad1 = 0; + + RD::get_singleton()->compute_list_set_push_constant(compute_list, &push_constant, sizeof(SkeletonShader::PushConstant)); + + //dispatch without barrier, so all is done at the same time + RD::get_singleton()->compute_list_dispatch_threads(compute_list, push_constant.vertex_count, 1, 1, 64, 1, 1); + } + + mi->dirty = false; + if (sk) { + mi->skeleton_version = sk->version; + } + dirty_mesh_instance_arrays.remove(&mi->array_update_list); + } + + RD::get_singleton()->compute_list_end(); +} + +void RendererStorageRD::_mesh_surface_generate_version_for_input_mask(Mesh::Surface::Version &v, Mesh::Surface *s, uint32_t p_input_mask, MeshInstance::Surface *mis) { Vector<RD::VertexAttribute> attributes; Vector<RID> buffers; @@ -2873,7 +3139,11 @@ void RendererStorageRD::_mesh_surface_generate_version_for_input_mask(Mesh::Surf stride += sizeof(float) * 3; } - buffer = s->vertex_buffer; + if (mis) { + buffer = mis->vertex_buffer; + } else { + buffer = s->vertex_buffer; + } } break; case RS::ARRAY_NORMAL: { @@ -2882,14 +3152,22 @@ void RendererStorageRD::_mesh_surface_generate_version_for_input_mask(Mesh::Surf vd.format = RD::DATA_FORMAT_A2B10G10R10_UNORM_PACK32; stride += sizeof(uint32_t); - buffer = s->vertex_buffer; + if (mis) { + buffer = mis->vertex_buffer; + } else { + buffer = s->vertex_buffer; + } } break; case RS::ARRAY_TANGENT: { vd.offset = stride; vd.format = RD::DATA_FORMAT_A2B10G10R10_UNORM_PACK32; stride += sizeof(uint32_t); - buffer = s->vertex_buffer; + if (mis) { + buffer = mis->vertex_buffer; + } else { + buffer = s->vertex_buffer; + } } break; case RS::ARRAY_COLOR: { vd.offset = attribute_stride; @@ -4847,6 +5125,7 @@ void RendererStorageRD::skeleton_allocate(RID p_skeleton, int p_bones, bool p_2d RD::get_singleton()->free(skeleton->buffer); skeleton->buffer = RID(); skeleton->data.resize(0); + skeleton->uniform_set_mi = RID(); } if (skeleton->size) { @@ -4855,6 +5134,18 @@ void RendererStorageRD::skeleton_allocate(RID p_skeleton, int p_bones, bool p_2d zeromem(skeleton->data.ptrw(), skeleton->data.size() * sizeof(float)); _skeleton_make_dirty(skeleton); + + { + Vector<RD::Uniform> uniforms; + { + RD::Uniform u; + u.binding = 0; + u.uniform_type = RD::UNIFORM_TYPE_STORAGE_BUFFER; + u.ids.push_back(skeleton->buffer); + uniforms.push_back(u); + } + skeleton->uniform_set_mi = RD::get_singleton()->uniform_set_create(uniforms, skeleton_shader.version_shader[0], SkeletonShader::UNIFORM_SET_SKELETON); + } } } @@ -4977,6 +5268,7 @@ void RendererStorageRD::_update_dirty_skeletons() { skeleton_dirty_list = skeleton->dirty_list; skeleton->instance_dependency.instance_notify_changed(true, false); + skeleton->version++; skeleton->dirty = false; skeleton->dirty_list = nullptr; @@ -7810,7 +8102,18 @@ bool RendererStorageRD::free(RID p_rid) { mesh_clear(p_rid); Mesh *mesh = mesh_owner.getornull(p_rid); mesh->instance_dependency.instance_notify_deleted(p_rid); + if (mesh->instances.size()) { + ERR_PRINT("deleting mesh with active instances"); + } mesh_owner.free(p_rid); + } else if (mesh_instance_owner.owns(p_rid)) { + MeshInstance *mi = mesh_instance_owner.getornull(p_rid); + _mesh_instance_clear(mi); + mi->mesh->instances.erase(mi->I); + mi->I = nullptr; + mesh_instance_owner.free(p_rid); + memdelete(mi); + } else if (multimesh_owner.owns(p_rid)) { _update_dirty_multimeshes(); multimesh_allocate(p_rid, 0, RS::MULTIMESH_TRANSFORM_2D); @@ -8517,6 +8820,30 @@ RendererStorageRD::RendererStorageRD() { rt_sdf.pipelines[i] = RD::get_singleton()->compute_pipeline_create(rt_sdf.shader.version_get_shader(rt_sdf.shader_version, i)); } } + { + Vector<String> skeleton_modes; + skeleton_modes.push_back("\n#define MODE_2D\n"); + skeleton_modes.push_back(""); + + skeleton_shader.shader.initialize(skeleton_modes); + skeleton_shader.version = skeleton_shader.shader.version_create(); + for (int i = 0; i < SkeletonShader::SHADER_MODE_MAX; i++) { + skeleton_shader.version_shader[i] = skeleton_shader.shader.version_get_shader(skeleton_shader.version, i); + skeleton_shader.pipeline[i] = RD::get_singleton()->compute_pipeline_create(skeleton_shader.version_shader[i]); + } + + { + Vector<RD::Uniform> uniforms; + { + RD::Uniform u; + u.binding = 0; + u.uniform_type = RD::UNIFORM_TYPE_STORAGE_BUFFER; + u.ids.push_back(default_rd_storage_buffer); + uniforms.push_back(u); + } + skeleton_shader.default_skeleton_uniform_set = RD::get_singleton()->uniform_set_create(uniforms, skeleton_shader.version_shader[0], SkeletonShader::UNIFORM_SET_SKELETON); + } + } } RendererStorageRD::~RendererStorageRD() { @@ -8546,6 +8873,8 @@ RendererStorageRD::~RendererStorageRD() { particles_shader.copy_shader.version_free(particles_shader.copy_shader_version); rt_sdf.shader.version_free(rt_sdf.shader_version); + skeleton_shader.shader.version_free(skeleton_shader.version); + RenderingServer::get_singleton()->free(particles_shader.default_material); RenderingServer::get_singleton()->free(particles_shader.default_shader); diff --git a/servers/rendering/renderer_rd/renderer_storage_rd.h b/servers/rendering/renderer_rd/renderer_storage_rd.h index b6a26fc9d0..ee4cf6da1a 100644 --- a/servers/rendering/renderer_rd/renderer_storage_rd.h +++ b/servers/rendering/renderer_rd/renderer_storage_rd.h @@ -31,6 +31,8 @@ #ifndef RENDERING_SERVER_STORAGE_RD_H #define RENDERING_SERVER_STORAGE_RD_H +#include "core/templates/list.h" +#include "core/templates/local_vector.h" #include "core/templates/rid_owner.h" #include "servers/rendering/renderer_compositor.h" #include "servers/rendering/renderer_rd/effects_rd.h" @@ -39,9 +41,9 @@ #include "servers/rendering/renderer_rd/shaders/giprobe_sdf.glsl.gen.h" #include "servers/rendering/renderer_rd/shaders/particles.glsl.gen.h" #include "servers/rendering/renderer_rd/shaders/particles_copy.glsl.gen.h" +#include "servers/rendering/renderer_rd/shaders/skeleton.glsl.gen.h" #include "servers/rendering/renderer_scene_render.h" #include "servers/rendering/rendering_device.h" - class RendererStorageRD : public RendererStorage { public: static _FORCE_INLINE_ void store_transform(const Transform &p_mtx, float *p_array) { @@ -377,6 +379,8 @@ private: /* Mesh */ + struct MeshInstance; + struct Mesh { struct Surface { RS::PrimitiveType primitive = RS::PRIMITIVE_POINTS; @@ -386,6 +390,8 @@ private: RID attribute_buffer; RID skin_buffer; uint32_t vertex_count = 0; + uint32_t vertex_buffer_size = 0; + uint32_t skin_buffer_size = 0; // A different pipeline needs to be allocated // depending on the inputs available in the @@ -433,6 +439,8 @@ private: uint32_t particles_render_index = 0; uint64_t particles_render_pass = 0; + + RID uniform_set; }; uint32_t blend_shape_count = 0; @@ -443,17 +451,90 @@ private: Vector<AABB> bone_aabbs; + bool has_bone_weights = false; + AABB aabb; AABB custom_aabb; Vector<RID> material_cache; + List<MeshInstance *> instances; + RendererStorage::InstanceDependency instance_dependency; }; mutable RID_Owner<Mesh> mesh_owner; - void _mesh_surface_generate_version_for_input_mask(Mesh::Surface *s, uint32_t p_input_mask); + struct MeshInstance { + Mesh *mesh; + RID skeleton; + struct Surface { + RID vertex_buffer; + RID uniform_set; + + Mesh::Surface::Version *versions = nullptr; //allocated on demand + uint32_t version_count = 0; + }; + LocalVector<Surface> surfaces; + LocalVector<float> blend_weights; + + RID blend_weights_buffer; + List<MeshInstance *>::Element *I = nullptr; //used to erase itself + uint64_t skeleton_version = 0; + bool dirty = false; + bool weights_dirty = false; + SelfList<MeshInstance> weight_update_list; + SelfList<MeshInstance> array_update_list; + MeshInstance() : + weight_update_list(this), array_update_list(this) {} + }; + + void _mesh_instance_clear(MeshInstance *mi); + void _mesh_instance_add_surface(MeshInstance *mi, Mesh *mesh, uint32_t p_surface); + + mutable RID_PtrOwner<MeshInstance> mesh_instance_owner; + + SelfList<MeshInstance>::List dirty_mesh_instance_weights; + SelfList<MeshInstance>::List dirty_mesh_instance_arrays; + + struct SkeletonShader { + struct PushConstant { + uint32_t has_normal; + uint32_t has_tangent; + uint32_t has_skeleton; + uint32_t has_blend_shape; + + uint32_t vertex_count; + uint32_t vertex_stride; + uint32_t skin_stride; + uint32_t skin_weight_offset; + + uint32_t blend_shape_count; + uint32_t normalized_blend_shapes; + uint32_t pad0; + uint32_t pad1; + }; + + enum { + UNIFORM_SET_INSTANCE = 0, + UNIFORM_SET_SURFACE = 1, + UNIFORM_SET_SKELETON = 2, + }; + enum { + SHADER_MODE_2D, + SHADER_MODE_3D, + SHADER_MODE_MAX + }; + + SkeletonShaderRD shader; + RID version; + RID version_shader[SHADER_MODE_MAX]; + RID pipeline[SHADER_MODE_MAX]; + + RID default_skeleton_uniform_set; + } skeleton_shader; + + void _mesh_surface_generate_version_for_input_mask(Mesh::Surface::Version &v, Mesh::Surface *s, uint32_t p_input_mask, MeshInstance::Surface *mis = nullptr); RID mesh_default_rd_buffers[DEFAULT_RD_BUFFER_MAX]; @@ -826,6 +907,9 @@ private: Transform2D base_transform_2d; RID uniform_set_3d; + RID uniform_set_mi; + + uint64_t version = 1; RendererStorage::InstanceDependency instance_dependency; }; @@ -1280,6 +1364,8 @@ public: virtual RID mesh_create(); + virtual void mesh_set_blend_shape_count(RID p_mesh, int p_blend_shape_count); + /// Return stride virtual void mesh_add_surface(RID p_mesh, const RS::SurfaceData &p_surface); @@ -1304,6 +1390,16 @@ public: virtual void mesh_clear(RID p_mesh); + virtual bool mesh_needs_instance(RID p_mesh, bool p_has_skeleton); + + /* MESH INSTANCE */ + + virtual RID mesh_instance_create(RID p_base); + virtual void mesh_instance_set_skeleton(RID p_mesh_instance, RID p_skeleton); + virtual void mesh_instance_set_blend_shape_weight(RID p_mesh_instance, int p_shape, float p_weight); + virtual void mesh_instance_check_for_update(RID p_mesh_instance); + virtual void update_mesh_instances(); + _FORCE_INLINE_ const RID *mesh_get_surface_count_and_materials(RID p_mesh, uint32_t &r_surface_count) { Mesh *mesh = mesh_owner.getornull(p_mesh); ERR_FAIL_COND_V(!mesh, nullptr); @@ -1353,9 +1449,11 @@ public: return; } - uint32_t version = s->version_count; //gets added at the end + uint32_t version = s->version_count; + s->version_count++; + s->versions = (Mesh::Surface::Version *)memrealloc(s->versions, sizeof(Mesh::Surface::Version) * s->version_count); - _mesh_surface_generate_version_for_input_mask(s, p_input_mask); + _mesh_surface_generate_version_for_input_mask(s->versions[version], s, p_input_mask); r_vertex_format = s->versions[version].vertex_format; r_vertex_array_rd = s->versions[version].vertex_array; @@ -1363,6 +1461,44 @@ public: s->version_lock.unlock(); } + _FORCE_INLINE_ void mesh_instance_surface_get_arrays_and_format(RID p_mesh_instance, uint32_t p_surface_index, uint32_t p_input_mask, RID &r_vertex_array_rd, RID &r_index_array_rd, RD::VertexFormatID &r_vertex_format) { + MeshInstance *mi = mesh_instance_owner.getornull(p_mesh_instance); + ERR_FAIL_COND(!mi); + Mesh *mesh = mi->mesh; + ERR_FAIL_UNSIGNED_INDEX(p_surface_index, mesh->surface_count); + + MeshInstance::Surface *mis = &mi->surfaces[p_surface_index]; + Mesh::Surface *s = mesh->surfaces[p_surface_index]; + + r_index_array_rd = s->index_array; + + s->version_lock.lock(); + + //there will never be more than, at much, 3 or 4 versions, so iterating is the fastest way + + for (uint32_t i = 0; i < mis->version_count; i++) { + if (mis->versions[i].input_mask != p_input_mask) { + continue; + } + //we have this version, hooray + r_vertex_format = mis->versions[i].vertex_format; + r_vertex_array_rd = mis->versions[i].vertex_array; + s->version_lock.unlock(); + return; + } + + uint32_t version = mis->version_count; + mis->version_count++; + mis->versions = (Mesh::Surface::Version *)memrealloc(mis->versions, sizeof(Mesh::Surface::Version) * mis->version_count); + + _mesh_surface_generate_version_for_input_mask(mis->versions[version], s, p_input_mask, mis); + + r_vertex_format = mis->versions[version].vertex_format; + r_vertex_array_rd = mis->versions[version].vertex_array; + + s->version_lock.unlock(); + } + _FORCE_INLINE_ RID mesh_get_default_rd_buffer(DefaultRDBuffer p_buffer) { ERR_FAIL_INDEX_V(p_buffer, DEFAULT_RD_BUFFER_MAX, RID()); return mesh_default_rd_buffers[p_buffer]; diff --git a/servers/rendering/renderer_rd/shaders/SCsub b/servers/rendering/renderer_rd/shaders/SCsub index 1fe43b25f6..cb62882deb 100644 --- a/servers/rendering/renderer_rd/shaders/SCsub +++ b/servers/rendering/renderer_rd/shaders/SCsub @@ -41,3 +41,4 @@ if "RD_GLSL" in env["BUILDERS"]: env.RD_GLSL("particles.glsl") env.RD_GLSL("particles_copy.glsl") env.RD_GLSL("sort.glsl") + env.RD_GLSL("skeleton.glsl") diff --git a/servers/rendering/renderer_rd/shaders/sdfgi_fields.glsl b/servers/rendering/renderer_rd/shaders/sdfgi_fields.glsl index eec0a90c0d..69d8824d8a 100644 --- a/servers/rendering/renderer_rd/shaders/sdfgi_fields.glsl +++ b/servers/rendering/renderer_rd/shaders/sdfgi_fields.glsl @@ -14,7 +14,7 @@ layout(local_size_x = OCT_RES, local_size_y = OCT_RES, local_size_z = 1) in; layout(rgba16f, set = 0, binding = 1) uniform restrict image2DArray irradiance_texture; layout(rg16f, set = 0, binding = 2) uniform restrict image2DArray depth_texture; -ayout(rgba32ui, set = 0, binding = 3) uniform restrict uimage2DArray irradiance_history_texture; +layout(rgba32ui, set = 0, binding = 3) uniform restrict uimage2DArray irradiance_history_texture; layout(rg32ui, set = 0, binding = 4) uniform restrict uimage2DArray depth_history_texture; struct CascadeData { diff --git a/servers/rendering/renderer_rd/shaders/skeleton.glsl b/servers/rendering/renderer_rd/shaders/skeleton.glsl new file mode 100644 index 0000000000..b19f5a9ad3 --- /dev/null +++ b/servers/rendering/renderer_rd/shaders/skeleton.glsl @@ -0,0 +1,199 @@ +#[compute] + +#version 450 + +VERSION_DEFINES + +layout(local_size_x = 64, local_size_y = 1, local_size_z = 1) in; + +layout(set = 0, binding = 1, std430) buffer restrict writeonly DstVertexData { + uint data[]; +} +dst_vertices; + +layout(set = 0, binding = 2, std430) buffer restrict readonly BlendShapeWeights { + float data[]; +} +blend_shape_weights; + +layout(set = 1, binding = 0, std430) buffer restrict readonly SrcVertexData { + uint data[]; +} +src_vertices; + +layout(set = 1, binding = 1, std430) buffer restrict readonly BoneWeightData { + uint data[]; +} +src_bone_weights; + +layout(set = 1, binding = 2, std430) buffer restrict readonly BlendShapeData { + uint data[]; +} +src_blend_shapes; + +layout(set = 2, binding = 0, std430) buffer restrict readonly SkeletonData { + vec4 data[]; +} +bone_transforms; + +layout(push_constant, binding = 0, std430) uniform Params { + bool has_normal; + bool has_tangent; + bool has_skeleton; + bool has_blend_shape; + + uint vertex_count; + uint vertex_stride; + uint skin_stride; + uint skin_weight_offset; + + uint blend_shape_count; + bool normalized_blend_shapes; + uint pad0; + uint pad1; +} +params; + +vec4 decode_abgr_2_10_10_10(uint base) { + uvec4 abgr_2_10_10_10 = (uvec4(base) >> uvec4(0, 10, 20, 30)) & uvec4(0x3FF, 0x3FF, 0x3FF, 0x3); + return vec4(abgr_2_10_10_10) / vec4(1023.0, 1023.0, 1023.0, 3.0) * 2.0 - 1.0; +} + +uint encode_abgr_2_10_10_10(vec4 base) { + uvec4 abgr_2_10_10_10 = uvec4(clamp(ivec4((base * 0.5 + 0.5) * vec4(1023.0, 1023.0, 1023.0, 3.0)), ivec4(0), ivec4(0x3FF, 0x3FF, 0x3FF, 0x3))) << uvec4(0, 10, 20, 30); + return abgr_2_10_10_10.x | abgr_2_10_10_10.y | abgr_2_10_10_10.z | abgr_2_10_10_10.w; +} + +void main() { + uint index = gl_GlobalInvocationID.x; + if (index >= params.vertex_count) { + return; + } + + uint src_offset = index * params.vertex_stride; + +#ifdef MODE_2D + vec2 vertex = uintBitsToFloat(uvec2(src_vertices.data[src_offset + 0], src_vertices.data[src_offset + 1])); +#else + vec3 vertex; + vec3 normal; + vec4 tangent; + + vertex = uintBitsToFloat(uvec3(src_vertices.data[src_offset + 0], src_vertices.data[src_offset + 1], src_vertices.data[src_offset + 2])); + + src_offset += 3; + + if (params.has_normal) { + normal = decode_abgr_2_10_10_10(src_vertices.data[src_offset]).rgb; + src_offset++; + } + + if (params.has_tangent) { + tangent = decode_abgr_2_10_10_10(src_vertices.data[src_offset]); + } + + if (params.has_blend_shape) { + float blend_total = 0.0; + vec3 blend_vertex = vec3(0.0); + vec3 blend_normal = vec3(0.0); + vec3 blend_tangent = vec3(0.0); + + for (uint i = 0; i < params.blend_shape_count; i++) { + float w = blend_shape_weights.data[i]; + if (w > 0.0001) { + uint base_offset = (params.vertex_count * i + index) * params.vertex_stride; + + blend_vertex += uintBitsToFloat(uvec3(src_blend_shapes.data[base_offset + 0], src_blend_shapes.data[base_offset + 1], src_blend_shapes.data[base_offset + 2])) * w; + + base_offset += 3; + + if (params.has_normal) { + blend_normal += decode_abgr_2_10_10_10(src_blend_shapes.data[base_offset]).rgb * w; + base_offset++; + } + + if (params.has_tangent) { + blend_tangent += decode_abgr_2_10_10_10(src_blend_shapes.data[base_offset]).rgb; + } + + blend_total += w; + } + } + + if (params.normalized_blend_shapes) { + vertex = (1.0 - blend_total) * vertex; + normal = (1.0 - blend_total) * normal; + tangent.rgb = (1.0 - blend_total) * tangent.rgb; + } + + vertex += blend_vertex; + normal += normalize(normal + blend_normal); + tangent.rgb += normalize(tangent.rgb + blend_tangent); + } + + if (params.has_skeleton) { + uint skin_offset = params.skin_stride * index; + + uvec2 bones = uvec2(src_bone_weights.data[skin_offset + 0], src_bone_weights.data[skin_offset + 1]); + uvec2 bones_01 = uvec2(bones.x & 0xFFFF, bones.x >> 16) * 3; //pre-add xform offset + uvec2 bones_23 = uvec2(bones.y & 0xFFFF, bones.y >> 16) * 3; + + skin_offset += params.skin_weight_offset; + + uvec2 weights = uvec2(src_bone_weights.data[skin_offset + 0], src_bone_weights.data[skin_offset + 1]); + + vec2 weights_01 = unpackUnorm2x16(weights.x); + vec2 weights_23 = unpackUnorm2x16(weights.y); + + mat4 m = mat4(bone_transforms.data[bones_01.x], bone_transforms.data[bones_01.x + 1], bone_transforms.data[bones_01.x + 2], vec4(0.0, 0.0, 0.0, 1.0)) * weights_01.x; + m += mat4(bone_transforms.data[bones_01.y], bone_transforms.data[bones_01.y + 1], bone_transforms.data[bones_01.y + 2], vec4(0.0, 0.0, 0.0, 1.0)) * weights_01.y; + m += mat4(bone_transforms.data[bones_23.x], bone_transforms.data[bones_23.x + 1], bone_transforms.data[bones_23.x + 2], vec4(0.0, 0.0, 0.0, 1.0)) * weights_23.x; + m += mat4(bone_transforms.data[bones_23.y], bone_transforms.data[bones_23.y + 1], bone_transforms.data[bones_23.y + 2], vec4(0.0, 0.0, 0.0, 1.0)) * weights_23.y; + + if (params.skin_weight_offset == 4) { + //using 8 bones/weights + skin_offset = params.skin_stride * index + 2; + + bones = uvec2(src_bone_weights.data[skin_offset + 0], src_bone_weights.data[skin_offset + 1]); + bones_01 = uvec2(bones.x & 0xFFFF, bones.x >> 16) * 3; //pre-add xform offset + bones_23 = uvec2(bones.y & 0xFFFF, bones.y >> 16) * 3; + + skin_offset += params.skin_weight_offset; + + weights = uvec2(src_bone_weights.data[skin_offset + 0], src_bone_weights.data[skin_offset + 1]); + + weights_01 = unpackUnorm2x16(weights.x); + weights_23 = unpackUnorm2x16(weights.y); + + m += mat4(bone_transforms.data[bones_01.x], bone_transforms.data[bones_01.x + 1], bone_transforms.data[bones_01.x + 2], vec4(0.0, 0.0, 0.0, 1.0)) * weights_01.x; + m += mat4(bone_transforms.data[bones_01.y], bone_transforms.data[bones_01.y + 1], bone_transforms.data[bones_01.y + 2], vec4(0.0, 0.0, 0.0, 1.0)) * weights_01.y; + m += mat4(bone_transforms.data[bones_23.x], bone_transforms.data[bones_23.x + 1], bone_transforms.data[bones_23.x + 2], vec4(0.0, 0.0, 0.0, 1.0)) * weights_23.x; + m += mat4(bone_transforms.data[bones_23.y], bone_transforms.data[bones_23.y + 1], bone_transforms.data[bones_23.y + 2], vec4(0.0, 0.0, 0.0, 1.0)) * weights_23.y; + } + + //reverse order because its transposed + vertex = (vec4(vertex, 1.0) * m).xyz; + normal = normalize((vec4(normal, 0.0) * m).xyz); + tangent.xyz = normalize((vec4(tangent.xyz, 0.0) * m).xyz); + } + + uint dst_offset = index * params.vertex_stride; + + uvec3 uvertex = floatBitsToUint(vertex); + dst_vertices.data[dst_offset + 0] = uvertex.x; + dst_vertices.data[dst_offset + 1] = uvertex.y; + dst_vertices.data[dst_offset + 2] = uvertex.z; + + dst_offset += 3; + + if (params.has_normal) { + dst_vertices.data[dst_offset] = encode_abgr_2_10_10_10(vec4(normal, 0.0)); + dst_offset++; + } + + if (params.has_tangent) { + dst_vertices.data[dst_offset] = encode_abgr_2_10_10_10(tangent); + } + +#endif +} |