From 0b28c1f8edf1e5bc812a29096ca8e0903c389458 Mon Sep 17 00:00:00 2001 From: clayjohn Date: Sat, 28 Jan 2023 20:36:17 -0800 Subject: [PATCH] Remove cap on number of items drawn in frame in 2D gl_compatibility render Also clean up some names to make the overall organization more clear Also remove cap on items per batch --- drivers/gles3/rasterizer_canvas_gles3.cpp | 154 +++++++++++++--------- drivers/gles3/rasterizer_canvas_gles3.h | 15 ++- servers/rendering_server.cpp | 2 +- 3 files changed, 99 insertions(+), 72 deletions(-) diff --git a/drivers/gles3/rasterizer_canvas_gles3.cpp b/drivers/gles3/rasterizer_canvas_gles3.cpp index fbe00d22afd2..0ca484bfc7d8 100644 --- a/drivers/gles3/rasterizer_canvas_gles3.cpp +++ b/drivers/gles3/rasterizer_canvas_gles3.cpp @@ -113,16 +113,19 @@ void RasterizerCanvasGLES3::canvas_render_items(RID p_to_render_target, Item *p_ // Clear out any state that may have been left from the 3D pass. reset_canvas(); - if (state.canvas_instance_data_buffers[state.current_buffer].fence != GLsync()) { + if (state.canvas_instance_data_buffers[state.current_data_buffer_index].fence != GLsync()) { GLint syncStatus; - glGetSynciv(state.canvas_instance_data_buffers[state.current_buffer].fence, GL_SYNC_STATUS, sizeof(GLint), nullptr, &syncStatus); + glGetSynciv(state.canvas_instance_data_buffers[state.current_data_buffer_index].fence, GL_SYNC_STATUS, sizeof(GLint), nullptr, &syncStatus); if (syncStatus == GL_UNSIGNALED) { // If older than 2 frames, wait for sync OpenGL can have up to 3 frames in flight, any more and we need to sync anyway. - if (state.canvas_instance_data_buffers[state.current_buffer].last_frame_used < RSG::rasterizer->get_frame_number() - 2) { + if (state.canvas_instance_data_buffers[state.current_data_buffer_index].last_frame_used < RSG::rasterizer->get_frame_number() - 2) { #ifndef WEB_ENABLED // On web, we do nothing as the glSubBufferData will force a sync anyway and WebGL does not like waiting. - glClientWaitSync(state.canvas_instance_data_buffers[state.current_buffer].fence, 0, 100000000); // wait for up to 100ms + glClientWaitSync(state.canvas_instance_data_buffers[state.current_data_buffer_index].fence, 0, 100000000); // wait for up to 100ms #endif + state.canvas_instance_data_buffers[state.current_data_buffer_index].last_frame_used = RSG::rasterizer->get_frame_number(); + glDeleteSync(state.canvas_instance_data_buffers[state.current_data_buffer_index].fence); + state.canvas_instance_data_buffers[state.current_data_buffer_index].fence = GLsync(); } else { // Used in last frame or frame before that. OpenGL can get up to two frames behind, so these buffers may still be in use // Allocate a new buffer and use that. @@ -130,9 +133,9 @@ void RasterizerCanvasGLES3::canvas_render_items(RID p_to_render_target, Item *p_ } } else { // Already finished all rendering commands, we can use it. - state.canvas_instance_data_buffers[state.current_buffer].last_frame_used = RSG::rasterizer->get_frame_number(); - glDeleteSync(state.canvas_instance_data_buffers[state.current_buffer].fence); - state.canvas_instance_data_buffers[state.current_buffer].fence = GLsync(); + state.canvas_instance_data_buffers[state.current_data_buffer_index].last_frame_used = RSG::rasterizer->get_frame_number(); + glDeleteSync(state.canvas_instance_data_buffers[state.current_data_buffer_index].fence); + state.canvas_instance_data_buffers[state.current_data_buffer_index].fence = GLsync(); } } @@ -279,7 +282,7 @@ void RasterizerCanvasGLES3::canvas_render_items(RID p_to_render_target, Item *p_ } if (light_count > 0) { - glBindBufferBase(GL_UNIFORM_BUFFER, LIGHT_UNIFORM_LOCATION, state.canvas_instance_data_buffers[state.current_buffer].light_ubo); + glBindBufferBase(GL_UNIFORM_BUFFER, LIGHT_UNIFORM_LOCATION, state.canvas_instance_data_buffers[state.current_data_buffer_index].light_ubo); #ifdef WEB_ENABLED glBufferSubData(GL_UNIFORM_BUFFER, 0, sizeof(LightUniform) * light_count, state.light_uniforms); @@ -361,7 +364,7 @@ void RasterizerCanvasGLES3::canvas_render_items(RID p_to_render_target, Item *p_ state_buffer.tex_to_sdf = 1.0 / ((canvas_scale.x + canvas_scale.y) * 0.5); - glBindBufferBase(GL_UNIFORM_BUFFER, BASE_UNIFORM_LOCATION, state.canvas_instance_data_buffers[state.current_buffer].state_ubo); + glBindBufferBase(GL_UNIFORM_BUFFER, BASE_UNIFORM_LOCATION, state.canvas_instance_data_buffers[state.current_data_buffer_index].state_ubo); glBufferData(GL_UNIFORM_BUFFER, sizeof(StateBuffer), &state_buffer, GL_STREAM_DRAW); GLuint global_buffer = material_storage->global_shader_parameters_get_uniform_buffer(); @@ -395,7 +398,7 @@ void RasterizerCanvasGLES3::canvas_render_items(RID p_to_render_target, Item *p_ Item *ci = p_item_list; Item *canvas_group_owner = nullptr; - uint32_t starting_index = 0; + state.last_item_index = 0; while (ci) { if (ci->copy_back_buffer && canvas_group_owner == nullptr) { @@ -454,7 +457,7 @@ void RasterizerCanvasGLES3::canvas_render_items(RID p_to_render_target, Item *p_ update_skeletons = false; } // Canvas group begins here, render until before this item - _render_items(p_to_render_target, item_count, canvas_transform_inverse, p_light_list, starting_index, r_sdf_used); + _render_items(p_to_render_target, item_count, canvas_transform_inverse, p_light_list, r_sdf_used); item_count = 0; if (ci->canvas_group_owner->canvas_group->mode != RS::CANVAS_GROUP_MODE_TRANSPARENT) { @@ -485,7 +488,7 @@ void RasterizerCanvasGLES3::canvas_render_items(RID p_to_render_target, Item *p_ mesh_storage->update_mesh_instances(); update_skeletons = false; } - _render_items(p_to_render_target, item_count, canvas_transform_inverse, p_light_list, starting_index, r_sdf_used, true); + _render_items(p_to_render_target, item_count, canvas_transform_inverse, p_light_list, r_sdf_used, true); item_count = 0; if (ci->canvas_group->blur_mipmaps) { @@ -504,7 +507,7 @@ void RasterizerCanvasGLES3::canvas_render_items(RID p_to_render_target, Item *p_ } //render anything pending, including clearing if no items - _render_items(p_to_render_target, item_count, canvas_transform_inverse, p_light_list, starting_index, r_sdf_used); + _render_items(p_to_render_target, item_count, canvas_transform_inverse, p_light_list, r_sdf_used); item_count = 0; texture_storage->render_target_copy_to_back_buffer(p_to_render_target, back_buffer_rect, backbuffer_gen_mipmaps); @@ -530,7 +533,7 @@ void RasterizerCanvasGLES3::canvas_render_items(RID p_to_render_target, Item *p_ mesh_storage->update_mesh_instances(); update_skeletons = false; } - _render_items(p_to_render_target, item_count, canvas_transform_inverse, p_light_list, starting_index, r_sdf_used); + _render_items(p_to_render_target, item_count, canvas_transform_inverse, p_light_list, r_sdf_used); //then reset item_count = 0; } @@ -542,14 +545,15 @@ void RasterizerCanvasGLES3::canvas_render_items(RID p_to_render_target, Item *p_ RenderingServerDefault::redraw_request(); } - state.canvas_instance_data_buffers[state.current_buffer].fence = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, 0); + state.canvas_instance_data_buffers[state.current_data_buffer_index].fence = glFenceSync(GL_SYNC_GPU_COMMANDS_COMPLETE, 0); // Clear out state used in 2D pass reset_canvas(); - state.current_buffer = (state.current_buffer + 1) % state.canvas_instance_data_buffers.size(); + state.current_data_buffer_index = (state.current_data_buffer_index + 1) % state.canvas_instance_data_buffers.size(); + state.current_instance_buffer_index = 0; } -void RasterizerCanvasGLES3::_render_items(RID p_to_render_target, int p_item_count, const Transform2D &p_canvas_transform_inverse, Light *p_lights, uint32_t &r_last_index, bool &r_sdf_used, bool p_to_backbuffer) { +void RasterizerCanvasGLES3::_render_items(RID p_to_render_target, int p_item_count, const Transform2D &p_canvas_transform_inverse, Light *p_lights, bool &r_sdf_used, bool p_to_backbuffer) { GLES3::MaterialStorage *material_storage = GLES3::MaterialStorage::get_singleton(); canvas_begin(p_to_render_target, p_to_backbuffer); @@ -565,17 +569,17 @@ void RasterizerCanvasGLES3::_render_items(RID p_to_render_target, int p_item_cou // Record Batches. // First item always forms its own batch. bool batch_broken = false; - _new_batch(batch_broken, index); + _new_batch(batch_broken); // Override the start position and index as we want to start from where we finished off last time. - state.canvas_instance_batches[state.current_batch_index].start = r_last_index; + state.canvas_instance_batches[state.current_batch_index].start = state.last_item_index; index = 0; for (int i = 0; i < p_item_count; i++) { Item *ci = items[i]; if (ci->final_clip_owner != state.canvas_instance_batches[state.current_batch_index].clip) { - _new_batch(batch_broken, index); + _new_batch(batch_broken); state.canvas_instance_batches[state.current_batch_index].clip = ci->final_clip_owner; current_clip = ci->final_clip_owner; } @@ -599,7 +603,7 @@ void RasterizerCanvasGLES3::_render_items(RID p_to_render_target, int p_item_cou GLES3::CanvasShaderData *shader_data_cache = nullptr; if (material != state.canvas_instance_batches[state.current_batch_index].material) { - _new_batch(batch_broken, index); + _new_batch(batch_broken); GLES3::CanvasMaterialData *material_data = nullptr; if (material.is_valid()) { @@ -629,12 +633,12 @@ void RasterizerCanvasGLES3::_render_items(RID p_to_render_target, int p_item_cou } // Copy over all data needed for rendering. - glBindBuffer(GL_ARRAY_BUFFER, state.canvas_instance_data_buffers[state.current_buffer].buffer); + glBindBuffer(GL_ARRAY_BUFFER, state.canvas_instance_data_buffers[state.current_data_buffer_index].instance_buffers[state.current_instance_buffer_index]); #ifdef WEB_ENABLED - glBufferSubData(GL_ARRAY_BUFFER, r_last_index * sizeof(InstanceData), sizeof(InstanceData) * index, state.instance_data_array); + glBufferSubData(GL_ARRAY_BUFFER, state.last_item_index * sizeof(InstanceData), sizeof(InstanceData) * index, state.instance_data_array); #else // On Desktop and mobile we map the memory without synchronizing for maximum speed. - void *buffer = glMapBufferRange(GL_ARRAY_BUFFER, r_last_index * sizeof(InstanceData), index * sizeof(InstanceData), GL_MAP_WRITE_BIT | GL_MAP_UNSYNCHRONIZED_BIT); + void *buffer = glMapBufferRange(GL_ARRAY_BUFFER, state.last_item_index * sizeof(InstanceData), index * sizeof(InstanceData), GL_MAP_WRITE_BIT | GL_MAP_UNSYNCHRONIZED_BIT); memcpy(buffer, state.instance_data_array, index * sizeof(InstanceData)); glUnmapBuffer(GL_ARRAY_BUFFER); #endif @@ -757,14 +761,14 @@ void RasterizerCanvasGLES3::_render_items(RID p_to_render_target, int p_item_cou state.current_batch_index = 0; state.canvas_instance_batches.clear(); - r_last_index += index; + state.last_item_index += index; } void RasterizerCanvasGLES3::_record_item_commands(const Item *p_item, RID p_render_target, const Transform2D &p_canvas_transform_inverse, Item *¤t_clip, GLES3::CanvasShaderData::BlendMode p_blend_mode, Light *p_lights, uint32_t &r_index, bool &r_batch_broken, bool &r_sdf_used) { RenderingServer::CanvasItemTextureFilter texture_filter = p_item->texture_filter == RS::CANVAS_ITEM_TEXTURE_FILTER_DEFAULT ? state.default_filter : p_item->texture_filter; if (texture_filter != state.canvas_instance_batches[state.current_batch_index].filter) { - _new_batch(r_batch_broken, r_index); + _new_batch(r_batch_broken); state.canvas_instance_batches[state.current_batch_index].filter = texture_filter; } @@ -772,7 +776,7 @@ void RasterizerCanvasGLES3::_record_item_commands(const Item *p_item, RID p_rend RenderingServer::CanvasItemTextureRepeat texture_repeat = p_item->texture_repeat == RS::CANVAS_ITEM_TEXTURE_REPEAT_DEFAULT ? state.default_repeat : p_item->texture_repeat; if (texture_repeat != state.canvas_instance_batches[state.current_batch_index].repeat) { - _new_batch(r_batch_broken, r_index); + _new_batch(r_batch_broken); state.canvas_instance_batches[state.current_batch_index].repeat = texture_repeat; } @@ -816,7 +820,7 @@ void RasterizerCanvasGLES3::_record_item_commands(const Item *p_item, RID p_rend bool lights_disabled = light_count == 0 && !state.using_directional_lights; if (lights_disabled != state.canvas_instance_batches[state.current_batch_index].lights_disabled) { - _new_batch(r_batch_broken, r_index); + _new_batch(r_batch_broken); state.canvas_instance_batches[state.current_batch_index].lights_disabled = lights_disabled; } @@ -864,7 +868,7 @@ void RasterizerCanvasGLES3::_record_item_commands(const Item *p_item, RID p_rend } if (blend_mode != state.canvas_instance_batches[state.current_batch_index].blend_mode || blend_color != state.canvas_instance_batches[state.current_batch_index].blend_color) { - _new_batch(r_batch_broken, r_index); + _new_batch(r_batch_broken); state.canvas_instance_batches[state.current_batch_index].blend_mode = blend_mode; state.canvas_instance_batches[state.current_batch_index].blend_color = blend_color; } @@ -874,12 +878,12 @@ void RasterizerCanvasGLES3::_record_item_commands(const Item *p_item, RID p_rend const Item::CommandRect *rect = static_cast(c); if (rect->flags & CANVAS_RECT_TILE && state.canvas_instance_batches[state.current_batch_index].repeat != RenderingServer::CanvasItemTextureRepeat::CANVAS_ITEM_TEXTURE_REPEAT_ENABLED) { - _new_batch(r_batch_broken, r_index); + _new_batch(r_batch_broken); state.canvas_instance_batches[state.current_batch_index].repeat = RenderingServer::CanvasItemTextureRepeat::CANVAS_ITEM_TEXTURE_REPEAT_ENABLED; } if (rect->texture != state.canvas_instance_batches[state.current_batch_index].tex || state.canvas_instance_batches[state.current_batch_index].command_type != Item::Command::TYPE_RECT) { - _new_batch(r_batch_broken, r_index); + _new_batch(r_batch_broken); state.canvas_instance_batches[state.current_batch_index].tex = rect->texture; state.canvas_instance_batches[state.current_batch_index].command_type = Item::Command::TYPE_RECT; state.canvas_instance_batches[state.current_batch_index].command = c; @@ -969,7 +973,7 @@ void RasterizerCanvasGLES3::_record_item_commands(const Item *p_item, RID p_rend const Item::CommandNinePatch *np = static_cast(c); if (np->texture != state.canvas_instance_batches[state.current_batch_index].tex || state.canvas_instance_batches[state.current_batch_index].command_type != Item::Command::TYPE_NINEPATCH) { - _new_batch(r_batch_broken, r_index); + _new_batch(r_batch_broken); state.canvas_instance_batches[state.current_batch_index].tex = np->texture; state.canvas_instance_batches[state.current_batch_index].command_type = Item::Command::TYPE_NINEPATCH; state.canvas_instance_batches[state.current_batch_index].command = c; @@ -1034,7 +1038,7 @@ void RasterizerCanvasGLES3::_record_item_commands(const Item *p_item, RID p_rend const Item::CommandPolygon *polygon = static_cast(c); // Polygon's can't be batched, so always create a new batch - _new_batch(r_batch_broken, r_index); + _new_batch(r_batch_broken); state.canvas_instance_batches[state.current_batch_index].tex = polygon->texture; state.canvas_instance_batches[state.current_batch_index].command_type = Item::Command::TYPE_POLYGON; @@ -1061,7 +1065,7 @@ void RasterizerCanvasGLES3::_record_item_commands(const Item *p_item, RID p_rend const Item::CommandPrimitive *primitive = static_cast(c); if (primitive->point_count != state.canvas_instance_batches[state.current_batch_index].primitive_points || state.canvas_instance_batches[state.current_batch_index].command_type != Item::Command::TYPE_PRIMITIVE) { - _new_batch(r_batch_broken, r_index); + _new_batch(r_batch_broken); state.canvas_instance_batches[state.current_batch_index].tex = primitive->texture; state.canvas_instance_batches[state.current_batch_index].primitive_points = primitive->point_count; state.canvas_instance_batches[state.current_batch_index].command_type = Item::Command::TYPE_PRIMITIVE; @@ -1086,10 +1090,7 @@ void RasterizerCanvasGLES3::_record_item_commands(const Item *p_item, RID p_rend if (primitive->point_count == 4) { // Reset base data. _update_transform_2d_to_mat2x3(base_transform * draw_transform, state.instance_data_array[r_index].world); - state.instance_data_array[r_index].color_texture_pixel_size[0] = 0.0; - state.instance_data_array[r_index].color_texture_pixel_size[1] = 0.0; - - state.instance_data_array[r_index].flags = base_flags | (state.instance_data_array[r_index - 1].flags & (FLAGS_DEFAULT_NORMAL_MAP_USED | FLAGS_DEFAULT_SPECULAR_MAP_USED)); //reset on each command for sanity, keep canvastexture binding config + _prepare_canvas_texture(state.canvas_instance_batches[state.current_batch_index].tex, state.canvas_instance_batches[state.current_batch_index].filter, state.canvas_instance_batches[state.current_batch_index].repeat, r_index, texpixel_size); for (uint32_t j = 0; j < 3; j++) { int offset = j == 0 ? 0 : 1; @@ -1111,7 +1112,7 @@ void RasterizerCanvasGLES3::_record_item_commands(const Item *p_item, RID p_rend case Item::Command::TYPE_MULTIMESH: case Item::Command::TYPE_PARTICLES: { // Mesh's can't be batched, so always create a new batch - _new_batch(r_batch_broken, r_index); + _new_batch(r_batch_broken); Color modulate(1, 1, 1, 1); state.canvas_instance_batches[state.current_batch_index].shader_variant = CanvasShaderGLES3::MODE_ATTRIBUTES; @@ -1183,7 +1184,7 @@ void RasterizerCanvasGLES3::_record_item_commands(const Item *p_item, RID p_rend const Item::CommandClipIgnore *ci = static_cast(c); if (current_clip) { if (ci->ignore != reclip) { - _new_batch(r_batch_broken, r_index); + _new_batch(r_batch_broken); if (ci->ignore) { state.canvas_instance_batches[state.current_batch_index].clip = nullptr; reclip = true; @@ -1227,7 +1228,7 @@ void RasterizerCanvasGLES3::_render_batch(Light *p_lights, uint32_t p_index) { case Item::Command::TYPE_RECT: case Item::Command::TYPE_NINEPATCH: { glBindVertexArray(data.indexed_quad_array); - glBindBuffer(GL_ARRAY_BUFFER, state.canvas_instance_data_buffers[state.current_buffer].buffer); + glBindBuffer(GL_ARRAY_BUFFER, state.canvas_instance_data_buffers[state.current_data_buffer_index].instance_buffers[state.canvas_instance_batches[p_index].instance_buffer_index]); uint32_t range_start = state.canvas_instance_batches[p_index].start * sizeof(InstanceData); _enable_attributes(range_start, false); @@ -1243,7 +1244,7 @@ void RasterizerCanvasGLES3::_render_batch(Light *p_lights, uint32_t p_index) { ERR_FAIL_COND(!pb); glBindVertexArray(pb->vertex_array); - glBindBuffer(GL_ARRAY_BUFFER, state.canvas_instance_data_buffers[state.current_buffer].buffer); + glBindBuffer(GL_ARRAY_BUFFER, state.canvas_instance_data_buffers[state.current_data_buffer_index].instance_buffers[state.canvas_instance_batches[p_index].instance_buffer_index]); uint32_t range_start = state.canvas_instance_batches[p_index].start * sizeof(InstanceData); _enable_attributes(range_start, false); @@ -1267,7 +1268,7 @@ void RasterizerCanvasGLES3::_render_batch(Light *p_lights, uint32_t p_index) { case Item::Command::TYPE_PRIMITIVE: { glBindVertexArray(data.canvas_quad_array); - glBindBuffer(GL_ARRAY_BUFFER, state.canvas_instance_data_buffers[state.current_buffer].buffer); + glBindBuffer(GL_ARRAY_BUFFER, state.canvas_instance_data_buffers[state.current_data_buffer_index].instance_buffers[state.canvas_instance_batches[p_index].instance_buffer_index]); uint32_t range_start = state.canvas_instance_batches[p_index].start * sizeof(InstanceData); _enable_attributes(range_start, true); @@ -1370,7 +1371,7 @@ void RasterizerCanvasGLES3::_render_batch(Light *p_lights, uint32_t p_index) { index_array_gl = mesh_storage->mesh_surface_get_index_buffer(surface, 0); bool use_index_buffer = false; glBindVertexArray(vertex_array_gl); - glBindBuffer(GL_ARRAY_BUFFER, state.canvas_instance_data_buffers[state.current_buffer].buffer); + glBindBuffer(GL_ARRAY_BUFFER, state.canvas_instance_data_buffers[state.current_data_buffer_index].instance_buffers[state.canvas_instance_batches[p_index].instance_buffer_index]); uint32_t range_start = state.canvas_instance_batches[p_index].start * sizeof(InstanceData); _enable_attributes(range_start, false, instance_count); @@ -1427,20 +1428,30 @@ void RasterizerCanvasGLES3::_render_batch(Light *p_lights, uint32_t p_index) { } void RasterizerCanvasGLES3::_add_to_batch(uint32_t &r_index, bool &r_batch_broken) { - if (r_index >= data.max_instances_per_buffer - 1) { - ERR_PRINT_ONCE("Trying to draw too many items. Please increase maximum number of items in the project settings 'rendering/gl_compatibility/item_buffer_size'"); - return; - } - - if (state.canvas_instance_batches[state.current_batch_index].instance_count >= data.max_instances_per_batch) { - _new_batch(r_batch_broken, r_index); - } - state.canvas_instance_batches[state.current_batch_index].instance_count++; r_index++; + if (r_index >= data.max_instances_per_buffer) { + // Copy over all data needed for rendering right away + // then go back to recording item commands. + glBindBuffer(GL_ARRAY_BUFFER, state.canvas_instance_data_buffers[state.current_data_buffer_index].instance_buffers[state.current_instance_buffer_index]); +#ifdef WEB_ENABLED + glBufferSubData(GL_ARRAY_BUFFER, state.last_item_index * sizeof(InstanceData), sizeof(InstanceData) * r_index, state.instance_data_array); +#else + // On Desktop and mobile we map the memory without synchronizing for maximum speed. + void *buffer = glMapBufferRange(GL_ARRAY_BUFFER, state.last_item_index * sizeof(InstanceData), r_index * sizeof(InstanceData), GL_MAP_WRITE_BIT | GL_MAP_UNSYNCHRONIZED_BIT); + memcpy(buffer, state.instance_data_array, r_index * sizeof(InstanceData)); + glUnmapBuffer(GL_ARRAY_BUFFER); +#endif + _allocate_instance_buffer(); + r_index = 0; + state.last_item_index = 0; + r_batch_broken = false; // Force a new batch to be created + _new_batch(r_batch_broken); + state.canvas_instance_batches[state.current_batch_index].start = 0; + } } -void RasterizerCanvasGLES3::_new_batch(bool &r_batch_broken, uint32_t &r_index) { +void RasterizerCanvasGLES3::_new_batch(bool &r_batch_broken) { if (state.canvas_instance_batches.size() == 0) { state.canvas_instance_batches.push_back(Batch()); return; @@ -1456,7 +1467,7 @@ void RasterizerCanvasGLES3::_new_batch(bool &r_batch_broken, uint32_t &r_index) Batch new_batch = state.canvas_instance_batches[state.current_batch_index]; new_batch.instance_count = 0; new_batch.start = state.canvas_instance_batches[state.current_batch_index].start + state.canvas_instance_batches[state.current_batch_index].instance_count; - + new_batch.instance_buffer_index = state.current_instance_buffer_index; state.current_batch_index++; state.canvas_instance_batches.push_back(new_batch); } @@ -2433,17 +2444,35 @@ void RasterizerCanvasGLES3::_allocate_instance_data_buffer() { glBindBuffer(GL_UNIFORM_BUFFER, new_buffers[2]); glBufferData(GL_UNIFORM_BUFFER, sizeof(StateBuffer), nullptr, GL_STREAM_DRAW); - state.current_buffer = (state.current_buffer + 1); + state.current_data_buffer_index = (state.current_data_buffer_index + 1); DataBuffer db; - db.buffer = new_buffers[0]; + db.instance_buffers.push_back(new_buffers[0]); db.light_ubo = new_buffers[1]; db.state_ubo = new_buffers[2]; db.last_frame_used = RSG::rasterizer->get_frame_number(); - state.canvas_instance_data_buffers.insert(state.current_buffer, db); - state.current_buffer = state.current_buffer % state.canvas_instance_data_buffers.size(); + state.canvas_instance_data_buffers.insert(state.current_data_buffer_index, db); + state.current_data_buffer_index = state.current_data_buffer_index % state.canvas_instance_data_buffers.size(); glBindBuffer(GL_ARRAY_BUFFER, 0); glBindBuffer(GL_UNIFORM_BUFFER, 0); } +void RasterizerCanvasGLES3::_allocate_instance_buffer() { + state.current_instance_buffer_index++; + + if (int(state.current_instance_buffer_index) < state.canvas_instance_data_buffers[state.current_data_buffer_index].instance_buffers.size()) { + // We already allocated another buffer in a previous frame, so we can just use it. + return; + } + + GLuint new_buffer; + glGenBuffers(1, &new_buffer); + + glBindBuffer(GL_ARRAY_BUFFER, new_buffer); + glBufferData(GL_ARRAY_BUFFER, data.max_instance_buffer_size, nullptr, GL_STREAM_DRAW); + + state.canvas_instance_data_buffers[state.current_data_buffer_index].instance_buffers.push_back(new_buffer); + + glBindBuffer(GL_ARRAY_BUFFER, 0); +} void RasterizerCanvasGLES3::set_time(double p_time) { state.time = p_time; @@ -2586,14 +2615,12 @@ RasterizerCanvasGLES3::RasterizerCanvasGLES3() { int uniform_max_size = config->max_uniform_buffer_size; if (uniform_max_size < 65536) { data.max_lights_per_render = 64; - data.max_instances_per_batch = 128; } else { data.max_lights_per_render = 256; - data.max_instances_per_batch = 2048; } // Reserve 3 Uniform Buffers for instance data Frame N, N+1 and N+2 - data.max_instances_per_buffer = MAX(data.max_instances_per_batch, uint32_t(GLOBAL_GET("rendering/gl_compatibility/item_buffer_size"))); + data.max_instances_per_buffer = uint32_t(GLOBAL_GET("rendering/gl_compatibility/item_buffer_size")); data.max_instance_buffer_size = data.max_instances_per_buffer * sizeof(InstanceData); // 16,384 instances * 128 bytes = 2,097,152 bytes = 2,048 kb state.canvas_instance_data_buffers.resize(3); state.canvas_instance_batches.reserve(200); @@ -2611,7 +2638,7 @@ RasterizerCanvasGLES3::RasterizerCanvasGLES3() { glBindBuffer(GL_UNIFORM_BUFFER, new_buffers[2]); glBufferData(GL_UNIFORM_BUFFER, sizeof(StateBuffer), nullptr, GL_STREAM_DRAW); DataBuffer db; - db.buffer = new_buffers[0]; + db.instance_buffers.push_back(new_buffers[0]); db.light_ubo = new_buffers[1]; db.state_ubo = new_buffers[2]; db.last_frame_used = 0; @@ -2638,7 +2665,6 @@ RasterizerCanvasGLES3::RasterizerCanvasGLES3() { String global_defines; global_defines += "#define MAX_GLOBAL_SHADER_UNIFORMS 256\n"; // TODO: this is arbitrary for now global_defines += "#define MAX_LIGHTS " + itos(data.max_lights_per_render) + "\n"; - global_defines += "#define MAX_DRAW_DATA_INSTANCES " + itos(data.max_instances_per_batch) + "\n"; GLES3::MaterialStorage::get_singleton()->shaders.canvas_shader.initialize(global_defines, 1); data.canvas_shader_default_version = GLES3::MaterialStorage::get_singleton()->shaders.canvas_shader.version_create(); diff --git a/drivers/gles3/rasterizer_canvas_gles3.h b/drivers/gles3/rasterizer_canvas_gles3.h index 916e12057cf8..1c14d0b46606 100644 --- a/drivers/gles3/rasterizer_canvas_gles3.h +++ b/drivers/gles3/rasterizer_canvas_gles3.h @@ -247,7 +247,6 @@ public: uint32_t max_lights_per_render = 256; uint32_t max_lights_per_item = 16; - uint32_t max_instances_per_batch = 512; uint32_t max_instances_per_buffer = 16384; uint32_t max_instance_buffer_size = 16384 * 128; } data; @@ -256,6 +255,7 @@ public: // Position in the UBO measured in bytes uint32_t start = 0; uint32_t instance_count = 0; + uint32_t instance_buffer_index = 0; RID tex; RS::CanvasItemTextureFilter filter = RS::CANVAS_ITEM_TEXTURE_FILTER_MAX; @@ -281,7 +281,7 @@ public: // We track them and ensure that they don't get reused until at least 2 frames have passed // to avoid the GPU stalling to wait for a resource to become available. struct DataBuffer { - GLuint buffer = 0; + Vector instance_buffers; GLuint light_ubo = 0; GLuint state_ubo = 0; uint64_t last_frame_used = -3; @@ -291,9 +291,10 @@ public: struct State { LocalVector canvas_instance_data_buffers; LocalVector canvas_instance_batches; - uint32_t current_buffer = 0; - uint32_t current_buffer_index = 0; + uint32_t current_data_buffer_index = 0; + uint32_t current_instance_buffer_index = 0; uint32_t current_batch_index = 0; + uint32_t last_item_index = 0; InstanceData *instance_data_array = nullptr; @@ -354,14 +355,14 @@ public: void _prepare_canvas_texture(RID p_texture, RS::CanvasItemTextureFilter p_base_filter, RS::CanvasItemTextureRepeat p_base_repeat, uint32_t &r_index, Size2 &r_texpixel_size); void canvas_render_items(RID p_to_render_target, Item *p_item_list, const Color &p_modulate, Light *p_light_list, Light *p_directional_list, const Transform2D &p_canvas_transform, RS::CanvasItemTextureFilter p_default_filter, RS::CanvasItemTextureRepeat p_default_repeat, bool p_snap_2d_vertices_to_pixel, bool &r_sdf_used) override; - void _render_items(RID p_to_render_target, int p_item_count, const Transform2D &p_canvas_transform_inverse, Light *p_lights, uint32_t &r_last_index, bool &r_sdf_used, bool p_to_backbuffer = false); + void _render_items(RID p_to_render_target, int p_item_count, const Transform2D &p_canvas_transform_inverse, Light *p_lights, bool &r_sdf_used, bool p_to_backbuffer = false); void _record_item_commands(const Item *p_item, RID p_render_target, const Transform2D &p_canvas_transform_inverse, Item *¤t_clip, GLES3::CanvasShaderData::BlendMode p_blend_mode, Light *p_lights, uint32_t &r_index, bool &r_break_batch, bool &r_sdf_used); void _render_batch(Light *p_lights, uint32_t p_index); bool _bind_material(GLES3::CanvasMaterialData *p_material_data, CanvasShaderGLES3::ShaderVariant p_variant, uint64_t p_specialization); - void _new_batch(bool &r_batch_broken, uint32_t &r_index); + void _new_batch(bool &r_batch_broken); void _add_to_batch(uint32_t &r_index, bool &r_batch_broken); void _allocate_instance_data_buffer(); - void _align_instance_data_buffer(uint32_t &r_index); + void _allocate_instance_buffer(); void _enable_attributes(uint32_t p_start, bool p_primitive, uint32_t p_rate = 1); void set_time(double p_time); diff --git a/servers/rendering_server.cpp b/servers/rendering_server.cpp index 4ef562adcb77..97907bed48fd 100644 --- a/servers/rendering_server.cpp +++ b/servers/rendering_server.cpp @@ -2875,7 +2875,7 @@ void RenderingServer::init() { GLOBAL_DEF("rendering/2d/shadow_atlas/size", 2048); // Number of commands that can be drawn per frame. - GLOBAL_DEF_RST(PropertyInfo(Variant::INT, "rendering/gl_compatibility/item_buffer_size", PROPERTY_HINT_RANGE, "1024,1048576,1"), 16384); + GLOBAL_DEF_RST(PropertyInfo(Variant::INT, "rendering/gl_compatibility/item_buffer_size", PROPERTY_HINT_RANGE, "128,1048576,1"), 16384); GLOBAL_DEF("rendering/shader_compiler/shader_cache/enabled", true); GLOBAL_DEF("rendering/shader_compiler/shader_cache/compress", true);