webrender 0.69.0

/* This Source Code Form is subject to the terms of the Mozilla Public
 * License, v. 2.0. If a copy of the MPL was not distributed with this
 * file, You can obtain one at http://mozilla.org/MPL/2.0/. */

//! Rendering logic related to the vertex shaders and their states, uncluding
//!  - Vertex Array Objects
//!  - vertex layout descriptors
//!  - textures bound at vertex stage

use std::{marker::PhantomData, mem, num::NonZeroUsize, ops};
use api::units::*;
use crate::{
    device::{
        Device, Texture, TextureFilter, TextureUploader, UploadPBOPool, VertexUsageHint, VAO,
    },
    frame_builder::Frame,
    gpu_types::{PrimitiveHeaderI, PrimitiveHeaderF},
    internal_types::Swizzle,
    render_task::RenderTaskData,
    transform::TransformData,
};

use crate::internal_types::FrameVec;

pub const VERTEX_TEXTURE_EXTRA_ROWS: i32 = 10;

pub const MAX_VERTEX_TEXTURE_WIDTH: usize = webrender_build::MAX_VERTEX_TEXTURE_WIDTH;

pub mod desc {
    use crate::device::{VertexAttribute, VertexAttributeKind, VertexDescriptor};

    pub const PRIM_INSTANCES: VertexDescriptor = VertexDescriptor {
        vertex_attributes: &[VertexAttribute::quad_instance_vertex()],
        instance_attributes: &[VertexAttribute {
            name: "aData",
            count: 4,
            kind: VertexAttributeKind::I32,
        }],
    };

    pub const BLUR: VertexDescriptor = VertexDescriptor {
        vertex_attributes: &[VertexAttribute::quad_instance_vertex()],
        instance_attributes: &[
            VertexAttribute::gpu_buffer_address("aBlurRenderTaskAddress"),
            VertexAttribute::gpu_buffer_address("aBlurSourceTaskAddress"),
            VertexAttribute::i32("aBlurDirection"),
            VertexAttribute::i32("aBlurEdgeMode"),
            VertexAttribute::f32x3("aBlurParams"),
        ],
    };

    pub const LINE: VertexDescriptor = VertexDescriptor {
        vertex_attributes: &[VertexAttribute::quad_instance_vertex()],
        instance_attributes: &[
            VertexAttribute::f32x4("aTaskRect"),
            VertexAttribute::f32x2("aLocalSize"),
            VertexAttribute::f32("aWavyLineThickness"),
            VertexAttribute::i32("aStyle"),
            VertexAttribute::f32("aAxisSelect"),
        ],
    };


    pub const BORDER: VertexDescriptor = VertexDescriptor {
        vertex_attributes: &[VertexAttribute::quad_instance_vertex()],
        instance_attributes: &[
            VertexAttribute::f32x2("aTaskOrigin"),
            VertexAttribute::f32x4("aRect"),
            VertexAttribute::f32x4("aColor0"),
            VertexAttribute::f32x4("aColor1"),
            VertexAttribute::i32("aFlags"),
            VertexAttribute::f32x2("aWidths"),
            VertexAttribute::f32x2("aRadii"),
            VertexAttribute::f32x4("aClipParams1"),
            VertexAttribute::f32x4("aClipParams2"),
        ],
    };

    pub const SCALE: VertexDescriptor = VertexDescriptor {
        vertex_attributes: &[VertexAttribute::quad_instance_vertex()],
        instance_attributes: &[
            VertexAttribute::f32x4("aScaleTargetRect"),
            VertexAttribute::f32x4("aScaleSourceRect"),
            VertexAttribute::f32("aSourceRectType"),
        ],
    };

    pub const CLIP_RECT: VertexDescriptor = VertexDescriptor {
        vertex_attributes: &[VertexAttribute::quad_instance_vertex()],
        instance_attributes: &[
            // common clip attributes
            VertexAttribute::f32x4("aClipDeviceArea"),
            VertexAttribute::f32x4("aClipOrigins"),
            VertexAttribute::f32("aDevicePixelScale"),
            VertexAttribute::i32x2("aTransformIds"),
            // specific clip attributes
            VertexAttribute::f32x2("aClipLocalPos"),
            VertexAttribute::f32x4("aClipLocalRect"),
            VertexAttribute::f32("aClipMode"),
            VertexAttribute::f32x4("aClipRect_TL"),
            VertexAttribute::f32x4("aClipRadii_TL"),
            VertexAttribute::f32x4("aClipRect_TR"),
            VertexAttribute::f32x4("aClipRadii_TR"),
            VertexAttribute::f32x4("aClipRect_BL"),
            VertexAttribute::f32x4("aClipRadii_BL"),
            VertexAttribute::f32x4("aClipRect_BR"),
            VertexAttribute::f32x4("aClipRadii_BR"),
        ],
    };

    pub const SVG_FILTER_NODE: VertexDescriptor = VertexDescriptor {
        vertex_attributes: &[VertexAttribute::quad_instance_vertex()],
        instance_attributes: &[
            VertexAttribute::f32x4("aFilterTargetRect"),
            VertexAttribute::f32x4("aFilterInput1ContentScaleAndOffset"),
            VertexAttribute::f32x4("aFilterInput2ContentScaleAndOffset"),
            VertexAttribute::gpu_buffer_address("aFilterInput1TaskAddress"),
            VertexAttribute::gpu_buffer_address("aFilterInput2TaskAddress"),
            VertexAttribute::u16("aFilterKind"),
            VertexAttribute::u16("aFilterInputCount"),
            VertexAttribute::gpu_buffer_address("aFilterExtraDataAddress"),
        ],
    };

    pub const MASK: VertexDescriptor = VertexDescriptor {
        vertex_attributes: &[VertexAttribute::quad_instance_vertex()],
        instance_attributes: &[
            VertexAttribute::i32x4("aData"),
            VertexAttribute::i32x4("aClipData"),
        ],
    };

    pub const COMPOSITE: VertexDescriptor = VertexDescriptor {
        vertex_attributes: &[VertexAttribute::quad_instance_vertex()],
        instance_attributes: &[
            VertexAttribute::f32x4("aDeviceRect"),
            VertexAttribute::f32x4("aDeviceClipRect"),
            VertexAttribute::f32x4("aColor"),
            VertexAttribute::f32x4("aParams"),
            VertexAttribute::f32x4("aUvRect0"),
            VertexAttribute::f32x4("aUvRect1"),
            VertexAttribute::f32x4("aUvRect2"),
            VertexAttribute::f32x2("aFlip"),
            VertexAttribute::f32x4("aDeviceRoundedClipRect"),
            VertexAttribute::f32x4("aDeviceRoundedClipRadii"),
        ],
    };

    pub const CLEAR: VertexDescriptor = VertexDescriptor {
        vertex_attributes: &[VertexAttribute::quad_instance_vertex()],
        instance_attributes: &[
            VertexAttribute::f32x4("aRect"),
            VertexAttribute::f32x4("aColor"),
        ],
    };

    pub const COPY: VertexDescriptor = VertexDescriptor {
        vertex_attributes: &[VertexAttribute::quad_instance_vertex()],
        instance_attributes: &[
            VertexAttribute::f32x4("a_src_rect"),
            VertexAttribute::f32x4("a_dst_rect"),
            VertexAttribute::f32x2("a_dst_texture_size"),
        ],
    };
}

#[derive(Debug, Copy, Clone, PartialEq)]
pub enum VertexArrayKind {
    Primitive,
    Blur,
    ClipRect,
    Border,
    Scale,
    LineDecoration,
    SvgFilterNode,
    Composite,
    Clear,
    Copy,
    Mask,
}

pub struct VertexDataTexture<T> {
    texture: Option<Texture>,
    format: api::ImageFormat,
    _marker: PhantomData<T>,
}

impl<T> VertexDataTexture<T> {
    pub fn new(format: api::ImageFormat) -> Self {
        Self {
            texture: None,
            format,
            _marker: PhantomData,
        }
    }

    /// Returns a borrow of the GPU texture. Panics if it hasn't been initialized.
    pub fn texture(&self) -> &Texture {
        self.texture.as_ref().unwrap()
    }

    /// Returns an estimate of the GPU memory consumed by this VertexDataTexture.
    pub fn size_in_bytes(&self) -> usize {
        self.texture.as_ref().map_or(0, |t| t.size_in_bytes())
    }

    pub fn update<'a>(
        &'a mut self,
        device: &mut Device,
        texture_uploader: &mut TextureUploader<'a>,
        data: &mut FrameVec<T>,
    ) {
        debug_assert!(mem::size_of::<T>() % 16 == 0);
        let texels_per_item = mem::size_of::<T>() / 16;
        let items_per_row = MAX_VERTEX_TEXTURE_WIDTH / texels_per_item;
        debug_assert_ne!(items_per_row, 0);

        // Ensure we always end up with a texture when leaving this method.
        let mut len = data.len();
        if len == 0 {
            if self.texture.is_some() {
                return;
            }
            data.reserve(items_per_row);
            len = items_per_row;
        } else {
            // Extend the data array to have enough capacity to upload at least
            // a multiple of the row size.  This ensures memory safety when the
            // array is passed to OpenGL to upload to the GPU.
            let extra = len % items_per_row;
            if extra != 0 {
                let padding = items_per_row - extra;
                data.reserve(padding);
                len += padding;
            }
        }

        let needed_height = (len / items_per_row) as i32;
        let existing_height = self
            .texture
            .as_ref()
            .map_or(0, |t| t.get_dimensions().height);

        // Create a new texture if needed.
        //
        // These textures are generally very small, which is why we don't bother
        // with incremental updates and just re-upload every frame. For most pages
        // they're one row each, and on stress tests like css-francine they end up
        // in the 6-14 range. So we size the texture tightly to what we need (usually
        // 1), and shrink it if the waste would be more than `VERTEX_TEXTURE_EXTRA_ROWS`
        // rows. This helps with memory overhead, especially because there are several
        // instances of these textures per Renderer.
        if needed_height > existing_height
            || needed_height + VERTEX_TEXTURE_EXTRA_ROWS < existing_height
        {
            // Drop the existing texture, if any.
            if let Some(t) = self.texture.take() {
                device.delete_texture(t);
            }

            let texture = device.create_texture(
                api::ImageBufferKind::Texture2D,
                self.format,
                MAX_VERTEX_TEXTURE_WIDTH as i32,
                // Ensure height is at least two to work around
                // https://bugs.chromium.org/p/angleproject/issues/detail?id=3039
                needed_height.max(2),
                TextureFilter::Nearest,
                None,
            );
            self.texture = Some(texture);
        }

        // Note: the actual width can be larger than the logical one, with a few texels
        // of each row unused at the tail. This is needed because there is still hardware
        // (like Intel iGPUs) that prefers power-of-two sizes of textures ([1]).
        //
        // [1] https://software.intel.com/en-us/articles/opengl-performance-tips-power-of-two-textures-have-better-performance
        let logical_width = if needed_height == 1 {
            data.len() * texels_per_item
        } else {
            MAX_VERTEX_TEXTURE_WIDTH - (MAX_VERTEX_TEXTURE_WIDTH % texels_per_item)
        };

        let rect = DeviceIntRect::from_size(
            DeviceIntSize::new(logical_width as i32, needed_height),
        );

        debug_assert!(len <= data.capacity(), "CPU copy will read out of bounds");
        texture_uploader.upload(
            device,
            self.texture(),
            rect,
            None,
            None,
            data.as_ptr(),
            len,
        );
    }

    pub fn deinit(mut self, device: &mut Device) {
        if let Some(t) = self.texture.take() {
            device.delete_texture(t);
        }
    }
}

pub struct VertexDataTextures {
    prim_header_f_texture: VertexDataTexture<PrimitiveHeaderF>,
    prim_header_i_texture: VertexDataTexture<PrimitiveHeaderI>,
    transforms_texture: VertexDataTexture<TransformData>,
    render_task_texture: VertexDataTexture<RenderTaskData>,
}

impl VertexDataTextures {
    pub fn new() -> Self {
        VertexDataTextures {
            prim_header_f_texture: VertexDataTexture::new(api::ImageFormat::RGBAF32),
            prim_header_i_texture: VertexDataTexture::new(api::ImageFormat::RGBAI32),
            transforms_texture: VertexDataTexture::new(api::ImageFormat::RGBAF32),
            render_task_texture: VertexDataTexture::new(api::ImageFormat::RGBAF32),
        }
    }

    pub fn update(&mut self, device: &mut Device, pbo_pool: &mut UploadPBOPool, frame: &mut Frame) {
        let mut texture_uploader = device.upload_texture(pbo_pool);
        self.prim_header_f_texture.update(
            device,
            &mut texture_uploader,
            &mut frame.prim_headers.headers_float,
        );
        self.prim_header_i_texture.update(
            device,
            &mut texture_uploader,
            &mut frame.prim_headers.headers_int,
        );
        self.transforms_texture
            .update(device, &mut texture_uploader, &mut frame.transform_palette);
        self.render_task_texture.update(
            device,
            &mut texture_uploader,
            &mut frame.render_tasks.task_data,
        );

        // Flush and drop the texture uploader now, so that
        // we can borrow the textures to bind them.
        texture_uploader.flush(device);

        device.bind_texture(
            super::TextureSampler::PrimitiveHeadersF,
            &self.prim_header_f_texture.texture(),
            Swizzle::default(),
        );
        device.bind_texture(
            super::TextureSampler::PrimitiveHeadersI,
            &self.prim_header_i_texture.texture(),
            Swizzle::default(),
        );
        device.bind_texture(
            super::TextureSampler::TransformPalette,
            &self.transforms_texture.texture(),
            Swizzle::default(),
        );
        device.bind_texture(
            super::TextureSampler::RenderTasks,
            &self.render_task_texture.texture(),
            Swizzle::default(),
        );
    }

    pub fn size_in_bytes(&self) -> usize {
        self.prim_header_f_texture.size_in_bytes()
            + self.prim_header_i_texture.size_in_bytes()
            + self.transforms_texture.size_in_bytes()
            + self.render_task_texture.size_in_bytes()
    }

    pub fn deinit(self, device: &mut Device) {
        self.transforms_texture.deinit(device);
        self.prim_header_f_texture.deinit(device);
        self.prim_header_i_texture.deinit(device);
        self.render_task_texture.deinit(device);
    }
}

pub struct RendererVAOs {
    prim_vao: VAO,
    blur_vao: VAO,
    clip_rect_vao: VAO,
    border_vao: VAO,
    line_vao: VAO,
    scale_vao: VAO,
    svg_filter_node_vao: VAO,
    composite_vao: VAO,
    clear_vao: VAO,
    copy_vao: VAO,
    mask_vao: VAO,
}

impl RendererVAOs {
    pub fn new(device: &mut Device, indexed_quads: Option<NonZeroUsize>) -> Self {
        const QUAD_INDICES: [u16; 6] = [0, 1, 2, 2, 1, 3];
        const QUAD_VERTICES: [[u8; 2]; 4] = [[0, 0], [0xFF, 0], [0, 0xFF], [0xFF, 0xFF]];

        let instance_divisor = if indexed_quads.is_some() { 0 } else { 1 };
        let prim_vao = device.create_vao(&desc::PRIM_INSTANCES, instance_divisor);

        device.bind_vao(&prim_vao);
        match indexed_quads {
            Some(count) => {
                assert!(count.get() < u16::MAX as usize);
                let quad_indices = (0 .. count.get() as u16)
                    .flat_map(|instance| QUAD_INDICES.iter().map(move |&index| instance * 4 + index))
                    .collect::<Vec<_>>();
                device.update_vao_indices(&prim_vao, &quad_indices, VertexUsageHint::Static);
                let quad_vertices = (0 .. count.get() as u16)
                    .flat_map(|_| QUAD_VERTICES.iter().cloned())
                    .collect::<Vec<_>>();
                device.update_vao_main_vertices(&prim_vao, &quad_vertices, VertexUsageHint::Static);
            }
            None => {
                device.update_vao_indices(&prim_vao, &QUAD_INDICES, VertexUsageHint::Static);
                device.update_vao_main_vertices(&prim_vao, &QUAD_VERTICES, VertexUsageHint::Static);
            }
        }

        RendererVAOs {
            blur_vao: device.create_vao_with_new_instances(&desc::BLUR, &prim_vao),
            clip_rect_vao: device.create_vao_with_new_instances(&desc::CLIP_RECT, &prim_vao),
            border_vao: device.create_vao_with_new_instances(&desc::BORDER, &prim_vao),
            scale_vao: device.create_vao_with_new_instances(&desc::SCALE, &prim_vao),
            line_vao: device.create_vao_with_new_instances(&desc::LINE, &prim_vao),
            svg_filter_node_vao: device.create_vao_with_new_instances(&desc::SVG_FILTER_NODE, &prim_vao),
            composite_vao: device.create_vao_with_new_instances(&desc::COMPOSITE, &prim_vao),
            clear_vao: device.create_vao_with_new_instances(&desc::CLEAR, &prim_vao),
            copy_vao: device.create_vao_with_new_instances(&desc::COPY, &prim_vao),
            mask_vao: device.create_vao_with_new_instances(&desc::MASK, &prim_vao),
            prim_vao,
        }
    }

    pub fn deinit(self, device: &mut Device) {
        device.delete_vao(self.prim_vao);
        device.delete_vao(self.clip_rect_vao);
        device.delete_vao(self.blur_vao);
        device.delete_vao(self.line_vao);
        device.delete_vao(self.border_vao);
        device.delete_vao(self.scale_vao);
        device.delete_vao(self.svg_filter_node_vao);
        device.delete_vao(self.composite_vao);
        device.delete_vao(self.clear_vao);
        device.delete_vao(self.copy_vao);
        device.delete_vao(self.mask_vao);
    }
}

impl ops::Index<VertexArrayKind> for RendererVAOs {
    type Output = VAO;
    fn index(&self, kind: VertexArrayKind) -> &VAO {
        match kind {
            VertexArrayKind::Primitive => &self.prim_vao,
            VertexArrayKind::ClipRect => &self.clip_rect_vao,
            VertexArrayKind::Blur => &self.blur_vao,
            VertexArrayKind::Border => &self.border_vao,
            VertexArrayKind::Scale => &self.scale_vao,
            VertexArrayKind::LineDecoration => &self.line_vao,
            VertexArrayKind::SvgFilterNode => &self.svg_filter_node_vao,
            VertexArrayKind::Composite => &self.composite_vao,
            VertexArrayKind::Clear => &self.clear_vao,
            VertexArrayKind::Copy => &self.copy_vao,
            VertexArrayKind::Mask => &self.mask_vao,
        }
    }
}