Adding upstream version 115.7.0esr.upstream/115.7.0esr

Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>

27 files changed, 10137 insertions, 0 deletions

diff --git a/third_party/rust/wpf-gpu-raster/.cargo-checksum.json b/third_party/rust/wpf-gpu-raster/.cargo-checksum.json
new file mode 100644
index 0000000000..90a4069198
--- /dev/null
+++ b/third_party/rust/wpf-gpu-raster/.cargo-checksum.json
@@ -0,0 +1 @@
+{"files":{".github/workflows/coverage.yml":"90aaa068c16cb778b24badaff78baf2a313637780a723be09596abde0f4c827a",".github/workflows/rust.yml":"905954be896d052ced621eedb9d5b9d35795490f27071ac1147e75ac3b3711ec","CHANGES.md":"5f54e553a1c4ef21c5be6109b25df9d1d63c4547627723fe044c73dbddf0db2f","Cargo.toml":"912553cc7b6024b0acbba42d6b64020fc98086596e99765ae95c1808ad0ab2d1","LICENSE":"ae48df11a335dc1a615f4f938b69cba73bcf4485c4f97af49b38efb0f216353b","README.md":"e14b7ddbd29b6f87d956921999da1cf7bc3add0166cacf21e8b1ac1d9092a90d","examples/draw.rs":"52fee9e2f2c11e1c891b30cb460be2a0ec65974f38dc0c08fd48391caf1e4247","examples/obj-output.rs":"6fc549022aa715eee74ea1cafb89ca33189e9dbe914ea6b2c46160049bda68f3","examples/simple.rs":"99fb566414cbd4a0eb69a2774c9780d7cd17e5cdaa14837b280fba319c053f22","notes":"48e636c646d697e213b3a79e31063e11b6ffc7493592d31f3929b1db495870b8","src/aacoverage.rs":"fdadadd208caa986cc386797f937a976b5a315174c7c0782b87c0334d6474a97","src/aarasterizer.rs":"283bed1e22917118f332b24731cb6bd11334a4f0ba0d88821cfeb6b607de12da","src/bezier.rs":"f089ab04e30077ce4e0fe59dfa602948b989aa53d51ad207fbc30c1edd24086b","src/c_bindings.rs":"9c5ab638cf0a14220d93528e37cdc0f6d83277eaa10acf9ce36f32a28e30c02b","src/fix.rs":"7ccf63db5bab4ab0135d92691f7c2272a27866b9792dd55ec98b2d1c1b7c0358","src/geometry_sink.rs":"9025569f77f475a1e47fd470e8f53dcdf88ef57e3a5b8a51268fff892da8b1a7","src/helpers.rs":"220294dac335943518f249c4a27ad803f8226ed62cd780f517e95be6343a1f2f","src/hwrasterizer.rs":"82b2d6d35488a6ad7de4d82f3ee38c6f09f4b6de06b4f98eea61b3abdd72eb62","src/hwvertexbuffer.rs":"f3dd54f17570eb530c9c827b24a53b755a2dfa6028e9b83f9d7a4ba9945c2ecf","src/lib.rs":"8cd659ff6e2165884eaa1657462e1a5bf15a68a83cb768c2dff2f433ff0a022c","src/matrix.rs":"1ac44bc5d073f96ab64b1b5c6077fd0d47fe61db8243bd9a55fc91d8eae1dd92","src/notes":"d50d49e0b5660bc6350d8055f25f26700c937558de0af690e1fc4f50ed7e05c9","src/nullable_ref.rs":"789fe0e59b7d4a925faecbf2362be93643ea8382b4424ca0e60866f9bf83c3cd","src/real.rs":"73a2d1a77613364e9514fd7ead4d708a554d2b7343645cdb4cb8a2b3b640e057","src/tri_rasterize.rs":"30821a3465cea3c5ac578590013b530c03ea3010225f580d6cf609e39910c412","src/types.rs":"b840212a99a212ef38211aaf1bd801ec83416569541941d15fd95285d1342b99"},"package":null}
\ No newline at end of file
diff --git a/third_party/rust/wpf-gpu-raster/.github/workflows/coverage.yml b/third_party/rust/wpf-gpu-raster/.github/workflows/coverage.yml
new file mode 100644
index 0000000000..3f2e6f2523
--- /dev/null
+++ b/third_party/rust/wpf-gpu-raster/.github/workflows/coverage.yml
@@ -0,0 +1,23 @@
+name: Coverage
+
+on: [pull_request, push]
+
+jobs:
+  coverage:
+    runs-on: ubuntu-latest
+    env:
+      CARGO_TERM_COLOR: always
+    steps:
+      - uses: actions/checkout@v3
+      - name: Install Rust
+        run: rustup toolchain install stable --component llvm-tools-preview
+      - name: Install cargo-llvm-cov
+        uses: taiki-e/install-action@cargo-llvm-cov
+      - name: Generate code coverage
+        run: cargo llvm-cov --all-features --workspace --lcov --output-path lcov.info
+      - name: Upload coverage to Codecov
+        uses: codecov/codecov-action@v3
+        with:
+          token: ${{ secrets.CODECOV_TOKEN }} # not required for public repos
+          files: lcov.info
+          fail_ci_if_error: true
diff --git a/third_party/rust/wpf-gpu-raster/.github/workflows/rust.yml b/third_party/rust/wpf-gpu-raster/.github/workflows/rust.yml
new file mode 100644
index 0000000000..1f38bfc1f7
--- /dev/null
+++ b/third_party/rust/wpf-gpu-raster/.github/workflows/rust.yml
@@ -0,0 +1,39 @@
+name: Rust
+
+on:
+  push:
+  pull_request:
+
+env:
+  CARGO_TERM_COLOR: always
+
+jobs:
+  build:
+
+    runs-on: ubuntu-latest
+
+    steps:
+    - uses: actions/checkout@v3
+    - name: Build
+      run: cargo build --verbose
+    - name: Run tests
+      run: cargo test --verbose
+
+  aarch64:
+    runs-on: ubuntu-20.04
+    steps:
+    - name: Checkout
+      uses: actions/checkout@v2
+
+    - name: Install toolchain
+      uses: actions-rs/toolchain@v1
+      with:
+        toolchain: stable
+        override: true
+        target: aarch64-unknown-linux-gnu
+
+    - name: Install cross
+      run: cargo install cross
+
+    - name: Run tests with Neon
+      run: cross test --target aarch64-unknown-linux-gnu
diff --git a/third_party/rust/wpf-gpu-raster/CHANGES.md b/third_party/rust/wpf-gpu-raster/CHANGES.md
new file mode 100644
index 0000000000..8de9816947
--- /dev/null
+++ b/third_party/rust/wpf-gpu-raster/CHANGES.md
@@ -0,0 +1,22 @@
+Changes for Safety
+------------------
+
+`CEdgeStore` is replaced by `typed_arena_nomut::Arena<CEdge>`. 
+
+`CEdgeStore` is an arena with built-in stack storage for the first allocation
+of the arena. It exposes the allocated buffers to support very fast allocation,
+and supports fast enumeration by returning pointers to each allocation.
+
+`CCoverageBuffer` also now uses a `typed_arena_nomut::Arena<CEdge>` but uses it
+to allocate `CCoverageIntervalBuffer`'s. We currently lack support for
+the builtin stack storage. Storing these in an Arena is not ideal, we'd rather
+just heap allocate them individually.
+
+
+Changes for performance
+-----------------------
+
+Switched from using triangle strips to triangle lists. This lets
+us use a single triangle to draw each line segement which reduces
+the amount of geometry per line segment from 6 vertices to 3.
+Direct2D also made this switch in later versions.
diff --git a/third_party/rust/wpf-gpu-raster/Cargo.toml b/third_party/rust/wpf-gpu-raster/Cargo.toml
new file mode 100644
index 0000000000..79ec71f76c
--- /dev/null
+++ b/third_party/rust/wpf-gpu-raster/Cargo.toml
@@ -0,0 +1,32 @@
+# THIS FILE IS AUTOMATICALLY GENERATED BY CARGO
+#
+# When uploading crates to the registry Cargo will automatically
+# "normalize" Cargo.toml files for maximal compatibility
+# with all versions of Cargo and also rewrite `path` dependencies
+# to registry (e.g., crates.io) dependencies.
+#
+# If you are reading this file be aware that the original Cargo.toml
+# will likely look very different (and much more reasonable).
+# See Cargo.toml.orig for the original contents.
+
+[package]
+edition = "2021"
+name = "wpf-gpu-raster"
+version = "0.1.0"
+readme = "README.md"
+license = "MIT"
+
+[profile.release]
+debug = true
+
+[dependencies]
+typed-arena-nomut = "0.1.0"
+
+[dev-dependencies]
+euclid = "0.22.6"
+png = "0.17.2"
+usvg = "0.4"
+
+[features]
+c_bindings = []
+default = ["c_bindings"]
diff --git a/third_party/rust/wpf-gpu-raster/LICENSE b/third_party/rust/wpf-gpu-raster/LICENSE
new file mode 100644
index 0000000000..a616ed188d
--- /dev/null
+++ b/third_party/rust/wpf-gpu-raster/LICENSE
@@ -0,0 +1,23 @@
+The MIT License (MIT)
+
+Copyright (c) .NET Foundation and Contributors
+
+All rights reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
\ No newline at end of file
diff --git a/third_party/rust/wpf-gpu-raster/README.md b/third_party/rust/wpf-gpu-raster/README.md
new file mode 100644
index 0000000000..1d4756b13f
--- /dev/null
+++ b/third_party/rust/wpf-gpu-raster/README.md
@@ -0,0 +1,22 @@
+This is a port of the WPF hardware rasterizer code to Rust. That
+rasterizer is predecessor to the Direct2D rasterizer. Direct2D still
+uses a similar technique when run on hardware that does not support
+Target Independent Rasterization.
+
+Design
+======
+
+The general algorithm used for rasterization is a vertical sweep of
+the shape that maintains an active edge list.  The sweep is done
+at a sub-scanline resolution and results in either:
+   1. Sub-scanlines being combined in the coverage buffer and output
+      as "complex scans". These are emitted as lines constructed out
+      of triangle strips.
+   2. Simple trapezoids being recognized in the active edge list
+      and output using a faster simple trapezoid path.
+
+Bezier flattening is done using an approach that uses forward differencing
+of the error metric to compute a flattened version that would match a traditional
+adaptive recursive flattening.
+
+
diff --git a/third_party/rust/wpf-gpu-raster/examples/draw.rs b/third_party/rust/wpf-gpu-raster/examples/draw.rs
new file mode 100644
index 0000000000..828449ee72
--- /dev/null
+++ b/third_party/rust/wpf-gpu-raster/examples/draw.rs
@@ -0,0 +1,354 @@
+/* The rasterization code here is based off of piglit/tests/general/triangle-rasterization.cpp:
+
+    /**************************************************************************
+     *
+     * Copyright 2012 VMware, Inc.
+     * All Rights Reserved.
+     *
+     * Permission is hereby granted, free of charge, to any person obtaining a
+     * copy of this software and associated documentation files (the
+     * "Software"), to deal in the Software without restriction, including
+     * without limitation the rights to use, copy, modify, merge, publish,
+     * distribute, sub license, and/or sell copies of the Software, and to
+     * permit persons to whom the Software is furnished to do so, subject to
+     * the following conditions:
+     *
+     * The above copyright notice and this permission notice (including the
+     * next paragraph) shall be included in all copies or substantial portions
+     * of the Software.
+     *
+     * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+     * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+     * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+     * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
+     * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+     * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+     * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+     *
+     **************************************************************************/
+
+*/
+
+#![allow(non_camel_case_types)]
+#![allow(non_snake_case)]
+use euclid::{default::Transform2D, point2};
+use wpf_gpu_raster::{PathBuilder};
+
+
+use std::ops::Index;
+
+
+const  WIDTH: u32  = 800;
+const HEIGHT: u32  = 800;
+
+
+fn over(src: u32, dst: u32) -> u32 {
+    let a = src >> 24;
+    let a = 255 - a;
+    let mask = 0xff00ff;
+    let t = (dst & mask) * a + 0x800080;
+    let mut rb = (t + ((t >> 8) & mask)) >> 8;
+    rb &= mask;
+
+    rb += src & mask;
+
+    // saturate
+    rb |= 0x1000100 - ((rb >> 8) & mask);
+    rb &= mask;
+
+    let t = ((dst >> 8) & mask) * a + 0x800080;
+    let mut ag = (t + ((t >> 8) & mask)) >> 8;
+    ag &= mask;
+    ag += (src >> 8) & mask;
+
+    // saturate
+    ag |= 0x1000100 - ((ag >> 8) & mask);
+    ag &= mask;
+
+    (ag << 8) + rb
+}
+
+pub fn alpha_mul(x: u32, a: u32) -> u32 {
+    let mask = 0xFF00FF;
+
+    let src_rb = ((x & mask) * a) >> 8;
+    let src_ag = ((x >> 8) & mask) * a;
+
+    (src_rb & mask) | (src_ag & !mask)
+}
+
+fn write_image(data: &[u32], path: &str) {
+    use std::path::Path;
+    use std::fs::File;
+    use std::io::BufWriter;
+
+    let mut png_data: Vec<u8> = vec![0; (WIDTH * HEIGHT * 3) as usize];
+    let mut i = 0;
+    for pixel in data {
+        png_data[i] = ((pixel >> 16) & 0xff) as u8;
+        png_data[i + 1] = ((pixel >> 8) & 0xff) as u8;
+        png_data[i + 2] = ((pixel >> 0) & 0xff) as u8;
+        i += 3;
+    }
+
+
+    let path = Path::new(path);
+    let file = File::create(path).unwrap();
+    let w = &mut BufWriter::new(file);
+
+    let mut encoder = png::Encoder::new(w, WIDTH, HEIGHT); // Width is 2 pixels and height is 1.
+    encoder.set_color(png::ColorType::Rgb);
+    encoder.set_depth(png::BitDepth::Eight);
+    let mut writer = encoder.write_header().unwrap();
+
+    writer.write_image_data(&png_data).unwrap(); // Save
+}
+
+#[derive(Debug)]
+struct Vertex {
+    x: f32,
+    y: f32,
+    coverage: f32
+}
+#[derive(Debug)]
+struct Triangle {
+    v: [Vertex; 3],
+}
+
+impl Index<usize> for Triangle  {
+    type Output = Vertex;
+
+    fn index(&self, index: usize) -> &Self::Output {
+        &self.v[index]
+    }
+}
+
+// D3D11 mandates 8 bit subpixel precision:
+// https://microsoft.github.io/DirectX-Specs/d3d/archive/D3D11_3_FunctionalSpec.htm#CoordinateSnapping
+const FIXED_SHIFT: i32 = 8;
+const FIXED_ONE: f32 = (1 << FIXED_SHIFT) as f32;
+
+/* Proper rounding of float to integer */
+fn iround(mut v: f32) -> i64 {
+    if v > 0.0 {
+        v += 0.5;
+    }
+    if v < 0.0 {
+        v -= 0.5;
+    }
+    return v as i64
+}
+
+/* Based on http://devmaster.net/forums/topic/1145-advanced-rasterization */
+fn rast_triangle(buffer: &mut [u32], stride: usize, tri: &Triangle, color: u32) {
+    let center_offset = -0.5;
+
+    let mut coverage1 = tri[0].coverage;
+    let mut coverage2 = tri[1].coverage;
+    let mut coverage3 = tri[2].coverage;
+
+    /* fixed point coordinates */
+    let mut x1 = iround(FIXED_ONE * (tri[0].x + center_offset));
+    let     x2 = iround(FIXED_ONE * (tri[1].x + center_offset));
+    let mut x3 = iround(FIXED_ONE * (tri[2].x + center_offset));
+
+    let mut y1 = iround(FIXED_ONE * (tri[0].y + center_offset));
+    let     y2 = iround(FIXED_ONE * (tri[1].y + center_offset));
+    let mut y3 = iround(FIXED_ONE * (tri[2].y + center_offset));
+
+
+    /* Force correct vertex order */
+    let cross = (x2 - x1) * (y3 - y2) - (y2 - y1) * (x3 - x2);
+    if cross > 0 {
+        std::mem::swap(&mut x1, &mut x3);
+        std::mem::swap(&mut y1, &mut y3);
+        // I don't understand why coverage 2 and 3 are swapped instead of 1 and 3
+        std::mem::swap(&mut coverage2, &mut coverage3);
+    } else {
+        std::mem::swap(&mut coverage1, &mut coverage3);
+    }
+    
+    /* Deltas */
+    let dx12 = x1 - x2;
+    let dx23 = x2 - x3;
+    let dx31 = x3 - x1;
+
+    let dy12 = y1 - y2;
+    let dy23 = y2 - y3;
+    let dy31 = y3 - y1;
+
+    /* Fixed-point deltas */
+    let fdx12 = dx12 << FIXED_SHIFT;
+    let fdx23 = dx23 << FIXED_SHIFT;
+    let fdx31 = dx31 << FIXED_SHIFT;
+
+    let fdy12 = dy12 << FIXED_SHIFT;
+    let fdy23 = dy23 << FIXED_SHIFT;
+    let fdy31 = dy31 << FIXED_SHIFT;
+
+    /* Bounding rectangle */
+    let mut minx = x1.min(x2).min(x3) >> FIXED_SHIFT;
+    let mut maxx = x1.max(x2).max(x3) >> FIXED_SHIFT;
+
+    let mut miny = y1.min(y2).min(y3) >> FIXED_SHIFT;
+    let mut maxy = y1.max(y2).max(y3) >> FIXED_SHIFT;
+
+    minx = minx.max(0);
+    maxx = maxx.min(WIDTH as i64 - 1);
+
+    miny = miny.max(0);
+    maxy = maxy.min(HEIGHT as i64 - 1);
+
+    /* Half-edge constants */
+    let mut c1 = dy12 * x1 - dx12 * y1;
+    let mut c2 = dy23 * x2 - dx23 * y2;
+    let mut c3 = dy31 * x3 - dx31 * y3;
+
+    /* Correct for top-left filling convention */
+    if dy12 < 0 || (dy12 == 0 && dx12 < 0) { c1 += 1 }
+    if dy23 < 0 || (dy23 == 0 && dx23 < 0) { c2 += 1 }
+    if dy31 < 0 || (dy31 == 0 && dx31 < 0) { c3 += 1 }
+
+    let mut cy1 = c1 + dx12 * (miny << FIXED_SHIFT) - dy12 * (minx << FIXED_SHIFT);
+    let mut cy2 = c2 + dx23 * (miny << FIXED_SHIFT) - dy23 * (minx << FIXED_SHIFT);
+    let mut cy3 = c3 + dx31 * (miny << FIXED_SHIFT) - dy31 * (minx << FIXED_SHIFT);
+
+    /* Perform rasterization */
+    let mut buffer = &mut buffer[miny as usize * stride..];
+    for _y in miny..=maxy {
+        let mut cx1 = cy1;
+        let mut cx2 = cy2;
+        let mut cx3 = cy3;
+
+        for x in minx..=maxx {
+            if cx1 > 0 && cx2 > 0 && cx3 > 0 {
+                // cross is equal to 2*area of the triangle.
+                // we can normalize cx by 2*area to get barycentric coords.
+                let area = cross.abs() as f32;
+                let bary = (cx1 as f32 / area, cx2 as f32/ area, cx3 as f32 / area);
+
+                let coverages = coverage1 * bary.0 + coverage2 * bary.1 + coverage3 * bary.2;
+
+                let color = alpha_mul(color, (coverages * 256. + 0.5) as u32);
+                buffer[x as usize] = over(color, buffer[x as usize]);
+            }
+
+            cx1 -= fdy12;
+            cx2 -= fdy23;
+            cx3 -= fdy31;
+        }
+
+        cy1 += fdx12;
+        cy2 += fdx23;
+        cy3 += fdx31;
+
+        buffer = &mut buffer[stride..];
+    }
+}
+
+
+fn main() {
+    let opt = usvg::Options::default();
+
+    let rtree = usvg::Tree::from_file("tiger.svg", &opt).unwrap();
+
+    let mut image = vec![0; (WIDTH * HEIGHT) as usize];
+    for _ in 0..1 {
+    let mut total_vertex_count = 0;
+    let mut total_time = std::time::Duration::default();
+    for node in rtree.root().descendants() {
+        use usvg::NodeExt;
+        let t = node.transform();
+        let transform = Transform2D::new(
+            t.a as f32, t.b as f32,
+            t.c as f32, t.d as f32,
+            t.e as f32, t.f as f32,
+        );
+
+
+        let s = 1.;
+        if let usvg::NodeKind::Path(ref usvg_path) = *node.borrow() {
+            let color = match usvg_path.fill {
+                Some(ref fill) => {
+                    match fill.paint {
+                        usvg::Paint::Color(c) => 0xff000000 | (c.red as u32) << 16 | (c.green as u32) << 8 | c.blue as u32,
+                        _ => 0xff00ff00,
+                    }
+                }
+                None => {
+                    continue;
+                }
+            };
+            let mut builder = PathBuilder::new();
+            //dbg!(&usvg_path.segments);
+            for segment in &usvg_path.segments {
+                match *segment {
+                    usvg::PathSegment::MoveTo { x, y } => {
+                        let p = transform.transform_point(point2(x as f32, y as f32)) * s;
+                        builder.move_to(p.x, p.y);
+                    }
+                    usvg::PathSegment::LineTo { x, y } => {
+                        let p = transform.transform_point(point2(x as f32, y as f32)) * s;
+                        builder.line_to(p.x, p.y);
+                    }
+                    usvg::PathSegment::CurveTo { x1, y1, x2, y2, x, y, } => {
+                        let c1 = transform.transform_point(point2(x1 as f32, y1 as f32)) * s;
+                        let c2 = transform.transform_point(point2(x2 as f32, y2 as f32)) * s;
+                        let p = transform.transform_point(point2(x as f32, y as f32)) * s;
+                        builder.curve_to(
+                            c1.x, c1.y,
+                            c2.x, c2.y,
+                            p.x, p.y,
+                        );
+                    }
+                    usvg::PathSegment::ClosePath => {
+                        builder.close();
+                    }
+                }
+            }
+            let start = std::time::Instant::now(); 
+            let result = builder.rasterize_to_tri_list(0, 0, WIDTH as i32, HEIGHT as i32);
+            let end = std::time::Instant::now();
+            total_time += end - start;
+
+            println!("vertices {}", result.len());
+            total_vertex_count += result.len();
+            if result.len() == 0 {
+                continue;
+            }
+
+            for n in (0..result.len()).step_by(3) {
+                let vertices =  {
+                    [&result[n], &result[n+1], &result[n+2]]
+                };
+
+                let src = color;
+                let tri = Triangle { v: [
+                    Vertex { x: vertices[0].x, y: vertices[0].y, coverage: vertices[0].coverage},
+                    Vertex { x: vertices[1].x, y: vertices[1].y, coverage: vertices[1].coverage},
+                    Vertex { x: vertices[2].x, y: vertices[2].y, coverage: vertices[2].coverage}
+                    ]
+                };
+                rast_triangle(&mut image, WIDTH as usize, &tri, src);
+            }
+        }
+    }
+
+    println!("total vertex count {}, took {}ms", total_vertex_count, total_time.as_secs_f32()*1000.);
+    }
+
+
+    write_image(&image, "out.png");
+    use std::{hash::{Hash, Hasher}, collections::hash_map::DefaultHasher};
+    use crate::*;
+    fn calculate_hash<T: Hash>(t: &T) -> u64 {
+        let mut s = DefaultHasher::new();
+        t.hash(&mut s);
+        s.finish()
+    }
+
+    assert_eq!(calculate_hash(&image),
+        if cfg!(debug_assertions) { 0x5973c52a1c0232f3 } else { 0xf15821a5bebc5ecf});
+
+
+}
diff --git a/third_party/rust/wpf-gpu-raster/examples/obj-output.rs b/third_party/rust/wpf-gpu-raster/examples/obj-output.rs
new file mode 100644
index 0000000000..c7ec8d7686
--- /dev/null
+++ b/third_party/rust/wpf-gpu-raster/examples/obj-output.rs
@@ -0,0 +1,26 @@
+// Output an .obj file of the generated mesh. Viewable at https://3dviewer.net/
+
+fn output_obj_file(data: &[OutputVertex]) {
+        for v in data {
+                let color = v.coverage;
+                println!("v {} {} {} {} {} {}", v.x, v.y, 0., color, color, color);
+        }
+
+        // output a standard triangle strip face list
+        for n in (1..data.len()-1).step_by(3) {
+                println!("f {} {} {}", n, n+1, n+2);
+        }
+}
+
+use wpf_gpu_raster::{PathBuilder, OutputVertex};
+fn main() {
+    let mut p = PathBuilder::new();
+    p.move_to(10., 10.0);
+    p.line_to(30., 10.);
+    p.line_to(50., 20.);
+    p.line_to(30., 30.);
+    p.line_to(10., 30.);
+    p.close();
+    let result = p.rasterize_to_tri_list(0, 0, 100, 100);
+    output_obj_file(&result)
+}
diff --git a/third_party/rust/wpf-gpu-raster/examples/simple.rs b/third_party/rust/wpf-gpu-raster/examples/simple.rs
new file mode 100644
index 0000000000..5b82cdd941
--- /dev/null
+++ b/third_party/rust/wpf-gpu-raster/examples/simple.rs
@@ -0,0 +1,11 @@
+use wpf_gpu_raster::PathBuilder;
+fn main() {
+    let mut p = PathBuilder::new();
+    p.move_to(10., 10.);
+    p.line_to(10., 30.);
+    p.line_to(30., 30.);
+    p.line_to(30., 10.);
+    p.close();
+    let _result = p.rasterize_to_tri_list(0, 0, 100, 100);
+    //dbg!(result);
+}
diff --git a/third_party/rust/wpf-gpu-raster/notes b/third_party/rust/wpf-gpu-raster/notes
new file mode 100644
index 0000000000..8550376eac
--- /dev/null
+++ b/third_party/rust/wpf-gpu-raster/notes
@@ -0,0 +1,8 @@
+bezier flattening
+-----------------
+if we make sure we flatten beziers to integer y values we can avoid having to hit
+the slow complex coverage path
+
+We can probably do this by using a skia style flattener.
+Normally we compute a series of line segments using partial differencing.
+I think we can adjust the line towards an integer y value by having small partial differences that we can move by.
diff --git a/third_party/rust/wpf-gpu-raster/src/aacoverage.rs b/third_party/rust/wpf-gpu-raster/src/aacoverage.rs
new file mode 100644
index 0000000000..7b165a9da0
--- /dev/null
+++ b/third_party/rust/wpf-gpu-raster/src/aacoverage.rs
@@ -0,0 +1,647 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+// See the LICENSE file in the project root for more information.
+
+
+//------------------------------------------------------------------------------
+//
+
+use std::cell::Cell;
+
+use typed_arena_nomut::Arena;
+
+//
+//  Description:
+//      Coverage buffer implementation
+#[cfg(debug_assertions)]
+use crate::aarasterizer::AssertActiveList;
+use crate::aarasterizer::CEdge;
+use crate::nullable_ref::Ref;
+use crate::types::*;
+//struct CEdge;
+//struct CInactiveEdge;
+
+//-------------------------------------------------------------------------
+//
+// TrapezoidalAA only supports 8x8 mode, so the shifts/masks are all
+// constants.  Also, since we must be symmetrical, x and y shifts are
+// merged into one shift unlike the implementation in aarasterizer.
+//
+//-------------------------------------------------------------------------
+
+pub const c_nShift: INT = 3; 
+pub const c_nShiftSize: INT = 8; 
+pub const c_nShiftSizeSquared: INT = c_nShiftSize * c_nShiftSize; 
+pub const c_nHalfShiftSize: INT = 4; 
+pub const c_nShiftMask: INT = 7; 
+//pub const c_rShiftSize: f32 = 8.0;
+//pub const c_rHalfShiftSize: f32 = 4.0;
+pub const c_rInvShiftSize: f32 = 1.0/8.0;
+pub const c_antiAliasMode: MilAntiAliasMode = MilAntiAliasMode::EightByEight;
+
+//
+// Interval coverage descriptor for our antialiased filler
+//
+
+pub struct CCoverageInterval<'a>
+{
+    pub m_pNext: Cell<Ref<'a, CCoverageInterval<'a>>>, // m_pNext interval (look for sentinel, not NULL)
+    pub m_nPixelX: Cell<INT>,              // Interval's left edge (m_pNext->X is the right edge)
+    pub m_nCoverage: Cell<INT>,            // Pixel coverage for interval
+}
+
+impl<'a> Default for CCoverageInterval<'a> {
+    fn default() -> Self {
+        Self { m_pNext: Cell::new(unsafe { Ref::null() } ), m_nPixelX: Default::default(), m_nCoverage: Default::default() }
+    }
+}
+
+// Define our on-stack storage use.  The 'free' versions are nicely tuned
+// to avoid allocations in most common scenarios, while at the same time
+// not chewing up toooo much stack space.  
+//
+// We make the debug versions small so that we hit the 'grow' cases more
+// frequently, for better testing:
+
+#[cfg(debug_assertions)]
+    // Must be at least 6 now: 4 for the "minus4" logic in hwrasterizer.*, and then 
+    // 1 each for the head and tail sentinels (since their allocation doesn't use Grow).
+    const INTERVAL_BUFFER_NUMBER: usize = 8;        
+#[cfg(not(debug_assertions))]
+    const INTERVAL_BUFFER_NUMBER: usize = 32;
+
+
+//
+// Allocator structure for the antialiased fill interval data
+//
+
+struct CCoverageIntervalBuffer<'a>
+{
+    m_pNext: Cell<Option<& 'a CCoverageIntervalBuffer<'a>>>,
+    m_interval: [CCoverageInterval<'a>; INTERVAL_BUFFER_NUMBER],
+}
+
+impl<'a>  Default for CCoverageIntervalBuffer<'a> {
+    fn default() -> Self {
+        Self { m_pNext: Cell::new(None), m_interval: Default::default() }
+    }
+}
+
+//------------------------------------------------------------------------------
+//
+//  Class: CCoverageBuffer
+//
+//  Description:
+//      Coverage buffer implementation that maintains coverage information
+//      for one scanline.  
+//
+//      This implementation will maintain a linked list of intervals consisting
+//      of x value in pixel space and a coverage value that applies for all pixels
+//      between pInterval->X and pInterval->Next->X.
+//
+//      For example, if we add the following interval (assuming 8x8 anti-aliasing)
+//      to the coverage buffer:
+//       _____ _____ _____ _____
+//      |     |     |     |     |
+//      |  -------------------  |
+//      |_____|_____|_____|_____|
+//    (0,0) (1,0) (2,0) (3,0) (4,0)
+//
+//      Then we will get the following coverage buffer:
+//
+//     m_nPixelX: INT_MIN  |  0  |  1  |  3  |  4  | INT_MAX
+//   m_nCoverage: 0        |  4  |  8  |  4  |  0  | 0xdeadbeef
+//       m_pNext: -------->|---->|---->|---->|---->| NULL
+//              
+//------------------------------------------------------------------------------
+pub struct CCoverageBuffer<'a>
+{
+    /*
+public:
+    //
+    // Init/Destroy methods
+    //
+
+    VOID Initialize();
+    VOID Destroy();
+
+    //
+    // Setup the buffer so that it can accept another scanline
+    //
+
+    VOID Reset();
+
+    //
+    // Add a subpixel interval to the coverage buffer
+    //
+
+    HRESULT FillEdgesAlternating(
+        __in_ecount(1) const CEdge *pEdgeActiveList,
+        INT nSubpixelYCurrent
+        );
+
+    HRESULT FillEdgesWinding(
+        __in_ecount(1) const CEdge *pEdgeActiveList,
+        INT nSubpixelYCurrent
+        );
+
+    HRESULT AddInterval(INT nSubpixelXLeft, INT nSubpixelXRight);
+
+private:
+
+    HRESULT Grow(
+        __deref_out_ecount(1) CCoverageInterval **ppIntervalNew, 
+        __deref_out_ecount(1) CCoverageInterval **ppIntervalEndMinus4
+        );
+
+public:*/
+    pub m_pIntervalStart: Cell<Ref<'a, CCoverageInterval<'a>>>,           // Points to list head entry
+
+//private:
+    m_pIntervalNew: Cell<Ref<'a, CCoverageInterval<'a>>>,
+    interval_new_index: Cell<usize>,
+
+    // The Minus4 in the below variable refers to the position at which
+    // we need to Grow the buffer.  The buffer is grown once before an
+    // AddInterval, so the Grow has to ensure that there are enough 
+    // intervals for the AddInterval worst case which is the following:
+    //
+    //  1     2           3     4
+    //  *_____*_____ _____*_____* 
+    //  |     |     |     |     |
+    //  |  ---|-----------|---  |
+    //  |_____|_____|_____|_____|
+    //
+    // Note that the *'s above mark potentional insert points in the list,
+    // so we need to ensure that at least 4 intervals can be allocated.
+    //
+
+    m_pIntervalEndMinus4:  Cell<Ref<'a, CCoverageInterval<'a>>>,
+
+    // Cache the next-to-last added interval to accelerate insertion.
+    m_pIntervalLast: Cell<Ref<'a, CCoverageInterval<'a>>>,
+
+    m_pIntervalBufferBuiltin: CCoverageIntervalBuffer<'a>,
+    m_pIntervalBufferCurrent: Cell<Ref<'a, CCoverageIntervalBuffer<'a>>>,
+
+    arena: Arena<CCoverageIntervalBuffer<'a>>
+       
+    // Disable instrumentation checks within all methods of this class
+    //SET_MILINSTRUMENTATION_FLAGS(MILINSTRUMENTATIONFLAGS_DONOTHING);
+}
+
+impl<'a> Default for CCoverageBuffer<'a> {
+    fn default() -> Self {
+        Self {
+            m_pIntervalStart: Cell::new(unsafe { Ref::null() }),
+            m_pIntervalNew: Cell::new(unsafe { Ref::null() }),
+            m_pIntervalEndMinus4: Cell::new(unsafe { Ref::null() }),
+            m_pIntervalLast: Cell::new(unsafe { Ref::null() }),
+            m_pIntervalBufferBuiltin: Default::default(),
+            m_pIntervalBufferCurrent: unsafe { Cell::new(Ref::null()) },
+            arena: Arena::new(),
+            interval_new_index: Cell::new(0),
+        }
+    }
+}
+
+
+//
+// Inlines
+//
+impl<'a> CCoverageBuffer<'a> {
+//-------------------------------------------------------------------------
+//
+//  Function:   CCoverageBuffer::AddInterval
+//
+//  Synopsis:   Add a subpixel resolution interval to the coverage buffer
+// 
+//-------------------------------------------------------------------------
+pub fn AddInterval(&'a self, nSubpixelXLeft: INT, nSubpixelXRight: INT) -> HRESULT
+{
+    let hr: HRESULT = S_OK;
+    let mut nPixelXNext: INT;
+    let nPixelXLeft: INT;
+    let nPixelXRight: INT;
+    let nCoverageLeft: INT;  // coverage from right edge of pixel for interval start
+    let nCoverageRight: INT; // coverage from left edge of pixel for interval end
+
+    let mut pInterval = self.m_pIntervalStart.get();
+    let mut pIntervalNew = self.m_pIntervalNew.get();
+    let mut interval_new_index = self.interval_new_index.get();
+    let mut pIntervalEndMinus4 = self.m_pIntervalEndMinus4.get();
+
+    // Make sure we have enough room to add two intervals if
+    // necessary:
+
+    if (pIntervalNew >= pIntervalEndMinus4)
+    {
+        IFC!(self.Grow(&mut pIntervalNew, &mut pIntervalEndMinus4, &mut interval_new_index));
+    }
+
+    // Convert interval to pixel space so that we can insert it 
+    // into the coverage buffer
+
+    debug_assert!(nSubpixelXLeft < nSubpixelXRight);
+    nPixelXLeft = nSubpixelXLeft >> c_nShift;
+    nPixelXRight = nSubpixelXRight >> c_nShift; 
+
+    // Try to resume searching from the last searched interval.
+    if self.m_pIntervalLast.get().m_nPixelX.get() < nPixelXLeft {
+        pInterval = self.m_pIntervalLast.get();
+    }
+
+    // Skip any intervals less than 'nPixelLeft':
+
+    loop {
+        let nextInterval = pInterval.m_pNext.get();
+        nPixelXNext = nextInterval.m_nPixelX.get();
+        if !(nPixelXNext < nPixelXLeft) { break }
+
+        pInterval = nextInterval;
+    }
+
+    // Remember the found interval.
+    self.m_pIntervalLast.set(pInterval);
+
+    // Insert a new interval if necessary:
+
+    if (nPixelXNext != nPixelXLeft)
+    {
+        pIntervalNew.m_nPixelX.set(nPixelXLeft);
+        pIntervalNew.m_nCoverage.set(pInterval.m_nCoverage.get());
+
+        pIntervalNew.m_pNext.set(pInterval.m_pNext.get());
+        pInterval.m_pNext.set(pIntervalNew);
+
+        pInterval = pIntervalNew;
+
+        interval_new_index += 1;
+        pIntervalNew = Ref::new(&Ref::get_ref(self.m_pIntervalBufferCurrent.get()).m_interval[interval_new_index])
+
+    }
+    else
+    {
+        pInterval = (*pInterval).m_pNext.get();
+    }
+
+    //
+    // Compute coverage for left segment as shown by the *'s below
+    //
+    //  |_____|_____|_____|_
+    //  |     |     |     |
+    //  |  ***----------  |
+    //  |_____|_____|_____|
+    //
+
+    nCoverageLeft = c_nShiftSize - (nSubpixelXLeft & c_nShiftMask);
+
+    // If nCoverageLeft == 0, then the value of nPixelXLeft is wrong
+    // and should have been equal to nPixelXLeft+1.
+    debug_assert!(nCoverageLeft > 0);
+
+    // If we have partial coverage, then ensure that we have a position
+    // for the end of the pixel 
+
+    if ((nCoverageLeft < c_nShiftSize || (nPixelXLeft == nPixelXRight))
+        && nPixelXLeft + 1 != pInterval.m_pNext.get().m_nPixelX.get())
+    {
+        pIntervalNew.m_nPixelX.set(nPixelXLeft + 1);
+        pIntervalNew.m_nCoverage.set(pInterval.m_nCoverage.get());
+
+        pIntervalNew.m_pNext.set(pInterval.m_pNext.get());
+        pInterval.m_pNext.set(pIntervalNew);
+
+        interval_new_index += 1;
+        pIntervalNew = Ref::new(&Ref::get_ref(self.m_pIntervalBufferCurrent.get()).m_interval[interval_new_index])
+    }
+    
+    //
+    // If the interval only includes one pixel, then the coverage is
+    // nSubpixelXRight - nSubpixelXLeft
+    //
+
+    if (nPixelXLeft == nPixelXRight)
+    {
+        pInterval.m_nCoverage.set(pInterval.m_nCoverage.get() + nSubpixelXRight - nSubpixelXLeft);
+        debug_assert!(pInterval.m_nCoverage.get() <= c_nShiftSize*c_nShiftSize);
+        //goto Cleanup;
+
+        //Cleanup:
+        // Update the coverage buffer new interval
+        self.interval_new_index.set(interval_new_index);
+        self.m_pIntervalNew.set(pIntervalNew);
+        return hr;
+    }
+
+    // Update coverage of current interval
+    pInterval.m_nCoverage.set(pInterval.m_nCoverage.get() + nCoverageLeft);
+    debug_assert!(pInterval.m_nCoverage.get() <= c_nShiftSize*c_nShiftSize);
+
+    // Increase the coverage for any intervals between 'nPixelXLeft'
+    // and 'nPixelXRight':
+
+    loop {
+        let nextInterval = pInterval.m_pNext.get();
+        (nPixelXNext = nextInterval.m_nPixelX.get());
+    
+        if !(nPixelXNext < nPixelXRight) {
+            break;
+        }
+        pInterval = nextInterval;
+        pInterval.m_nCoverage.set(pInterval.m_nCoverage.get() + c_nShiftSize);
+        debug_assert!(pInterval.m_nCoverage.get() <= c_nShiftSize*c_nShiftSize);
+    }
+
+    // Remember the found interval.
+    self.m_pIntervalLast.set(pInterval);
+
+    // Insert another new interval if necessary:
+
+    if (nPixelXNext != nPixelXRight)
+    {
+        pIntervalNew.m_nPixelX.set(nPixelXRight);
+        pIntervalNew.m_nCoverage.set(pInterval.m_nCoverage.get() - c_nShiftSize);
+
+        pIntervalNew.m_pNext.set(pInterval.m_pNext.get());
+        pInterval.m_pNext.set(pIntervalNew);
+
+        pInterval = pIntervalNew;
+
+        interval_new_index += 1;
+        pIntervalNew = Ref::new(&Ref::get_ref(self.m_pIntervalBufferCurrent.get()).m_interval[interval_new_index])
+    }
+    else
+    {
+        pInterval = pInterval.m_pNext.get();
+    }
+
+    //
+    // Compute coverage for right segment as shown by the *'s below
+    //
+    //  |_____|_____|_____|_
+    //  |     |     |     |
+    //  |  ---------****  |
+    //  |_____|_____|_____|
+    //
+
+    nCoverageRight = nSubpixelXRight & c_nShiftMask;
+    if (nCoverageRight > 0)
+    {
+        if (nPixelXRight + 1 != (*(*pInterval).m_pNext.get()).m_nPixelX.get())
+        {
+            pIntervalNew.m_nPixelX.set(nPixelXRight + 1);
+            pIntervalNew.m_nCoverage.set(pInterval.m_nCoverage.get());
+
+            pIntervalNew.m_pNext.set(pInterval.m_pNext.get());
+            pInterval.m_pNext.set(pIntervalNew);
+
+            interval_new_index += 1;
+            pIntervalNew = Ref::new(&Ref::get_ref(self.m_pIntervalBufferCurrent.get()).m_interval[interval_new_index])
+        }
+
+        pInterval.m_nCoverage.set((*pInterval).m_nCoverage.get() + nCoverageRight);
+        debug_assert!(pInterval.m_nCoverage.get() <= c_nShiftSize*c_nShiftSize);
+    }
+
+//Cleanup:
+    // Update the coverage buffer new interval
+    self.interval_new_index.set(interval_new_index);
+    self.m_pIntervalNew.set(pIntervalNew);
+
+    return hr;
+}
+
+
+//-------------------------------------------------------------------------
+//
+//  Function:   CCoverageBuffer::FillEdgesAlternating
+//
+//  Synopsis:   
+//      Given the active edge list for the current scan, do an alternate-mode
+//      antialiased fill.
+//
+//-------------------------------------------------------------------------
+pub fn FillEdgesAlternating(&'a self,
+    pEdgeActiveList: Ref<CEdge>,
+    nSubpixelYCurrent: INT
+    ) -> HRESULT
+{
+
+    let hr: HRESULT = S_OK;
+    let mut pEdgeStart: Ref<CEdge> = (*pEdgeActiveList).Next.get();
+    let mut pEdgeEnd: Ref<CEdge>;
+    let mut nSubpixelXLeft: INT;
+    let mut nSubpixelXRight: INT;
+
+    ASSERTACTIVELIST!(pEdgeActiveList, nSubpixelYCurrent);
+
+    while (pEdgeStart.X.get() != INT::MAX)
+    {
+        pEdgeEnd = pEdgeStart.Next.get();
+
+        // We skip empty pairs:
+        (nSubpixelXLeft = pEdgeStart.X.get());
+        if (nSubpixelXLeft != pEdgeEnd.X.get())
+        {
+            // We now know we have a non-empty interval.  Skip any
+            // empty interior pairs:
+
+            while ({(nSubpixelXRight = pEdgeEnd.X.get()); pEdgeEnd.X == pEdgeEnd.Next.get().X})
+            {
+                pEdgeEnd = pEdgeEnd.Next.get().Next.get();
+            }
+
+            debug_assert!((nSubpixelXLeft < nSubpixelXRight) && (nSubpixelXRight < INT::MAX));
+
+            IFC!(self.AddInterval(nSubpixelXLeft, nSubpixelXRight));
+        }
+
+        // Prepare for the next iteration:
+        pEdgeStart = pEdgeEnd.Next.get();
+    }
+
+//Cleanup:
+    return hr
+
+}
+
+//-------------------------------------------------------------------------
+//
+//  Function:   CCoverageBuffer::FillEdgesWinding
+//
+//  Synopsis:   
+//      Given the active edge list for the current scan, do an alternate-mode
+//      antialiased fill.
+//
+//-------------------------------------------------------------------------
+pub fn FillEdgesWinding(&'a self,
+    pEdgeActiveList: Ref<CEdge>,
+    nSubpixelYCurrent: INT
+    ) -> HRESULT
+{
+
+    let hr: HRESULT = S_OK;
+    let mut pEdgeStart: Ref<CEdge> = pEdgeActiveList.Next.get();
+    let mut pEdgeEnd: Ref<CEdge>;
+    let mut nSubpixelXLeft: INT;
+    let mut nSubpixelXRight: INT;
+    let mut nWindingValue: INT;
+
+    ASSERTACTIVELIST!(pEdgeActiveList, nSubpixelYCurrent);
+
+    while (pEdgeStart.X.get() != INT::MAX)
+    {
+        pEdgeEnd = pEdgeStart.Next.get();
+
+        nWindingValue = pEdgeStart.WindingDirection;
+        while ({nWindingValue += pEdgeEnd.WindingDirection; nWindingValue != 0})
+        {
+            pEdgeEnd = pEdgeEnd.Next.get();
+        }
+
+        debug_assert!(pEdgeEnd.X.get() != INT::MAX);
+
+        // We skip empty pairs:
+
+        if ({nSubpixelXLeft = pEdgeStart.X.get(); nSubpixelXLeft != pEdgeEnd.X.get()})
+        {
+            // We now know we have a non-empty interval.  Skip any
+            // empty interior pairs:
+
+            while ({nSubpixelXRight = pEdgeEnd.X.get(); nSubpixelXRight == pEdgeEnd.Next.get().X.get()})
+            {
+                pEdgeStart = pEdgeEnd.Next.get();
+                pEdgeEnd = pEdgeStart.Next.get();
+
+                nWindingValue = pEdgeStart.WindingDirection;
+                while ({nWindingValue += pEdgeEnd.WindingDirection; nWindingValue != 0})
+                {
+                    pEdgeEnd = pEdgeEnd.Next.get();
+                }
+            }
+
+            debug_assert!((nSubpixelXLeft < nSubpixelXRight) && (nSubpixelXRight < INT::MAX));
+
+            IFC!(self.AddInterval(nSubpixelXLeft, nSubpixelXRight));
+        }
+
+        // Prepare for the next iteration:
+
+        pEdgeStart = pEdgeEnd.Next.get();
+    } 
+
+//Cleanup:
+    return hr;//RRETURN(hr);
+}
+
+//-------------------------------------------------------------------------
+//
+//  Function:   CCoverageBuffer::Initialize
+//
+//  Synopsis:   Set the coverage buffer to a valid initial state
+// 
+//-------------------------------------------------------------------------
+pub fn Initialize(&'a self) 
+{
+    self.m_pIntervalBufferBuiltin.m_interval[0].m_nPixelX.set(INT::MIN);
+    self.m_pIntervalBufferBuiltin.m_interval[0].m_nCoverage.set(0);
+    self.m_pIntervalBufferBuiltin.m_interval[0].m_pNext.set(Ref::new(&self.m_pIntervalBufferBuiltin.m_interval[1]));
+
+    self.m_pIntervalBufferBuiltin.m_interval[1].m_nPixelX.set(INT::MAX);
+    self.m_pIntervalBufferBuiltin.m_interval[1].m_nCoverage.set(0xdeadbeef);
+    self.m_pIntervalBufferBuiltin.m_interval[1].m_pNext.set(unsafe { Ref::null() });
+
+    self.m_pIntervalBufferBuiltin.m_pNext.set(None);
+    self.m_pIntervalBufferCurrent.set(Ref::new(&self.m_pIntervalBufferBuiltin));
+
+    self.m_pIntervalStart.set(Ref::new(&self.m_pIntervalBufferBuiltin.m_interval[0]));
+    self.m_pIntervalNew.set(Ref::new(&self.m_pIntervalBufferBuiltin.m_interval[2]));
+    self.interval_new_index.set(2);
+    self.m_pIntervalEndMinus4.set(Ref::new(&self.m_pIntervalBufferBuiltin.m_interval[INTERVAL_BUFFER_NUMBER - 4]));
+    self.m_pIntervalLast.set(Ref::new(&self.m_pIntervalBufferBuiltin.m_interval[1]));
+}
+
+//-------------------------------------------------------------------------
+//
+//  Function:   CCoverageBuffer::Destroy
+//
+//  Synopsis:   Free all allocated buffers
+// 
+//-------------------------------------------------------------------------
+pub fn Destroy(&mut self)
+{
+    // Free the linked-list of allocations (skipping 'm_pIntervalBufferBuiltin',
+    // which is built into the class):
+
+
+}
+
+
+//-------------------------------------------------------------------------
+//
+//  Function:   CCoverageBuffer::Reset
+//
+//  Synopsis:   Reset the coverage buffer
+// 
+//-------------------------------------------------------------------------
+pub fn Reset(&'a self)
+{
+    // Reset our coverage structure.  Point the head back to the tail,
+    // and reset where the next new entry will be placed:
+
+    self.m_pIntervalBufferBuiltin.m_interval[0].m_pNext.set(Ref::new(&self.m_pIntervalBufferBuiltin.m_interval[1]));
+
+    self.m_pIntervalBufferCurrent.set(Ref::new(&self.m_pIntervalBufferBuiltin));
+    self.m_pIntervalNew.set(Ref::new(&self.m_pIntervalBufferBuiltin.m_interval[2]));
+    self.interval_new_index.set(2);
+    self.m_pIntervalEndMinus4.set(Ref::new(&self.m_pIntervalBufferBuiltin.m_interval[INTERVAL_BUFFER_NUMBER - 4]));
+    self.m_pIntervalLast.set(Ref::new(&self.m_pIntervalBufferBuiltin.m_interval[1]));
+}
+
+//-------------------------------------------------------------------------
+//
+//  Function:   CCoverageBuffer::Grow
+//
+//  Synopsis:   
+//      Grow our interval buffer.
+//
+//-------------------------------------------------------------------------
+fn Grow(&'a self,
+    ppIntervalNew: &mut Ref<'a, CCoverageInterval<'a>>, 
+    ppIntervalEndMinus4: &mut Ref<'a, CCoverageInterval<'a>>,
+    interval_new_index: &mut usize
+    ) -> HRESULT
+{
+    let hr: HRESULT = S_OK;
+    let pIntervalBufferNew = (*self.m_pIntervalBufferCurrent.get()).m_pNext.get();
+
+    let pIntervalBufferNew = pIntervalBufferNew.unwrap_or_else(||
+    {
+        let pIntervalBufferNew = self.arena.alloc(Default::default());
+
+        (*pIntervalBufferNew).m_pNext.set(None);
+        (*self.m_pIntervalBufferCurrent.get()).m_pNext.set(Some(pIntervalBufferNew));
+        pIntervalBufferNew
+    });
+
+    self.m_pIntervalBufferCurrent.set(Ref::new(pIntervalBufferNew));
+
+    self.m_pIntervalNew.set(Ref::new(&(*pIntervalBufferNew).m_interval[2]));
+    self.interval_new_index.set(2);
+    self.m_pIntervalEndMinus4.set(Ref::new(&(*pIntervalBufferNew).m_interval[INTERVAL_BUFFER_NUMBER - 4]));
+
+    *ppIntervalNew = self.m_pIntervalNew.get();
+    *ppIntervalEndMinus4 = self.m_pIntervalEndMinus4.get();
+    *interval_new_index = 2;
+
+    return hr;
+}
+
+}
+/* 
+impl<'a> Drop for CCoverageBuffer<'a> {
+    fn drop(&mut self) {
+        self.Destroy();
+    }
+}*/
diff --git a/third_party/rust/wpf-gpu-raster/src/aarasterizer.rs b/third_party/rust/wpf-gpu-raster/src/aarasterizer.rs
new file mode 100644
index 0000000000..ad9617a42d
--- /dev/null
+++ b/third_party/rust/wpf-gpu-raster/src/aarasterizer.rs
@@ -0,0 +1,1768 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+// See the LICENSE file in the project root for more information.
+
+#![allow(unused_parens)]
+
+use std::cell::Cell;
+
+use crate::aacoverage::c_nShift;
+use crate::bezier::CMILBezier;
+use crate::helpers::Int32x32To64;
+use crate::matrix::CMILMatrix;
+use crate::nullable_ref::Ref;
+use crate::real::CFloatFPU;
+//use crate::types::PathPointType::*;
+use crate::types::*;
+use typed_arena_nomut::Arena;
+
+const S_OK: HRESULT = 0;
+
+#[cfg(debug_assertions)]
+macro_rules! EDGE_STORE_STACK_NUMBER {
+    () => {
+        10
+    };
+}
+#[cfg(debug_assertions)]
+macro_rules! EDGE_STORE_ALLOCATION_NUMBER { () => { 11 }; }
+#[cfg(debug_assertions)]
+macro_rules! INACTIVE_LIST_NUMBER { () => { 12 }; }
+#[cfg(debug_assertions)]
+macro_rules! ENUMERATE_BUFFER_NUMBER { () => { 15 }; }
+
+#[cfg(not(debug_assertions))]
+macro_rules! EDGE_STORE_STACK_NUMBER { () => { (1600 / std::mem::size_of::<CEdge>()) }; }
+#[cfg(not(debug_assertions))]
+macro_rules! EDGE_STORE_ALLOCATION_NUMBER { () => { (4032 / std::mem::size_of::<CEdge>()) as u32 }; }
+#[cfg(not(debug_assertions))]
+macro_rules! INACTIVE_LIST_NUMBER { () => { EDGE_STORE_STACK_NUMBER!() }; }
+#[cfg(not(debug_assertions))]
+macro_rules! ENUMERATE_BUFFER_NUMBER { () => { 32 }; }
+
+macro_rules! ASSERTACTIVELIST {
+    ($list: expr, $y: expr) => {
+        // make sure we use y even in non debug builds
+        _ = $y;
+        #[cfg(debug_assertions)]
+        AssertActiveList($list, $y);
+    };
+}
+pub struct CEdge<'a> {
+    pub Next: Cell<Ref<'a, CEdge<'a>>>, // Next active edge (don't check for NULL,
+    //   look for tail sentinel instead)
+    pub X: Cell<INT>,                // Current X location
+    pub Dx: INT,               // X increment
+    pub Error: Cell<INT>,            // Current DDA error
+    pub ErrorUp: INT,          // Error increment
+    pub ErrorDown: INT,        // Error decrement when the error rolls over
+    pub StartY: INT,           // Y-row start
+    pub EndY: INT,             // Y-row end
+    pub WindingDirection: INT, // -1 or 1
+}
+
+impl<'a> std::default::Default for CEdge<'a> {
+    fn default() -> Self {
+        Self {
+            Next: Cell::new(unsafe { Ref::null() }),
+            X: Default::default(),
+            Dx: Default::default(),
+            Error: Default::default(),
+            ErrorUp: Default::default(),
+            ErrorDown: Default::default(),
+            StartY: Default::default(),
+            EndY: Default::default(),
+            WindingDirection: Default::default(),
+        }
+    }
+}
+
+// We the inactive-array separate from the edge allocations so that
+// we can more easily do in-place sorts on it:
+#[derive(Clone)]
+pub struct CInactiveEdge<'a> {
+    Edge: Ref<'a, CEdge<'a>>, // Associated edge
+    Yx: LONGLONG,     // Sorting key, StartY and X packed into an lword
+}
+
+impl<'a> Default for CInactiveEdge<'a> {
+    fn default() -> Self {
+        Self {
+            Edge: unsafe { Ref::null() },
+            Yx: Default::default(),
+        }
+    }
+}
+macro_rules! ASSERTACTIVELISTORDER {
+    ($list: expr) => {
+        #[cfg(debug_assertions)]
+        AssertActiveListOrder($list)
+    };
+}
+
+/**************************************************************************\
+*
+* Function Description:
+*
+* Advance DDA and update active edge list
+*
+* Created:
+*
+*   06/20/2003 ashrafm
+*
+\**************************************************************************/
+pub fn AdvanceDDAAndUpdateActiveEdgeList(nSubpixelYCurrent: INT, pEdgeActiveList: Ref<CEdge>) {
+        let mut outOfOrder = false;
+        let mut pEdgePrevious: Ref<CEdge> = pEdgeActiveList;
+        let mut pEdgeCurrent: Ref<CEdge> = pEdgeActiveList.Next.get();
+        let mut prevX = pEdgePrevious.X.get();
+
+        // Advance DDA and update edge list
+
+        loop {
+            if (pEdgeCurrent.EndY <= nSubpixelYCurrent) {
+                // If we've hit the sentinel, our work here is done:
+
+                if (pEdgeCurrent.EndY == INT::MIN) {
+                    break; // ============>
+                }
+                // This edge is stale, remove it from the list:
+
+                pEdgeCurrent = pEdgeCurrent.Next.get();
+                pEdgePrevious.Next.set(pEdgeCurrent);
+                continue; // ============>
+            }
+
+            // Advance the DDA:
+
+            let mut x = pEdgeCurrent.X.get() + pEdgeCurrent.Dx;
+            let mut error = pEdgeCurrent.Error.get() + pEdgeCurrent.ErrorUp;
+            if (error >= 0) {
+                error -= pEdgeCurrent.ErrorDown;
+                x += 1;
+            }
+            pEdgeCurrent.X.set(x);
+            pEdgeCurrent.Error.set(error);
+
+            // Is this entry out-of-order with respect to the previous one?
+            outOfOrder |= (prevX > x);
+
+            // Advance:
+
+            pEdgePrevious = pEdgeCurrent;
+            pEdgeCurrent = pEdgeCurrent.Next.get();
+            prevX = x;
+        }
+
+        // It turns out that having any out-of-order edges at this point
+        // is extremely rare in practice, so only call the bubble-sort
+        // if it's truly needed.
+        //
+        // NOTE: If you're looking at this code trying to fix a bug where
+        //       the edges are out of order when the filler is called, do
+        //       NOT simply change the code to always do the bubble-sort!
+        //       Instead, figure out what caused our 'outOfOrder' logic
+        //       above to get messed up.
+
+        if (outOfOrder) {
+            SortActiveEdges(pEdgeActiveList);
+        }
+        ASSERTACTIVELISTORDER!(pEdgeActiveList);
+
+}
+
+//+----------------------------------------------------------------------------
+//
+
+//
+//  Description:  Code for rasterizing the fill of a path.
+//
+//  >>>> Note that some of this code is duplicated in hw\hwrasterizer.cpp,
+//  >>>> so changes to this file may need to propagate.
+//
+//   pursue reduced code duplication
+//
+
+// This option may potentially increase performance for many
+// paths that have edges adjacent at their top point and cover
+// more than one span.  The code has been tested, but performance
+// has not been thoroughly investigated.
+const SORT_EDGES_INCLUDING_SLOPE: bool = false;
+
+/////////////////////////////////////////////////////////////////////////
+// The x86 C compiler insists on making a divide and modulus operation
+// into two DIVs, when it can in fact be done in one.  So we use this
+// macro.
+//
+// Note: QUOTIENT_REMAINDER implicitly takes unsigned arguments.
+//
+// QUOTIENT_REMAINDER_64_32 takes a 64-bit numerator and produces 32-bit
+// results.
+
+macro_rules! QUOTIENT_REMAINDER {
+    ($ulNumerator: ident, $ulDenominator: ident, $ulQuotient: ident, $ulRemainder: ident) => {
+        $ulQuotient = (($ulNumerator as ULONG) / ($ulDenominator as ULONG)) as _;
+        $ulRemainder = (($ulNumerator as ULONG) % ($ulDenominator as ULONG)) as _;
+    };
+}
+
+macro_rules! QUOTIENT_REMAINDER_64_32 {
+    ($ulNumerator: ident, $ulDenominator: ident, $ulQuotient: ident, $ulRemainder: ident) => {
+        $ulQuotient = (($ulNumerator as ULONGLONG) / (($ulDenominator as ULONG) as ULONGLONG)) as _;
+        $ulRemainder =
+            (($ulNumerator as ULONGLONG) % (($ulDenominator as ULONG) as ULONGLONG)) as _;
+    };
+}
+
+// SWAP macro:
+macro_rules! SWAP {
+    ($temp: ident, $a: expr, $b: expr) => {
+        $temp = $a;
+        $a = $b;
+        $b = $temp;
+    };
+}
+
+struct CEdgeAllocation {
+    Next: *mut CEdgeAllocation, // Next allocation batch (may be NULL)
+    /*__field_range(<=, EDGE_STORE_ALLOCATION_NUMBER)*/ Count: UINT,
+    EdgeArray: [CEdge<'static>; EDGE_STORE_STACK_NUMBER!()],
+}
+
+impl Default for CEdgeAllocation {
+    fn default() -> Self {
+        Self { Next: NULL(), Count: Default::default(), EdgeArray: [(); EDGE_STORE_STACK_NUMBER!()].map(|_| Default::default()) }
+    }
+}
+/* 
+pub struct CEdgeStore {
+    /* __field_range(<=, UINT_MAX - 2) */ TotalCount: UINT, // Total edge count in store
+    /* __field_range(<=, CurrentBuffer->Count) */
+    CurrentRemaining: UINT, // How much room remains in current buffer
+    CurrentBuffer: *mut CEdgeAllocation, // Current buffer
+    CurrentEdge: *mut CEdge<'static>, // Current edge in current buffer
+    Enumerator: *mut CEdgeAllocation, // For enumerating all the edges
+    EdgeHead: CEdgeAllocation, // Our built-in allocation
+}
+
+impl Default for CEdgeStore {
+    fn default() -> Self {
+        Self { TotalCount: Default::default(), CurrentRemaining: Default::default(), CurrentBuffer: NULL(), CurrentEdge: NULL(), Enumerator: NULL(), EdgeHead: Default::default() }
+    }
+}
+
+impl CEdgeStore {
+    pub fn init(&mut self) {
+        self.TotalCount = 0;
+        self.CurrentBuffer = NULL();
+        self.CurrentEdge = NULL();
+        self.Enumerator = NULL();
+        self.CurrentRemaining = EDGE_STORE_STACK_NUMBER!() as u32;
+
+        self.EdgeHead = CEdgeAllocation {
+                Count: EDGE_STORE_STACK_NUMBER!() as u32,
+                // hack to work around limited Default implementation for arrays
+                EdgeArray: [(); EDGE_STORE_STACK_NUMBER!()].map(|_| Default::default()),
+                Next: NULL(),
+            };
+        self.CurrentBuffer = &mut self.EdgeHead;
+        self.CurrentEdge = &mut self.EdgeHead.EdgeArray[0];
+    }
+}
+
+impl Drop for CEdgeStore {
+    fn drop(&mut self) {
+        // Free our allocation list, skipping the head, which is not
+        // dynamically allocated:
+
+        let mut allocation: *mut CEdgeAllocation = self.EdgeHead.Next;
+        while (allocation != NULL()) {
+            let next = unsafe { (*allocation).Next };
+            drop(unsafe { Box::from_raw(allocation) });
+            allocation = next;
+        }
+    }
+}
+
+impl CEdgeStore {
+    pub fn StartEnumeration(&mut self) -> UINT {
+        unsafe {
+            self.Enumerator = &mut self.EdgeHead;
+
+            // Update the count and make sure nothing more gets added (in
+            // part because this Count would have to be re-computed):
+
+            (*self.CurrentBuffer).Count -= self.CurrentRemaining;
+
+            // This will never overflow because NextAddBuffer always ensures that TotalCount has
+            // space remaining to describe the capacity of all new buffers added to the edge list.
+            self.TotalCount += (*self.CurrentBuffer).Count;
+
+            // Prevent this from being called again, because bad things would
+            // happen:
+
+            self.CurrentBuffer = NULL();
+
+            return self.TotalCount;
+        }
+    }
+
+    fn Enumerate(
+        &mut self,
+        /*__deref_out_ecount(*ppEndEdge - *ppStartEdge)*/ ppStartEdge: &mut *mut CEdge,
+        /* __deref_out_ecount(0) */ ppEndEdge: &mut *mut CEdge,
+    ) -> bool {
+        /* 
+        unsafe {
+            let enumerator: *mut CEdgeAllocation = self.Enumerator;
+
+            // Might return startEdge == endEdge:
+
+            *ppStartEdge = &mut (*enumerator).EdgeArray[0];
+            *ppEndEdge = (*ppStartEdge).offset((*enumerator).Count as isize);
+
+            self.Enumerator = (*enumerator).Next;
+            return (self.Enumerator != NULL());
+        }*/
+        return true;
+    }
+
+    fn StartAddBuffer(
+        &self,
+        /*__deref_out_ecount(*puRemaining)*/ ppCurrentEdge: &mut *mut CEdge,
+        /* __deref_out_range(==, (this->CurrentRemaining)) */ puRemaining: &mut UINT,
+    ) {
+        panic!()
+        // *ppCurrentEdge = self.CurrentEdge;
+        // *puRemaining = self.CurrentRemaining;
+    }
+
+    fn EndAddBuffer(
+        &mut self,
+        /*__in_ecount(remaining) */ pCurrentEdge: *mut CEdge,
+        /* __range(0, (this->CurrentBuffer->Count)) */ remaining: UINT,
+    ) {
+        panic!();
+        //self.CurrentEdge = pCurrentEdge;
+        //self.CurrentRemaining = remaining;
+    }
+
+    // Disable instrumentation checks within all methods of this class
+    //SET_MILINSTRUMENTATION_FLAGS(MILINSTRUMENTATIONFLAGS_DONOTHING);
+}
+
+/**************************************************************************\
+*
+* Function Description:
+*
+*   The edge initializer is out of room in its current 'store' buffer;
+*   get it a new one.
+*
+* Created:
+*
+*   03/25/2000 andrewgo
+*
+\**************************************************************************/
+
+impl CEdgeStore {
+    fn NextAddBuffer(
+        &mut self,
+        /*__deref_out_ecount(*puRemaining)*/ ppCurrentEdge: &mut *mut CEdge,
+        puRemaining: &mut UINT,
+    ) -> HRESULT {
+        panic!()
+        /* 
+        unsafe {
+            let hr = S_OK;
+
+            let mut cNewTotalCount: u32 = 0;
+
+            // The caller has completely filled up this chunk:
+
+            assert!(*puRemaining == 0);
+
+            // Check to make sure that "TotalCount" will be able to represent the current capacity
+            cNewTotalCount = self.TotalCount + (*self.CurrentBuffer).Count;
+
+            if (cNewTotalCount < self.TotalCount) {
+                return WINCODEC_ERR_VALUEOVERFLOW;
+            }
+
+            // And that it can represent the new capacity as well, with at least 2 to spare.
+            // This "magic" 2 comes from the fact that the usage pattern of this class has callers
+            // needing to allocate space for TotalCount + 2 edges.
+            if (cNewTotalCount + ((EDGE_STORE_ALLOCATION_NUMBER!() + 2) as UINT) < cNewTotalCount) {
+                return WINCODEC_ERR_VALUEOVERFLOW;
+            }
+
+            // We have to grow our data structure by adding a new buffer
+            // and adding it to the list:
+
+            let newBuffer: *mut CEdgeAllocation = Box::into_raw(Box::<CEdgeAllocation>::new(Default::default()));/*static_cast<CEdgeAllocation*>
+                                                            (GpMalloc(Mt(MAARasterizerEdge),
+                                                                      sizeof(CEdgeAllocation) +
+                                                                      sizeof(CEdge) * (EDGE_STORE_ALLOCATION_NUMBER
+                                                                                      - EDGE_STORE_STACK_NUMBER)));*/
+            IFCOOM!(newBuffer);
+
+            (*newBuffer).Next = NULL();
+            (*newBuffer).Count = EDGE_STORE_STACK_NUMBER!() as u32;//EDGE_STORE_ALLOCATION_NUMBER!() as u32;
+
+            self.TotalCount = cNewTotalCount;
+
+            (*self.CurrentBuffer).Next = newBuffer;
+            self.CurrentBuffer = newBuffer;
+
+            self.CurrentEdge = &mut (*newBuffer).EdgeArray[0];
+            *ppCurrentEdge = panic!();//self.CurrentEdge;
+            self.CurrentRemaining = EDGE_STORE_STACK_NUMBER!() as u32;//EDGE_STORE_ALLOCATION_NUMBER!();
+            *puRemaining = EDGE_STORE_STACK_NUMBER!() as u32; //EDGE_STORE_ALLOCATION_NUMBER!();
+
+            return hr;
+        }*/
+    }
+}
+*/
+/**************************************************************************\
+*
+* Function Description:
+*
+*   Some debug code for verifying the state of the active edge list.
+*
+* Created:
+*
+*   03/25/2000 andrewgo
+*
+\**************************************************************************/
+
+pub fn AssertActiveList(mut list: Ref<CEdge>, yCurrent: INT) -> bool {
+
+    let mut b = true;
+    let mut activeCount = 0;
+
+    assert!((*list).X.get() == INT::MIN);
+    b &= ((*list).X.get() == INT::MIN);
+
+    // Skip the head sentinel:
+
+    list = (*list).Next.get();
+
+    while ((*list).X.get() != INT::MAX) {
+        assert!((*list).X.get() != INT::MIN);
+        b &= ((*list).X.get() != INT::MIN);
+
+        assert!((*list).X <= (*(*list).Next.get()).X);
+        b &= ((*list).X <= (*(*list).Next.get()).X);
+
+        assert!(((*list).StartY <= yCurrent) && (yCurrent < (*list).EndY));
+        b &= (((*list).StartY <= yCurrent) && (yCurrent < (*list).EndY));
+
+        activeCount += 1;
+        list = (*list).Next.get();
+    }
+
+    assert!((*list).X.get() == INT::MAX);
+    b &= ((*list).X.get() == INT::MAX);
+
+    // There should always be a multiple of 2 edges in the active list.
+    //
+    // NOTE: If you hit this assert, do NOT simply comment it out!
+    //       It usually means that all the edges didn't get initialized
+    //       properly.  For every scan-line, there has to be a left edge
+    //       and a right edge (or a multiple thereof).  So if you give
+    //       even a single bad edge to the edge initializer (or you miss
+    //       one), you'll probably hit this assert.
+
+    assert!((activeCount & 1) == 0);
+    b &= ((activeCount & 1) == 0);
+
+    return (b);
+
+}
+
+/**************************************************************************\
+*
+* Function Description:
+*
+*   Some debug code for verifying the state of the active edge list.
+*
+* Created:
+*
+*   03/25/2000 andrewgo
+*
+\**************************************************************************/
+
+fn AssertActiveListOrder(mut list:  Ref<CEdge>) {
+
+    assert!((*list).X.get() == INT::MIN);
+
+    // Skip the head sentinel:
+
+    list = (*list).Next.get();
+
+    while ((*list).X.get() != INT::MAX) {
+        assert!((*list).X.get() != INT::MIN);
+        assert!((*list).X <= (*(*list).Next.get()).X);
+
+        list = (*list).Next.get();
+    }
+
+    assert!((*list).X.get() == INT::MAX);
+}
+
+/**************************************************************************\
+*
+* Function Description:
+*
+*   Clip the edge vertically.
+*
+*   We've pulled this routine out-of-line from InitializeEdges mainly
+*   because it needs to call inline Asm, and when there is in-line
+*   Asm in a routine the compiler generally does a much less efficient
+*   job optimizing the whole routine.  InitializeEdges is rather
+*   performance critical, so we avoid polluting the whole routine
+*   by having this functionality out-of-line.
+*
+* Created:
+*
+*   03/25/2000 andrewgo
+*
+\**************************************************************************/
+fn ClipEdge(edgeBuffer: &mut CEdge, yClipTopInteger: INT, dMOriginal: INT) {
+    let mut xDelta: INT;
+    let mut error: INT;
+
+    // Cases where bigNumerator will exceed 32-bits in precision
+    // will be rare, but could happen, and we can't fall over
+    // in those cases.
+
+    let dN: INT = edgeBuffer.ErrorDown;
+    let mut bigNumerator: LONGLONG = Int32x32To64(dMOriginal, yClipTopInteger - edgeBuffer.StartY)
+        + (edgeBuffer.Error.get() + dN) as LONGLONG;
+    if (bigNumerator >= 0) {
+        QUOTIENT_REMAINDER_64_32!(bigNumerator, dN, xDelta, error);
+    } else {
+        bigNumerator = -bigNumerator;
+        QUOTIENT_REMAINDER_64_32!(bigNumerator, dN, xDelta, error);
+
+        xDelta = -xDelta;
+        if (error != 0) {
+            xDelta -= 1;
+            error = dN - error;
+        }
+    }
+
+    // Update the edge data structure with the results:
+
+    edgeBuffer.StartY = yClipTopInteger;
+    edgeBuffer.X.set(edgeBuffer.X.get() + xDelta);
+    edgeBuffer.Error.set(error - dN); // Renormalize error
+}
+
+pub fn CheckValidRange28_4(x: f32, y: f32) -> bool {
+    //
+    // We want coordinates in the 28.4 range in the end.  The matrix we get
+    // as input includes the scale by 16 to get to 28.4, so we want to
+    // ensure that we are in integer range.  Assuming a sign bit and
+    // five bits for the rasterizer working range, we want coordinates in the
+    // -2^26 to 2^26.
+    //
+    // Note that the 5-bit requirement comes from the
+    // implementation of InitializeEdges.
+    // (See line with "error -= dN * (16 - (xStart & 15))")
+    //
+    // Anti-aliasing uses another c_nShift bits, so we get a
+    // desired range of -2^(26-c_nShift) to 2^(26-c_nShift)
+    //
+    let rPixelCoordinateMax = (1 << (26 - c_nShift)) as f32;
+    let rPixelCoordinateMin = -rPixelCoordinateMax;
+    return x <= rPixelCoordinateMax && x >= rPixelCoordinateMin
+            && y <= rPixelCoordinateMax && y >= rPixelCoordinateMin;
+}
+
+//+-----------------------------------------------------------------------------
+//
+//  Function:  TransformRasterizerPointsTo28_4
+//
+//  Synopsis:
+//      Transform rasterizer points to 28.4.  If overflow occurs, return that
+//      information.
+//
+//------------------------------------------------------------------------------
+fn TransformRasterizerPointsTo28_4(
+    pmat: &CMILMatrix,
+    // Transform to take us to 28.4
+    mut pPtsSource: &[MilPoint2F],
+    // Source points
+    mut cPoints: UINT,
+    // Count of points
+    mut pPtsDest: &mut [POINT], // Destination points
+) -> HRESULT {
+    let hr = S_OK;
+
+    debug_assert!(cPoints > 0);
+
+    while {
+        //
+        // Transform coordinates
+        //
+
+        let rPixelX =
+            (pmat.GetM11() * pPtsSource[0].X) + (pmat.GetM21() * pPtsSource[0].Y) + pmat.GetDx();
+        let rPixelY =
+            (pmat.GetM12() * pPtsSource[0].X) + (pmat.GetM22() * pPtsSource[0].Y) + pmat.GetDy();
+
+        //
+        // Check for NaNs or overflow
+        //
+
+        if !CheckValidRange28_4(rPixelX, rPixelY) {
+            return WGXERR_BADNUMBER;
+        }
+
+        //
+        // Assign coordinates
+        //
+
+        pPtsDest[0].x = CFloatFPU::Round(rPixelX);
+        pPtsDest[0].y = CFloatFPU::Round(rPixelY);
+
+        pPtsDest = &mut pPtsDest[1..];
+        pPtsSource = &pPtsSource[1..];
+        cPoints -= 1;
+        cPoints != 0
+    } {}
+
+    return hr;
+}
+
+pub fn AppendScaleToMatrix(pmat: &mut CMILMatrix, scaleX: REAL, scaleY: REAL) {
+    pmat.SetM11(pmat.GetM11() * scaleX);
+    pmat.SetM21(pmat.GetM21() * scaleX);
+    pmat.SetM12(pmat.GetM12() * scaleY);
+    pmat.SetM22(pmat.GetM22() * scaleY);
+    pmat.SetDx(pmat.GetDx() * scaleX);
+    pmat.SetDy(pmat.GetDy() * scaleY);
+}
+
+/**************************************************************************\
+*
+* Function Description:
+*
+*   Add edges to the edge list.
+*
+* Created:
+*
+*   03/25/2000 andrewgo
+*
+\**************************************************************************/
+
+pub struct CInitializeEdgesContext<'a> {
+    pub MaxY: INT, // Maximum 'y' found, should be INT_MIN on
+    //   first call to 'InitializeEdges'
+    pub ClipRect: Option<&'a RECT>, // Bounding clip rectangle in 28.4 format
+    pub Store: &'a Arena<CEdge<'a>>,  // Where to stick the edges
+    pub AntiAliasMode: MilAntiAliasMode,
+}
+
+impl<'a> CInitializeEdgesContext<'a> {
+    pub fn new(store: &'a Arena<CEdge<'a>>) -> Self {
+        CInitializeEdgesContext { MaxY: Default::default(), ClipRect: Default::default(), Store: store, AntiAliasMode: MilAntiAliasMode::None }
+    }
+}
+
+fn InitializeEdges(
+    pEdgeContext: &mut CInitializeEdgesContext,
+    /*__inout_ecount(vertexCount)*/
+    mut pointArray: &mut [POINT], // Points to a 28.4 array of size 'vertexCount'
+    //   Note that we may modify the contents!
+    /*__in_range(>=, 2)*/ vertexCount: UINT,
+) -> HRESULT {
+    // Disable instrumentation checks for this function
+    //SET_MILINSTRUMENTATION_FLAGS(MILINSTRUMENTATIONFLAGS_DONOTHING);
+
+    let hr = S_OK;
+
+    let mut xStart;
+    let mut yStart;
+    let mut yStartInteger;
+    let mut yEndInteger;
+    let mut dMOriginal;
+    let mut dM: i32;
+    let mut dN: i32;
+    let mut dX: i32;
+    let mut errorUp: i32;
+    let mut quotient: i32;
+    let mut remainder: i32;
+    let mut error: i32;
+    let mut windingDirection;
+    //let mut edgeBuffer: *mut CEdge = NULL();
+    let bufferCount: UINT = 0;
+    let mut yClipTopInteger;
+    let mut yClipTop;
+    let mut yClipBottom;
+    let mut xClipLeft;
+    let mut xClipRight;
+
+    let mut yMax = pEdgeContext.MaxY;
+    let _store = &mut pEdgeContext.Store;
+    let clipRect = pEdgeContext.ClipRect;
+
+    let mut edgeCount = vertexCount - 1;
+    assert!(edgeCount >= 1);
+
+    if let Some(clipRect) = clipRect {
+        yClipTopInteger = clipRect.top >> 4;
+        yClipTop = clipRect.top;
+        yClipBottom = clipRect.bottom;
+        xClipLeft = clipRect.left;
+        xClipRight = clipRect.right;
+
+        assert!(yClipBottom > 0);
+        assert!(yClipTop <= yClipBottom);
+    } else {
+        yClipBottom = 0;
+        yClipTopInteger = INT::MIN >> c_nShift;
+
+        // These 3 values are only used when clipRect is non-NULL
+        yClipTop = 0;
+        xClipLeft = 0;
+        xClipRight = 0;
+    }
+
+    if (pEdgeContext.AntiAliasMode != MilAntiAliasMode::None) {
+        // If antialiasing, apply the supersampling scaling here before we
+        // calculate the DDAs.  We do this here and not in the Matrix
+        // transform we give to FixedPointPathEnumerate mainly so that the
+        // Bezier flattener can continue to operate in its optimal 28.4
+        // format.
+        //
+        // PS#856364-2003/07/01-JasonHa  Remove pixel center fixup
+        //
+        // We also apply a half-pixel offset here so that the antialiasing
+        // code can assume that the pixel centers are at half-pixel
+        // coordinates, not on the integer coordinates.
+
+        let mut point = &mut *pointArray;
+        let mut i = vertexCount;
+
+        while {
+            point[0].x = (point[0].x + 8) << c_nShift;
+            point[0].y = (point[0].y + 8) << c_nShift;
+            point = &mut point[1..];
+            i -= 1;
+            i != 0
+        } {}
+
+        yClipTopInteger <<= c_nShift;
+        yClipTop <<= c_nShift;
+        yClipBottom <<= c_nShift;
+        xClipLeft <<= c_nShift;
+        xClipRight <<= c_nShift;
+    }
+
+    // Make 'yClipBottom' inclusive by subtracting off one pixel
+    // (keeping in mind that we're in 28.4 device space):
+
+    yClipBottom -= 16;
+
+    // Warm up the store where we keep the edge data:
+
+    //store.StartAddBuffer(&mut edgeBuffer, &mut bufferCount);
+
+    'outer: loop { loop { 
+        // Handle trivial rejection:
+
+        if (yClipBottom >= 0) {
+            // Throw out any edges that are above or below the clipping.
+            // This has to be a precise check, because we assume later
+            // on that every edge intersects in the vertical dimension
+            // with the clip rectangle.  That asssumption is made in two
+            // places:
+            //
+            // 1.  When we sort the edges, we assume either zero edges,
+            //     or two or more.
+            // 2.  When we start the DDAs, we assume either zero edges,
+            //     or that there's at least one scan of DDAs to output.
+            //
+            // Plus, of course, it's less efficient if we let things
+            // through.
+            //
+            // Note that 'yClipBottom' is inclusive:
+
+            let clipHigh = ((pointArray[0]).y <= yClipTop) && ((pointArray[1]).y <= yClipTop);
+
+            let clipLow = ((pointArray[0]).y > yClipBottom) && ((pointArray[1]).y > yClipBottom);
+
+            #[cfg(debug_assertions)]
+            {
+                let (mut yRectTop, mut yRectBottom, y0, y1, yTop, yBottom);
+
+                // Getting the trivial rejection code right is tricky.
+                // So on checked builds let's verify that we're doing it
+                // correctly, using a different approach:
+
+                let mut clipped = false;
+                if let Some(clipRect) = clipRect {
+                    yRectTop = clipRect.top >> 4;
+                    yRectBottom = clipRect.bottom >> 4;
+                    if (pEdgeContext.AntiAliasMode != MilAntiAliasMode::None) {
+                        yRectTop <<= c_nShift;
+                        yRectBottom <<= c_nShift;
+                    }
+                    y0 = ((pointArray[0]).y + 15) >> 4;
+                    y1 = ((pointArray[1]).y + 15) >> 4;
+                    yTop = y0.min(y1);
+                    yBottom = y0.max(y1);
+
+                    clipped = ((yTop >= yRectBottom) || (yBottom <= yRectTop));
+                }
+
+                assert!(clipped == (clipHigh || clipLow));
+            }
+
+            if (clipHigh || clipLow) {
+                break; // ======================>
+            }
+
+            if (edgeCount > 1) {
+                // Here we'll collapse two edges down to one if both are
+                // to the left or to the right of the clipping rectangle.
+
+                if (((pointArray[0]).x < xClipLeft)
+                    && ((pointArray[1]).x < xClipLeft)
+                    && ((pointArray[2]).x < xClipLeft))
+                {
+                    // Note this is one reason why 'pointArray' can't be 'const':
+
+                    pointArray[1] = pointArray[0];
+
+                    break; // ======================>
+                }
+
+                if (((pointArray[0]).x > xClipRight)
+                    && ((pointArray[1]).x > xClipRight)
+                    && ((pointArray[2]).x > xClipRight))
+                {
+                    // Note this is one reason why 'pointArray' can't be 'const':
+
+                    pointArray[1] = pointArray[0];
+
+                    break; // ======================>
+                }
+            }
+        }
+
+        dM = (pointArray[1]).x - (pointArray[0]).x;
+        dN = (pointArray[1]).y - (pointArray[0]).y;
+
+        if (dN >= 0) {
+            // The vector points downward:
+
+            xStart = (pointArray[0]).x;
+            yStart = (pointArray[0]).y;
+
+            yStartInteger = (yStart + 15) >> 4;
+            yEndInteger = ((pointArray[1]).y + 15) >> 4;
+
+            windingDirection = 1;
+        } else {
+            // The vector points upward, so we have to essentially
+            // 'swap' the end points:
+
+            dN = -dN;
+            dM = -dM;
+
+            xStart = (pointArray[1]).x;
+            yStart = (pointArray[1]).y;
+
+            yStartInteger = (yStart + 15) >> 4;
+            yEndInteger = ((pointArray[0]).y + 15) >> 4;
+
+            windingDirection = -1;
+        }
+
+        // The edgeBuffer must span an integer y-value in order to be
+        // added to the edgeBuffer list.  This serves to get rid of
+        // horizontal edges, which cause trouble for our divides.
+
+        if (yEndInteger > yStartInteger) {
+            yMax = yMax.max(yEndInteger);
+
+            dMOriginal = dM;
+            if (dM < 0) {
+                dM = -dM;
+                if (dM < dN)
+                // Can't be '<='
+                {
+                    dX = -1;
+                    errorUp = dN - dM;
+                } else {
+                    QUOTIENT_REMAINDER!(dM, dN, quotient, remainder);
+
+                    dX = -quotient;
+                    errorUp = remainder;
+                    if (remainder > 0) {
+                        dX = -quotient - 1;
+                        errorUp = dN - remainder;
+                    }
+                }
+            } else {
+                if (dM < dN) {
+                    dX = 0;
+                    errorUp = dM;
+                } else {
+                    QUOTIENT_REMAINDER!(dM, dN, quotient, remainder);
+
+                    dX = quotient;
+                    errorUp = remainder;
+                }
+            }
+
+            error = -1; // Error is initially zero (add dN - 1 for
+                        //   the ceiling, but subtract off dN so that
+                        //   we can check the sign instead of comparing
+                        //   to dN)
+
+            if ((yStart & 15) != 0) {
+                // Advance to the next integer y coordinate
+
+                let mut i = 16 - (yStart & 15);
+                while i != 0 {
+                    xStart += dX;
+                    error += errorUp;
+                    if (error >= 0)
+                    {
+                        error -= dN;
+                        xStart += 1;
+                    }
+                    i -= 1;
+                }
+            }
+
+            if ((xStart & 15) != 0) {
+                error -= dN * (16 - (xStart & 15));
+                xStart += 15; // We'll want the ceiling in just a bit...
+            }
+
+            xStart >>= 4;
+            error >>= 4;
+
+            if (bufferCount == 0) {
+                //IFC!(store.NextAddBuffer(&mut edgeBuffer, &mut bufferCount));
+            }
+
+            let mut edge = CEdge {
+                Next: Cell::new(unsafe { Ref::null() } ),
+                X: Cell::new(xStart),
+                Dx: dX,
+                Error: Cell::new(error),
+                ErrorUp: errorUp,
+                ErrorDown: dN,
+                WindingDirection: windingDirection,
+                StartY: yStartInteger,
+                EndY: yEndInteger,// Exclusive of end
+            };
+
+            assert!(error < 0);
+
+            // Here we handle the case where the edge starts above the
+            // clipping rectangle, and we need to jump down in the 'y'
+            // direction to the first unclipped scan-line.
+            //
+            // Consequently, we advance the DDA here:
+
+            if (yClipTopInteger > yStartInteger) {
+                assert!(edge.EndY  > yClipTopInteger);
+
+                ClipEdge(&mut edge, yClipTopInteger, dMOriginal);
+            }
+
+            // Advance to handle the next edge:
+
+            //edgeBuffer = unsafe { edgeBuffer.offset(1) };
+            pEdgeContext.Store.alloc(edge);
+            //bufferCount -= 1;
+        }
+        break;
+    }
+    pointArray = &mut pointArray[1..];
+    edgeCount -= 1;
+    if edgeCount == 0 {
+        break 'outer;
+    }
+    }
+
+    // We're done with this batch.  Let the store know how many edges
+    // we ended up with:
+
+    //store.EndAddBuffer(edgeBuffer, bufferCount);
+
+    pEdgeContext.MaxY = yMax;
+
+    return hr;
+}
+
+/**************************************************************************\
+*
+* Function Description:
+*
+*   Does complete parameter checking on the 'types' array of a path.
+*
+* Created:
+*
+*   03/25/2000 andrewgo
+*
+\**************************************************************************/
+fn ValidatePathTypes(typesArray: &[BYTE], mut count: INT) -> bool {
+    let mut types = typesArray;
+
+    if (count == 0) {
+        return (true);
+    }
+
+    loop {
+        // The first point in every subpath has to be an unadorned
+        // 'start' point:
+
+        if ((types[0] & PathPointTypePathTypeMask) != PathPointTypeStart) {
+            TraceTag!((tagMILWarning, "Bad subpath start"));
+            return (false);
+        }
+
+        // Advance to the first point after the 'start' point:
+        count -= 1;
+        if (count == 0) {
+            TraceTag!((tagMILWarning, "Path ended after start-path"));
+            return (false);
+        }
+
+        if ((types[1] & PathPointTypePathTypeMask) == PathPointTypeStart) {
+            TraceTag!((tagMILWarning, "Can't have a start followed by a start!"));
+            return (false);
+        }
+
+        // Process runs of lines and Bezier curves:
+
+        loop {
+            match (types[1] & PathPointTypePathTypeMask) {
+                PathPointTypeLine => {
+                    types = &types[1..];
+                    count -= 1;
+                    if (count == 0) {
+                        return (true);
+                    }
+                }
+
+                PathPointTypeBezier => {
+                    if (count < 3) {
+                        TraceTag!((
+                            tagMILWarning,
+                            "Path ended before multiple of 3 Bezier points"
+                        ));
+                        return (false);
+                    }
+
+                    if ((types[1] & PathPointTypePathTypeMask) != PathPointTypeBezier) {
+                        TraceTag!((tagMILWarning, "Bad subpath start"));
+                        return (false);
+                    }
+
+                    types = &types[1..];
+                    count -= 3;
+                    if (count == 0) {
+                        return (true);
+                    }
+                }
+
+                _ => {
+                    TraceTag!((tagMILWarning, "Illegal type"));
+                    return (false);
+                }
+            }
+
+            // A close-subpath marker or a start-subpath marker marks the
+            // end of a subpath:
+            if !(!((types[0] & PathPointTypeCloseSubpath) != 0)
+                && ((types[1] & PathPointTypePathTypeMask) != PathPointTypeStart)) {
+                    types = &types[1..];
+                    break;
+                }
+        }
+    }
+}
+
+/**************************************************************************\
+*
+* Function Description:
+*
+*   Some debug code for verifying the path.
+*
+* Created:
+*
+*   03/25/2000 andrewgo
+*
+\**************************************************************************/
+macro_rules! ASSERTPATH {
+    ($types: expr, $points: expr) => {
+        #[cfg(debug_assertions)]
+        AssertPath($types, $points)
+    };
+}
+fn AssertPath(rgTypes: &[BYTE], cPoints: UINT) {
+    // Make sure that the 'types' array is well-formed, otherwise we
+    // may fall over in the FixedPointPathEnumerate function.
+    //
+    // NOTE: If you hit this assert, DO NOT SIMPLY COMMENT THIS Assert OUT!
+    //
+    //       Instead, fix the ValidatePathTypes code if it's letting through
+    //       valid paths, or (more likely) fix the code that's letting bogus
+    //       paths through.  The FixedPointPathEnumerate routine has some
+    //       subtle assumptions that require the path to be perfectly valid!
+    //
+    //       No internal code should be producing invalid paths, and all
+    //       paths created by the application must be parameter checked!
+    assert!(ValidatePathTypes(rgTypes, cPoints as INT));
+}
+
+//+----------------------------------------------------------------------------
+//
+//  Member:
+//      FixedPointPathEnumerate
+//
+//  Synopsis:
+//
+//      Enumerate the path.
+//
+//      NOTE: The 'enumerateFunction' function is allowed to modify the
+//            contents of our call-back buffer!  (This is mainly done to allow
+//            'InitializeEdges' to be simpler for some clipping trivial
+//            rejection cases.)
+//
+//      NOTICE-2006/03/22-milesc  This function was initially built to be a
+//      general path enumeration function. However, we were only using it for
+//      one specific purpose... for Initializing edges of a path to be filled.
+//      In doing security work, I simplified this function to just do edge
+//      initialization. The name is therefore now overly general. I have kept
+//      the name to be a reminder that this function has been written to be
+//      more general than would otherwise be evident.
+//
+
+pub fn FixedPointPathEnumerate(
+    rgpt: &[POINT],
+    rgTypes: &[BYTE],
+    cPoints: UINT,
+    _matrix: &CMILMatrix,
+    clipRect: Option<&RECT>, // In scaled 28.4 format
+    enumerateContext: &mut CInitializeEdgesContext,
+) -> HRESULT {
+    let hr = S_OK;
+    let mut bufferStart: [POINT; ENUMERATE_BUFFER_NUMBER!()] = [(); ENUMERATE_BUFFER_NUMBER!()].map(|_| Default::default());
+    let mut bezierBuffer: [POINT; 4] = Default::default();
+    let mut buffer: &mut [POINT];
+    let mut bufferSize: usize;
+    let mut startFigure: [POINT; 1] = Default::default();
+    // The current point offset in rgpt
+    let mut iPoint: usize;
+    // The current type offset in rgTypes
+    let mut iType: usize;
+    let mut runSize: usize;
+    let mut thisCount: usize;
+    let mut isMore: bool = false;
+    let mut xLast: INT;
+    let mut yLast: INT;
+
+    ASSERTPATH!(rgTypes, cPoints);
+
+    // Every valid subpath has at least two vertices in it, hence the
+    // check of 'cPoints - 1':
+
+    iPoint = 0;
+    iType = 0;
+
+    assert!(cPoints > 1);
+    while (iPoint < cPoints as usize - 1) {
+        assert!((rgTypes[iType] & PathPointTypePathTypeMask) == PathPointTypeStart);
+        assert!((rgTypes[iType + 1] & PathPointTypePathTypeMask) != PathPointTypeStart);
+
+        // Add the start point to the beginning of the batch, and
+        // remember it for handling the close figure:
+
+        startFigure[0] = rgpt[iPoint];
+
+        bufferStart[0].x = startFigure[0].x;
+        bufferStart[0].y = startFigure[0].y;
+        let bufferStartPtr = bufferStart.as_ptr();
+        buffer = &mut bufferStart[1..];
+        bufferSize = ENUMERATE_BUFFER_NUMBER!() - 1;
+
+        // We need to enter our loop with 'iType' pointing one past
+        // the start figure:
+
+        iPoint += 1;
+        iType += 1;
+
+        while {
+            // Try finding a run of lines:
+
+            if ((rgTypes[iType] & PathPointTypePathTypeMask) == PathPointTypeLine) {
+                runSize = 1;
+
+                while ((iPoint + runSize < cPoints as usize)
+                    && ((rgTypes[iType + runSize] & PathPointTypePathTypeMask) == PathPointTypeLine))
+                {
+                    runSize += 1;
+                }
+
+                // Okay, we've found a run of lines.  Break it up into our
+                // buffer size:
+
+                loop {
+                    thisCount = bufferSize.min(runSize);
+
+                    buffer[0 .. thisCount].copy_from_slice(&rgpt[iPoint .. iPoint + thisCount]);
+
+                    __analysis_assume!(
+                        buffer + bufferSize == bufferStart + ENUMERATE_BUFFER_NUMBER
+                    );
+                    assert!(buffer.as_ptr().wrapping_offset(bufferSize as isize) == bufferStartPtr.wrapping_offset(ENUMERATE_BUFFER_NUMBER!()) );
+
+                    iPoint += thisCount;
+                    iType += thisCount;
+                    buffer = &mut buffer[thisCount..];
+                    runSize -= thisCount;
+                    bufferSize -= thisCount;
+
+                    if (bufferSize > 0) {
+                        break;
+                    }
+
+                    xLast = bufferStart[ENUMERATE_BUFFER_NUMBER!() - 1].x;
+                    yLast = bufferStart[ENUMERATE_BUFFER_NUMBER!() - 1].y;
+                    IFR!(InitializeEdges(
+                        enumerateContext,
+                        &mut bufferStart,
+                        ENUMERATE_BUFFER_NUMBER!()
+                    ));
+
+                    // Continue the last vertex as the first in the new batch:
+
+                    bufferStart[0].x = xLast;
+                    bufferStart[0].y = yLast;
+                    buffer = &mut bufferStart[1..];
+                    bufferSize = ENUMERATE_BUFFER_NUMBER!() - 1;
+                    if !(runSize != 0) {
+                        break;
+                    }
+                }
+            } else {
+                assert!(iPoint + 3 <= cPoints as usize);
+                assert!((rgTypes[iType] & PathPointTypePathTypeMask) == PathPointTypeBezier);
+
+                bezierBuffer.copy_from_slice(&rgpt[(iPoint - 1) .. iPoint + 3]);
+
+                // Prepare for the next iteration:
+
+                iPoint += 3;
+                iType += 1;
+
+                // Process the Bezier:
+
+                let mut bezier = CMILBezier::new(&bezierBuffer, clipRect);
+                loop {
+                    thisCount = bezier.Flatten(buffer, &mut isMore) as usize;
+
+                    __analysis_assume!(
+                        buffer + bufferSize == bufferStart + ENUMERATE_BUFFER_NUMBER!()
+                    );
+                    assert!(buffer.as_ptr().wrapping_offset(bufferSize as isize) == bufferStartPtr.wrapping_offset(ENUMERATE_BUFFER_NUMBER!()));
+
+                    buffer = &mut buffer[thisCount..];
+                    bufferSize -= thisCount;
+
+                    if (bufferSize > 0) {
+                        break;
+                    }
+
+                    xLast = bufferStart[ENUMERATE_BUFFER_NUMBER!() - 1].x;
+                    yLast = bufferStart[ENUMERATE_BUFFER_NUMBER!() - 1].y;
+                    IFR!(InitializeEdges(
+                        enumerateContext,
+                        &mut bufferStart,
+                        ENUMERATE_BUFFER_NUMBER!()
+                    ));
+
+                    // Continue the last vertex as the first in the new batch:
+
+                    bufferStart[0].x = xLast;
+                    bufferStart[0].y = yLast;
+                    buffer = &mut bufferStart[1..];
+                    bufferSize = ENUMERATE_BUFFER_NUMBER!() - 1;
+                    if !isMore {
+                        break;
+                    }
+                }
+            }
+
+            ((iPoint < cPoints as usize)
+                && ((rgTypes[iType] & PathPointTypePathTypeMask) != PathPointTypeStart))
+        } {}
+
+        // Okay, the subpath is done.  But we still have to handle the
+        // 'close figure' (which is implicit for a fill):
+        // Add the close-figure point:
+
+        buffer[0].x = startFigure[0].x;
+        buffer[0].y = startFigure[0].y;
+        bufferSize -= 1;
+
+        // We have to flush anything we might have in the batch, unless
+        // there's only one vertex in there!  (The latter case may happen
+        // for the stroke case with no close figure if we just flushed a
+        // batch.)
+        // If we're flattening, we must call the one additional time to
+        // correctly handle closing the subpath, even if there is only
+        // one entry in the batch. The flattening callback handles the
+        // one point case and closes the subpath properly without adding
+        // extraneous points.
+
+        let verticesInBatch = ENUMERATE_BUFFER_NUMBER!() - bufferSize;
+        if (verticesInBatch > 1) {
+            IFR!(InitializeEdges(
+                enumerateContext,
+                &mut bufferStart,
+                (verticesInBatch) as UINT
+            ));
+        }
+    }
+
+    return hr;
+}
+
+/**************************************************************************\
+*
+* Function Description:
+*
+*   We want to sort in the inactive list; the primary key is 'y', and
+*   the secondary key is 'x'.  This routine creates a single LONGLONG
+*   key that represents both.
+*
+* Created:
+*
+*   03/25/2000 andrewgo
+*
+\**************************************************************************/
+
+fn YX(x: INT, y: INT, p: &mut LONGLONG) {
+    // Bias 'x' by INT_MAX so that it's effectively unsigned:
+    /*
+    reinterpret_cast<LARGE_INTEGER*>(p)->HighPart = y;
+    reinterpret_cast<LARGE_INTEGER*>(p)->LowPart = x + INT_MAX;
+    */
+    *p = (((y as u64) << 32) | (((x as i64 + i32::MAX as i64) as u64) & 0xffffffff)) as i64;
+}
+
+/**************************************************************************\
+*
+* Function Description:
+*
+*   Recursive function to quick-sort our inactive edge list.  Note that
+*   for performance, the results are not completely sorted; an insertion
+*   sort has to be run after the quicksort in order to do a lighter-weight
+*   sort of the subtables.
+*
+* Created:
+*
+*   03/25/2000 andrewgo
+*
+\**************************************************************************/
+
+const QUICKSORT_THRESHOLD: isize = 8;
+
+fn QuickSortEdges(inactive: &mut [CInactiveEdge],
+    /*__inout_xcount(f - l + 1 elements)*/ f: usize,
+    /*__inout_xcount(array starts at f)*/ l: usize,
+) {
+    let mut e: Ref<CEdge>;
+    let mut y: LONGLONG;
+    let mut first: LONGLONG;
+    let mut second: LONGLONG;
+    let mut last: LONGLONG;
+
+    // Find the median of the first, middle, and last elements:
+
+    let m = f + ((l - f) >> 1);
+
+    SWAP!(y, inactive[f + 1].Yx, inactive[m].Yx);
+    SWAP!(e, inactive[f + 1].Edge, inactive[m].Edge);
+
+    if {second = inactive[f + 1].Yx; second > {last = inactive[l].Yx; last}} {
+        inactive[f + 1].Yx = last;
+        inactive[l].Yx = second;
+
+        SWAP!(e, inactive[f + 1].Edge, inactive[l].Edge);
+    }
+    if {first = inactive[f].Yx; first} > {last = inactive[l].Yx; last} {
+        inactive[f].Yx = last;
+        inactive[l].Yx = first;
+
+        SWAP!(e, inactive[f].Edge, inactive[l].Edge);
+    }
+    if {second = inactive[f + 1].Yx; second} > {first = inactive[f].Yx; first} {
+        inactive[f + 1].Yx = first;
+        inactive[f].Yx = second;
+
+        SWAP!(e, inactive[f + 1].Edge, inactive[f].Edge);
+    }
+
+    // f->Yx is now the desired median, and (f + 1)->Yx <= f->Yx <= l->Yx
+
+    debug_assert!((inactive[f + 1].Yx <= inactive[f].Yx) && (inactive[f].Yx <= inactive[l].Yx));
+
+    let median = inactive[f].Yx;
+
+    let mut i = f + 2;
+    while (inactive[i].Yx < median) {
+        i += 1;
+    }
+
+    let mut j = l - 1;
+    while (inactive[j].Yx > median) {
+        j -= 1;
+    }
+
+    while (i < j) {
+        SWAP!(y, inactive[i].Yx, inactive[j].Yx);
+        SWAP!(e, inactive[i].Edge, inactive[j].Edge);
+
+        while {
+            i = i + 1;
+            inactive[i].Yx < median
+        } {}
+
+        while {
+            j = j - 1 ;
+            inactive[j].Yx > median
+        } {}
+    }
+
+    SWAP!(y, inactive[f].Yx, inactive[j].Yx);
+    SWAP!(e, inactive[f].Edge, inactive[j].Edge);
+
+    let a = j - f;
+    let b = l - j;
+
+    // Use less stack space by recursing on the shorter subtable.  Also,
+    // have the less-overhead insertion-sort handle small subtables.
+
+    if (a <= b) {
+        if (a > QUICKSORT_THRESHOLD as usize) {
+            // 'a' is the smallest, so do it first:
+
+            QuickSortEdges(inactive, f, j - 1);
+            QuickSortEdges(inactive, j + 1, l);
+        } else if (b > QUICKSORT_THRESHOLD as usize) {
+            QuickSortEdges(inactive, j + 1, l);
+        }
+    } else {
+        if (b > QUICKSORT_THRESHOLD as usize) {
+            // 'b' is the smallest, so do it first:
+
+            QuickSortEdges(inactive, j + 1 , l);
+            QuickSortEdges(inactive, f, j + 1);
+        } else if (a > QUICKSORT_THRESHOLD as usize) {
+            QuickSortEdges(inactive, f, j -1);
+        }
+    }
+}
+
+/**************************************************************************\
+*
+* Function Description:
+*
+*   Do a sort of the inactive table using an insertion-sort.  Expects
+*   large tables to have already been sorted via quick-sort.
+*
+* Created:
+*
+*   03/25/2000 andrewgo
+*
+\**************************************************************************/
+
+fn InsertionSortEdges(
+    /* __inout_xcount(count forward & -1 back)*/ mut inactive: &mut [CInactiveEdge],
+    mut count: INT,
+) {
+    let mut e: Ref<CEdge>;
+    let mut y: LONGLONG;
+    let mut yPrevious: LONGLONG;
+
+    assert!(inactive[0].Yx == i64::MIN);
+    assert!(count >= 2);
+    //inactive = &mut inactive[1..];
+
+    let mut indx = 2; // Skip first entry (by definition it's already in order!)
+    count -= 1;
+
+    while {
+        let mut p = indx;
+
+        // Copy the current stuff to temporary variables to make a hole:
+
+        e = (inactive[indx]).Edge;
+        y = (inactive[indx]).Yx;
+
+        // Shift everything one slot to the right (effectively moving
+        // the hole one position to the left):
+
+        while (y < {yPrevious = inactive[p-1].Yx; yPrevious}) {
+            inactive[p].Yx = yPrevious;
+            inactive[p].Edge = inactive[p-1].Edge;
+            p -= 1;
+        }
+
+        // Drop the temporary stuff into the final hole:
+
+        inactive[p].Yx = y;
+        inactive[p].Edge = e;
+
+        // The quicksort should have ensured that we don't have to move
+        // any entry terribly far:
+
+        assert!((indx - p) <= QUICKSORT_THRESHOLD as usize);
+
+        indx += 1;
+        count -= 1;
+        count != 0
+    } {}
+}
+
+/**************************************************************************\
+*
+* Function Description:
+*
+*   Assert the state of the inactive array.
+*
+* Created:
+*
+*   03/25/2000 andrewgo
+*
+\**************************************************************************/
+macro_rules! ASSERTINACTIVEARRAY {
+    ($inactive: expr, $count: expr) => {
+        #[cfg(debug_assertions)]
+        AssertInactiveArray($inactive, $count);
+    };
+}
+fn AssertInactiveArray(
+    /*__in_ecount(count)*/
+    mut inactive: &[CInactiveEdge], // Annotation should allow the -1 element
+    mut count: INT,
+) {
+    // Verify the head:
+
+    /*#if !ANALYSIS*/
+    // #if needed because prefast don't know that the -1 element is avaliable
+    assert!(inactive[0].Yx == i64::MIN);
+    /*#endif*/
+    assert!(inactive[1].Yx != i64::MIN);
+
+    while {
+        let mut yx: LONGLONG = 0;
+        YX((*inactive[1].Edge).X.get(), (*inactive[1].Edge).StartY, &mut yx);
+
+        assert!(inactive[1].Yx == yx);
+        /*#if !ANALYSIS*/
+        // #if needed because tools don't know that the -1 element is avaliable
+        assert!(inactive[1].Yx >= inactive[0].Yx);
+        /*#endif*/
+        inactive = &inactive[1..];
+        count -= 1;
+        count != 0
+    } {}
+
+    // Verify that the tail is setup appropriately:
+
+    assert!((*inactive[1].Edge).StartY == INT::MAX);
+}
+
+/**************************************************************************\
+*
+* Function Description:
+*
+*   Initialize and sort the inactive array.
+*
+* Returns:
+*
+*   'y' value of topmost edge.
+*
+* Created:
+*
+*   03/25/2000 andrewgo
+*
+\**************************************************************************/
+
+pub fn InitializeInactiveArray<'a>(
+    pEdgeStore: &'a Arena<CEdge<'a>>,
+    /*__in_ecount(count+2)*/ rgInactiveArray: &mut [CInactiveEdge<'a>],
+    count: UINT,
+    tailEdge: Ref<'a, CEdge<'a>> // Tail sentinel for inactive list
+) -> INT {
+    let rgInactiveArrayPtr = rgInactiveArray.as_mut_ptr();
+
+    // First initialize the inactive array.  Skip the first entry,
+    // which we reserve as a head sentinel for the insertion sort:
+
+    let mut pInactiveEdge = &mut rgInactiveArray[1..];
+
+    for e in pEdgeStore.iter() {
+
+            pInactiveEdge[0].Edge = Ref::new(e);
+            YX(e.X.get(), e.StartY, &mut pInactiveEdge[0].Yx);
+            pInactiveEdge = &mut pInactiveEdge[1..];
+    }
+
+    assert!(unsafe { pInactiveEdge.as_mut_ptr().offset_from(rgInactiveArrayPtr) } as UINT == count + 1);
+
+    // Add the tail, which is used when reading back the array.  This
+    // is why we had to allocate the array as 'count + 1':
+
+    pInactiveEdge[0].Edge = tailEdge;
+
+    // Add the head, which is used for the insertion sort.  This is why
+    // we had to allocate the array as 'count + 2':
+
+    rgInactiveArray[0].Yx = i64::MIN;
+
+    // Only invoke the quicksort routine if it's worth the overhead:
+
+    if (count as isize > QUICKSORT_THRESHOLD) {
+        // Quick-sort this, skipping the first and last elements,
+        // which are sentinels.
+        //
+        // We do 'inactiveArray + count' to be inclusive of the last
+        // element:
+
+        QuickSortEdges(rgInactiveArray, 1, count as usize);
+    }
+
+    // Do a quick sort to handle the mostly sorted result:
+
+    InsertionSortEdges(rgInactiveArray, count as i32);
+
+    ASSERTINACTIVEARRAY!(rgInactiveArray, count as i32);
+
+    // Return the 'y' value of the topmost edge:
+
+    return (*rgInactiveArray[1].Edge).StartY;
+
+}
+
+/**************************************************************************\
+*
+* Function Description:
+*
+*   Insert edges into the active edge list.
+*
+* Created:
+*
+*   03/25/2000 andrewgo
+*
+\**************************************************************************/
+
+pub fn InsertNewEdges<'a>(
+    mut pActiveList: Ref<'a, CEdge<'a>>,
+    iCurrentY: INT,
+    /*__deref_inout_xcount(array terminated by an edge with StartY != iCurrentY)*/
+    ppInactiveEdge: &'a mut [CInactiveEdge<'a>],
+    pYNextInactive: &mut INT, // will be INT_MAX when no more
+) -> &'a mut [CInactiveEdge<'a>] {
+
+    let mut inactive: &mut [CInactiveEdge] = ppInactiveEdge;
+
+    assert!((*inactive[0].Edge).StartY == iCurrentY);
+
+    while {
+        let newActive: Ref<CEdge> = inactive[0].Edge;
+
+        // The activeList edge list sentinel has X = INT_MAX, so this always
+        // terminates:
+
+        while ((*(*pActiveList).Next.get()).X < (*newActive).X) {
+            pActiveList = (*pActiveList).Next.get();
+        }
+
+        if SORT_EDGES_INCLUDING_SLOPE {
+            // The activeList edge list sentinel has Dx = INT_MAX, so this always
+            // terminates:
+
+            while (((*(*pActiveList).Next.get()).X == (*newActive).X) && ((*(*pActiveList).Next.get()).Dx < (*newActive).Dx)) {
+                pActiveList = (*pActiveList).Next.get();
+            }
+        }
+
+        (*newActive).Next.set((*pActiveList).Next.get());
+        (*pActiveList).Next.set(newActive);
+
+        inactive = &mut inactive[1..];
+        (*(inactive[0]).Edge).StartY == iCurrentY
+    } {}
+
+    *pYNextInactive = (*(inactive[0]).Edge).StartY;
+    return inactive;
+
+}
+
+/**************************************************************************\
+*
+* Function Description:
+*
+*   Sort the edges so that they're in ascending 'x' order.
+*
+*   We use a bubble-sort for this stage, because edges maintain good
+*   locality and don't often switch ordering positions.
+*
+* Created:
+*
+*   03/25/2000 andrewgo
+*
+\**************************************************************************/
+
+fn SortActiveEdges(list: Ref<CEdge>) {
+
+    let mut swapOccurred: bool;
+    let mut tmp: Ref<CEdge>;
+
+    // We should never be called with an empty active edge list:
+
+    assert!((*(*list).Next.get()).X.get() != INT::MAX);
+
+    while {
+        swapOccurred = false;
+
+        let mut previous = list;
+        let mut current = (*list).Next.get();
+        let mut next = (*current).Next.get();
+        let mut nextX = (*next).X.get();
+
+        while {
+            if (nextX < (*current).X.get()) {
+                swapOccurred = true;
+
+                (*previous).Next.set(next);
+                (*current).Next.set((*next).Next.get());
+                (*next).Next.set(current);
+
+                SWAP!(tmp, next, current);
+            }
+
+            previous = current;
+            current = next;
+            next = (*next).Next.get();
+            nextX = (*next).X.get();
+            nextX != INT::MAX
+        } {}
+        swapOccurred
+    } {}
+
+}
diff --git a/third_party/rust/wpf-gpu-raster/src/bezier.rs b/third_party/rust/wpf-gpu-raster/src/bezier.rs
new file mode 100644
index 0000000000..fe54628a40
--- /dev/null
+++ b/third_party/rust/wpf-gpu-raster/src/bezier.rs
@@ -0,0 +1,990 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+// See the LICENSE file in the project root for more information.
+
+//+-----------------------------------------------------------------------------
+//
+//  class Bezier32
+//
+//  Bezier cracker.
+//
+//  A hybrid cubic Bezier curve flattener based on KirkO's error factor.
+//  Generates line segments fast without using the stack.  Used to flatten a
+//  path.
+//
+//  For an understanding of the methods used, see:
+//
+//  Kirk Olynyk, "..."
+//  Goossen and Olynyk, "System and Method of Hybrid Forward
+//      Differencing to Render Bezier Splines"
+//  Lien, Shantz and Vaughan Pratt, "Adaptive Forward Differencing for
+//  Rendering Curves and Surfaces", Computer Graphics, July 1987
+//  Chang and Shantz, "Rendering Trimmed NURBS with Adaptive Forward
+//      Differencing", Computer Graphics, August 1988
+//  Foley and Van Dam, "Fundamentals of Interactive Computer Graphics"
+//
+//  Public Interface:
+//      bInit(pptfx)                - pptfx points to 4 control points of
+//                                    Bezier.  Current point is set to the first
+//                                    point after the start-point.
+//      Bezier32(pptfx)             - Constructor with initialization.
+//      vGetCurrent(pptfx)          - Returns current polyline point.
+//      bCurrentIsEndPoint()        - TRUE if current point is end-point.
+//      vNext()                     - Moves to next polyline point.
+//
+
+
+#![allow(unused_parens)]
+#![allow(non_upper_case_globals)]
+//+-----------------------------------------------------------------------------
+//
+
+//
+//  $TAG ENGR
+
+//      $Module:    win_mil_graphics_geometry
+//      $Keywords:
+//
+//  $Description:
+//      Class for flattening a bezier.
+//
+//  $ENDTAG
+//
+//------------------------------------------------------------------------------
+
+// First conversion from original 28.4 to 18.14 format
+const HFD32_INITIAL_SHIFT: i32 = 10;
+
+// Second conversion to 15.17 format
+const HFD32_ADDITIONAL_SHIFT: i32 = 3;
+
+
+// BEZIER_FLATTEN_GDI_COMPATIBLE:
+//
+// Don't turn on this switch without testing carefully. It's more for
+// documentation's sake - to show the values that GDI used - for an error
+// tolerance of 2/3.
+
+// It turns out that 2/3 produces very noticable artifacts on antialiased lines,
+// so we want to use 1/4 instead.
+/* 
+#ifdef BEZIER_FLATTEN_GDI_COMPATIBLE
+
+// Flatten to an error of 2/3.  During initial phase, use 18.14 format.
+
+#define TEST_MAGNITUDE_INITIAL    (6 * 0x00002aa0L)
+
+// Error of 2/3.  During normal phase, use 15.17 format.
+
+#define TEST_MAGNITUDE_NORMAL     (TEST_MAGNITUDE_INITIAL << 3)
+
+#else
+*/
+use crate::types::*;
+/* 
+// Flatten to an error of 1/4.  During initial phase, use 18.14 format.
+
+const TEST_MAGNITUDE_INITIAL: i32 =   (6 * 0x00001000);
+
+// Error of 1/4.  During normal phase, use 15.17 format.
+
+const TEST_MAGNITUDE_NORMAL: i32 =    (TEST_MAGNITUDE_INITIAL << 3);
+*/
+
+// I have modified the constants for HFD32 as part of fixing accuracy errors
+// (Bug 816015).  Something similar could be done for the 64 bit hfd, but it ain't
+// broke so I'd rather not fix it.
+
+// The shift to the steady state 15.17 format
+const HFD32_SHIFT: LONG = HFD32_INITIAL_SHIFT + HFD32_ADDITIONAL_SHIFT;
+
+// Added to output numbers before rounding back to original representation
+const HFD32_ROUND: LONG = 1 << (HFD32_SHIFT - 1);
+
+// The error is tested on max(|e2|, |e3|), which represent 6 times the actual error.
+// The flattening tolerance is hard coded to 1/4 in the original geometry space,
+// which translates to 4 in 28.4 format.  So 6 times that is:
+
+const HFD32_TOLERANCE: LONGLONG = 24;
+
+// During the initial phase, while working in 18.14 format
+const HFD32_INITIAL_TEST_MAGNITUDE: LONGLONG = HFD32_TOLERANCE << HFD32_INITIAL_SHIFT; 
+
+// During the steady state, while working in 15.17 format
+const HFD32_TEST_MAGNITUDE: LONGLONG = HFD32_INITIAL_TEST_MAGNITUDE << HFD32_ADDITIONAL_SHIFT; 
+
+// We will stop halving the segment with basis e1, e2, e3, e4 when max(|e2|, |e3|)
+// is less than HFD32_TOLERANCE.  The operation e2 = (e2 + e3) >> 3 in vHalveStepSize() may
+// eat up 3 bits of accuracy. HfdBasis32 starts off with a pad of HFD32_SHIFT zeros, so
+// we can stay exact up to HFD32_SHIFT/3 subdivisions.  Since every subdivision is guaranteed
+// to shift max(|e2|, |e3|) at least by 2, we will subdivide no more than n times if the 
+// initial max(|e2|, |e3|) is less than than HFD32_TOLERANCE << 2n.  But if the initial 
+// max(|e2|, |e3|) is greater than HFD32_TOLERANCE >> (HFD32_SHIFT / 3) then we may not be
+// able to flatten with the 32 bit hfd, so we need to resort to the 64 bit hfd. 
+
+const HFD32_MAX_ERROR: INT = (HFD32_TOLERANCE as i32) << ((2 * HFD32_INITIAL_SHIFT) / 3);
+
+// The maximum size of coefficients that can be handled by HfdBasis32.
+const HFD32_MAX_SIZE: LONGLONG = 0xffffc000;
+
+// Michka 9/12/03: I found this number in the the body of the code witout any explanation.
+// My analysis suggests that we could get away with larger numbers, but if I'm wrong we 
+// could be in big trouble, so let us stay conservative.
+//
+// In bInit() we subtract Min(Bezier coeffients) from the original coefficients, so after
+// that 0 <= coefficients <= Bound, and the test will be Bound < HFD32_MAX_SIZE. When 
+// switching to the HFD basis in bInit():
+//   * e0 is the first Bezier coeffient, so abs(e0) <= Bound.
+//   * e1 is a difference of non-negative coefficients so abs(e1) <= Bound. 
+//   * e2 and e3 can be written as 12*(p - (q + r)/2) where p,q and r are coefficients.
+//     0 <=(q + r)/2 <= Bound, so abs(p - (q + r)/2) <= 2*Bound, hence 
+//     abs(e2), abs(e3) <= 12*Bound.
+//
+// During vLazyHalveStepSize we add e2 + e3, resulting in absolute value <= 24*Bound.
+// Initially HfdBasis32 shifts the numbers by HFD32_INITIAL_SHIFT, so we need to handle 
+// 24*bounds*(2^HFD32_SHIFT), and that needs to be less than 2^31. So the bounds need to
+// be less than 2^(31-HFD32_INITIAL_SHIFT)/24).
+//
+// For speed, the algorithm uses & rather than < for comparison.  To facilitate that we 
+// replace 24 by 32=2^5, and then the binary representation of the number is of the form
+// 0...010...0 with HFD32_SHIFT+5 trailing zeros.  By subtracting that from 2^32 = 0xffffffff+1 
+// we get a number that is 1..110...0 with the same number of trailing zeros, and that can be 
+// used with an & for comparison.  So the number should be:
+//
+//      0xffffffffL - (1L << (31 - HFD32_INITIAL_SHIFT - 5)) + 1 = (1L << 16) + 1 = 0xffff0000
+//
+// For the current values of HFD32_INITIAL_SHIFT=10 and HFD32_ADDITIONAL_SHIFT=3, the steady
+// state doesn't pose additional requirements, as shown below. 
+//
+// For some reason the current code uses 0xfffc0000 = (1L << 14) + 1.
+//
+// Here is why the steady state doesn't pose additional requirements:
+//
+// In vSteadyState we multiply e0 and e1 by 8, so the requirement is Bounds*2^13 < 2^31,
+// or Bounds < 2^18, less stringent than the above.
+//
+// In vLazyHalveStepSize we cut the error down by subdivision, making abs(e2) and abs(e3) 
+// less than HFD32_TEST_MAGNITUDE = 24*2^13,  well below 2^31.
+//
+// During all the steady-state operations - vTakeStep, vHalveStepSize and vDoubleStepSize, 
+// e0 is on the curve and e1 is a difference of 2 points on the curve, so
+// abs(e0), abs(e1) < Bounds * 2^13, which requires Bound < 2^(31-13) = 2^18.  e2 and e3
+// are errors, kept below 6*HFD32_TEST_MAGNITUDE = 216*2^13.  Details:
+//
+// In vTakeStep e2 = 2e2 - e3 keeps abs(e2) < 3*HFD32_TEST_MAGNITUDE =  72*2^13,  
+// well below 2^31
+//
+// In vHalveStepSize we add e2 + e3 when their absolute is < 3*HFD32_TEST_MAGNITUDE (because
+// this comes after a step), so that keeps the result below 6*HFD32_TEST_MAGNITUDE = 216*2^13.
+// 
+// In vDoubleStepSize we know that abs(e2), abs(e3) < HFD32_TEST_MAGNITUDE/4, otherwise we
+// would not have doubled the step.
+
+#[derive(Default)]
+struct HfdBasis32
+{
+    e0: LONG,
+    e1: LONG,
+    e2: LONG,
+    e3: LONG,
+}
+
+impl HfdBasis32 {
+    fn lParentErrorDividedBy4(&self) -> LONG { 
+        self.e3.abs().max((self.e2 + self.e2 - self.e3).abs())
+    }
+
+    fn lError(&self) -> LONG             
+    { 
+        self.e2.abs().max(self.e3.abs())
+    }
+
+    fn fxValue(&self) -> INT
+    { 
+        return((self.e0 + HFD32_ROUND) >> HFD32_SHIFT); 
+    }
+
+    fn bInit(&mut self, p1: INT, p2: INT, p3: INT, p4: INT) -> bool
+    {
+    // Change basis and convert from 28.4 to 18.14 format:
+    
+        self.e0 = (p1                     ) << HFD32_INITIAL_SHIFT;
+        self.e1 = (p4 - p1                ) << HFD32_INITIAL_SHIFT;
+        
+        self.e2 = 6 * (p2 - p3 - p3 + p4);
+        self.e3 = 6 * (p1 - p2 - p2 + p3);
+
+        if (self.lError() >= HFD32_MAX_ERROR)
+        {
+            // Large error, will require too many subdivision for this 32 bit hfd
+            return false;
+        }
+        
+        self.e2 <<= HFD32_INITIAL_SHIFT;
+        self.e3 <<= HFD32_INITIAL_SHIFT;
+
+        return true; 
+    }
+    
+    fn vLazyHalveStepSize(&mut self, cShift: LONG)
+    {
+        self.e2 = self.ExactShiftRight(self.e2 + self.e3,  1);
+        self.e1 = self.ExactShiftRight(self.e1 - self.ExactShiftRight(self.e2, cShift), 1);
+    }
+    
+    fn vSteadyState(&mut self, cShift: LONG)
+    {
+    // We now convert from 18.14 fixed format to 15.17:
+    
+        self.e0 <<= HFD32_ADDITIONAL_SHIFT;
+        self.e1 <<= HFD32_ADDITIONAL_SHIFT;
+    
+        let mut lShift = cShift - HFD32_ADDITIONAL_SHIFT;
+    
+        if (lShift < 0)
+        {
+            lShift = -lShift;
+            self.e2 <<= lShift;
+            self.e3 <<= lShift;
+        }
+        else
+        {
+            self.e2 >>= lShift;
+            self.e3 >>= lShift;
+        }
+    }
+    
+    fn vHalveStepSize(&mut self)
+    {
+        self.e2 = self.ExactShiftRight(self.e2 + self.e3, 3);
+        self.e1 = self.ExactShiftRight(self.e1 - self.e2, 1);
+        self.e3 = self.ExactShiftRight(self.e3, 2);
+    }
+    
+    fn vDoubleStepSize(&mut self)
+    {
+        self.e1 += self.e1 + self.e2;
+        self.e3 <<= 2;
+        self.e2 = (self.e2 << 3) - self.e3;
+    }
+    
+    fn vTakeStep(&mut self)
+    {
+        self.e0 += self.e1;
+        let lTemp = self.e2;
+        self.e1 += lTemp;
+        self.e2 += lTemp - self.e3;
+        self.e3 = lTemp;
+    }
+
+    fn ExactShiftRight(&self, num: i32, shift: i32) -> i32
+    {
+        // Performs a shift to the right while asserting that we're not 
+        // losing significant bits
+     
+        assert!(num == (num >> shift) << shift); 
+        return num >> shift;
+    }
+}
+
+fn vBoundBox(
+    aptfx: &[POINT; 4]) -> RECT
+{
+    let mut left = aptfx[0].x;
+    let mut right = aptfx[0].x;
+    let mut top = aptfx[0].y;
+    let mut bottom = aptfx[0].y;
+
+    for i in 1..4
+    {
+        left = left.min(aptfx[i].x);
+        top = top.min(aptfx[i].y);
+        right = right.max(aptfx[i].x);
+        bottom = bottom.max(aptfx[i].y);
+    }
+
+    // We make the bounds one pixel loose for the nominal width 
+    // stroke case, which increases the bounds by half a pixel 
+    // in every dimension:
+
+    RECT { left: left - 16, top: top - 16, right: right + 16, bottom: bottom + 16}
+}
+
+
+
+fn bIntersect(
+    a: &RECT,
+    b: &RECT) -> bool
+{
+    return((a.left < b.right) &&
+           (a.top < b.bottom) &&
+           (a.right > b.left) &&
+           (a.bottom > b.top));
+}
+
+#[derive(Default)]
+pub struct Bezier32
+{
+    cSteps: LONG,
+    x: HfdBasis32,
+    y: HfdBasis32,
+    rcfxBound: RECT
+}
+impl Bezier32 {
+    
+fn bInit(&mut self,
+    aptfxBez: &[POINT; 4],
+        // Pointer to 4 control points
+    prcfxClip: Option<&RECT>) -> bool
+        // Bound box of visible region (optional)
+{
+    let mut aptfx;
+    let mut cShift = 0;    // Keeps track of 'lazy' shifts
+
+    self.cSteps = 1;         // Number of steps to do before reach end of curve
+
+    self.rcfxBound = vBoundBox(aptfxBez);
+
+    aptfx = aptfxBez.clone();
+
+    {
+        let mut fxOr;
+        let mut fxOffset;
+
+        // find out if the coordinates minus the bounding box
+        // exceed 10 bits
+        fxOffset = self.rcfxBound.left;
+        fxOr  = {aptfx[0].x -= fxOffset; aptfx[0].x};
+        fxOr |= {aptfx[1].x -= fxOffset; aptfx[1].x};
+        fxOr |= {aptfx[2].x -= fxOffset; aptfx[2].x};
+        fxOr |= {aptfx[3].x -= fxOffset; aptfx[3].x};
+
+        fxOffset = self.rcfxBound.top;
+        fxOr |= {aptfx[0].y -= fxOffset; aptfx[0].y};
+        fxOr |= {aptfx[1].y -= fxOffset; aptfx[1].y};
+        fxOr |= {aptfx[2].y -= fxOffset; aptfx[2].y};
+        fxOr |= {aptfx[3].y -= fxOffset; aptfx[3].y};
+
+    // This 32 bit cracker can only handle points in a 10 bit space:
+
+        if ((fxOr as i64 & HFD32_MAX_SIZE) != 0) {
+            return false;
+        }
+    }
+
+    if (!self.x.bInit(aptfx[0].x, aptfx[1].x, aptfx[2].x, aptfx[3].x))
+    {
+        return false;
+    }
+    if (!self.y.bInit(aptfx[0].y, aptfx[1].y, aptfx[2].y, aptfx[3].y))
+    {
+        return false;
+    }
+    
+
+    if (match prcfxClip { None => true, Some(clip) => bIntersect(&self.rcfxBound, clip)})
+    {
+        
+        loop {
+            let lTestMagnitude = (HFD32_INITIAL_TEST_MAGNITUDE << cShift) as LONG;
+
+            if (self.x.lError() <= lTestMagnitude && self.y.lError() <= lTestMagnitude) {
+                break;
+            }
+
+            cShift += 2;
+            self.x.vLazyHalveStepSize(cShift);
+            self.y.vLazyHalveStepSize(cShift);
+            self.cSteps <<= 1;
+        }
+    }
+
+    self.x.vSteadyState(cShift);
+    self.y.vSteadyState(cShift);
+
+// Note that this handles the case where the initial error for
+// the Bezier is already less than HFD32_TEST_MAGNITUDE:
+
+    self.x.vTakeStep();
+    self.y.vTakeStep();
+    self.cSteps-=1;
+
+    return true;
+}
+
+
+fn cFlatten(&mut self,
+    mut pptfx: &mut [POINT],
+    pbMore: &mut bool) -> i32
+{
+    let mut cptfx = pptfx.len();
+    assert!(cptfx > 0);
+
+    let cptfxOriginal = cptfx;
+
+    while {
+    // Return current point:
+    
+        pptfx[0].x = self.x.fxValue() + self.rcfxBound.left;
+        pptfx[0].y = self.y.fxValue() + self.rcfxBound.top;
+        pptfx = &mut pptfx[1..];
+    
+    // If cSteps == 0, that was the end point in the curve!
+    
+        if (self.cSteps == 0)
+        {
+            *pbMore = false;
+
+            // '+1' because we haven't decremented 'cptfx' yet:
+
+            return(cptfxOriginal - cptfx + 1) as i32;
+        }
+    
+    // Okay, we have to step:
+    
+        if (self.x.lError().max(self.y.lError()) > HFD32_TEST_MAGNITUDE as LONG)
+        {
+            self.x.vHalveStepSize();
+            self.y.vHalveStepSize();
+            self.cSteps <<= 1;
+        }
+    
+        // We are here after vTakeStep.  Before that the error max(|e2|,|e3|) was less
+        // than HFD32_TEST_MAGNITUDE.  vTakeStep changed e2 to 2e2-e3. Since 
+        // |2e2-e3| < max(|e2|,|e3|) << 2 and vHalveStepSize is guaranteed to reduce 
+        // max(|e2|,|e3|) by >> 2, no more than one subdivision should be required to 
+        // bring the new max(|e2|,|e3|) back to within HFD32_TEST_MAGNITUDE, so:
+        assert!(self.x.lError().max(self.y.lError()) <= HFD32_TEST_MAGNITUDE as LONG);
+    
+        while (!(self.cSteps & 1 != 0) &&
+               self.x.lParentErrorDividedBy4() <= (HFD32_TEST_MAGNITUDE as LONG >> 2) &&
+               self.y.lParentErrorDividedBy4() <= (HFD32_TEST_MAGNITUDE as LONG >> 2))
+        {
+            self.x.vDoubleStepSize();
+            self.y.vDoubleStepSize();
+            self.cSteps >>= 1;
+        }
+    
+        self.cSteps -=1 ;
+        self.x.vTakeStep();
+        self.y.vTakeStep();
+        cptfx -= 1;
+        cptfx != 0
+    } {}
+
+    *pbMore = true;
+    return cptfxOriginal as i32;
+}
+}
+
+
+///////////////////////////////////////////////////////////////////////////
+// Bezier64
+//
+// All math is done using 64 bit fixed numbers in a 36.28 format.
+//
+// All drawing is done in a 31 bit space, then a 31 bit window offset
+// is applied.  In the initial transform where we change to the HFD
+// basis, e2 and e3 require the most bits precision: e2 = 6(p2 - 2p3 + p4).
+// This requires an additional 4 bits precision -- hence we require 36 bits
+// for the integer part, and the remaining 28 bits is given to the fraction.
+//
+// In rendering a Bezier, every 'subdivide' requires an extra 3 bits of
+// fractional precision.  In order to be reversible, we can allow no
+// error to creep in.  Since a INT coordinate is 32 bits, and we
+// require an additional 4 bits as mentioned above, that leaves us
+// 28 bits fractional precision -- meaning we can do a maximum of
+// 9 subdivides.  Now, the maximum absolute error of a Bezier curve in 27
+// bit integer space is 2^29 - 1.  But 9 subdivides reduces the error by a
+// guaranteed factor of 2^18, meaning we can subdivide down only to an error
+// of 2^11 before we overflow, when in fact we want to reduce error to less
+// than 1.
+//
+// So what we do is HFD until we hit an error less than 2^11, reverse our
+// basis transform to get the four control points of this smaller curve
+// (rounding in the process to 32 bits), then invoke another copy of HFD
+// on the reduced Bezier curve.  We again have enough precision, but since
+// its starting error is less than 2^11, we can reduce error to 2^-7 before
+// overflowing!  We'll start a low HFD after every step of the high HFD.
+////////////////////////////////////////////////////////////////////////////
+#[derive(Default)]
+struct HfdBasis64
+{
+    e0: LONGLONG,
+    e1: LONGLONG,
+    e2: LONGLONG,
+    e3: LONGLONG,
+}
+
+impl HfdBasis64 {
+fn vParentError(&self) -> LONGLONG
+{
+    (self.e3 << 2).abs().max(((self.e2 << 3) - (self.e3 << 2)).abs())
+}
+
+fn vError(&self) -> LONGLONG
+{
+    self.e2.abs().max(self.e3.abs())
+}
+
+fn fxValue(&self) -> INT
+{
+// Convert from 36.28 and round:
+
+    let mut eq = self.e0;
+    eq += (1 << (BEZIER64_FRACTION - 1));
+    eq >>= BEZIER64_FRACTION;
+    return eq as LONG as INT;
+}
+
+fn vInit(&mut self, p1: INT, p2: INT, p3: INT, p4: INT)
+{
+    let mut eqTmp;
+    let eqP2 = p2 as LONGLONG;
+    let eqP3 = p3 as LONGLONG;
+
+// e0 = p1
+// e1 = p4 - p1
+// e2 = 6(p2 - 2p3 + p4)
+// e3 = 6(p1 - 2p2 + p3)
+
+// Change basis:
+
+    self.e0 = p1 as LONGLONG;                                        // e0 = p1
+    self.e1 = p4 as LONGLONG;
+    self.e2 = eqP2; self.e2 -= eqP3; self.e2 -= eqP3; self.e2 += self.e1;    // e2 = p2 - 2*p3 + p4
+    self.e3 = self.e0;   self.e3 -= eqP2; self.e3 -= eqP2; self.e3 += eqP3;  // e3 = p1 - 2*p2 + p3
+    self.e1 -= self.e0;                                       // e1 = p4 - p1
+
+// Convert to 36.28 format and multiply e2 and e3 by six:
+
+    self.e0 <<= BEZIER64_FRACTION;
+    self.e1 <<= BEZIER64_FRACTION;
+    eqTmp = self.e2; self.e2 += eqTmp; self.e2 += eqTmp; self.e2 <<= (BEZIER64_FRACTION + 1);
+    eqTmp = self.e3; self.e3 += eqTmp; self.e3 += eqTmp; self.e3 <<= (BEZIER64_FRACTION + 1);
+}
+
+fn vUntransform<F: Fn(&mut POINT) -> &mut LONG>(&self,
+    afx: &mut [POINT; 4], field: F)
+{
+// Declare some temps to hold our operations, since we can't modify e0..e3.
+
+    let mut eqP0;
+    let mut eqP1;
+    let mut eqP2;
+    let mut eqP3;
+
+// p0 = e0
+// p1 = e0 + (6e1 - e2 - 2e3)/18
+// p2 = e0 + (12e1 - 2e2 - e3)/18
+// p3 = e0 + e1
+
+    eqP0 = self.e0;
+
+// NOTE PERF: Convert this to a multiply by 6: [andrewgo]
+
+    eqP2 = self.e1;
+    eqP2 += self.e1;
+    eqP2 += self.e1;
+    eqP1 = eqP2;
+    eqP1 += eqP2;           // 6e1
+    eqP1 -= self.e2;             // 6e1 - e2
+    eqP2 = eqP1;
+    eqP2 += eqP1;           // 12e1 - 2e2
+    eqP2 -= self.e3;             // 12e1 - 2e2 - e3
+    eqP1 -= self.e3;
+    eqP1 -= self.e3;             // 6e1 - e2 - 2e3
+
+// NOTE: May just want to approximate these divides! [andrewgo]
+// Or can do a 64 bit divide by 32 bit to get 32 bits right here.
+
+    eqP1 /= 18;
+    eqP2 /= 18;
+    eqP1 += self.e0;
+    eqP2 += self.e0;
+
+    eqP3 = self.e0;
+    eqP3 += self.e1;
+
+// Convert from 36.28 format with rounding:
+
+    eqP0 += (1 << (BEZIER64_FRACTION - 1)); eqP0 >>= BEZIER64_FRACTION; *field(&mut afx[0]) = eqP0 as LONG;
+    eqP1 += (1 << (BEZIER64_FRACTION - 1)); eqP1 >>= BEZIER64_FRACTION; *field(&mut afx[1]) = eqP1 as LONG;
+    eqP2 += (1 << (BEZIER64_FRACTION - 1)); eqP2 >>= BEZIER64_FRACTION; *field(&mut afx[2]) = eqP2 as LONG;
+    eqP3 += (1 << (BEZIER64_FRACTION - 1)); eqP3 >>= BEZIER64_FRACTION; *field(&mut afx[3]) = eqP3 as LONG;
+}
+
+fn vHalveStepSize(&mut self)
+{
+// e2 = (e2 + e3) >> 3
+// e1 = (e1 - e2) >> 1
+// e3 >>= 2
+
+    self.e2 += self.e3; self.e2 >>= 3;
+    self.e1 -= self.e2; self.e1 >>= 1;
+    self.e3 >>= 2;
+}
+
+fn vDoubleStepSize(&mut self)
+{
+// e1 = 2e1 + e2
+// e3 = 4e3;
+// e2 = 8e2 - e3
+
+    self.e1 <<= 1; self.e1 += self.e2;
+    self.e3 <<= 2;
+    self.e2 <<= 3; self.e2 -= self.e3;
+}
+
+fn vTakeStep(&mut self)
+{
+    self.e0 += self.e1;
+    let eqTmp = self.e2;
+    self.e1 += self.e2;
+    self.e2 += eqTmp; self.e2 -= self.e3;
+    self.e3 = eqTmp;
+}
+}
+
+const BEZIER64_FRACTION: LONG  = 28;
+
+// The following is our 2^11 target error encoded as a 36.28 number
+// (don't forget the additional 4 bits of fractional precision!) and
+// the 6 times error multiplier:
+
+const geqErrorHigh: LONGLONG  = (6 * (1 << 15) >> (32 - BEZIER64_FRACTION)) << 32;
+
+/*#ifdef BEZIER_FLATTEN_GDI_COMPATIBLE
+
+// The following is the default 2/3 error encoded as a 36.28 number,
+// multiplied by 6, and leaving 4 bits for fraction:
+
+const LONGLONG geqErrorLow = (LONGLONG)(4) << 32;
+
+#else*/
+
+// The following is the default 1/4 error encoded as a 36.28 number,
+// multiplied by 6, and leaving 4 bits for fraction:
+
+use crate::types::POINT;
+
+const geqErrorLow: LONGLONG = (3) << 31;
+
+//#endif
+#[derive(Default)]
+pub struct Bezier64
+{
+    xLow: HfdBasis64,
+    yLow: HfdBasis64,
+    xHigh: HfdBasis64,
+    yHigh: HfdBasis64,
+
+    eqErrorLow: LONGLONG,
+    rcfxClip: Option<RECT>,
+
+    cStepsHigh: LONG,
+    cStepsLow: LONG
+}
+
+impl Bezier64 {
+
+fn vInit(&mut self, 
+    aptfx: &[POINT; 4],
+        // Pointer to 4 control points
+    prcfxVis: Option<&RECT>,
+        // Pointer to bound box of visible area (may be NULL)
+    eqError: LONGLONG)
+        // Fractional maximum error (32.32 format)
+{
+    self.cStepsHigh = 1;
+    self.cStepsLow  = 0;
+
+    self.xHigh.vInit(aptfx[0].x, aptfx[1].x, aptfx[2].x, aptfx[3].x);
+    self.yHigh.vInit(aptfx[0].y, aptfx[1].y, aptfx[2].y, aptfx[3].y);
+
+// Initialize error:
+
+    self.eqErrorLow = eqError;
+
+    self.rcfxClip = prcfxVis.cloned();
+
+    while (((self.xHigh.vError()) > geqErrorHigh) ||
+           ((self.yHigh.vError()) > geqErrorHigh))
+    {
+        self.cStepsHigh <<= 1;
+        self.xHigh.vHalveStepSize();
+        self.yHigh.vHalveStepSize();
+    }
+}
+
+fn cFlatten(
+    &mut self,
+    mut pptfx: &mut [POINT],
+    pbMore: &mut bool) -> INT
+{
+    let mut aptfx: [POINT; 4] = Default::default();
+    let mut cptfx = pptfx.len();
+    let mut rcfxBound: RECT;
+    let cptfxOriginal = cptfx;
+
+    assert!(cptfx > 0);
+
+    while {
+        if (self.cStepsLow == 0)
+        {
+        // Optimization that if the bound box of the control points doesn't
+        // intersect with the bound box of the visible area, render entire
+        // curve as a single line:
+    
+            self.xHigh.vUntransform(&mut aptfx, |p| &mut p.x);
+            self.yHigh.vUntransform(&mut aptfx, |p| &mut p.y);
+    
+            self.xLow.vInit(aptfx[0].x, aptfx[1].x, aptfx[2].x, aptfx[3].x);
+            self.yLow.vInit(aptfx[0].y, aptfx[1].y, aptfx[2].y, aptfx[3].y);
+            self.cStepsLow = 1;
+    
+            if (match &self.rcfxClip { None => true, Some(clip) => {rcfxBound = vBoundBox(&aptfx); bIntersect(&rcfxBound, &clip)}})
+            {
+                while (((self.xLow.vError()) > self.eqErrorLow) ||
+                       ((self.yLow.vError()) > self.eqErrorLow))
+                {
+                    self.cStepsLow <<= 1;
+                    self.xLow.vHalveStepSize();
+                    self.yLow.vHalveStepSize();
+                }
+            }
+    
+        // This 'if' handles the case where the initial error for the Bezier
+        // is already less than the target error:
+    
+            if ({self.cStepsHigh -= 1; self.cStepsHigh} != 0)
+            {
+                self.xHigh.vTakeStep();
+                self.yHigh.vTakeStep();
+    
+                if (((self.xHigh.vError()) > geqErrorHigh) ||
+                    ((self.yHigh.vError()) > geqErrorHigh))
+                {
+                    self.cStepsHigh <<= 1;
+                    self.xHigh.vHalveStepSize();
+                    self.yHigh.vHalveStepSize();
+                }
+    
+                while (!(self.cStepsHigh & 1 != 0) &&
+                       ((self.xHigh.vParentError()) <= geqErrorHigh) &&
+                       ((self.yHigh.vParentError()) <= geqErrorHigh))
+                {
+                    self.xHigh.vDoubleStepSize();
+                    self.yHigh.vDoubleStepSize();
+                    self.cStepsHigh >>= 1;
+                }
+            }
+        }
+    
+        self.xLow.vTakeStep();
+        self.yLow.vTakeStep();
+    
+        pptfx[0].x = self.xLow.fxValue();
+        pptfx[0].y = self.yLow.fxValue();
+        pptfx = &mut pptfx[1..];
+    
+        self.cStepsLow-=1;
+        if (self.cStepsLow == 0 && self.cStepsHigh == 0)
+        {
+            *pbMore = false;
+
+            // '+1' because we haven't decremented 'cptfx' yet:
+
+            return(cptfxOriginal - cptfx + 1) as INT;
+        }
+    
+        if ((self.xLow.vError() > self.eqErrorLow) ||
+            (self.yLow.vError() > self.eqErrorLow))
+        {
+            self.cStepsLow <<= 1;
+            self.xLow.vHalveStepSize();
+            self.yLow.vHalveStepSize();
+        }
+    
+        while (!(self.cStepsLow & 1 != 0) &&
+               ((self.xLow.vParentError()) <= self.eqErrorLow) &&
+               ((self.yLow.vParentError()) <= self.eqErrorLow))
+        {
+            self.xLow.vDoubleStepSize();
+            self.yLow.vDoubleStepSize();
+            self.cStepsLow >>= 1;
+        }
+        cptfx -= 1;
+        cptfx != 0
+    } {};
+
+    *pbMore = true;
+    return(cptfxOriginal) as INT;
+}
+}
+
+//+-----------------------------------------------------------------------------
+//
+//  class CMILBezier
+//
+//  Bezier cracker.  Flattens any Bezier in our 28.4 device space down to a
+//  smallest 'error' of 2^-7 = 0.0078.  Will use fast 32 bit cracker for small
+//  curves and slower 64 bit cracker for big curves.
+//
+//  Public Interface:
+//      vInit(aptfx, prcfxClip, peqError)
+//          - pptfx points to 4 control points of Bezier.  The first point
+//            retrieved by bNext() is the the first point in the approximation
+//            after the start-point.
+//
+//          - prcfxClip is an optional pointer to the bound box of the visible
+//            region.  This is used to optimize clipping of Bezier curves that
+//            won't be seen.  Note that this value should account for the pen's
+//            width!
+//
+//          - optional maximum error in 32.32 format, corresponding to Kirko's
+//            error factor.
+//
+//      bNext(pptfx)
+//          - pptfx points to where next point in approximation will be
+//            returned.  Returns FALSE if the point is the end-point of the
+//            curve.
+//
+pub (crate) enum CMILBezier
+{
+    Bezier64(Bezier64),
+    Bezier32(Bezier32)
+}
+
+impl CMILBezier {
+    // All coordinates must be in 28.4 format:
+    pub fn new(aptfxBez: &[POINT; 4], prcfxClip: Option<&RECT>) -> Self {
+        let mut bez32 = Bezier32::default();
+        let bBez32 = bez32.bInit(aptfxBez, prcfxClip);
+        if bBez32 {
+            CMILBezier::Bezier32(bez32)
+        } else {
+            let mut bez64 = Bezier64::default();
+            bez64.vInit(aptfxBez, prcfxClip, geqErrorLow);
+            CMILBezier::Bezier64(bez64)
+        }
+    }
+
+    // Returns the number of points filled in. This will never be zero.
+    //
+    // The last point returned may not be exactly the last control
+    //            point. The workaround is for calling code to add an extra
+    //            point if this is the case.
+    pub fn Flatten(    &mut self,
+        pptfx: &mut [POINT],
+        pbMore: &mut bool) -> INT {
+            match self {
+                CMILBezier::Bezier32(bez) => bez.cFlatten(pptfx, pbMore),
+                CMILBezier::Bezier64(bez) => bez.cFlatten(pptfx, pbMore)
+            }
+        }
+}
+
+#[test]
+fn flatten() {
+    let curve: [POINT; 4] = [
+    POINT{x: 1715, y: 6506},
+    POINT{x: 1692, y: 6506},
+    POINT{x: 1227, y: 5148},
+    POINT{x: 647, y: 5211}];
+    let mut bez = CMILBezier::new(&curve, None);
+    let mut result: [POINT; 32] = Default::default();
+    let mut more: bool = false;
+    let count = bez.Flatten(&mut result, &mut more);
+    assert_eq!(count, 21);
+    assert_eq!(more, false);
+}
+
+#[test]
+fn split_flatten32() {
+    // make sure that flattening a curve into two small buffers matches
+    // doing it into a large buffer
+    let curve: [POINT; 4] = [
+    POINT{x: 1795, y: 8445},
+    POINT{x: 1795, y: 8445},
+    POINT{x: 1908, y: 8683},
+    POINT{x: 2043, y: 8705}];
+
+    let mut bez = CMILBezier::new(&curve, None);
+    let mut result: [POINT; 8] = Default::default();
+    let mut more: bool = false;
+    let count = bez.Flatten(&mut result[..5], &mut more);
+    assert_eq!(count, 5);
+    assert_eq!(more, true);
+    let count = bez.Flatten(&mut result[5..], &mut more);
+    assert_eq!(count, 3);
+    assert_eq!(more, false);
+
+    let mut bez = CMILBezier::new(&curve, None);
+    let mut full_result: [POINT; 8] = Default::default();
+    let mut more: bool = false;
+    let count = bez.Flatten(&mut full_result, &mut more);
+    assert_eq!(count, 8);
+    assert_eq!(more, false);
+    assert!(result == full_result);
+}
+
+#[test]
+fn flatten32() {
+    let curve: [POINT; 4] = [
+    POINT{x: 100, y: 100},
+    POINT{x: 110, y: 100},
+    POINT{x: 110, y: 110},
+    POINT{x: 110, y: 100}];
+    let mut bez = CMILBezier::new(&curve, None);
+    let mut result: [POINT; 32] = Default::default();
+    let mut more: bool = false;
+    let count = bez.Flatten(&mut result, &mut more);
+    assert_eq!(count, 3);
+    assert_eq!(more, false);
+}
+
+#[test]
+fn flatten32_double_step_size() {
+    let curve: [POINT; 4] = [
+    POINT{x: 1761, y: 8152},
+    POINT{x: 1761, y: 8152},
+    POINT{x: 1750, y: 8355},
+    POINT{x: 1795, y: 8445}];
+    let mut bez = CMILBezier::new(&curve, None);
+    let mut result: [POINT; 32] = Default::default();
+    let mut more: bool = false;
+    let count = bez.Flatten(&mut result, &mut more);
+    assert_eq!(count, 7);
+    assert_eq!(more, false);
+}
+
+#[test]
+fn bezier64_init_high_num_steps() {
+    let curve: [POINT; 4] = [
+    POINT{x: 33, y: -1},
+    POINT{x: -1, y: -1},
+    POINT{x: -1, y: -16385},
+    POINT{x: -226, y: 10}];
+    let mut bez = CMILBezier::new(&curve, None);
+    let mut result: [POINT; 32] = Default::default();
+    let mut more: bool = false;
+    let count = bez.Flatten(&mut result, &mut more);
+    assert_eq!(count, 32);
+    assert_eq!(more, true);
+}
+
+#[test]
+fn bezier64_high_error() {
+    let curve: [POINT; 4] = [
+    POINT{x: -1, y: -1},
+    POINT{x: -4097, y: -1},
+    POINT{x: 65471, y: -256},
+    POINT{x: -1, y: 0}];
+    let mut bez = CMILBezier::new(&curve, None);
+    let mut result: [POINT; 32] = Default::default();
+    let mut more: bool = false;
+    let count = bez.Flatten(&mut result, &mut more);
+    assert_eq!(count, 32);
+    assert_eq!(more, true);
+}
\ No newline at end of file
diff --git a/third_party/rust/wpf-gpu-raster/src/c_bindings.rs b/third_party/rust/wpf-gpu-raster/src/c_bindings.rs
new file mode 100644
index 0000000000..1016287200
--- /dev/null
+++ b/third_party/rust/wpf-gpu-raster/src/c_bindings.rs
@@ -0,0 +1,158 @@
+use crate::{PathBuilder, OutputPath, OutputVertex, FillMode, rasterize_to_tri_list};
+use crate::types::{BYTE, POINT};
+
+#[no_mangle]
+pub extern "C" fn wgr_new_builder() -> *mut PathBuilder {
+    let pb = PathBuilder::new();
+    Box::into_raw(Box::new(pb))
+}
+
+#[no_mangle]
+pub extern "C" fn wgr_builder_move_to(pb: &mut PathBuilder, x: f32, y: f32) {
+    pb.move_to(x, y);
+}
+
+#[no_mangle]
+pub extern "C" fn wgr_builder_line_to(pb: &mut PathBuilder, x: f32, y: f32) {
+    pb.line_to(x, y);
+}
+
+#[no_mangle]
+pub extern "C" fn wgr_builder_curve_to(pb: &mut PathBuilder, c1x: f32, c1y: f32, c2x: f32, c2y: f32, x: f32, y: f32) {
+    pb.curve_to(c1x, c1y, c2x, c2y, x, y);
+}
+
+#[no_mangle]
+pub extern "C" fn wgr_builder_quad_to(pb: &mut PathBuilder, cx: f32, cy: f32, x: f32, y: f32) {
+    pb.quad_to(cx, cy, x, y);
+}
+
+#[no_mangle]
+pub extern "C" fn wgr_builder_close(pb: &mut PathBuilder) {
+    pb.close();
+}
+
+#[no_mangle]
+pub extern "C" fn wgr_builder_set_fill_mode(pb: &mut PathBuilder, fill_mode: FillMode) {
+    pb.set_fill_mode(fill_mode)
+}
+
+#[repr(C)]
+pub struct Path {
+    fill_mode: FillMode,
+    points: *const POINT,
+    num_points: usize,
+    types: *const BYTE,
+    num_types: usize,
+}
+
+impl From<OutputPath> for Path {
+    fn from(output_path: OutputPath) -> Self {
+        let path = Self {
+            fill_mode: output_path.fill_mode,
+            points: output_path.points.as_ptr(),
+            num_points: output_path.points.len(),
+            types: output_path.types.as_ptr(),
+            num_types: output_path.types.len(),
+        };
+        std::mem::forget(output_path);
+        path
+    }
+}
+
+impl Into<OutputPath> for Path {
+    fn into(self) -> OutputPath {
+        OutputPath {
+            fill_mode: self.fill_mode,
+            points: unsafe {
+                if self.points == std::ptr::null() {
+                    Default::default()
+                } else {
+                    Box::from_raw(std::slice::from_raw_parts_mut(self.points as *mut POINT, self.num_points))
+                }
+            },
+            types: unsafe {
+                if self.types == std::ptr::null() {
+                    Default::default()
+                } else {
+                    Box::from_raw(std::slice::from_raw_parts_mut(self.types as *mut BYTE, self.num_types))
+                }
+            },
+        }
+    }
+}
+
+#[no_mangle]
+pub extern "C" fn wgr_builder_get_path(pb: &mut PathBuilder) -> Path {
+    Path::from(pb.get_path().unwrap_or_default())
+}
+
+#[repr(C)]
+pub struct VertexBuffer {
+    data: *const OutputVertex,
+    len: usize
+}
+
+#[no_mangle]
+pub extern "C" fn wgr_path_rasterize_to_tri_list(
+    path: &Path,
+    clip_x: i32,
+    clip_y: i32,
+    clip_width: i32,
+    clip_height: i32,
+    need_inside: bool,
+    need_outside: bool,
+    rasterization_truncates: bool,
+    output_ptr: *mut OutputVertex,
+    output_capacity: usize,
+) -> VertexBuffer {
+    let output_buffer = if output_ptr != std::ptr::null_mut() {
+        unsafe { Some(std::slice::from_raw_parts_mut(output_ptr, output_capacity)) }
+    } else {
+        None
+    };
+    let mut result = rasterize_to_tri_list(
+        path.fill_mode,
+        unsafe { std::slice::from_raw_parts(path.types, path.num_types) },
+        unsafe { std::slice::from_raw_parts(path.points, path.num_points) },
+        clip_x, clip_y, clip_width, clip_height,
+        need_inside, need_outside,
+        rasterization_truncates,
+        output_buffer
+    );
+    if let Some(output_buffer_size) = result.get_output_buffer_size() {
+        VertexBuffer {
+            data: std::ptr::null(),
+            len: output_buffer_size,
+        }
+    } else {
+        let slice = result.flush_output();
+        let vb = VertexBuffer {
+            data: slice.as_ptr(),
+            len: slice.len(),
+        };
+        std::mem::forget(slice);
+        vb
+    }
+}
+
+#[no_mangle]
+pub extern "C" fn wgr_path_release(path: Path) {
+    let output_path: OutputPath = path.into();
+    drop(output_path);
+}
+
+#[no_mangle]
+pub extern "C" fn wgr_vertex_buffer_release(vb: VertexBuffer)
+{
+    if vb.data != std::ptr::null() {
+        unsafe {
+            drop(Box::from_raw(std::slice::from_raw_parts_mut(vb.data as *mut OutputVertex, vb.len)));
+        }
+    }
+}
+
+#[no_mangle]
+pub unsafe extern "C" fn wgr_builder_release(pb: *mut PathBuilder) {
+    drop(Box::from_raw(pb));
+}
diff --git a/third_party/rust/wpf-gpu-raster/src/fix.rs b/third_party/rust/wpf-gpu-raster/src/fix.rs
new file mode 100644
index 0000000000..bf4741e260
--- /dev/null
+++ b/third_party/rust/wpf-gpu-raster/src/fix.rs
@@ -0,0 +1,9 @@
+use crate::types::*;
+type FIX4 = INT;       // 28.4 fixed point value
+
+// constants for working with 28.4 fixed point values
+macro_rules! FIX4_SHIFT { () => { 4 } }
+macro_rules! FIX4_PRECISION { () => { 4 } }
+macro_rules! FIX4_ONE { () => { (1 << FIX4_PRECISION!()) } }
+macro_rules! FIX4_HALF { () => { (1 << (FIX4_PRECISION!()-1)) } }
+macro_rules! FIX4_MASK { () => { (FIX4_ONE!() - 1) } }
\ No newline at end of file
diff --git a/third_party/rust/wpf-gpu-raster/src/geometry_sink.rs b/third_party/rust/wpf-gpu-raster/src/geometry_sink.rs
new file mode 100644
index 0000000000..3282f10a15
--- /dev/null
+++ b/third_party/rust/wpf-gpu-raster/src/geometry_sink.rs
@@ -0,0 +1,92 @@
+use crate::aacoverage::CCoverageInterval;
+use crate::nullable_ref::Ref;
+use crate::types::*;
+
+pub trait IGeometrySink
+{
+    //
+    // Aliased geometry output
+    //
+/*
+    virtual HRESULT AddVertex(
+        __in_ecount(1) const MilPoint2F &ptPosition,
+            // In: Vertex coordinates
+        __out_ecount(1) WORD *pidxOut
+            // Out: Index of vertex
+        ) PURE;
+
+    virtual HRESULT AddIndexedVertices(
+        UINT cVertices,
+            // In: number of vertices
+        __in_bcount(cVertices*uVertexStride) const void *pVertexBuffer,
+            // In: vertex buffer containing the vertices
+        UINT uVertexStride,
+            // In: size of each vertex
+        MilVertexFormat mvfFormat,
+            // In: format of each vertex
+        UINT cIndices,
+            // In: Number of indices
+        __in_ecount(cIndices) const UINT *puIndexBuffer
+            // In: index buffer                                                             
+        ) PURE;
+
+    virtual void SetTransformMapping(
+        __in_ecount(1) const MILMatrix3x2 &mat2DTransform
+        ) PURE;
+
+    virtual HRESULT AddTriangle(
+        DWORD idx1,
+            // In: Index of triangle's first vertex
+        DWORD idx2,
+            // In: Index of triangle's second vertex
+        DWORD idx3
+            // In: Index of triangle's third vertex
+        ) PURE;
+
+    //
+    // Trapezoidal AA geometry output
+    //
+*/
+    fn AddComplexScan(&mut self,
+        nPixelY: INT,
+            // In: y coordinate in pixel space
+            pIntervalSpanStart: Ref<CCoverageInterval>
+            // In: coverage segments
+        ) -> HRESULT;
+    
+    fn AddTrapezoid(
+        &mut self,
+        rYMin: f32,
+            // In: y coordinate of top of trapezoid
+        rXLeftYMin: f32,
+            // In: x coordinate for top left
+        rXRightYMin: f32,
+            // In: x coordinate for top right
+        rYMax: f32,
+            // In: y coordinate of bottom of trapezoid
+        rXLeftYMax: f32,
+            // In: x coordinate for bottom left
+        rXRightYMax: f32,
+            // In: x coordinate for bottom right
+        rXDeltaLeft: f32,
+            // In: trapezoid expand radius
+        rXDeltaRight: f32
+            // In: trapezoid expand radius
+        ) -> HRESULT;
+
+    fn IsEmpty(&self) -> bool;
+    /*
+    virtual HRESULT AddParallelogram(
+        __in_ecount(4) const MilPoint2F *rgPosition
+        ) PURE;
+    
+    //
+    // Query sink status
+    //
+
+    // Some geometry generators don't actually know if they have output
+    // any triangles, so they need to get this information from the geometry sink.
+
+    virtual BOOL IsEmpty() PURE;
+*/
+}
diff --git a/third_party/rust/wpf-gpu-raster/src/helpers.rs b/third_party/rust/wpf-gpu-raster/src/helpers.rs
new file mode 100644
index 0000000000..27a594831e
--- /dev/null
+++ b/third_party/rust/wpf-gpu-raster/src/helpers.rs
@@ -0,0 +1,55 @@
+pub fn Int32x32To64(a: i32, b: i32) -> i64 { a as i64 * b as i64 }
+
+macro_rules! IsTagEnabled {
+    ($e: expr) => {
+        false
+    }
+}
+
+macro_rules! TraceTag {
+    (($e: expr, $s: expr)) => {
+        dbg!($s)
+    }
+}
+
+macro_rules! IFC {
+    ($e: expr) => {
+        assert_eq!($e, S_OK);
+    }
+}
+
+macro_rules! IFR {
+    ($e: expr) => {
+        let hresult = $e;
+        if (hresult != S_OK) { return hresult }
+    }
+}
+
+macro_rules! __analysis_assume {
+    ($e: expr) => {
+    }
+}
+
+macro_rules! IFCOOM {
+    ($e: expr) => {
+        assert_ne!($e, NULL());
+    }
+}
+
+macro_rules! RRETURN1 {
+    ($e: expr, $s1: expr) => {
+        if $e == $s1 {
+        } else {
+            assert_eq!($e, S_OK);
+        }
+        return $e;
+    }
+}
+
+macro_rules! RRETURN {
+    ($e: expr) => {
+        assert_eq!($e, S_OK);
+        return $e;
+    }
+}
+
diff --git a/third_party/rust/wpf-gpu-raster/src/hwrasterizer.rs b/third_party/rust/wpf-gpu-raster/src/hwrasterizer.rs
new file mode 100644
index 0000000000..49fed1a1bf
--- /dev/null
+++ b/third_party/rust/wpf-gpu-raster/src/hwrasterizer.rs
@@ -0,0 +1,1455 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+// See the LICENSE file in the project root for more information.
+
+#![allow(unused_parens)]
+
+use crate::aacoverage::{CCoverageBuffer, c_rInvShiftSize, c_antiAliasMode, c_nShift, CCoverageInterval, c_nShiftMask, c_nShiftSize, c_nHalfShiftSize};
+use crate::hwvertexbuffer::CHwVertexBufferBuilder;
+use crate::matrix::{CMILMatrix, CMatrix};
+use crate::nullable_ref::Ref;
+use crate::aarasterizer::*;
+use crate::geometry_sink::IGeometrySink;
+use crate::helpers::Int32x32To64;
+use crate::types::*;
+use typed_arena_nomut::Arena;
+
+//-----------------------------------------------------------------------------
+//
+
+//
+//  Description:
+//      Trapezoidal anti-aliasing implementation
+//
+//  >>>> Note that some of this code is duplicated in sw\aarasterizer.cpp,
+//  >>>> so changes to this file may need to propagate.
+//
+//   pursue reduced code duplication
+//
+
+macro_rules! MIL_THR {
+    ($e: expr) => {
+        $e//assert_eq!($e, S_OK);
+    }
+}
+
+
+//
+// Optimize for speed instead of size for these critical methods
+//
+
+
+//-------------------------------------------------------------------------
+//
+// Coordinate system encoding
+//
+// All points/coordinates are named as follows:
+//
+//    <HungarianType><CoordinateSystem>[X|Y][Left|Right|Top|Bottom]VariableName
+//
+//    Common hungarian types:
+//        n - INT
+//        u - UINT
+//        r - FLOAT
+//
+//    Coordinate systems:
+//        Pixel - Device pixel space assuming integer coordinates in the pixel top left corner.
+//        Subpixel - Overscaled space.
+//
+//        To convert between Pixel to Subpixel, we have:
+//            nSubpixelCoordinate = nPixelCoordinate << c_nShift;
+//            nPixelCoordinate = nSubpixelCoordinate >> c_nShift;
+//
+//        Note that the conversion to nPixelCoordinate needs to also track
+//        (nSubpixelCoordinate & c_nShiftMask) to maintain the full value.
+//
+//        Note that since trapezoidal only supports 8x8, c_nShiftSize is always equal to 8.  So,
+//        (1, 2) in pixel space would become (8, 16) in subpixel space.
+//
+//    [X|Y]
+//        Indicates which coordinate is being referred to.
+//
+//    [Left|Right|Top|Bottom]
+//        When referring to trapezoids or rectangular regions, this
+//        component indicates which edge is being referred to.
+//
+//    VariableName
+//       Descriptive portion of the variable name
+//
+//-------------------------------------------------------------------------
+
+
+//-------------------------------------------------------------------------
+//
+//  Function:   IsFractionGreaterThan
+//
+//  Synopsis:
+//     Determine if nNumeratorA/nDenominatorA > nNumeratorB/nDenominatorB
+//
+//     Note that we assume all denominators are strictly greater than zero.
+//
+//-------------------------------------------------------------------------
+fn IsFractionGreaterThan(
+    nNumeratorA: INT,                    // Left hand side numerator
+   /* __in_range(>=, 1) */ nDenominatorA: INT, // Left hand side denominator
+    nNumeratorB: INT,                    // Right hand side numerator
+   /* __in_range(>=, 1) */ nDenominatorB: INT,  // Right hand side denominator
+    ) -> bool
+{
+    //
+    // nNumeratorA/nDenominatorA > nNumeratorB/nDenominatorB
+    // iff nNumeratorA*nDenominatorB/nDenominatorA > nNumeratorB, since nDenominatorB > 0
+    // iff nNumeratorA*nDenominatorB > nNumeratorB*nDenominatorA, since nDenominatorA > 0
+    //
+    // Now, all input parameters are 32-bit integers, so we need to use
+    // a 64-bit result to compute the product.
+    //
+
+    let lNumeratorAxDenominatorB = Int32x32To64(nNumeratorA, nDenominatorB);
+    let lNumeratorBxDenominatorA = Int32x32To64(nNumeratorB, nDenominatorA);
+
+    return (lNumeratorAxDenominatorB > lNumeratorBxDenominatorA);
+}
+
+//-------------------------------------------------------------------------
+//
+//  Function:   IsFractionLessThan
+//
+//  Synopsis:
+//     Determine if nNumeratorA/nDenominatorA < nNumeratorB/nDenominatorB
+//
+//     Note that we assume all denominators are strictly greater than zero.
+//
+//-------------------------------------------------------------------------
+fn
+IsFractionLessThan(
+    nNumeratorA: INT,                    // Left hand side numerator
+    /* __in_range(>=, 1) */ nDenominatorA: INT, // Left hand side denominator
+    nNumeratorB: INT,                    // Right hand side numerator
+    /* __in_range(>=, 1) */ nDenominatorB: INT,  // Right hand side denominator
+) -> bool
+{
+    //
+    // Same check as previous function with less than comparision instead of
+    // a greater than comparison.
+    //
+
+    let lNumeratorAxDenominatorB = Int32x32To64(nNumeratorA, nDenominatorB);
+    let lNumeratorBxDenominatorA = Int32x32To64(nNumeratorB, nDenominatorA);
+
+    return (lNumeratorAxDenominatorB < lNumeratorBxDenominatorA);
+}
+
+
+//-------------------------------------------------------------------------
+//
+//  Function:   AdvanceDDAMultipleSteps
+//
+//  Synopsis:
+//     Advance the DDA by multiple steps
+//
+//-------------------------------------------------------------------------
+fn
+AdvanceDDAMultipleSteps(
+    pEdgeLeft: &CEdge,         // Left edge from active edge list
+    pEdgeRight: &CEdge,        // Right edge from active edge list
+    nSubpixelYAdvance: INT,                    // Number of steps to advance the DDA
+    nSubpixelXLeftBottom: &mut INT,     // Resulting left x position
+    nSubpixelErrorLeftBottom: &mut INT, // Resulting left x position error
+    nSubpixelXRightBottom: &mut INT,    // Resulting right x position
+    nSubpixelErrorRightBottom: &mut INT // Resulting right x position error
+    )
+{
+    //
+    // In this method, we need to be careful of overflow.  Expected input ranges for values are:
+    //
+    //      edge points: x and y subpixel space coordinates are between [-2^26, 2^26]
+    //                   since we start with 28.4 space (and are now in subpixel space,
+    //                   i.e., no 16x scale) and assume 2 bits of working space.
+    //
+    //                   This assumption is ensured by TransformRasterizerPointsTo28_4.
+    //
+    #[cfg(debug_assertions)]
+    {
+    let nDbgPixelCoordinateMax = (1 << 26);
+    let nDbgPixelCoordinateMin = -nDbgPixelCoordinateMax;
+
+    assert!(pEdgeLeft.X.get() >= nDbgPixelCoordinateMin && pEdgeLeft.X.get() <= nDbgPixelCoordinateMax);
+    assert!(pEdgeLeft.EndY >= nDbgPixelCoordinateMin && pEdgeLeft.EndY <= nDbgPixelCoordinateMax);
+    assert!(pEdgeRight.X.get() >= nDbgPixelCoordinateMin && pEdgeRight.X.get() <= nDbgPixelCoordinateMax);
+    assert!(pEdgeRight.EndY >= nDbgPixelCoordinateMin && pEdgeRight.EndY <= nDbgPixelCoordinateMax);
+
+    //
+    //        errorDown: (0, 2^30)
+    //                   Since errorDown is the edge delta y in 28.4 space (not subpixel space
+    //                   like the end points), we have a larger range of (0, 2^32) for the positive
+    //                   error down.  With 2 bits of work space (which TransformRasterizerPointsTo28_4
+    //                   ensures), we know we are between (0, 2^30)
+    //
+
+    let nDbgErrorDownMax: INT = (1 << 30);
+    assert!(pEdgeLeft.ErrorDown  > 0 && pEdgeLeft.ErrorDown  < nDbgErrorDownMax);
+    assert!(pEdgeRight.ErrorDown > 0 && pEdgeRight.ErrorDown < nDbgErrorDownMax);
+
+    //
+    //          errorUp: [0, errorDown)
+    //
+    assert!(pEdgeLeft.ErrorUp  >= 0 && pEdgeLeft.ErrorUp  < pEdgeLeft.ErrorDown);
+    assert!(pEdgeRight.ErrorUp >= 0 && pEdgeRight.ErrorUp < pEdgeRight.ErrorDown);
+    }
+
+    //
+    // Advance the left edge
+    //
+
+    // Since each point on the edge is withing 28.4 space, the following computation can't overflow.
+    *nSubpixelXLeftBottom = pEdgeLeft.X.get() + nSubpixelYAdvance*pEdgeLeft.Dx;
+
+    // Since the error values can be close to 2^30, we can get an overflow by multiplying with yAdvance.
+    // So, we need to use a 64-bit temporary in this case.
+    let mut llSubpixelErrorBottom: LONGLONG = pEdgeLeft.Error.get() as LONGLONG + Int32x32To64(nSubpixelYAdvance, pEdgeLeft.ErrorUp);
+    if (llSubpixelErrorBottom >= 0)
+    {
+        let llSubpixelXLeftDelta = llSubpixelErrorBottom / (pEdgeLeft.ErrorDown as LONGLONG);
+
+        // The delta should remain in range since it still represents a delta along the edge which
+        // we know fits entirely in 28.4.  Note that we add one here since the error must end up
+        // less than 0.
+        assert!(llSubpixelXLeftDelta < INT::MAX as LONGLONG);
+        let nSubpixelXLeftDelta: INT = (llSubpixelXLeftDelta as INT) + 1;
+
+        *nSubpixelXLeftBottom += nSubpixelXLeftDelta;
+        llSubpixelErrorBottom -= Int32x32To64(pEdgeLeft.ErrorDown, nSubpixelXLeftDelta);
+    }
+
+    // At this point, the subtraction above should have generated an error that is within
+    // (-pLeft->ErrorDown, 0)
+
+    assert!((llSubpixelErrorBottom > -pEdgeLeft.ErrorDown as LONGLONG) && (llSubpixelErrorBottom < 0));
+    *nSubpixelErrorLeftBottom = (llSubpixelErrorBottom as INT);
+
+    //
+    // Advance the right edge
+    //
+
+    // Since each point on the edge is withing 28.4 space, the following computation can't overflow.
+    *nSubpixelXRightBottom = pEdgeRight.X.get() + nSubpixelYAdvance*pEdgeRight.Dx;
+
+    // Since the error values can be close to 2^30, we can get an overflow by multiplying with yAdvance.
+    // So, we need to use a 64-bit temporary in this case.
+    llSubpixelErrorBottom = pEdgeRight.Error.get() as LONGLONG + Int32x32To64(nSubpixelYAdvance, pEdgeRight.ErrorUp);
+    if (llSubpixelErrorBottom >= 0)
+    {
+        let llSubpixelXRightDelta: LONGLONG = llSubpixelErrorBottom / (pEdgeRight.ErrorDown as LONGLONG);
+
+        // The delta should remain in range since it still represents a delta along the edge which
+        // we know fits entirely in 28.4.  Note that we add one here since the error must end up
+        // less than 0.
+        assert!(llSubpixelXRightDelta < INT::MAX as LONGLONG);
+        let nSubpixelXRightDelta: INT = (llSubpixelXRightDelta as INT) + 1;
+
+        *nSubpixelXRightBottom += nSubpixelXRightDelta;
+        llSubpixelErrorBottom -= Int32x32To64(pEdgeRight.ErrorDown, nSubpixelXRightDelta);
+    }
+
+    // At this point, the subtraction above should have generated an error that is within
+    // (-pRight->ErrorDown, 0)
+
+    assert!((llSubpixelErrorBottom > -pEdgeRight.ErrorDown as LONGLONG) && (llSubpixelErrorBottom < 0));
+    *nSubpixelErrorRightBottom = (llSubpixelErrorBottom as INT);
+}
+
+//-------------------------------------------------------------------------
+//
+//  Function:   ComputeDeltaUpperBound
+//
+//  Synopsis:
+//     Compute some value that is >= nSubpixelAdvanceY*|1/m| where m is the
+//     slope defined by the edge below.
+//
+//-------------------------------------------------------------------------
+fn
+ComputeDeltaUpperBound(
+    pEdge: &CEdge,  // Edge containing 1/m value used for computation
+    nSubpixelYAdvance: INT          // Multiplier in synopsis expression
+    ) -> INT
+{
+    let nSubpixelDeltaUpperBound: INT;
+
+    //
+    // Compute the delta bound
+    //
+
+    if (pEdge.ErrorUp == 0)
+    {
+        //
+        // No errorUp, so simply compute bound based on dx value
+        //
+
+        nSubpixelDeltaUpperBound = nSubpixelYAdvance*(pEdge.Dx).abs();
+    }
+    else
+    {
+        let nAbsDx: INT;
+        let nAbsErrorUp: INT;
+
+        //
+        // Compute abs of (dx, error)
+        //
+        // Here, we can assume errorUp > 0
+        //
+
+        assert!(pEdge.ErrorUp > 0);
+
+        if (pEdge.Dx >= 0)
+        {
+            nAbsDx = pEdge.Dx;
+            nAbsErrorUp = pEdge.ErrorUp;
+        }
+        else
+        {
+            //
+            // Dx < 0, so negate (dx, errorUp)
+            //
+            // Note that since errorUp > 0, we know -errorUp < 0 and that
+            // we need to add errorDown to get an errorUp >= 0 which
+            // also means substracting one from dx.
+            //
+
+            nAbsDx = -pEdge.Dx - 1;
+            nAbsErrorUp = -pEdge.ErrorUp + pEdge.ErrorDown;
+        }
+
+        //
+        // Compute the bound of nSubpixelAdvanceY*|1/m|
+        //
+        // Note that the +1 below is included to bound any left over errorUp that we are dropping here.
+        //
+
+        nSubpixelDeltaUpperBound = nSubpixelYAdvance*nAbsDx + (nSubpixelYAdvance*nAbsErrorUp)/pEdge.ErrorDown + 1;
+    }
+
+    return nSubpixelDeltaUpperBound;
+}
+
+//-------------------------------------------------------------------------
+//
+//  Function:   ComputeDistanceLowerBound
+//
+//  Synopsis:
+//     Compute some value that is <= distance between
+//     (pEdgeLeft->X, pEdgeLeft->Error) and (pEdgeRight->X, pEdgeRight->Error)
+//
+//-------------------------------------------------------------------------
+fn
+ComputeDistanceLowerBound(
+    pEdgeLeft: &CEdge, // Left edge containing the position for the distance computation
+    pEdgeRight: &CEdge // Right edge containing the position for the distance computation
+    ) -> INT
+{
+    //
+    // Note: In these comments, error1 and error2 are theoretical. The actual Error members
+    // are biased by -1.
+    //
+    // distance = (x2 + error2/errorDown2) - (x1 + error1/errorDown1)
+    //          = x2 - x1 + error2/errorDown2 - error1/errorDown1
+    //          >= x2 - x1 + error2/errorDown2   , since error1 < 0
+    //          >= x2 - x1 - 1                   , since error2 < 0
+    //          = pEdgeRight->X - pEdgeLeft->X - 1
+    //
+    // In the special case where error2/errorDown2 >= error1/errorDown1, we
+    // can get a tigher bound of:
+    //
+    //          pEdgeRight->X - pEdgeLeft->X
+    //
+    // This case occurs often in thin strokes, so we check for it here.
+    //
+
+    assert!(pEdgeLeft.Error.get()  < 0);
+    assert!(pEdgeRight.Error.get() < 0);
+    assert!(pEdgeLeft.X <= pEdgeRight.X);
+
+    let mut nSubpixelXDistanceLowerBound: INT = pEdgeRight.X.get() - pEdgeLeft.X.get();
+
+    //
+    // If error2/errorDown2 < error1/errorDown1, we need to subtract one from the bound.
+    // Note that error's are actually baised by -1, we so we have to add one before
+    // we do the comparison.
+    //
+
+    if (IsFractionLessThan(
+             pEdgeRight.Error.get()+1,
+             pEdgeRight.ErrorDown,
+             pEdgeLeft.Error.get()+1,
+             pEdgeLeft.ErrorDown
+        ))
+    {
+            // We can't use the tighter lower bound described above, so we need to subtract one to
+            // ensure we have a lower bound.
+
+            nSubpixelXDistanceLowerBound -= 1;
+    }
+
+    return nSubpixelXDistanceLowerBound;
+}
+pub struct CHwRasterizer<'x, 'y, 'z> {
+    m_rcClipBounds: MilPointAndSizeL,
+    m_matWorldToDevice: CMILMatrix,
+    m_pIGeometrySink: &'x mut CHwVertexBufferBuilder<'y, 'z>,
+    m_fillMode: MilFillMode,
+    /* 
+DynArray<MilPoint2F> *m_prgPoints;
+DynArray<BYTE>       *m_prgTypes;
+MilPointAndSizeL      m_rcClipBounds;
+CMILMatrix            m_matWorldToDevice;
+IGeometrySink        *m_pIGeometrySink;
+MilFillMode::Enum     m_fillMode;
+
+//
+// Complex scan coverage buffer
+//
+
+CCoverageBuffer m_coverageBuffer;
+
+CD3DDeviceLevel1 * m_pDeviceNoRef;*/
+    //m_coverageBuffer: CCoverageBuffer,
+}
+
+//-------------------------------------------------------------------------
+//
+//  Function:   CHwRasterizer::ConvertSubpixelXToPixel
+//
+//  Synopsis:
+//      Convert from our subpixel coordinate (x + error/errorDown)
+//      to a floating point value.
+//
+//-------------------------------------------------------------------------
+fn ConvertSubpixelXToPixel(
+    x: INT,
+    error: INT,
+    rErrorDown: f32
+    ) -> f32
+{
+    assert!(rErrorDown > f32::EPSILON);
+    return ((x as f32) + (error as f32)/rErrorDown)*c_rInvShiftSize;
+}
+
+//-------------------------------------------------------------------------
+//
+//  Function:   CHwRasterizer::ConvertSubpixelYToPixel
+//
+//  Synopsis:
+//      Convert from our subpixel space to pixel space assuming no
+//      error.
+//
+//-------------------------------------------------------------------------
+fn ConvertSubpixelYToPixel(
+    nSubpixel: i32
+    ) -> f32
+{
+    return (nSubpixel as f32)*c_rInvShiftSize;
+}
+
+impl<'x, 'y, 'z> CHwRasterizer<'x, 'y, 'z> {
+//-------------------------------------------------------------------------
+//
+//  Function:   CHwRasterizer::RasterizePath
+//
+//  Synopsis:
+//      Internal rasterizer fill path.  Note that this method follows the
+//      same basic structure as the software rasterizer in aarasterizer.cpp.
+//
+//      The general algorithm used for rasterization is a vertical sweep of
+//      the shape that maintains an active edge list.  The sweep is done
+//      at a sub-scanline resolution and results in either:
+//          1. Sub-scanlines being combined in the coverage buffer and output
+//             as "complex scans".
+//          2. Simple trapezoids being recognized in the active edge list
+//             and output using a faster simple trapezoid path.
+//
+//      This method consists of the setup to the main rasterization loop
+//      which includes:
+//
+//          1. Setup of the clip rectangle
+//          2. Calling FixedPointPathEnumerate to populate our inactive
+//             edge list.
+//          3. Delegating to RasterizePath to execute the main loop.
+//
+//-------------------------------------------------------------------------
+pub fn RasterizePath(
+    &mut self,
+    rgpt: &[POINT],
+    rgTypes: &[BYTE],
+    cPoints: UINT,
+    pmatWorldTransform: &CMILMatrix
+    ) -> HRESULT
+{
+    let mut hr;
+    // Default is not implemented for arrays of size 40 so we need to use map
+    let mut inactiveArrayStack: [CInactiveEdge; INACTIVE_LIST_NUMBER!()] = [(); INACTIVE_LIST_NUMBER!()].map(|_| Default::default());
+    let mut pInactiveArray: &mut [CInactiveEdge];
+    let mut pInactiveArrayAllocation: Vec<CInactiveEdge>;
+    let mut edgeHead: CEdge = Default::default();
+    let mut edgeTail: CEdge = Default::default();
+    let pEdgeActiveList: Ref<CEdge>;
+    let mut edgeStore = Arena::new();
+    //edgeStore.init();
+    let mut edgeContext: CInitializeEdgesContext = CInitializeEdgesContext::new(&mut edgeStore);
+
+    edgeContext.ClipRect = None;
+
+    edgeTail.X.set(i32::MAX);       // Terminator to active list
+    edgeTail.StartY = i32::MAX;  // Terminator to inactive list
+
+    edgeTail.EndY = i32::MIN;
+    edgeHead.X.set(i32::MIN);       // Beginning of active list
+    edgeContext.MaxY = i32::MIN;
+
+    edgeHead.Next.set(Ref::new(&edgeTail));
+    pEdgeActiveList = Ref::new(&mut edgeHead);
+    //edgeContext.Store = &mut edgeStore;
+
+    edgeContext.AntiAliasMode = c_antiAliasMode;
+    assert!(edgeContext.AntiAliasMode != MilAntiAliasMode::None);
+
+    // If the path contains 0 or 1 points, we can ignore it.
+    if (cPoints < 2)
+    {
+        return S_OK;
+    }
+
+    let nPixelYClipBottom: INT = self.m_rcClipBounds.Y + self.m_rcClipBounds.Height;
+
+    // Scale the clip bounds rectangle by 16 to account for our
+    // scaling to 28.4 coordinates:
+
+    let mut clipBounds : RECT = Default::default();
+    clipBounds.left   = self.m_rcClipBounds.X * FIX4_ONE!();
+    clipBounds.top    = self.m_rcClipBounds.Y * FIX4_ONE!();
+    clipBounds.right  = (self.m_rcClipBounds.X + self.m_rcClipBounds.Width) * FIX4_ONE!();
+    clipBounds.bottom = (self.m_rcClipBounds.Y + self.m_rcClipBounds.Height) * FIX4_ONE!();
+
+    edgeContext.ClipRect = Some(&clipBounds);
+
+    //////////////////////////////////////////////////////////////////////////
+    // Convert all our points to 28.4 fixed point:
+
+    let mut matrix: CMILMatrix = (*pmatWorldTransform).clone();
+    AppendScaleToMatrix(&mut matrix, TOREAL!(16), TOREAL!(16));
+
+    let coverageBuffer: CCoverageBuffer = Default::default();
+    // Initialize the coverage buffer
+    coverageBuffer.Initialize();
+
+    // Enumerate the path and construct the edge table:
+
+    hr = MIL_THR!(FixedPointPathEnumerate(
+        rgpt,
+        rgTypes,
+        cPoints,
+        &matrix,
+        edgeContext.ClipRect,
+        &mut edgeContext
+        ));
+
+    if (FAILED(hr))
+    {
+        if (hr == WGXERR_VALUEOVERFLOW)
+        {
+            // Draw nothing on value overflow and return
+            hr = S_OK;
+        }
+        return hr;
+    }
+
+    let nTotalCount: UINT; nTotalCount = edgeContext.Store.len() as u32;
+    if (nTotalCount == 0)
+    {
+        hr = S_OK;     // We're outta here (empty path or entirely clipped)
+        return hr;
+    }
+
+    // At this point, there has to be at least two edges.  If there's only
+    // one, it means that we didn't do the trivially rejection properly.
+
+    assert!((nTotalCount >= 2) && (nTotalCount <= (UINT::MAX - 2)));
+
+    pInactiveArray = &mut inactiveArrayStack[..];
+    if (nTotalCount > (INACTIVE_LIST_NUMBER!() as u32 - 2))
+    {
+        pInactiveArrayAllocation = vec![Default::default(); nTotalCount as usize + 2];
+
+        pInactiveArray = &mut pInactiveArrayAllocation;
+    }
+
+    // Initialize and sort the inactive array:
+
+    let nSubpixelYCurrent = InitializeInactiveArray(
+        edgeContext.Store,
+        pInactiveArray,
+        nTotalCount,
+        Ref::new(&edgeTail)
+        );
+
+    let mut nSubpixelYBottom = edgeContext.MaxY;
+
+    assert!(nSubpixelYBottom > 0);
+
+    // Skip the head sentinel on the inactive array:
+
+    pInactiveArray = &mut pInactiveArray[1..];
+
+    //
+    // Rasterize the path
+    //
+
+    // 'nPixelYClipBottom' is in screen space and needs to be converted to the
+    // format we use for antialiasing.
+
+    nSubpixelYBottom = nSubpixelYBottom.min(nPixelYClipBottom << c_nShift);
+
+    // 'nTotalCount' should have been zero if all the edges were
+    // clipped out (RasterizeEdges assumes there's at least one edge
+    // to be drawn):
+
+    assert!(nSubpixelYBottom > nSubpixelYCurrent);
+
+    IFC!(self.RasterizeEdges(
+        pEdgeActiveList,
+        pInactiveArray,
+        &coverageBuffer,
+        nSubpixelYCurrent,
+        nSubpixelYBottom
+        ));
+
+    return hr;
+}
+
+//-------------------------------------------------------------------------
+//
+//  Function:   CHwRasterizer::new
+//
+//  Synopsis:
+//      1. Ensure clean state
+//      2. Convert path to internal format
+//
+//-------------------------------------------------------------------------
+pub fn new(
+    pIGeometrySink: &'x mut CHwVertexBufferBuilder<'y, 'z>,
+    fillMode: MilFillMode,
+    pmatWorldToDevice: Option<CMatrix<CoordinateSpace::Shape,CoordinateSpace::Device>>,
+    clipRect: MilPointAndSizeL,
+    ) -> Self
+{
+    //
+    // PS#856364-2003/07/01-ashrafm  Remove pixel center fixup
+    //
+    // Incoming coordinate space uses integers at upper-left of pixel (pixel
+    // center are half integers) at device level.
+    //
+    // Rasterizer uses the coordinate space with integers at pixel center.
+    //
+    // To convert from center (1/2, 1/2) to center (0, 0) we need to subtract
+    // 1/2 from each coordinate in device space.
+    //
+    // See InitializeEdges in aarasterizer.ccp to see how we unconvert for
+    // antialiased rendering.
+    //
+
+    let mut matWorldHPCToDeviceIPC = pmatWorldToDevice.unwrap_or(CMatrix::Identity());
+    matWorldHPCToDeviceIPC.SetDx(matWorldHPCToDeviceIPC.GetDx() - 0.5);
+    matWorldHPCToDeviceIPC.SetDy(matWorldHPCToDeviceIPC.GetDy() - 0.5);
+
+    //
+    // Set local state.
+    //
+
+    //  There's an opportunity for early clipping here
+    //
+    // However, since the rasterizer itself does a reasonable job of clipping some
+    // cases, we don't early clip yet.
+
+    Self {
+        m_fillMode: fillMode,
+        m_rcClipBounds: clipRect,
+        m_pIGeometrySink: pIGeometrySink,
+        m_matWorldToDevice: matWorldHPCToDeviceIPC,
+    }
+}
+
+//-------------------------------------------------------------------------
+//
+//  Function:   CHwRasterizer::SendGeometry
+//
+//  Synopsis:
+//     Tessellate and send geometry to the pipeline
+//
+//-------------------------------------------------------------------------
+pub fn SendGeometry(&mut self,
+    points: &[POINT],
+    types: &[BYTE],
+    ) -> HRESULT
+{
+    let mut hr = S_OK;
+
+    //
+    // Rasterize the path
+    //
+    let count = points.len() as u32;
+    IFR!(self.RasterizePath(
+        points,
+        types,
+        count,
+        &self.m_matWorldToDevice.clone(),
+        ));
+        /* 
+    IFC!(self.RasterizePath(
+        self.m_prgPoints.as_ref().unwrap().GetDataBuffer(),
+        self.m_prgTypes.as_ref().unwrap().GetDataBuffer(),
+        self.m_prgPoints.as_ref().unwrap().GetCount() as u32,
+        &self.m_matWorldToDevice,
+        self.m_fillMode
+        ));*/
+
+    //
+    // It's possible that we output no triangles.  For example, if we tried to fill a
+    // line instead of stroke it.  Since we have no efficient way to detect all these cases
+    // up front, we simply rasterize and see if we generated anything.
+    //
+
+    if (self.m_pIGeometrySink.IsEmpty())
+    {
+        hr = WGXHR_EMPTYFILL;
+    }
+
+    RRETURN1!(hr, WGXHR_EMPTYFILL);
+}
+/*
+//-------------------------------------------------------------------------
+//
+//  Function:   CHwRasterizer::SendGeometryModifiers
+//
+//  Synopsis:   Send an AA color source to the pipeline.
+//
+//-------------------------------------------------------------------------
+fn SendGeometryModifiers(&self,
+    pPipelineBuilder: &mut CHwPipelineBuilder
+    ) -> HRESULT
+{
+    let hr = S_OK;
+
+    let pAntiAliasColorSource = None;
+
+    self.m_pDeviceNoRef.GetColorComponentSource(
+        CHwColorComponentSource::Diffuse,
+        &pAntiAliasColorSource
+        );
+
+    IFC!(pPipelineBuilder.Set_AAColorSource(
+        pAntiAliasColorSource
+        ));
+
+    return hr;
+}*/
+
+//-------------------------------------------------------------------------
+//
+//  Function:   CHwRasterizer::GenerateOutputAndClearCoverage
+//
+//  Synopsis:
+//      Collapse output and generate span data
+//
+//-------------------------------------------------------------------------
+fn
+GenerateOutputAndClearCoverage<'a>(&mut self, coverageBuffer: &'a CCoverageBuffer<'a>,
+    nSubpixelY: INT
+    ) -> HRESULT
+{
+    let hr = S_OK;
+    let nPixelY = nSubpixelY >> c_nShift;
+
+    let pIntervalSpanStart: Ref<CCoverageInterval> = coverageBuffer.m_pIntervalStart.get();
+
+    IFC!(self.m_pIGeometrySink.AddComplexScan(nPixelY, pIntervalSpanStart));
+
+    coverageBuffer.Reset();
+
+    return hr;
+}
+
+//-------------------------------------------------------------------------
+//
+//  Function:   CHwRasterizer::ComputeTrapezoidsEndScan
+//
+//  Synopsis:
+//      This methods takes the current active edge list (and ycurrent)
+//      and will determine:
+//
+//      1. Can we output some list of simple trapezoids for this active
+//         edge list?  If the answer is no, then we simply return
+//         nSubpixelYCurrent indicating this condition.
+//
+//      2. If we can output some set of trapezoids, then what is the
+//         next ycurrent, i.e., how tall are our trapezoids.
+//
+//     Note that all trapezoids output for a particular active edge list
+//     are all the same height.
+//
+//     To further understand the conditions for making this decision, it
+//     is important to consider the simple trapezoid tessellation:
+//
+//           ___+_________________+___
+//          /  +  /             \  +  \        '+' marks active edges
+//         /  +  /               \  +  \
+//        /  +  /                 \  +  \
+//       /__+__/___________________\__+__\
+//       1+1/m                         +
+//
+//      Note that 1+1/edge_slope is the required expand distance to ensure
+//      that we cover all pixels required.
+//
+//      Now, we can fail to output any trapezoids under the following conditions:
+//         1. The expand regions along the top edge of the trapezoid overlap.
+//         2. The expand regions along the bottom edge of the trapezoid overlap
+//            within the current scanline.  Note that if the bottom edges overlap
+//            at some later point, we can shorten our trapezoid to remove the
+//            overlapping.
+//
+//      The key to the algorithm at this point is to detect the above condition
+//      in our active edge list and either update the returned end y position
+//      or reject all together based on overlapping.
+//
+//-------------------------------------------------------------------------
+
+fn ComputeTrapezoidsEndScan(&mut self,
+    pEdgeCurrent: Ref<CEdge>,
+    nSubpixelYCurrent: INT,
+    nSubpixelYNextInactive: INT
+    ) -> INT
+{
+
+    let mut nSubpixelYBottomTrapezoids;
+    let mut pEdgeLeft: Ref<CEdge>;
+    let mut pEdgeRight: Ref<CEdge>;
+
+    //
+    // Trapezoids should always start at scanline boundaries
+    //
+
+    assert!((nSubpixelYCurrent & c_nShiftMask) == 0);
+
+    //
+    // If we are doing a winding mode fill, check that we can ignore mode and do an
+    // alternating fill in OutputTrapezoids.  This condition occurs when winding is
+    // equivalent to alternating which happens if the pairwise edges have different
+    // winding directions.
+    //
+
+    if (self.m_fillMode == MilFillMode::Winding)
+    {
+        let mut pEdge = pEdgeCurrent;
+        while pEdge.EndY != INT::MIN {
+            // The active edge list always has an even number of edges which we actually
+            // assert in ASSERTACTIVELIST.
+
+            assert!(pEdge.Next.get().EndY != INT::MIN);
+
+            // If not alternating winding direction, we can't fill with alternate mode
+
+            if (pEdge.WindingDirection == pEdge.Next.get().WindingDirection)
+            {
+                // Give up until we handle winding mode
+                nSubpixelYBottomTrapezoids = nSubpixelYCurrent;
+                return nSubpixelYBottomTrapezoids;
+            }
+
+            pEdge = pEdge.Next.get().Next.get();
+        }
+    }
+
+    //
+    // For each edge, we:
+    //
+    //    1. Set the new trapezoid bottom to the min of the current
+    //       one and the edge EndY
+    //
+    //    2. Check if edges will intersect during trapezoidal shrink/expand
+    //
+
+    nSubpixelYBottomTrapezoids = nSubpixelYNextInactive;
+
+    let mut pEdge = pEdgeCurrent;
+    while pEdge.EndY != INT::MIN {
+        //
+        // Step 1
+        //
+        // Updated nSubpixelYBottomTrapezoids based on edge EndY.
+        //
+        // Since edges are clipped to the current clip rect y bounds, we also know
+        // that pEdge->EndY <= nSubpixelYBottom so there is no need to check for that here.
+        //
+
+        nSubpixelYBottomTrapezoids = nSubpixelYBottomTrapezoids.min(pEdge.EndY);
+
+        //
+        // Step 2
+        //
+        // Check that edges will not overlap during trapezoid shrink/expand.
+        //
+
+        pEdgeLeft = pEdge;
+        pEdgeRight = pEdge.Next.get();
+
+        if (pEdgeRight.EndY != INT::MIN)
+        {
+            //
+            //        __A__A'___________________B'_B__
+            //        \  +  \                  /  +  /       '+' marks active edges
+            //         \  +  \                /  +  /
+            //          \  +  \              /  +  /
+            //           \__+__\____________/__+__/
+            //       1+1/m   C  C'         D' D
+            //
+            // We need to determine if position A' <= position B' and that position C' <= position D'
+            // in the above diagram.  So, we need to ensure that both the distance between
+            // A and B and the distance between C and D is greater than or equal to:
+            //
+            //    0.5 + |0.5/m1| + 0.5 + |0.5/m2|               (pixel space)
+            //  = shiftsize + halfshiftsize*(|1/m1| + |1/m2|)   (subpixel space)
+            //
+            // So, we'll start by computing this distance.  Note that we can compute a distance
+            // that is too large here since the self-intersection detection is simply used to
+            // recognize trapezoid opportunities and isn't required for visual correctness.
+            //
+
+            let nSubpixelExpandDistanceUpperBound: INT =
+                c_nShiftSize
+                + ComputeDeltaUpperBound(&*pEdgeLeft, c_nHalfShiftSize)
+                + ComputeDeltaUpperBound(&*pEdgeRight, c_nHalfShiftSize);
+
+            //
+            // Compute a top edge distance that is <= to the distance between A' and B' as follows:
+            //   lowerbound(distance(A, B)) - nSubpixelExpandDistanceUpperBound
+            //
+
+            let nSubpixelXTopDistanceLowerBound: INT =
+                ComputeDistanceLowerBound(&*pEdgeLeft, &*pEdgeRight) - nSubpixelExpandDistanceUpperBound;
+
+            //
+            // Check if the top edges cross
+            //
+
+            if (nSubpixelXTopDistanceLowerBound < 0)
+            {
+                // The top edges have crossed, so we are out of luck.  We can't
+                // start a trapezoid on this scanline
+
+                nSubpixelYBottomTrapezoids = nSubpixelYCurrent;
+                return nSubpixelYBottomTrapezoids;
+            }
+
+            //
+            // If the edges are converging, we need to check if they cross at
+            // nSubpixelYBottomTrapezoids
+            //
+            //
+            //  1) \       /    2) \    \       3)   /   /
+            //      \     /          \   \          /  /
+            //       \   /             \  \        / /
+            //
+            // The edges converge iff (dx1 > dx2 || (dx1 == dx2 && errorUp1/errorDown1 > errorUp2/errorDown2).
+            //
+            // Note that in the case where the edges do not converge, the code below will end up computing
+            // the DDA at the end points and checking for intersection again.  This code doesn't rely on
+            // the fact that the edges don't converge, so we can be too conservative here.
+            //
+
+            if (pEdgeLeft.Dx > pEdgeRight.Dx
+                || ((pEdgeLeft.Dx == pEdgeRight.Dx)
+                    && IsFractionGreaterThan(pEdgeLeft.ErrorUp, pEdgeLeft.ErrorDown, pEdgeRight.ErrorUp, pEdgeRight.ErrorDown)))
+            {
+
+                let nSubpixelYAdvance: INT =  nSubpixelYBottomTrapezoids - nSubpixelYCurrent;
+                assert!(nSubpixelYAdvance > 0);
+
+                //
+                // Compute the edge position at nSubpixelYBottomTrapezoids
+                //
+
+                let mut nSubpixelXLeftAdjustedBottom = 0;
+                let mut nSubpixelErrorLeftBottom = 0;
+                let mut nSubpixelXRightBottom = 0;
+                let mut nSubpixelErrorRightBottom = 0;
+
+                AdvanceDDAMultipleSteps(
+                    &*pEdgeLeft,
+                    &*pEdgeRight,
+                    nSubpixelYAdvance,
+                    &mut nSubpixelXLeftAdjustedBottom,
+                    &mut nSubpixelErrorLeftBottom,
+                    &mut nSubpixelXRightBottom,
+                    &mut nSubpixelErrorRightBottom
+                    );
+
+                //
+                // Adjust the bottom left position by the expand distance for all the math
+                // that follows.  Note that since we adjusted the top distance by that
+                // same expand distance, this adjustment is equivalent to moving the edges
+                // nSubpixelExpandDistanceUpperBound closer together.
+                //
+
+                nSubpixelXLeftAdjustedBottom += nSubpixelExpandDistanceUpperBound;
+
+                //
+                // Check if the bottom edge crosses.
+                //
+                // To avoid checking error1/errDown1 and error2/errDown2, we assume the
+                // edges cross if nSubpixelXLeftAdjustedBottom == nSubpixelXRightBottom
+                // and thus produce a result that is too conservative.
+                //
+
+                if (nSubpixelXLeftAdjustedBottom >= nSubpixelXRightBottom)
+                {
+
+                    //
+                    // At this point, we have the following scenario
+                    //
+                    //            ____d1____
+                    //            \        /   |   |
+                    //              \    /     h1  |
+                    //                \/       |   | nSubpixelYAdvance
+                    //               /  \          |
+                    //             /__d2__\        |
+                    //
+                    // We want to compute h1.  We know that:
+                    //
+                    //     h1 / nSubpixelYAdvance = d1 / (d1 + d2)
+                    //     h1 = nSubpixelYAdvance * d1 / (d1 + d2)
+                    //
+                    // Now, if we approximate d1 with some d1' <= d1, we get
+                    //
+                    //     h1 = nSubpixelYAdvance * d1 / (d1 + d2)
+                    //     h1 >= nSubpixelYAdvance * d1' / (d1' + d2)
+                    //
+                    // Similarly, if we approximate d2 with some d2' >= d2, we get
+                    //
+                    //     h1 >= nSubpixelYAdvance * d1' / (d1' + d2)
+                    //        >= nSubpixelYAdvance * d1' / (d1' + d2')
+                    //
+                    // Since we are allowed to be too conservative with h1 (it can be
+                    // less than the actual value), we'll construct such approximations
+                    // for simplicity.
+                    //
+                    // Note that d1' = nSubpixelXTopDistanceLowerBound which we have already
+                    // computed.
+                    //
+                    //      d2 = (x1 + error1/errorDown1) - (x2 + error2/errorDown2)
+                    //         = x1 - x2 + error1/errorDown1 - error2/errorDown2
+                    //         <= x1 - x2 - error2/errorDown2   , since error1 < 0
+                    //         <= x1 - x2 + 1                   , since error2 < 0
+                    //         = nSubpixelXLeftAdjustedBottom - nSubpixelXRightBottom + 1
+                    //
+
+                    let nSubpixelXBottomDistanceUpperBound: INT = nSubpixelXLeftAdjustedBottom - nSubpixelXRightBottom + 1;
+
+                    assert!(nSubpixelXTopDistanceLowerBound >= 0);
+                    assert!(nSubpixelXBottomDistanceUpperBound > 0);
+
+                    #[cfg(debug_assertions)]
+                    let nDbgPreviousSubpixelXBottomTrapezoids: INT = nSubpixelYBottomTrapezoids;
+
+
+                    nSubpixelYBottomTrapezoids =
+                        nSubpixelYCurrent +
+                        (nSubpixelYAdvance * nSubpixelXTopDistanceLowerBound) /
+                        (nSubpixelXTopDistanceLowerBound + nSubpixelXBottomDistanceUpperBound);
+
+                    #[cfg(debug_assertions)]
+                    assert!(nDbgPreviousSubpixelXBottomTrapezoids >= nSubpixelYBottomTrapezoids);
+
+                    if (nSubpixelYBottomTrapezoids < nSubpixelYCurrent + c_nShiftSize)
+                    {
+                        // We no longer have a trapezoid that is at least one scanline high, so
+                        // abort
+
+                        nSubpixelYBottomTrapezoids = nSubpixelYCurrent;
+                        return nSubpixelYBottomTrapezoids;
+                    }
+                }
+            }
+        }
+
+        pEdge = pEdge.Next.get();
+    }
+
+    //
+    // Snap to pixel boundary
+    //
+
+    nSubpixelYBottomTrapezoids = nSubpixelYBottomTrapezoids & (!c_nShiftMask);
+
+    //
+    // Ensure that we are never less than nSubpixelYCurrent
+    //
+
+    assert!(nSubpixelYBottomTrapezoids >= nSubpixelYCurrent);
+
+    //
+    // Return trapezoid end scan
+    //
+
+//Cleanup:
+    return nSubpixelYBottomTrapezoids;
+}
+
+
+//-------------------------------------------------------------------------
+//
+//  Function:   CHwRasterizer::OutputTrapezoids
+//
+//  Synopsis:
+//      Given the current active edge list, output a list of
+//      trapezoids.
+//
+//      _________________________
+//     /     /             \     \
+//    /     /               \     \
+//   /     /                 \     \
+//  /_____/___________________\_____\
+//  1+1/m
+//
+// We output a trapezoid where the distance in X is 1+1/m slope on either edge.
+// Note that we actually do a linear interpolation for coverage along the
+// entire falloff region which comes within 12.5% error when compared to our
+// 8x8 coverage output for complex scans.  What is happening here is
+// that we are applying a linear approximation to the coverage function
+// based on slope.  It is possible to get better linear interpolations
+// by varying the expanded region, but it hasn't been necessary to apply
+// these quality improvements yet.
+//
+//-------------------------------------------------------------------------
+fn 
+OutputTrapezoids(&mut self,
+    pEdgeCurrent: Ref<CEdge>,
+    nSubpixelYCurrent: INT, // inclusive
+    nSubpixelYNext: INT     // exclusive
+    ) -> HRESULT
+{
+
+    let hr = S_OK;
+    let nSubpixelYAdvance: INT;
+    let mut rSubpixelLeftErrorDown: f32;
+    let mut rSubpixelRightErrorDown: f32;
+    let mut rPixelXLeft: f32;
+    let mut rPixelXRight: f32;
+    let mut rSubpixelLeftInvSlope: f32;
+    let mut rSubpixelLeftAbsInvSlope: f32;
+    let mut rSubpixelRightInvSlope: f32;
+    let mut rSubpixelRightAbsInvSlope: f32;
+    let mut rPixelXLeftDelta: f32;
+    let mut rPixelXRightDelta: f32;
+
+    let mut pEdgeLeft = pEdgeCurrent;
+    let mut pEdgeRight = (*pEdgeCurrent).Next.get();
+
+    assert!((nSubpixelYCurrent & c_nShiftMask) == 0);
+    assert!(pEdgeLeft.EndY != INT::MIN);
+    assert!(pEdgeRight.EndY != INT::MIN);
+
+    //
+    // Compute the height our trapezoids
+    //
+
+    nSubpixelYAdvance = nSubpixelYNext - nSubpixelYCurrent;
+
+    //
+    // Output each trapezoid
+    //
+
+    loop
+    {
+        //
+        // Compute x/error for end of trapezoid
+        //
+
+        let mut nSubpixelXLeftBottom: INT = 0;
+        let mut nSubpixelErrorLeftBottom: INT = 0;
+        let mut nSubpixelXRightBottom: INT = 0;
+        let mut nSubpixelErrorRightBottom: INT = 0;
+
+        AdvanceDDAMultipleSteps(
+            &*pEdgeLeft,
+            &*pEdgeRight,
+            nSubpixelYAdvance,
+            &mut nSubpixelXLeftBottom,
+            &mut nSubpixelErrorLeftBottom,
+            &mut nSubpixelXRightBottom,
+            &mut nSubpixelErrorRightBottom
+            );
+
+        // The above computation should ensure that we are a simple
+        // trapezoid at this point
+
+        assert!(nSubpixelXLeftBottom <= nSubpixelXRightBottom);
+
+        // We know we have a simple trapezoid now.  Now, compute the end of our current trapezoid
+
+        assert!(nSubpixelYAdvance > 0);
+
+        //
+        // Computation of edge data
+        //
+
+        rSubpixelLeftErrorDown  = pEdgeLeft.ErrorDown as f32;
+        rSubpixelRightErrorDown = pEdgeRight.ErrorDown as f32;
+        rPixelXLeft  = ConvertSubpixelXToPixel(pEdgeLeft.X.get(), pEdgeLeft.Error.get(), rSubpixelLeftErrorDown);
+        rPixelXRight = ConvertSubpixelXToPixel(pEdgeRight.X.get(), pEdgeRight.Error.get(), rSubpixelRightErrorDown);
+
+        rSubpixelLeftInvSlope     = pEdgeLeft.Dx as f32 + pEdgeLeft.ErrorUp as f32/rSubpixelLeftErrorDown;
+        rSubpixelLeftAbsInvSlope  = rSubpixelLeftInvSlope.abs();
+        rSubpixelRightInvSlope    = pEdgeRight.Dx as f32 + pEdgeRight.ErrorUp as f32/rSubpixelRightErrorDown;
+        rSubpixelRightAbsInvSlope = rSubpixelRightInvSlope.abs();
+
+        rPixelXLeftDelta  = 0.5 + 0.5 * rSubpixelLeftAbsInvSlope;
+        rPixelXRightDelta = 0.5 + 0.5 * rSubpixelRightAbsInvSlope;
+
+        let rPixelYTop         = ConvertSubpixelYToPixel(nSubpixelYCurrent);
+        let rPixelYBottom      = ConvertSubpixelYToPixel(nSubpixelYNext);
+
+        let rPixelXBottomLeft  = ConvertSubpixelXToPixel(
+                                        nSubpixelXLeftBottom,
+                                        nSubpixelErrorLeftBottom,
+                                        pEdgeLeft.ErrorDown as f32
+                                        );
+
+        let rPixelXBottomRight = ConvertSubpixelXToPixel(
+                                        nSubpixelXRightBottom,
+                                        nSubpixelErrorRightBottom,
+                                        pEdgeRight.ErrorDown as f32
+                                        );
+
+        //
+        // Output the trapezoid
+        //
+
+        IFC!(self.m_pIGeometrySink.AddTrapezoid(
+            rPixelYTop,              // In: y coordinate of top of trapezoid
+            rPixelXLeft,             // In: x coordinate for top left
+            rPixelXRight,            // In: x coordinate for top right
+            rPixelYBottom,           // In: y coordinate of bottom of trapezoid
+            rPixelXBottomLeft,       // In: x coordinate for bottom left
+            rPixelXBottomRight,      // In: x coordinate for bottom right
+            rPixelXLeftDelta,        // In: trapezoid expand radius for left edge
+            rPixelXRightDelta        // In: trapezoid expand radius for right edge
+            ));
+
+        //
+        // Update the edge data
+        //
+
+        //  no need to do this if edges are stale
+
+        pEdgeLeft.X.set(nSubpixelXLeftBottom);
+        pEdgeLeft.Error.set(nSubpixelErrorLeftBottom);
+        pEdgeRight.X.set(nSubpixelXRightBottom);
+        pEdgeRight.Error.set(nSubpixelErrorRightBottom);
+
+        //
+        // Check for termination
+        //
+
+        if (pEdgeRight.Next.get().EndY == INT::MIN)
+        {
+            break;
+        }
+
+        //
+        // Advance edge data
+        //
+
+        pEdgeLeft  = pEdgeRight.Next.get();
+        pEdgeRight = pEdgeLeft.Next.get();
+
+    }
+
+    return hr;
+
+}
+
+//-------------------------------------------------------------------------
+//
+//  Function:   CHwRasterizer::RasterizeEdges
+//
+//  Synopsis:
+//      Rasterize using trapezoidal AA
+//
+//-------------------------------------------------------------------------
+fn
+RasterizeEdges<'a, 'b>(&mut self,
+    pEdgeActiveList: Ref<'a, CEdge<'a>>,
+    mut pInactiveEdgeArray: &'a mut [CInactiveEdge<'a>],
+    coverageBuffer: &'b CCoverageBuffer<'b>,
+    mut nSubpixelYCurrent: INT,
+    nSubpixelYBottom: INT
+    ) -> HRESULT
+{
+    let hr: HRESULT = S_OK;
+    let mut pEdgePrevious: Ref<CEdge>;
+    let mut pEdgeCurrent: Ref<CEdge>;
+    let mut nSubpixelYNextInactive: INT = 0;
+    let mut nSubpixelYNext: INT;
+
+    pInactiveEdgeArray = InsertNewEdges(
+        pEdgeActiveList,
+        nSubpixelYCurrent,
+        pInactiveEdgeArray,
+        &mut nSubpixelYNextInactive
+        );
+
+    while (nSubpixelYCurrent < nSubpixelYBottom)
+    {
+        ASSERTACTIVELIST!(pEdgeActiveList, nSubpixelYCurrent);
+
+        //
+        // Detect trapezoidal case
+        //
+
+        pEdgePrevious = pEdgeActiveList;
+        pEdgeCurrent = pEdgeActiveList.Next.get();
+
+        nSubpixelYNext = nSubpixelYCurrent;
+
+        if (!IsTagEnabled!(tagDisableTrapezoids)
+            && (nSubpixelYCurrent & c_nShiftMask) == 0
+            && pEdgeCurrent.EndY != INT::MIN
+            && nSubpixelYNextInactive >= nSubpixelYCurrent + c_nShiftSize
+            )
+        {
+            // Edges are paired, so we can assert we have another one
+            assert!(pEdgeCurrent.Next.get().EndY != INT::MIN);
+
+            //
+            // Given an active edge list, we compute the furthest we can go in the y direction
+            // without creating self-intersection or going past the edge EndY.  Note that if we
+            // can't even go one scanline, then nSubpixelYNext == nSubpixelYCurrent
+            //
+
+            nSubpixelYNext = self.ComputeTrapezoidsEndScan(Ref::new(&*pEdgeCurrent), nSubpixelYCurrent, nSubpixelYNextInactive);
+            assert!(nSubpixelYNext >= nSubpixelYCurrent);
+
+            //
+            // Attempt to output a trapezoid.  If it turns out we don't have any
+            // potential trapezoids, then nSubpixelYNext == nSubpixelYCurent
+            // indicating that we need to fall back to complex scans.
+            //
+
+            if (nSubpixelYNext >= nSubpixelYCurrent + c_nShiftSize)
+            {
+                IFC!(self.OutputTrapezoids(
+                    pEdgeCurrent,
+                    nSubpixelYCurrent,
+                    nSubpixelYNext
+                    ));
+            }
+        }
+
+        //
+        // Rasterize simple trapezoid or a complex scanline
+        //
+
+        if (nSubpixelYNext > nSubpixelYCurrent)
+        {
+            // If we advance, it must be by at least one scan line
+
+            assert!(nSubpixelYNext - nSubpixelYCurrent >= c_nShiftSize);
+
+            // Advance nSubpixelYCurrent
+
+            nSubpixelYCurrent = nSubpixelYNext;
+
+            // Remove stale edges.  Note that the DDA is incremented in OutputTrapezoids.
+
+            while (pEdgeCurrent.EndY != INT::MIN)
+            {
+                if (pEdgeCurrent.EndY <= nSubpixelYCurrent)
+                {
+                    // Unlink and advance
+
+                    pEdgeCurrent = pEdgeCurrent.Next.get();
+                    pEdgePrevious.Next.set(pEdgeCurrent);
+                }
+                else
+                {
+                    // Advance
+
+                    pEdgePrevious = pEdgeCurrent;
+                    pEdgeCurrent = pEdgeCurrent.Next.get();
+                }
+            }
+        }
+        else
+        {
+            //
+            // Trapezoid rasterization failed, so
+            //   1) Handle case with no active edges, or
+            //   2) fall back to scan rasterization
+            //
+
+            if (pEdgeCurrent.EndY == INT::MIN)
+            {
+                nSubpixelYNext = nSubpixelYNextInactive;
+            }
+            else
+            {
+                nSubpixelYNext = nSubpixelYCurrent + 1;
+                if (self.m_fillMode == MilFillMode::Alternate)
+                {
+                    IFC!(coverageBuffer.FillEdgesAlternating(pEdgeActiveList, nSubpixelYCurrent));
+                }
+                else
+                {
+                    IFC!(coverageBuffer.FillEdgesWinding(pEdgeActiveList, nSubpixelYCurrent));
+                }
+            }
+
+            // If the next scan is done, output what's there:
+            if (nSubpixelYNext > (nSubpixelYCurrent | c_nShiftMask))
+            {
+                IFC!(self.GenerateOutputAndClearCoverage(coverageBuffer, nSubpixelYCurrent));
+            }
+
+            // Advance nSubpixelYCurrent
+            nSubpixelYCurrent = nSubpixelYNext;
+
+            // Advance DDA and update edge list
+            AdvanceDDAAndUpdateActiveEdgeList(nSubpixelYCurrent, pEdgeActiveList);
+        }
+
+        //
+        // Update edge list
+        //
+
+        if (nSubpixelYCurrent == nSubpixelYNextInactive)
+        {
+            pInactiveEdgeArray = InsertNewEdges(
+                pEdgeActiveList,
+                nSubpixelYCurrent,
+                pInactiveEdgeArray,
+                &mut nSubpixelYNextInactive
+                );
+        }
+    }
+
+    //
+    // Output the last scanline that has partial coverage
+    //
+
+    if ((nSubpixelYCurrent & c_nShiftMask) != 0)
+    {
+        IFC!(self.GenerateOutputAndClearCoverage(coverageBuffer, nSubpixelYCurrent));
+    }
+
+    RRETURN!(hr);
+}
+
+}
diff --git a/third_party/rust/wpf-gpu-raster/src/hwvertexbuffer.rs b/third_party/rust/wpf-gpu-raster/src/hwvertexbuffer.rs
new file mode 100644
index 0000000000..6b01b5fdb2
--- /dev/null
+++ b/third_party/rust/wpf-gpu-raster/src/hwvertexbuffer.rs
@@ -0,0 +1,3075 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+// See the LICENSE file in the project root for more information.
+
+
+//-----------------------------------------------------------------------------
+//
+
+//
+//  Description:
+//      Contains HW Vertex Buffer and Builder class implementations
+//
+//
+//  Notes:
+//        
+//        +--------------------------------------+
+//        |                                      |
+//        |           Start Stratum              |
+//     1  |                                      |
+//        |                                      |
+//        +--------------------------------------+
+//     2  |======================================|
+//        +--------------------------------------+
+//        |          /   \             / \       |
+//        |         /     \           /   \      |
+//        |   A    /   B   \    C    /  D  \  E  |
+//     3  |       /         \       /       \    |
+//        |      /           \     /         \   |
+//        |     /             \   /           \  |
+//        |    /               \ /             \ |
+//        +--------------------------------------+
+//        |    \               / \             / |
+//        |     \             /   \           /  |
+//     4  |  F   \     G     /  H  \    I    / J |
+//        |       \         /       \       /    |
+//        +--------------------------------------+
+//     5  |======================================|
+//        +--------------------------------------+
+//     6  |======================================|
+//        +--------------------------------------+
+//        |                                      |
+//        |                                      |
+//     7  |           Stop Stratum               |
+//        |                                      |
+//        |                                      |
+//        +--------------------------------------+
+//        
+//  
+//  Strata & complement mode.
+//  
+//  The anti-aliased HW rasterizer produces a series of "strata" where
+//  each strata can be a complex span rendered using lines (#'s 2,5,6) or
+//  a series of trapezoids (#'s 3 & 4.)  In normal mode the trapezoid
+//  regions B,D,G,I are filled in.
+//  
+//  Complement mode complicates things.  Complex spans are relatively easy
+//  because we get the whole line's worth of data at once.  Trapezoids are
+//  more complex because we get B,D,G and I separately.  We handle this by
+//  tracking the current stratum and finishing the last incomplete
+//  trapezoid stratum when a new stratum begins.  Regions E & J finish
+//  trapezoid strata.  We also need to add rectangles at the beginning and
+//  end of the geometry (start and stop) to fill out the complement
+//  region.
+//  
+//  This is implemented like so:
+//  
+//    1. Strata are generated from top to bottom without gaps.
+//    2. Before drawing any lines or trapezoids call
+//       PrepareStratum(a, b, fTrapezoid) where a & b are the extent of
+//       the current stratum and fTrapezoid is true if you are drawing
+//       a trapezoid.  This will take care of creating the start
+//       stratum and/or finishing a trapezoid stratum if necessary.
+//    3. When completely done call EndBuildingOutside() which will
+//       close a pending trapezoid and/or produce the stop stratum.
+//  
+//-----------------------------------------------------------------------------
+
+const FORCE_TRIANGLES: bool = true;
+
+//+----------------------------------------------------------------------------
+//
+//  Constants to control when we stop waffling because the tiles are too
+//  small to make a difference.
+//
+//  Future Consideration:  can produce an excessive number of triangles.
+//   How we mitigate or handle this could be improved.  Right now we stop
+//   waffling if the waffle size is less than a quarter-pixel.
+//   Two big improvements that could be made are:
+//    - multipacking very small textures (but note that we cannot rely
+//      on prefiltering to ensure that small screen space means small texture
+//      source)
+//    - clipping primitives to approximately the screen size
+//
+//-----------------------------------------------------------------------------
+//const c_rMinWaffleWidthPixels: f32 = 0.25;
+
+
+const FLOAT_ZERO: f32 = 0.;
+const FLOAT_ONE: f32 = 1.;
+
+//+----------------------------------------------------------------------------
+//
+//  Class:     CHwVertexBuffer and CHwTVertexBuffer<class TVertex>
+//
+//  Synopsis:  This class accumulates geometry data for a primitive
+//
+//-----------------------------------------------------------------------------
+
+use crate::{types::*, geometry_sink::IGeometrySink, aacoverage::c_nShiftSizeSquared, OutputVertex, nullable_ref::Ref};
+
+
+//+----------------------------------------------------------------------------
+//
+//  Class:     CHwVertexBuffer::Builder
+//
+//  Synopsis:  Base vertex builder class
+//
+//  Responsibilities:
+//    - Given ordered basic vertex information expand/convert/pass-thru
+//      to vertex buffer  (Basic vertex information is minimal vertex
+//      information sent from the caller that may or may not have been
+//      passed thru a tessellator.)
+//    - Choosing vertex format from a minimal required vertex format
+//
+//  Not responsible for:
+//    - Allocating space in vertex buffer
+//
+//  Inputs required:
+//    - Key and data to translate input basic vertex info to full vertex data
+//    - Vertex info from tessellation (or other Geometry Generator)
+//    - Vertex buffer to send output to
+//
+
+/*pub struct CHwVertexBufferBuilder /* : public IGeometrySink */
+{
+    /* 
+public:
+
+    static HRESULT Create(
+        MilVertexFormat vfIn,
+        MilVertexFormat vfOut,
+        MilVertexFormatAttribute vfaAntiAliasScaleLocation,
+        __in_ecount_opt(1) CHwPipeline *pPipeline,
+        __in_ecount_opt(1) CD3DDeviceLevel1 *pDevice,
+        __in_ecount(1) CBufferDispenser *pBufferDispenser,
+        __deref_out_ecount(1) CHwVertexBuffer::Builder **ppVertexBufferBuilder
+        );
+
+    virtual ~Builder()
+    {
+#if DBG
+        Assert(!m_fDbgDestroyed);
+        m_fDbgDestroyed = true;
+#endif DBG
+    }
+
+    //+------------------------------------------------------------------------
+    //
+    //  Member:    SetConstantMapping
+    //
+    //  Synopsis:  Use this method to specify that the given color source for
+    //             the given vertex destination is constant (won't differ per
+    //             vertex)
+    //
+    //-------------------------------------------------------------------------
+
+    virtual HRESULT SetConstantMapping(
+        MilVertexFormatAttribute mvfaDestination,
+        __in_ecount(1) const CHwConstantColorSource *pConstCS
+        ) PURE;
+
+
+    //+------------------------------------------------------------------------
+    //
+    //  Member:    FinalizeMappings
+    //
+    //  Synopsis:  Use this method to let builder know that all mappings have
+    //             been sent
+    //
+    //-------------------------------------------------------------------------
+
+    virtual HRESULT FinalizeMappings(
+        ) PURE;
+
+    //+------------------------------------------------------------------------
+    //
+    //  Member:    SetOutsideBounds
+    //
+    //  Synopsis:  Enables rendering zero-alpha geometry outside of the input
+    //             shape but within the given bounding rectangle, if fNeedInside
+    //             isn't true then it doesn't render geometry with full alpha.
+    //
+    //-------------------------------------------------------------------------
+    virtual void SetOutsideBounds(
+        __in_ecount_opt(1) const CMILSurfaceRect *prcBounds,
+        bool fNeedInside
+        ) PURE;
+
+    //+------------------------------------------------------------------------
+    //
+    //  Member:    HasOutsideBounds
+    //
+    //  Synopsis:  Returns true if outside bounds have been set.
+    //
+    //-------------------------------------------------------------------------
+    virtual bool HasOutsideBounds() const PURE;
+
+    //+------------------------------------------------------------------------
+    //
+    //  Member:    BeginBuilding
+    //
+    //  Synopsis:  This method lets the builder know it should start from a
+    //             clean slate
+    //
+    //-------------------------------------------------------------------------
+
+    virtual HRESULT BeginBuilding(
+        ) PURE;
+
+    //+------------------------------------------------------------------------
+    //
+    //  Member:    EndBuilding
+    //
+    //  Synopsis:  Use this method to let the builder know that all of the
+    //             vertex data has been sent
+    //
+    //-------------------------------------------------------------------------
+
+    virtual HRESULT EndBuilding(
+        __deref_opt_out_ecount(1) CHwVertexBuffer **ppVertexBuffer
+        ) PURE;
+
+    //+------------------------------------------------------------------------
+    //
+    //  Member:    FlushReset
+    //
+    //  Synopsis:  Send pending state and geometry to the device and reset
+    //             the vertex buffer.
+    //
+    //-------------------------------------------------------------------------
+
+    MIL_FORCEINLINE HRESULT FlushReset()
+    {
+        return FlushInternal(NULL);
+    }
+        
+    //
+    // Currently all CHwVertexBuffer::Builder are supposed to be allocated via
+    // a CBufferDispenser.
+    //
+
+    DECLARE_BUFFERDISPENSER_DELETE
+
+protected:
+
+    Builder()
+    {
+        m_mvfIn = MILVFAttrNone;
+
+#if DBG
+        m_mvfDbgOut = MILVFAttrNone;
+#endif
+
+        m_mvfaAntiAliasScaleLocation = MILVFAttrNone;
+
+        m_pPipelineNoRef = NULL;
+        m_pDeviceNoRef = NULL;
+        
+#if DBG
+        m_fDbgDestroyed = false;
+#endif DBG
+    }
+
+    //+------------------------------------------------------------------------
+    //
+    //  Member:    FlushInternal
+    //
+    //  Synopsis:  Send any pending state and geometry to the device.
+    //             If the optional argument is NULL then reset the
+    //             vertex buffer.
+    //             If the optional argument is non-NULL AND we have
+    //             not yet flushed the vertex buffer return the vertex
+    //             buffer.
+    //
+    //-------------------------------------------------------------------------
+
+    virtual HRESULT FlushInternal(
+        __deref_opt_out_ecount_opt(1) CHwVertexBuffer **ppVertexBuffer
+        ) PURE;
+
+
+    CHwPipeline *m_pPipelineNoRef;
+    CD3DDeviceLevel1 *m_pDeviceNoRef;
+
+    MilVertexFormat m_mvfIn;         // Vertex fields that are pre-generated
+
+#if DBG
+    MilVertexFormat m_mvfDbgOut;     // Output format of the vertex
+#endif
+
+    MilVertexFormat m_mvfGenerated;  // Vertex fields that are dynamically
+                                     // generated by this builder
+
+    MilVertexFormatAttribute m_mvfaAntiAliasScaleLocation;  // Vertex field that
+                                                            // contains PPAA
+                                                            // falloff factor
+
+#if DBG
+private:
+
+    bool m_fDbgDestroyed;     // Used to check single Release pattern
+
+#endif DBG
+*/
+}*/
+#[derive(Default)]
+pub struct CD3DVertexXYZDUV2 {
+    x: f32,
+    y: f32,
+    //Z: f32,
+    coverage: f32,
+    /*U0: f32, V0: f32,
+    U1: f32, V1: f32,*/
+}
+pub type CHwVertexBuffer<'z> = CHwTVertexBuffer<'z, OutputVertex>;
+#[derive(Default)]
+pub struct CHwTVertexBuffer<'z, TVertex>
+{
+    //m_rgIndices: DynArray<WORD>,     // Dynamic array of indices
+
+
+    //m_pBuilder: Rc<CHwTVertexBufferBuilder<TVertex>>,
+
+    /* 
+#if DBG
+public:
+    
+    CHwTVertexBuffer()
+    {
+        m_fDbgNonLineSegmentTriangleStrip = false;
+    }
+#endif
+
+protected:
+
+    //+------------------------------------------------------------------------
+    //
+    //  Member:    Reset
+    //
+    //  Synopsis:  Mark the beginning of a new list of vertices; the existing
+    //             list is discarded
+    //
+    //-------------------------------------------------------------------------
+
+    MIL_FORCEINLINE void Reset(
+        __in_ecount(1) Builder *pVBB
+        )
+    {
+#if DBG
+        m_fDbgNonLineSegmentTriangleStrip = false;
+#endif
+        m_rgIndices.SetCount(0);
+        m_rgVerticesTriList.SetCount(0);
+        m_rgVerticesTriStrip.SetCount(0);
+        m_rgVerticesLineList.SetCount(0);
+        m_rgVerticesNonIndexedTriList.SetCount(0);
+
+        m_pBuilder = pVBB;
+    }
+
+    //+------------------------------------------------------------------------
+    //
+    //  Member:    AddNonIndexedTriListVertices
+    //
+    //  Synopsis:  Reserve space for consecutive vertices and return start
+    //             index
+    //
+    //-------------------------------------------------------------------------
+
+    MIL_FORCEINLINE HRESULT AddNonIndexedTriListVertices(
+        UINT uCount,
+        __deref_ecount(uCount) TVertex **ppVertices
+        );
+
+    //+------------------------------------------------------------------------
+    //
+    //  Member:    AddTriListVertices
+    //
+    //  Synopsis:  Reserve space for consecutive vertices and return start
+    //             index
+    //
+    //-------------------------------------------------------------------------
+
+    MIL_FORCEINLINE HRESULT AddTriListVertices(
+        UINT uDelta,
+        __deref_ecount(uDelta) TVertex **ppVertices,
+        __out_ecount(1) WORD *pwIndexStart
+        );
+
+    //+------------------------------------------------------------------------
+    //
+    //  Member:    AddTriStripVertices
+    //
+    //  Synopsis:  Reserve space for consecutive vertices and return start
+    //             index
+    //
+    //-------------------------------------------------------------------------
+
+    MIL_FORCEINLINE HRESULT AddTriStripVertices(
+        UINT uCount,
+        __deref_ecount(uCount) TVertex **ppVertices
+        );
+
+    //+------------------------------------------------------------------------
+    //
+    //  Member:    AddLineListVertices
+    //
+    //  Synopsis:  Reserve space for consecutive vertices and return start
+    //             index
+    //
+    //-------------------------------------------------------------------------
+
+    MIL_FORCEINLINE HRESULT AddLineListVertices(
+        UINT uCount,
+        __deref_ecount(uCount) TVertex **ppVertices
+        );
+
+public:
+
+    //+------------------------------------------------------------------------
+    //
+    //  Member:    AddLine implements ILineSink<PointXYA>
+    //
+    //  Synopsis:  Add a line given two points with x, y, & alpha.
+    //
+    //-------------------------------------------------------------------------
+    HRESULT AddLine(
+        __in_ecount(1) const PointXYA &v0,
+        __in_ecount(1) const PointXYA &v1
+        );
+
+    //+------------------------------------------------------------------------
+    //
+    //  Member:    AddTriangle implements ITriangleSink<PointXYA>
+    //
+    //  Synopsis:  Add a triangle given three points with x, y, & alpha.
+    //
+    //-------------------------------------------------------------------------
+
+    HRESULT AddTriangle(
+        __in_ecount(1) const PointXYA &v0,
+        __in_ecount(1) const PointXYA &v1,
+        __in_ecount(1) const PointXYA &v2
+        );
+
+    // Re-introduce parent AddTriangle(WORD,WORD,WORD) into this scope.
+    using CHwVertexBuffer::AddTriangle;
+    
+    //+------------------------------------------------------------------------
+    //
+    //  Member:    AddLineAsTriangleStrip
+    //
+    //  Synopsis:  Add a horizontal line using a trinagle strip
+    //
+    //-------------------------------------------------------------------------
+    HRESULT AddLineAsTriangleStrip(
+        __in_ecount(1) const TVertex *pBegin, // Begin
+        __in_ecount(1) const TVertex *pEnd    // End
+        );
+
+    //+------------------------------------------------------------------------
+    //
+    //  Member:    SendVertexFormat
+    //
+    //  Synopsis:  Send contained vertex format to device
+    //
+    //-------------------------------------------------------------------------
+
+    HRESULT SendVertexFormat(
+        __inout_ecount(1) CD3DDeviceLevel1 *pDevice
+        ) const;
+
+    //+------------------------------------------------------------------------
+    //
+    //  Member:    DrawPrimitive
+    //
+    //  Synopsis:  Send the geometry data to the device and execute rendering
+    //
+    //-------------------------------------------------------------------------
+
+    HRESULT DrawPrimitive(
+        __inout_ecount(1) CD3DDeviceLevel1 *pDevice
+        ) const;
+
+protected:
+    //+------------------------------------------------------------------------
+    //
+    //  Member:    GetNumTriListVertices
+    //
+    //  Synopsis:  Return current number of vertices
+    //
+    //-------------------------------------------------------------------------
+
+    MIL_FORCEINLINE DWORD GetNumTriListVertices() const
+    {
+        return m_rgVerticesTriList.GetCount();
+    }
+
+    //+------------------------------------------------------------------------
+    //
+    //  Member:    GetTriListVertices
+    //
+    //  Synopsis:  Return pointer to beginning of vertex list and their count
+    //
+    //-------------------------------------------------------------------------
+
+    MIL_FORCEINLINE void GetTriListVertices(
+        __deref_out_ecount_full(*puNumVertices) TVertex **ppVertices,
+        __out_ecount(1) UINT * puNumVertices
+        )
+    {
+        *ppVertices = m_rgVerticesTriList.GetDataBuffer();
+        *puNumVertices = m_rgVerticesTriList.GetCount();
+    }
+
+    //+------------------------------------------------------------------------
+    //
+    //  Member:    GetNumNonIndexedTriListVertices
+    //
+    //  Synopsis:  Return current number of vertices
+    //
+    //-------------------------------------------------------------------------
+
+    MIL_FORCEINLINE DWORD GetNumNonIndexedTriListVertices() const
+    {
+        return m_rgVerticesNonIndexedTriList.GetCount();
+    }
+
+    //+------------------------------------------------------------------------
+    //
+    //  Member:    GetNonIndexedTriListVertices
+    //
+    //  Synopsis:  Return pointer to beginning of vertex list and their count
+    //
+    //-------------------------------------------------------------------------
+
+    MIL_FORCEINLINE void GetNonIndexedTriListVertices(
+        __deref_out_ecount_full(*puNumVertices) TVertex **ppVertices,
+        __out_ecount(1) UINT * puNumVertices
+        )
+    {
+        *ppVertices = m_rgVerticesNonIndexedTriList.GetDataBuffer();
+        *puNumVertices = m_rgVerticesNonIndexedTriList.GetCount();
+    }
+
+    //+------------------------------------------------------------------------
+    //
+    //  Member:    GetNumTriStripVertices
+    //
+    //  Synopsis:  Return current number of vertices
+    //
+    //-------------------------------------------------------------------------
+
+    MIL_FORCEINLINE DWORD GetNumTriStripVertices() const
+    {
+        return m_rgVerticesTriStrip.GetCount();
+    }
+
+    //+------------------------------------------------------------------------
+    //
+    //  Member:    GetTriStripVertices
+    //
+    //  Synopsis:  Return pointer to beginning of vertex list and their count
+    //
+    //-------------------------------------------------------------------------
+
+    MIL_FORCEINLINE void GetTriStripVertices(
+        __deref_out_ecount_full(*puNumVertices) TVertex **ppVertices,
+        __out_ecount(1) UINT *puNumVertices
+        )
+    {
+        *ppVertices = m_rgVerticesTriStrip.GetDataBuffer();
+        *puNumVertices = m_rgVerticesTriStrip.GetCount();
+    }
+
+    //+------------------------------------------------------------------------
+    //
+    //  Member:    GetNumLineListVertices
+    //
+    //  Synopsis:  Return current number of vertices
+    //
+    //-------------------------------------------------------------------------
+
+    MIL_FORCEINLINE DWORD GetNumLineListVertices() const
+    {
+        return m_rgVerticesLineList.GetCount();
+    }
+
+    //+------------------------------------------------------------------------
+    //
+    //  Member:    GetLineListVertices
+    //
+    //  Synopsis:  Return pointer to beginning of vertex list and their count
+    //
+    //-------------------------------------------------------------------------
+
+    MIL_FORCEINLINE void GetLineListVertices(
+        __deref_out_ecount_full(*puNumVertices) TVertex **ppVertices,
+        __out_ecount(1) UINT * puNumVertices
+        )
+    {
+        *ppVertices = m_rgVerticesLineList.GetDataBuffer();
+        *puNumVertices = m_rgVerticesLineList.GetCount();
+    }
+
+    //+------------------------------------------------------------------------
+    //
+    //  Member:    GetLineListVertices
+    //
+    //  Synopsis:  Return pointer to beginning of vertex list
+    //
+    //-------------------------------------------------------------------------
+
+
+
+*/
+
+    // Dynamic array of vertices for which all allocations are zeroed.
+    // XXX: the zero has been removed
+    //m_rgVerticesTriList: DynArray<TVertex>,             // Indexed triangle list vertices
+    //m_rgVerticesNonIndexedTriList: DynArray<TVertex>,   // Non-indexed triangle list vertices
+    m_rgVerticesTriList: DynArray<TVertex>,            // Triangle strip vertices
+    //m_rgVerticesLineList: DynArray<TVertex>,            // Linelist vertices
+
+    m_rgVerticesBuffer: Option<&'z mut [TVertex]>,
+    m_rgVerticesBufferOffset: usize,
+
+    #[cfg(debug_assertions)]
+    // In debug make a note if we add a triangle strip that doesn't have 6 vertices
+    // so that we can ensure that we only waffle 6-vertex tri strips.
+    m_fDbgNonLineSegmentTriangleStrip: bool,
+    subpixel_bias: f32,
+}
+
+impl<'z, TVertex: Default> CHwTVertexBuffer<'z, TVertex> {
+    pub fn new(rasterization_truncates: bool, output_buffer: Option<&'z mut [TVertex]>) -> Self {
+        CHwTVertexBuffer::<TVertex> {
+            subpixel_bias: if rasterization_truncates {
+                // 1/512 is 0.5 of a subpixel when using 8 bits of subpixel precision.
+                1./512.
+            } else {
+                0.
+            },
+            m_rgVerticesBuffer: output_buffer,
+            m_rgVerticesBufferOffset: 0,
+            ..Default::default()
+        }
+    }
+
+    pub fn flush_output(&mut self) -> Box<[TVertex]> {
+        std::mem::take(&mut self.m_rgVerticesTriList).into_boxed_slice()
+    }
+
+    pub fn get_output_buffer_size(&self) -> Option<usize> {
+        if self.m_rgVerticesBuffer.is_some() {
+            Some(self.m_rgVerticesBufferOffset)
+        } else {
+            None
+        }
+    }
+}
+
+//+----------------------------------------------------------------------------
+//
+//  Class:     CHwTVertexMappings<class TVertex>
+//
+//  Synopsis:  Helper class that knows how to populate a vertex from the
+//             incoming basic per vertex data, like just X and Y
+//
+//-----------------------------------------------------------------------------
+#[derive(Default)]
+struct CHwTVertexMappings<TVertex>
+{/* 
+public:
+
+    CHwTVertexMappings();
+
+    void SetPositionTransform(
+        __in_ecount(1) const MILMatrix3x2 &matPositionTransform
+        );
+
+    HRESULT SetConstantMapping(
+        MilVertexFormatAttribute mvfaDestination,
+        __in_ecount(1) const CHwConstantColorSource *pConstCS
+        );
+
+    void PointToUV(
+        __in_ecount(1) const MilPoint2F &ptIn,
+        __bound UINT uIndex,
+        __out_ecount(1) TVertex *pvOut
+        );
+    
+    MIL_FORCEINLINE bool AreWaffling() const
+    {
+        return false;
+    }
+
+private:
+    static const size_t s_numOfVertexTextureCoords
+        = NUM_OF_VERTEX_TEXTURE_COORDS(TVertex);
+public:
+
+    MilVertexFormat m_mvfMapped;
+
+    MilColorF m_colorStatic;
+
+    MILMatrix3x2 m_matPos2DTransform;
+
+    MILMatrix3x2 m_rgmatPointToUV[s_numOfVertexTextureCoords];
+    CMilPointAndSizeF m_rgSubrect[s_numOfVertexTextureCoords];
+    WaffleModeFlags m_rgWaffleMode[s_numOfVertexTextureCoords];
+
+*/
+    m_vStatic: TVertex,
+    subpixel_bias: f32,
+}
+
+impl<TVertex> CHwTVertexBuffer<'_, TVertex> {
+    pub fn Reset(&mut self,
+        /*pVBB: &mut CHwTVertexBufferBuilder<TVertex>*/
+        )
+    {
+        #[cfg(debug_assertions)]
+        {
+            self.m_fDbgNonLineSegmentTriangleStrip = false;
+        }
+
+        //self.m_rgIndices.SetCount(0);
+        //self.m_rgVerticesTriList.SetCount(0);
+        self.m_rgVerticesTriList.SetCount(0);
+        self.m_rgVerticesBufferOffset = 0;
+        //self.m_rgVerticesLineList.SetCount(0);
+        //self.m_rgVerticesNonIndexedTriList.SetCount(0);
+
+        //self.m_pBuilder = pVBB;
+    }
+
+    fn IsEmpty(&self) -> bool
+    {
+        return true
+             //  && (self.m_rgIndices.GetCount() == 0)
+            //&& (self.m_rgVerticesLineList.GetCount() == 0)
+            && (self.m_rgVerticesTriList.GetCount() == 0)
+            && self.m_rgVerticesBufferOffset == 0
+            //&& (self.m_rgVerticesNonIndexedTriList.GetCount() == 0);
+    }
+
+}
+
+//+----------------------------------------------------------------------------
+//
+//  Class:     CHwTVertexBuffer<class TVertex>::Builder
+//
+//  Synopsis:  Implements CHwVertexBuffer::Builder for a particular vertex
+//             format
+//
+//-----------------------------------------------------------------------------
+
+pub struct CHwTVertexBufferBuilder<'y, 'z, TVertex>
+{
+    m_mvfIn: MilVertexFormat,         // Vertex fields that are pre-generated
+
+    #[cfg(debug_assertions)]
+    m_mvfDbgOut: MilVertexFormat,     // Output format of the vertex
+    
+    m_mvfGenerated: MilVertexFormat,  // Vertex fields that are dyn
+
+    m_mvfaAntiAliasScaleLocation: MilVertexFormatAttribute,  // Vertex field that
+                                                             // contains PPAA
+                                                             // falloff factor
+
+    /*
+public:
+
+    static MilVertexFormat GetOutVertexFormat();
+
+    static HRESULT Create(
+        __in_ecount(1) CHwTVertexBuffer<TVertex> *pVertexBuffer,
+        MilVertexFormat mvfIn,
+        MilVertexFormat mvfOut,
+        MilVertexFormatAttribute mvfaAntiAliasScaleLocation,
+        __inout_ecount(1) CBufferDispenser *pBufferDispenser,
+        __deref_out_ecount(1) typename CHwTVertexBuffer<TVertex>::Builder **ppVertexBufferBuilder
+        );
+
+    HRESULT SetConstantMapping(
+        MilVertexFormatAttribute mvfaDestination,
+        __in_ecount(1) const CHwConstantColorSource *pConstCS
+        );
+
+    void SetTransformMapping(
+        __in_ecount(1) const MILMatrix3x2 &mat2DTransform
+        );
+
+    HRESULT FinalizeMappings(
+        );
+
+    void SetOutsideBounds(
+        __in_ecount_opt(1) const CMILSurfaceRect *prcBounds,
+        bool fNeedInside
+        );
+
+    bool HasOutsideBounds() const
+    {
+        return NeedOutsideGeometry();
+    }
+
+    HRESULT BeginBuilding(
+        );
+
+    HRESULT AddVertex(
+        __in_ecount(1) const MilPoint2F &ptPosition,
+            // In: Vertex coordinates
+        __out_ecount(1) WORD *pIndex
+            // Out: The index of the new vertex
+        );
+
+    HRESULT AddIndexedVertices(
+        UINT cVertices,                                                  // In: number of vertices                                                                                            
+        __in_bcount(cVertices*uVertexStride) const void *pVertexBuffer,  // In: vertex buffer containing the vertices                                                                         
+        UINT uVertexStride,                                              // In: size of each vertex                                                                                           
+        MilVertexFormat mvfFormat,                                       // In: format of each vertex                                                                                         
+        UINT cIndices,                                                   // In: Number of indices                                                                                             
+        __in_ecount(cIndices) const UINT *puIndexBuffer                  // In: index buffer                                                             
+        );
+
+    HRESULT AddTriangle(
+        DWORD i1,                    // In: Index of triangle's first vertex
+        DWORD i2,                    // In: Index of triangle's second vertex
+        DWORD i3                     // In: Index of triangle's third vertex
+        );
+
+    HRESULT AddComplexScan(
+        INT nPixelY,
+            // In: y coordinate in pixel space
+        __in_ecount(1) const CCoverageInterval *pIntervalSpanStart
+            // In: coverage segments
+        );
+
+   HRESULT AddParallelogram(
+        __in_ecount(4)  const MilPoint2F *rgPosition
+        );
+   
+    HRESULT AddTrapezoid(
+        float rPixelYTop,               // In: y coordinate of top of trapezoid
+        float rPixelXTopLeft,           // In: x coordinate for top left
+        float rPixelXTopRight,          // In: x coordinate for top right
+        float rPixelYBottom,            // In: y coordinate of bottom of trapezoid
+        float rPixelXBottomLeft,        // In: x coordinate for bottom left
+        float rPixelXBottomRight,       // In: x coordinate for bottom right
+        float rPixelXLeftDelta,         // In: trapezoid expand radius for left edge
+        float rPixelXRightDelta         // In: trapezoid expand radius for right edge
+        );
+
+    BOOL IsEmpty();
+
+    HRESULT EndBuilding(
+        __deref_opt_out_ecount(1) CHwVertexBuffer **ppVertexBuffer
+        );
+
+    HRESULT FlushInternal(
+        __deref_opt_out_ecount_opt(1) CHwVertexBuffer **ppVertexBuffer
+        );
+            
+private:
+
+    // Helpers that do AddTrapezoid.  Same parameters
+    HRESULT AddTrapezoidStandard( float, float, float, float, float, float, float, float );
+    HRESULT AddTrapezoidWaffle( float, float, float, float, float, float, float, float );
+
+
+    
+    HRESULT PrepareStratumSlow(
+        float rStratumTop,
+        float rStratumBottom,
+        bool fTrapezoid,
+        float rTrapezoidLeft,
+        float rTrapezoidRight
+        );
+    
+    // Wrap up building of outside geometry.
+    HRESULT EndBuildingOutside();
+
+    DECLARE_BUFFERDISPENSER_NEW(CHwTVertexBuffer<TVertex>::Builder,
+                                Mt(CHwTVertexBuffer_Builder));
+
+    Builder(
+        __in_ecount(1) CHwTVertexBuffer<TVertex> *pVertexBuffer
+    );
+
+    HRESULT SetupConverter(
+        MilVertexFormat mvfIn,
+        MilVertexFormat mvfOut,
+        MilVertexFormatAttribute mvfaAntiAliasScaleLocation
+        );
+
+    HRESULT RenderPrecomputedIndexedTriangles(
+        __range(1, SHORT_MAX) UINT cVertices,
+        __in_ecount(cVertices) const TVertex *rgoVertices,
+        __range(1, UINT_MAX) UINT cIndices,
+        __in_ecount(cIndices) const UINT *rguIndices
+        );
+
+
+    // Expands all vertices in the buffer.
+    void ExpandVertices();
+    
+    // Has never been successfully used to declare a method or derived type...
+/*    typedef void (CHwTVertexBuffer<TVertex>::Builder::FN_ExpandVertices)(
+        UINT uCount,
+        TVertex *pVertex
+        );*/
+
+    // error C2143: syntax error : missing ';' before '*'
+//    typedef FN_ExpandVertices *PFN_ExpandVertices;
+
+    typedef void (CHwTVertexBuffer<TVertex>::Builder::* PFN_ExpandVertices)(
+        __range(1,UINT_MAX) UINT uCount,
+        __inout_ecount_full(uCount) TVertex *rgVertices
+        );
+
+    //
+    // Table of vertex expansion routines for common expansion cases:
+    //  - There are entries for Z, Diffuse, and one set texture coordinates for
+    //    a total of eight combinations.
+    //  - Additionally there is a second set of entries for anti-aliasing
+    //    falloff applied thru diffuse.
+    //
+
+    static const PFN_ExpandVertices sc_pfnExpandVerticesTable[8*2];
+
+    MIL_FORCEINLINE
+    void TransferAndOrExpandVerticesInline(
+        __range(1,UINT_MAX) UINT uCount,
+        __in_ecount(uCount) TVertex const * rgInputVertices,
+        __out_ecount(uCount) TVertex *rgOutputVertices,
+        MilVertexFormat mvfOut,
+        MilVertexFormatAttribute mvfaScaleByFalloff,
+        bool fInputOutputAreSameBuffer,
+        bool fTransformPosition
+        );
+
+    // FN_ExpandVertices ExpandVerticesFast
+    template <MilVertexFormat mvfOut, MilVertexFormatAttribute mvfaScaleByFalloff>
+    void ExpandVerticesFast(
+        __range(1,UINT_MAX) UINT uCount,
+        __inout_ecount_full(uCount) TVertex *rgVertices
+        )
+    {
+        TransferAndOrExpandVerticesInline(
+            uCount, 
+            rgVertices, 
+            rgVertices, 
+            mvfOut, 
+            mvfaScaleByFalloff,
+            true, // => fInputOutputAreSameBuffer
+            false // => fTransformPosition
+            );
+    }
+
+    // error C2146: syntax error : missing ';' before identifier 'ExpandVerticesGeneral'
+    // error C2501: 'CHwTVertexBufferBuilder<TVertex>::FN_ExpandVertices' : missing storage-class or type specifiers
+//    FN_ExpandVertices ExpandVerticesGeneral
+//    typename FN_ExpandVertices ExpandVerticesGeneral
+    // error C4346: 'CHwTVertexBufferBuilder<TVertex>::FN_ExpandVertices' : dependent name is not a type
+//    CHwTVertexBufferBuilder<TVertex>::FN_ExpandVertices ExpandVerticesGeneral
+    // Can't define methos here (unless not parameters are used).
+//    typename CHwTVertexBufferBuilder<TVertex>::FN_ExpandVertices ExpandVerticesGeneral
+    // FN_ExpandVertices ExpandVerticesGeneral
+    void ExpandVerticesGeneral(
+        __range(1,UINT_MAX) UINT uCount,
+        __inout_ecount_full(uCount) TVertex *rgVertices
+        )
+    {
+        TransferAndOrExpandVerticesInline(
+            uCount, 
+            rgVertices,
+            rgVertices,
+            m_mvfGenerated, 
+            m_mvfaAntiAliasScaleLocation,
+            true, // => fInputOutputAreSameBuffer
+            false // => fTransformPosition
+            );
+    }
+
+    void TransferAndExpandVerticesGeneral(
+        __range(1,UINT_MAX) UINT uCount,
+        __in_ecount(uCount) TVertex const *rgInputVertices,
+        __out_ecount_full(uCount) TVertex *rgOutputVertices,
+        bool fTransformPosition
+        )
+    {
+        TransferAndOrExpandVerticesInline(
+            uCount, 
+            rgInputVertices,
+            rgOutputVertices,
+            m_mvfGenerated, 
+            m_mvfaAntiAliasScaleLocation,
+            false,              // => fInputOutputAreSameBuffer
+            fTransformPosition  // => fTransformPosition
+            );
+    }
+
+    // FN_ExpandVertices ExpandVerticesInvalid
+    void ExpandVerticesInvalid(
+        __range(1,UINT_MAX) UINT uCount,
+        __inout_ecount_full(uCount) TVertex *rgVertices
+        )
+    {
+        RIP("Invalid ExpandVertices routine.");
+    }
+
+    //+------------------------------------------------------------------------
+    //
+    //  Member:    NeedCoverageGeometry
+    //
+    //  Synopsis:  True if we should create geometry for a particular
+    //             coverage value.
+    //
+    //-------------------------------------------------------------------------
+    bool NeedCoverageGeometry(INT nCoverage) const;
+
+
+
+
+
+    //+------------------------------------------------------------------------
+    //
+    //  Member:    ReinterpretFloatAsDWORD
+    //
+    //  Synopsis:  Quicky helper to convert a float to a DWORD bitwise.
+    //
+    //-------------------------------------------------------------------------
+    static MIL_FORCEINLINE DWORD ReinterpretFloatAsDWORD(float c)
+    {
+        return reinterpret_cast<DWORD &>(c);
+    }
+
+private:
+    MIL_FORCEINLINE bool AreWaffling() const
+    {
+        return m_map.AreWaffling();
+    }
+   
+    void ViewportToPackedCoordinates(
+        __range(1,UINT_MAX / uGroupSize) UINT uGroupCount,        
+        __inout_ecount(uGroupCount * uGroupSize) TVertex *pVertex,
+        __range(2,6) UINT uGroupSize,
+        /*__range(0,NUM_OF_VERTEX_TEXTURE_COORDS(TVertex)-1)*/ __bound UINT uIndex
+        );
+    
+    void ViewportToPackedCoordinates(
+        __range(1,UINT_MAX / uGroupSize) UINT uGroupCount,
+        __inout_ecount(uGroupCount * uGroupSize) TVertex *pVertex,
+        __range(2,6) UINT uGroupSize
+        );
+
+    template<class TWaffler>
+    __out_ecount(1) typename TWaffler::ISink *
+    BuildWafflePipeline(
+        __out_xcount(NUM_OF_VERTEX_TEXTURE_COORDS(TVertex) * 2) TWaffler *wafflers,
+        __out_ecount(1) bool &fWafflersUsed
+        ) const;
+
+    
+    template<class TWaffler>
+    typename TWaffler::ISink *
+    BuildWafflePipeline(
+        __out_xcount(NUM_OF_VERTEX_TEXTURE_COORDS(TVertex) * 2) TWaffler *wafflers
+        ) const
+    {
+        bool fNotUsed;
+        return BuildWafflePipeline(wafflers, fNotUsed);
+    }*/
+
+    m_pVB: &'y mut CHwTVertexBuffer<'z, TVertex>,
+
+    //m_pfnExpandVertices: PFN_ExpandVertices,  // Method for expanding vertices
+
+    //m_rgoPrecomputedTriListVertices: *const TVertex,
+    //m_cPrecomputedTriListVertices: UINT,
+
+    //m_rguPrecomputedTriListIndices: *const UINT,
+    //m_cPrecomputedTriListIndices: UINT,
+
+    //m_map: CHwTVertexMappings<TVertex>,
+
+    // This is true if we had to flush the pipeline as we were getting
+    // geometry rather than just filling up a single vertex buffer.
+    m_fHasFlushed: bool,
+
+    // The next two members control the generation of the zero-alpha geometry
+    // outside the input geometry.
+    m_fNeedOutsideGeometry: bool,
+    m_fNeedInsideGeometry: bool,
+    m_rcOutsideBounds: CMILSurfaceRect, // Bounds for creation of outside geometry
+
+    /* 
+    // Helpful m_rcOutsideBounds casts.
+    float OutsideLeft() const { return static_cast<float>(m_rcOutsideBounds.left); }
+    float OutsideRight() const { return static_cast<float>(m_rcOutsideBounds.right); }
+    float OutsideTop() const { return static_cast<float>(m_rcOutsideBounds.top); }
+    float OutsideBottom() const { return static_cast<float>(m_rcOutsideBounds.bottom); }
+    */
+    // This interval (if we are doing outside) shows the location
+    // of the current stratum.  It is initialized to [FLT_MAX, -FLT_MAX].
+    //
+    // If the current stratum is a complex span then
+    // m_rCurStratumBottom is set to the bottom of the stratum and
+    // m_rCurStratumTop is set to FLT_MAX.
+    //
+    // If the current stratum is a trapezoidal one, then
+    // m_rCurStratumBottom is its bottom and m_rCurStratumTop is its
+    // top.
+    m_rCurStratumTop: f32,
+    m_rCurStratumBottom: f32,
+
+    // If the current stratum is a trapezoidal one, following var stores
+    // right boundary of the last trapezoid handled by PrepareStratum.
+    // We need it to cloze the stratus properly.
+    m_rLastTrapezoidRight: f32,
+
+    // These are needed to implement outside geometry using triangle lists
+    m_rLastTrapezoidTopRight: f32,
+    m_rLastTrapezoidBottomRight: f32,
+}
+
+/*
+//+----------------------------------------------------------------------------
+//
+//  Member:    CHwVertexBuffer::AddTriangle
+//
+//  Synopsis:  Add a triangle using the three indices given to the list
+//
+impl CHwVertexBuffer {
+
+fn AddTriangle(
+    i1: WORD,         // In: Index of triangle's first vertex
+    i2: WORD,         // In: Index of triangle's second vertex
+    i3: WORD          // In: Index of triangle's third vertex
+    ) -> HRESULT
+{
+    let hr: HRESULT = S_OK;
+
+    // Asserting indices < max vertex requires a debug only pure virtual method
+    // which is too much of a functionality change between retail and debug.
+    //
+    //
+    // Assert(i1 < GetNumTriListVertices());
+    // Assert(i2 < GetNumTriListVertices());
+    // Assert(i3 < GetNumTriListVertices());
+
+    WORD *pIndices;
+
+    IFC(m_rgIndices.AddMultiple(3, &pIndices));
+
+    pIndices[0] = i1;
+    pIndices[1] = i2;
+    pIndices[2] = i3;
+
+Cleanup:
+    RRETURN(hr);
+}
+
+//+----------------------------------------------------------------------------
+//
+//  Member:    CHwTVertexBuffer<TVertex>::AddTriangle
+//
+//  Synopsis:  Add a triangle using given three points to the list
+//
+//-----------------------------------------------------------------------------
+template <class TVertex>
+HRESULT
+CHwTVertexBuffer<TVertex>::AddTriangle(
+    __in_ecount(1) const PointXYA &v0,
+    __in_ecount(1) const PointXYA &v1,
+    __in_ecount(1) const PointXYA &v2)
+{
+    let hr: HRESULT = S_OK;
+    
+    TVertex *pVertices;
+    hr = AddNonIndexedTriListVertices(3,&pVertices);
+
+    if (hr == E_OUTOFMEMORY)
+    {
+        DebugBreak ();
+    }
+    IFC(hr);
+    
+    pVertices[0].ptPt.X = v0.x;
+    pVertices[0].ptPt.Y = v0.y;
+    pVertices[0].Diffuse = reinterpret_cast<const DWORD &>(v0.a);
+    pVertices[1].ptPt.X = v1.x;
+    pVertices[1].ptPt.Y = v1.y;
+    pVertices[1].Diffuse = reinterpret_cast<const DWORD &>(v1.a);
+    pVertices[2].ptPt.X = v2.x;
+    pVertices[2].ptPt.Y = v2.y;
+    pVertices[2].Diffuse = reinterpret_cast<const DWORD &>(v2.a);
+
+Cleanup:
+    RRETURN(hr);
+}
+*/
+
+impl CHwVertexBuffer<'_> {
+//+----------------------------------------------------------------------------
+//
+//  Member:    CHwTVertexBuffer<TVertex>::AddLine
+//
+//  Synopsis:  Add a nominal width line using given two points to the list
+//
+//-----------------------------------------------------------------------------
+fn AddLine(&mut self,
+    v0: &PointXYA,
+    v1: &PointXYA
+    ) -> HRESULT
+{
+    type TVertex = CD3DVertexXYZDUV2;
+    let hr = S_OK;
+
+    let pVertices: &mut [TVertex];
+    let mut rgScratchVertices: [TVertex; 2] = Default::default();
+
+    assert!(!(v0.y != v1.y));
+    
+    let fUseTriangles = /*(v0.y < m_pBuilder->GetViewportTop() + 1) ||*/ FORCE_TRIANGLES;
+
+    //if (fUseTriangles)
+    //{
+        pVertices = &mut rgScratchVertices;
+    //}
+    //else
+    //{
+        //IFC!(AddLineListVertices(2, &pVertices));
+    //}
+    
+    pVertices[0].x = v0.x;
+    pVertices[0].y = v0.y;
+    pVertices[0].coverage = v0.a;
+    pVertices[1].x = v1.x;
+    pVertices[1].y = v1.y;
+    pVertices[1].coverage = v1.a;
+
+    if (fUseTriangles)
+    {
+        IFC!(self.AddLineAsTriangleList(&pVertices[0],&pVertices[1]));
+    }
+    
+    RRETURN!(hr);
+}
+}
+/* 
+//+----------------------------------------------------------------------------
+//
+//  Member:    CHwTVertexBuffer<TVertex>::AddTriListVertices
+//
+//  Synopsis:  Reserve space for consecutive vertices and return start index
+//
+
+template <class TVertex>
+MIL_FORCEINLINE
+HRESULT
+CHwTVertexBuffer<TVertex>::AddTriListVertices(
+    UINT uDelta,
+    __deref_ecount(uDelta) TVertex **ppVertices,
+    __out_ecount(1) WORD *pwIndexStart
+    )
+{
+    HRESULT hr = S_OK;
+
+    Assert(ppVertices);
+
+    UINT uCount = static_cast<UINT>(m_rgVerticesTriList.GetCount());
+    if (uCount > SHRT_MAX)
+    {
+        IFC(WGXERR_INVALIDPARAMETER);
+    }
+    UINT newCount;
+    newCount = uDelta + uCount;
+
+    if (newCount > SHRT_MAX)
+    {
+        IFC(m_pBuilder->FlushReset());
+        uCount = 0;
+        newCount = uDelta;
+    }
+
+    if (newCount > m_rgVerticesTriList.GetCapacity())
+    {
+        IFC(m_rgVerticesTriList.ReserveSpace(uDelta));
+    }
+
+    m_rgVerticesTriList.SetCount(newCount);
+    *pwIndexStart = static_cast<WORD>(uCount);
+    *ppVertices = &m_rgVerticesTriList[uCount];
+
+  Cleanup:
+    RRETURN(hr);
+}
+*/
+
+impl<TVertex: Clone + Default> CHwTVertexBuffer<'_, TVertex> {
+
+fn AddTriVertices(&mut self, v0: TVertex, v1: TVertex, v2: TVertex) {
+    if let Some(output_buffer) = &mut self.m_rgVerticesBuffer {
+        let offset = self.m_rgVerticesBufferOffset;
+        if offset + 3 <= output_buffer.len() {
+            output_buffer[offset] = v0;
+            output_buffer[offset + 1] = v1;
+            output_buffer[offset + 2] = v2;
+        }
+        self.m_rgVerticesBufferOffset = offset + 3;
+    } else {
+        self.m_rgVerticesTriList.reserve(3);
+        self.m_rgVerticesTriList.push(v0);
+        self.m_rgVerticesTriList.push(v1);
+        self.m_rgVerticesTriList.push(v2);
+    }
+}
+
+fn AddTrapezoidVertices(&mut self, v0: TVertex, v1: TVertex, v2: TVertex, v3: TVertex) {
+    if let Some(output_buffer) = &mut self.m_rgVerticesBuffer {
+        let offset = self.m_rgVerticesBufferOffset;
+        if offset + 6 <= output_buffer.len() {
+            output_buffer[offset] = v0;
+            output_buffer[offset + 1] = v1.clone();
+            output_buffer[offset + 2] = v2.clone();
+
+            output_buffer[offset + 3] = v1;
+            output_buffer[offset + 4] = v2;
+            output_buffer[offset + 5] = v3;
+        }
+        self.m_rgVerticesBufferOffset = offset + 6;
+    } else {
+        self.m_rgVerticesTriList.reserve(6);
+
+        self.m_rgVerticesTriList.push(v0);
+        self.m_rgVerticesTriList.push(v1.clone());
+        self.m_rgVerticesTriList.push(v2.clone());
+
+        self.m_rgVerticesTriList.push(v1);
+        self.m_rgVerticesTriList.push(v2);
+        self.m_rgVerticesTriList.push(v3);
+    }
+}
+
+fn AddedNonLineSegment(&mut self) {
+    #[cfg(debug_assertions)]
+    {
+        self.m_fDbgNonLineSegmentTriangleStrip = true;
+    }
+}
+
+}
+
+/* 
+//+----------------------------------------------------------------------------
+//
+//  Member:    CHwTVertexBuffer<TVertex>::AddNonIndexedTriListVertices
+//
+//  Synopsis:  Reserve space for triangle list vertices.
+//
+
+template <class TVertex>
+MIL_FORCEINLINE
+HRESULT
+CHwTVertexBuffer<TVertex>::AddNonIndexedTriListVertices(
+    UINT uCount,
+    __deref_ecount(uCount) TVertex **ppVertices
+    )
+{
+    HRESULT hr = S_OK;
+
+    UINT Count = static_cast<UINT>(m_rgVerticesNonIndexedTriList.GetCount());
+    UINT newCount = Count + uCount;
+
+    if (newCount > m_rgVerticesNonIndexedTriList.GetCapacity())
+    {
+        IFC(m_rgVerticesNonIndexedTriList.ReserveSpace(uCount));
+    }
+
+    m_rgVerticesNonIndexedTriList.SetCount(newCount);
+    *ppVertices = &m_rgVerticesNonIndexedTriList[Count];
+
+Cleanup:
+    RRETURN(hr);
+}
+
+//+----------------------------------------------------------------------------
+//
+//  Member:    CHwTVertexBuffer<TVertex>::AddLineListVertices
+//
+//  Synopsis:  Reserve space for consecutive vertices
+//
+
+template <class TVertex>
+MIL_FORCEINLINE
+HRESULT
+CHwTVertexBuffer<TVertex>::AddLineListVertices(
+    UINT uCount,
+    __deref_ecount(uCount) TVertex **ppVertices
+    )
+{
+    HRESULT hr = S_OK;
+
+    Assert(ppVertices);
+
+    UINT Count = static_cast<UINT>(m_rgVerticesLineList.GetCount());
+    UINT newCount = Count + uCount;
+
+    if (newCount > m_rgVerticesLineList.GetCapacity())
+    {
+        IFC(m_rgVerticesLineList.ReserveSpace(uCount));
+    }
+
+    m_rgVerticesLineList.SetCount(newCount);
+    *ppVertices = &m_rgVerticesLineList[Count];
+
+Cleanup:
+    RRETURN(hr);
+}
+
+//+----------------------------------------------------------------------------
+//
+//  Class:     CHwVertexBuffer::Builder
+//
+//-----------------------------------------------------------------------------
+
+//+----------------------------------------------------------------------------
+//
+//  Member:    CHwTVertexBuffer<TVertex>::Builder::GetOutVertexFormat
+//
+//  Synopsis:  Return MIL vertex format covered by specific builders
+//
+//-----------------------------------------------------------------------------
+
+template <>
+MilVertexFormat
+CHwTVertexBuffer<CD3DVertexXYZDUV2>::Builder::GetOutVertexFormat()
+{
+    return (MILVFAttrXYZ | MILVFAttrDiffuse | MILVFAttrUV2);
+}
+
+template <>
+MilVertexFormat
+CHwTVertexBuffer<CD3DVertexXYZDUV8>::Builder::GetOutVertexFormat()
+{
+    return (MILVFAttrXYZ | MILVFAttrDiffuse | MILVFAttrUV8);
+}
+
+template <>
+MilVertexFormat
+CHwTVertexBuffer<CD3DVertexXYZDUV6>::Builder::GetOutVertexFormat()
+{
+    return (MILVFAttrXYZ | MILVFAttrDiffuse | MILVFAttrUV6);
+}
+
+template <>
+MilVertexFormat
+CHwTVertexBuffer<CD3DVertexXYZNDSUV4>::Builder::GetOutVertexFormat()
+{
+    return (MILVFAttrXYZ |
+            MILVFAttrNormal |
+            MILVFAttrDiffuse |
+            MILVFAttrSpecular |
+            MILVFAttrUV4);
+}
+
+
+//+----------------------------------------------------------------------------
+//
+//  Member:    CHwVertexBuffer::Builder::Create
+//
+//  Synopsis:  Choose the appropriate final vertex format and instantiate the
+//             matching vertex builder
+//
+*/
+pub type CHwVertexBufferBuilder<'y, 'z> = CHwTVertexBufferBuilder<'y, 'z, OutputVertex>;
+impl<'y, 'z> CHwVertexBufferBuilder<'y, 'z> {
+pub fn Create(
+     vfIn: MilVertexFormat,
+     vfOut: MilVertexFormat,
+     mvfaAntiAliasScaleLocation: MilVertexFormatAttribute,
+    pVertexBuffer: &'y mut CHwVertexBuffer<'z>,
+    /*pBufferDispenser: &CBufferDispenser*/
+    ) -> CHwVertexBufferBuilder<'y, 'z>
+{
+    CHwVertexBufferBuilder::CreateTemplate(pVertexBuffer, vfIn, vfOut, mvfaAntiAliasScaleLocation)
+    //let hr: HRESULT = S_OK;
+
+    //assert!(ppVertexBufferBuilder);
+
+    //*ppVertexBufferBuilder = None;
+/* 
+    if (!(vfOut & ~CHwTVertexBuffer<CD3DVertexXYZDUV2>::Builder::GetOutVertexFormat()))
+    {
+        CHwTVertexBuffer<CD3DVertexXYZDUV2> *pVB = pDevice->GetVB_XYZDUV2();
+        CHwTVertexBuffer<CD3DVertexXYZDUV2>::Builder *pVBB = NULL;
+
+        IFC(CHwTVertexBuffer<CD3DVertexXYZDUV2>::Builder::Create(
+            pVB,
+            vfIn,
+            vfOut,
+            mvfaAntiAliasScaleLocation,
+            pBufferDispenser,
+            &pVBB
+            ));
+        
+        *ppVertexBufferBuilder = pVBB;
+    }
+    else if (!(vfOut & ~CHwTVertexBuffer<CD3DVertexXYZDUV8>::Builder::GetOutVertexFormat()))
+    {
+        CHwTVertexBuffer<CD3DVertexXYZDUV8> *pVB = pDevice->GetVB_XYZRHWDUV8();
+        CHwTVertexBuffer<CD3DVertexXYZDUV8>::Builder *pVBB = NULL;
+
+        IFC(CHwTVertexBuffer<CD3DVertexXYZDUV8>::Builder::Create(
+            pVB,
+            vfIn,
+            vfOut,
+            mvfaAntiAliasScaleLocation,
+            pBufferDispenser,
+            &pVBB
+            ));
+
+        *ppVertexBufferBuilder = pVBB;
+    }
+    else
+    {
+        // NOTE-2004/03/22-chrisra Adding another vertexbuffer type requires updating enum
+        //
+        // If we add another buffer builder type kMaxVertexBuilderSize enum in hwvertexbuffer.h file
+        // needs to be updated to reflect possible changes to the maximum size of buffer builders.
+        //
+        IFC(E_NOTIMPL);
+    }
+
+    // Store the pipeline, if any, which this VBB can use to spill the vertex buffer to if it
+    // overflows.
+    (**ppVertexBufferBuilder).m_pPipelineNoRef = pPipeline;
+    (**ppVertexBufferBuilder).m_pDeviceNoRef = pDevice;
+
+
+Cleanup:
+    RRETURN(hr);*/
+    //hr
+}
+    /*fn AreWafffling(&self) -> bool {
+        false
+    }*/
+
+        // Helpful m_rcOutsideBounds casts.
+        fn OutsideLeft(&self) -> f32  { return self.m_rcOutsideBounds.left as f32; }
+        fn OutsideRight(&self) -> f32 { return self.m_rcOutsideBounds.right as f32; }
+        fn OutsideTop(&self) -> f32 { return self.m_rcOutsideBounds.top as f32; }
+        fn OutsideBottom(&self) -> f32 { return self.m_rcOutsideBounds.bottom as f32; }
+}
+
+//+----------------------------------------------------------------------------
+//
+//  Class:     THwTVertexMappings<class TVertex>
+//
+//-----------------------------------------------------------------------------
+
+//+----------------------------------------------------------------------------
+//
+//  Member:    THwTVertexMappings<TVertex>::THwTVertexMappings
+//
+//  Synopsis:  ctor
+//
+//-----------------------------------------------------------------------------
+/* 
+template <class TVertex>
+CHwTVertexMappings<TVertex>::CHwTVertexMappings()
+ :
+    m_mvfMapped(MILVFAttrNone)
+{
+    for (int i = 0; i < ARRAY_SIZE(m_rgWaffleMode); ++i)
+    {
+        m_rgWaffleMode[i] = WaffleModeNone;
+    }
+
+    m_matPos2DTransform.SetIdentity();
+}
+
+
+//+----------------------------------------------------------------------------
+//
+//  Member:    THwTVertexMappings<TVertex>::SetPositionTransform
+//
+//  Synopsis:  Sets the position transform that needs to be applied.
+//
+//-----------------------------------------------------------------------------
+template <class TVertex>
+void 
+CHwTVertexMappings<TVertex>::SetPositionTransform(
+    __in_ecount(1) const MILMatrix3x2 &matPositionTransform
+    )
+{
+    m_matPos2DTransform = matPositionTransform;
+}
+
+//+----------------------------------------------------------------------------
+//
+//  Member:    CHwTVertexMappings<TVertex>::SetConstantMapping
+//
+//  Synopsis:  Remember the static color for the given vertex field
+//
+
+template <class TVertex>
+HRESULT
+CHwTVertexMappings<TVertex>::SetConstantMapping(
+    MilVertexFormatAttribute mvfaLocation,
+    __in_ecount(1) const CHwConstantColorSource *pConstCS
+    )
+{
+    HRESULT hr = S_OK;
+
+    Assert(!(m_mvfMapped & mvfaLocation));
+    pConstCS->GetColor(m_colorStatic);
+    m_mvfMapped |= mvfaLocation;    // Remember this field has been mapped
+
+    RRETURN(hr);
+}
+
+//+----------------------------------------------------------------------------
+//
+//  Function:  GetMILVFAttributeOfTextureCoord
+//
+//  Synopsis:  Compute MilVertexFormatAttribute for a texture coordinate index
+//
+
+MIL_FORCEINLINE
+MilVertexFormat
+GetMILVFAttributeOfTextureCoord(
+    DWORD dwCoordIndex
+    )
+{
+    return MILVFAttrUV1 << dwCoordIndex;
+}
+
+//+----------------------------------------------------------------------------
+//
+//  Member:    CHwTVertexMappings<TVertex>::PointToUV
+//
+//  Synopsis:  Helper function to populate the texture coordinates at the given
+//             index using the given point
+//
+
+template <class TVertex>
+MIL_FORCEINLINE void
+CHwTVertexMappings<TVertex>::PointToUV(
+    __in_ecount(1) const MilPoint2F &ptIn,
+    __bound UINT uIndex,
+    __out_ecount(1) TVertex *pvOut
+    )
+{
+    m_rgmatPointToUV[uIndex].TransformPoint(
+        &pvOut->ptTx[uIndex],
+        ptIn.X,
+        ptIn.Y
+        );
+}
+
+
+
+
+
+//+----------------------------------------------------------------------------
+//
+//  Class:     CHwTVertexBuffer<TVertex>::Builder
+//
+//-----------------------------------------------------------------------------
+
+
+*/
+
+impl<'y, 'z, TVertex: Default> CHwTVertexBufferBuilder<'y, 'z, TVertex> {
+
+//+----------------------------------------------------------------------------
+//
+//  Member:    CHwTVertexBuffer<TVertex>::Builder::Create
+//
+//  Synopsis:  Instantiate a specific type of vertex builder
+//
+
+fn CreateTemplate(
+     pVertexBuffer: &'y mut CHwTVertexBuffer<'z, TVertex>,
+     mvfIn: MilVertexFormat,
+     mvfOut: MilVertexFormat,
+     mvfaAntiAliasScaleLocation: MilVertexFormatAttribute,
+     /*pBufferDispenser: __inout_ecount(1) CBufferDispenser *,*/
+    ) -> Self
+{
+
+
+
+    let mut pVertexBufferBuilder = CHwTVertexBufferBuilder::<TVertex>::new(pVertexBuffer);
+
+    IFC!(pVertexBufferBuilder.SetupConverter(
+        mvfIn,
+        mvfOut,
+        mvfaAntiAliasScaleLocation
+        ));
+
+    return pVertexBufferBuilder;
+}
+
+//+----------------------------------------------------------------------------
+//
+//  Member:    CHwTVertexBuffer<TVertex>::Builder::Builder
+//
+//  Synopsis:  ctor
+//
+//-----------------------------------------------------------------------------
+
+fn new(pVertexBuffer: &'y mut CHwTVertexBuffer<'z, TVertex>) -> Self
+{
+    Self {
+    m_pVB: pVertexBuffer,
+
+
+    //m_rgoPrecomputedTriListVertices: NULL(),
+    //m_cPrecomputedTriListVertices: 0,
+
+    //m_rguPrecomputedTriListIndices: NULL(),
+    //m_cPrecomputedTriListIndices: 0,
+
+    // These two track the Y extent of the shape this builder is producing.
+    m_rCurStratumTop: f32::MAX,
+    m_rCurStratumBottom:  -f32::MAX,
+    m_fNeedOutsideGeometry: false,
+    m_fNeedInsideGeometry: true,
+
+    m_rLastTrapezoidRight: -f32::MAX,
+    m_rLastTrapezoidTopRight: -f32::MAX,
+    m_rLastTrapezoidBottomRight: -f32::MAX,
+
+    m_fHasFlushed: false,
+    //m_map: Default::default(),
+    m_rcOutsideBounds: Default::default(),
+        #[cfg(debug_assertions)]
+        m_mvfDbgOut: MilVertexFormatAttribute::MILVFAttrNone as MilVertexFormat,
+        m_mvfIn: MilVertexFormatAttribute::MILVFAttrNone as MilVertexFormat,
+        m_mvfGenerated: MilVertexFormatAttribute::MILVFAttrNone  as MilVertexFormat,
+        m_mvfaAntiAliasScaleLocation: MilVertexFormatAttribute::MILVFAttrNone,
+    }
+}
+
+//+----------------------------------------------------------------------------
+//
+//  Member:    CHwTVertexBuffer<TVertex>::Builder::SetupConverter
+//
+//  Synopsis:  Choose the appropriate conversion method
+//
+
+fn SetupConverter(&mut self,
+     mvfIn: MilVertexFormat,
+     mvfOut: MilVertexFormat,
+     mvfaAntiAliasScaleLocation: MilVertexFormatAttribute,
+     ) -> HRESULT
+{
+    let hr = S_OK;
+
+    self.m_mvfIn = mvfIn;
+
+    #[cfg(debug_assertions)]
+    {
+    self.m_mvfDbgOut = mvfOut;
+    }
+
+    self.m_mvfGenerated = mvfOut & !self.m_mvfIn;
+    self.m_mvfaAntiAliasScaleLocation = mvfaAntiAliasScaleLocation;
+
+    assert!((self.m_mvfGenerated & MilVertexFormatAttribute::MILVFAttrXY as MilVertexFormat) == 0);
+
+    RRETURN!(hr);
+}
+}
+/* 
+
+//+----------------------------------------------------------------------------
+//
+//  Member:    CHwTVertexBuffer<TVertex>::Builder::SetTransformMapping
+//
+//  Synopsis:  Delegate mapping sets to CHwTVertexMappings
+//
+//-----------------------------------------------------------------------------
+
+template <class TVertex>
+void
+CHwTVertexBuffer<TVertex>::Builder::SetTransformMapping(
+    __in_ecount(1) const MILMatrix3x2 &mat2DPositionTransform
+    )
+{
+    m_map.SetPositionTransform(mat2DPositionTransform);
+}
+                                                                    
+template <class TVertex>
+HRESULT
+CHwTVertexBuffer<TVertex>::Builder::SetConstantMapping(
+    MilVertexFormatAttribute mvfaLocation,
+    __in_ecount(1) const CHwConstantColorSource *pConstCS
+    )
+{
+    HRESULT hr = S_OK;
+
+    IFC(m_map.SetConstantMapping(mvfaLocation, pConstCS));
+
+Cleanup:
+    RRETURN(hr);
+}
+
+//+----------------------------------------------------------------------------
+//
+//  Member:    CHwTVertexBuffer<TVertex>::Builder::FinalizeMappings
+//
+//  Synopsis:  Complete setup of vertex mappings
+//
+
+template <class TVertex>
+HRESULT
+CHwTVertexBuffer<TVertex>::Builder::FinalizeMappings(
+    )
+{
+    HRESULT hr = S_OK;
+
+    //
+    // Set default Z if required.
+    //
+
+    if (m_mvfGenerated & MILVFAttrZ)
+    {
+        if (!(m_map.m_mvfMapped & MILVFAttrZ))
+        {
+            m_map.m_vStatic.Z = 0.5f;
+        }
+    }
+
+    //
+    // If AA falloff is not going to scale the diffuse color and it is
+    // generated then see if the color is constant such that we can do any
+    // complex conversions just once here instead of in every iteration of the
+    // expansion loop.  If AA falloff is going to scale the diffuse color then
+    // we can still optimize for the falloff = 1.0 case by precomputing that
+    // color now and checking for 1.0 during generation.  Such a precomputation
+    // has shown significant to performance.
+    //
+
+    if (m_mvfGenerated & MILVFAttrDiffuse)
+    {
+        if (m_map.m_mvfMapped & MILVFAttrDiffuse)
+        {
+
+            // Assumes diffuse color is constant
+            m_map.m_vStatic.Diffuse =
+                Convert_MilColorF_scRGB_To_Premultiplied_MilColorB_sRGB(&m_map.m_colorStatic);
+        }
+        else
+        {
+            // Set default Diffuse value: White
+            m_map.m_vStatic.Diffuse = MIL_COLOR(0xFF,0xFF,0xFF,0xFF);
+        }
+    }
+
+    RRETURN(hr);
+}*/
+impl<TVertex> CHwTVertexBufferBuilder<'_, '_, TVertex> {
+
+//+----------------------------------------------------------------------------
+//
+//  Member:    CHwTVertexBuffer<TVertex>::Builder::SetOutsideBounds
+//
+//
+//  Synopsis:  Enables rendering geometry for areas outside the shape but
+//             within the bounds.  These areas will be created with
+//             zero alpha.
+//
+
+pub fn SetOutsideBounds(&mut self,
+    prcOutsideBounds: Option<&CMILSurfaceRect>,
+    fNeedInside: bool,
+    )
+{
+    // Waffling and outside bounds is not currently implemented.  It's
+    // not difficult to do but currently there is no need.
+    //assert!(!(self.AreWaffling() && self.prcOutsideBounds));
+
+    if let Some(prcOutsideBounds) = prcOutsideBounds
+    {
+        self.m_rcOutsideBounds = prcOutsideBounds.clone();
+        self.m_fNeedOutsideGeometry = true;
+        self.m_fNeedInsideGeometry = fNeedInside;
+    }
+    else
+    {
+        self.m_fNeedOutsideGeometry = false;
+        self.m_fNeedInsideGeometry = true;
+    }
+}
+
+//+----------------------------------------------------------------------------
+//
+//  Member:    CHwTVertexBuffer<TVertex>::Builder::BeginBuilding
+//
+//  Synopsis:  Prepare for a new primitive by resetting the vertex buffer
+//
+pub fn BeginBuilding(&mut self,
+    ) -> HRESULT
+{
+    
+    let hr: HRESULT = S_OK;
+
+    self.m_fHasFlushed = false;
+    self.m_pVB.Reset(/*self*/);
+
+    RRETURN!(hr);
+}
+}
+impl IGeometrySink for CHwVertexBufferBuilder<'_, '_> {
+
+    fn AddTrapezoid(&mut self,
+        rPixelYTop: f32,              // In: y coordinate of top of trapezoid
+        rPixelXTopLeft: f32,          // In: x coordinate for top left
+        rPixelXTopRight: f32,         // In: x coordinate for top right
+        rPixelYBottom: f32,           // In: y coordinate of bottom of trapezoid
+        rPixelXBottomLeft: f32,       // In: x coordinate for bottom left
+        rPixelXBottomRight: f32,      // In: x coordinate for bottom right
+        rPixelXLeftDelta: f32,        // In: trapezoid expand radius for left edge
+        rPixelXRightDelta: f32        // In: trapezoid expand radius for right edge
+        ) -> HRESULT
+    {
+        let hr = S_OK;
+    
+        if (/*self.AreWaffling()*/ false)
+        {
+            /*IFC(AddTrapezoidWaffle(
+                    rPixelYTop,
+                    rPixelXTopLeft,
+                    rPixelXTopRight,
+                    rPixelYBottom,
+                    rPixelXBottomLeft,
+                    rPixelXBottomRight,
+                    rPixelXLeftDelta,
+                    rPixelXRightDelta));*/
+        }
+        else
+        {
+            IFC!(self.AddTrapezoidStandard(
+                    rPixelYTop,
+                    rPixelXTopLeft,
+                    rPixelXTopRight,
+                    rPixelYBottom,
+                    rPixelXBottomLeft,
+                    rPixelXBottomRight,
+                    rPixelXLeftDelta,
+                    rPixelXRightDelta));
+        }
+    
+    //Cleanup:
+        RRETURN!(hr);
+    }
+    
+
+    fn IsEmpty(&self) -> bool {
+        self.m_pVB.IsEmpty()
+    }
+
+/* 
+
+//+----------------------------------------------------------------------------
+//
+//  Member:    CHwTVertexBuffer<TVertex>::Builder::AddVertex
+//
+//  Synopsis:  Add a vertex to the vertex buffer
+//
+//             Remember just the given vertex information now and convert later
+//             in a single, more optimal pass.
+//
+
+template <class TVertex>
+HRESULT
+CHwTVertexBuffer<TVertex>::Builder::AddVertex(
+    __in_ecount(1) const MilPoint2F &ptPosition,
+        // Vertex coordinates
+    __out_ecount(1) WORD *pIndex
+        // The index of the new vertex
+    )
+{
+    HRESULT hr = S_OK;
+
+    Assert(!NeedOutsideGeometry());
+    Assert(m_mvfIn == MILVFAttrXY);
+
+    TVertex *pVertex;
+
+    IFC(m_pVB->AddTriListVertices(1, &pVertex, pIndex));
+
+    pVertex->ptPt = ptPosition;
+
+    //  store coverage as a DWORD instead of float
+
+    pVertex->Diffuse = FLOAT_ONE;
+
+Cleanup:
+    RRETURN(hr);
+}
+
+//+----------------------------------------------------------------------------
+//
+//  Member:    CHwTVertexBuffer<TVertex>::Builder::AddIndexedVertices, IGeometrySink
+//
+//  Synopsis:  Add a fully computed, indexed vertex to the vertex buffer
+//
+
+template <class TVertex>
+HRESULT
+CHwTVertexBuffer<TVertex>::Builder::AddIndexedVertices(
+    UINT cVertices,
+        // In: number of vertices                                                       
+    __in_bcount(cVertices*uVertexStride) const void *pVertexBufferNoRef,
+        // In: vertex buffer containing the vertices                                    
+    UINT uVertexStride,
+        // In: size of each vertex                                                      
+    MilVertexFormat mvfFormat,
+        // In: format of each vertex                                                    
+    UINT cIndices,
+        // In: Number of indices                                                        
+    __in_ecount(cIndices) const UINT *puIndexBuffer
+        // In: index buffer                                                             
+    )
+{
+    Assert(m_mvfIn & (MILVFAttrXYZ | MILVFAttrDiffuse | MILVFAttrUV2));
+    Assert(mvfFormat == (MILVFAttrXYZ | MILVFAttrDiffuse | MILVFAttrUV2));
+
+    Assert(uVertexStride == sizeof(TVertex));
+
+    m_rgoPrecomputedTriListVertices = reinterpret_cast<const TVertex *>(pVertexBufferNoRef);
+    m_cPrecomputedTriListVertices = cVertices;
+
+    m_rguPrecomputedTriListIndices = puIndexBuffer;
+    m_cPrecomputedTriListIndices = cIndices;
+
+    return S_OK;
+}
+
+//+----------------------------------------------------------------------------
+//
+//  Member:    CHwTVertexBuffer<TVertex>::Builder::AddTriangle
+//
+//  Synopsis:  Add a triangle to the vertex buffer
+//
+
+template <class TVertex>
+HRESULT
+CHwTVertexBuffer<TVertex>::Builder::AddTriangle(
+    DWORD i1,                    // In: Index of triangle's first vertex
+    DWORD i2,                    // In: Index of triangle's second vertex
+    DWORD i3                     // In: Index of triangle's third vertex
+    )
+{
+    HRESULT hr = S_OK;
+
+    Assert(!NeedOutsideGeometry());
+
+    if (AreWaffling())
+    {
+        TVertex *pVertex;
+        UINT uNumVertices;
+        m_pVB->GetTriListVertices(&pVertex, &uNumVertices);
+
+        Assert(i1 < uNumVertices);
+        Assert(i2 < uNumVertices);
+        Assert(i3 < uNumVertices);
+
+        PointXYA rgPoints[3];
+        rgPoints[0].x = pVertex[i1].ptPt.X;
+        rgPoints[0].y = pVertex[i1].ptPt.Y;
+        rgPoints[0].a = 1;
+        rgPoints[1].x = pVertex[i2].ptPt.X;
+        rgPoints[1].y = pVertex[i2].ptPt.Y;
+        rgPoints[1].a = 1;
+        rgPoints[2].x = pVertex[i3].ptPt.X;
+        rgPoints[2].y = pVertex[i3].ptPt.Y;
+        rgPoints[2].a = 1;
+        
+        TriangleWaffler<PointXYA> wafflers[NUM_OF_VERTEX_TEXTURE_COORDS(TVertex) * 2];
+        TriangleWaffler<PointXYA>::ISink *pWaffleSinkNoRef = BuildWafflePipeline(wafflers);
+        IFC(pWaffleSinkNoRef->AddTriangle(rgPoints[0], rgPoints[1], rgPoints[2]));
+    }
+    else
+    {
+        IFC(m_pVB->AddTriangle(
+                static_cast<WORD>(i1),
+                static_cast<WORD>(i2),
+                static_cast<WORD>(i3)
+                ));
+    }
+    
+Cleanup:
+    RRETURN(hr);
+}
+*/
+
+
+//+----------------------------------------------------------------------------
+//
+//  Member:    CHwTVertexBuffer<TVertex>::Builder::AddComplexScan
+//
+//  Synopsis:  Add a coverage span to the vertex buffer
+//
+//-----------------------------------------------------------------------------
+    fn AddComplexScan(&mut self,
+        nPixelY: INT,
+            // In: y coordinate in pixel space
+            mut pIntervalSpanStart: Ref<crate::aacoverage::CCoverageInterval>
+            // In: coverage segments
+        ) -> HRESULT {
+
+    let hr: HRESULT = S_OK;
+    //let pVertex: *mut CD3DVertexXYZDUV2 = NULL();
+
+    IFC!(self.PrepareStratum((nPixelY) as f32,
+                  (nPixelY+1) as f32, 
+                  false, /* Not a trapezoid. */ 
+                  0., 0.,
+                0., 0., 0., 0.));
+
+    let rPixelY: f32;
+    rPixelY = (nPixelY) as f32 + 0.5;
+
+    //LineWaffler<PointXYA> wafflers[NUM_OF_VERTEX_TEXTURE_COORDS(TVertex) * 2];
+
+    // Use sink for waffling & the first line fix up (aka the complicated cases.)
+    //ILineSink<PointXYA> *pLineSink = NULL;
+    let mut pLineSink = None;
+
+    /*if (self.AreWaffling())
+    {
+        bool fWafflersUsed;
+        pLineSink = BuildWafflePipeline(wafflers, OUT fWafflersUsed);
+        if (!fWafflersUsed)
+        {
+            pLineSink = NULL;
+        }
+    }*/
+    
+    // Use triangles instead of lines, for lines too close to the top of the viewport
+    // because lines are clipped (before rasterization) against a viewport that only
+    // includes half of the top pixel row.  Waffling will take care of this separately.
+    if (/*pLineSink.is_none() && rPixelY < self.GetViewportTop() + 1 ||*/ FORCE_TRIANGLES)
+    {
+        pLineSink = Some(&mut self.m_pVB);
+    }
+
+    //
+    // Output all segments if creating outside geometry, otherwise only output segments
+    // with non-zero coverage.
+    //
+
+    if (pLineSink.is_none())
+    {
+        /* 
+        UINT nSegmentCount = 0;
+
+        for (const CCoverageInterval *pIntervalSpanTemp = pIntervalSpanStart;
+             pIntervalSpanTemp->m_nPixelX != INT_MAX;
+             pIntervalSpanTemp = pIntervalSpanTemp->m_pNext
+             )
+        {
+            if (NeedCoverageGeometry(pIntervalSpanTemp->m_nCoverage))
+            {
+                ++nSegmentCount;
+            }
+        }
+
+        //
+        // Add vertices
+        //
+        if (nSegmentCount)
+        {
+            IFC(m_pVB->AddLineListVertices(nSegmentCount*2, &pVertex));
+        }*/
+    }
+
+    //
+    // Having allocated space (if not using sink), now let's actually output the vertices.
+    //
+
+    while ((*pIntervalSpanStart).m_nPixelX.get() != INT::MAX)
+    {
+        assert!(!(*pIntervalSpanStart).m_pNext.get().is_null());
+
+        //
+        // Output line list segments
+        //
+        // Note that line segments light pixels by going through the the
+        // "diamond" interior of a pixel.  While we could accomplish this
+        // by going from left edge to right edge of pixel, D3D10 uses the
+        // convention that the LASTPIXEL is never lit.  We respect that now
+        // by setting D3DRS_LASTPIXEL to FALSE and use line segments that
+        // start in center of first pixel and end in center of one pixel
+        // beyond last.
+        //
+        // Since our top left corner is integer, we add 0.5 to get to the
+        // pixel center.
+        //
+        if (self.NeedCoverageGeometry((*pIntervalSpanStart).m_nCoverage.get()))
+        {
+            let rCoverage: f32 = ((*pIntervalSpanStart).m_nCoverage.get() as f32)/(c_nShiftSizeSquared as f32);
+            
+            let mut iBegin: LONG = (*pIntervalSpanStart).m_nPixelX.get();
+            let mut iEnd: LONG = (*(*pIntervalSpanStart).m_pNext.get()).m_nPixelX.get();
+            if (self.NeedOutsideGeometry())
+            {
+                // Intersect the interval with the outside bounds to create
+                // start and stop lines.  The scan begins (ends) with an
+                // interval starting (ending) at -inf (+inf).
+
+                // The given geometry is not guaranteed to be within m_rcOutsideBounds but
+                // the additional inner min and max (in that order) produce empty spans
+                // for intervals not intersecting m_rcOutsideBounds.
+                //
+                // We could cull here but that should really be done by the geometry
+                // generator.
+
+                iBegin = iBegin.max(iEnd.min(self.m_rcOutsideBounds.left));
+                iEnd = iEnd.min(iBegin.max(self.m_rcOutsideBounds.right));
+            }
+            let rPixelXBegin: f32= (iBegin as f32) + 0.5;
+            let rPixelXEnd: f32 = (iEnd as f32) + 0.5;
+
+            //
+            // Output line (linelist or tristrip) for a pixel
+            //
+
+            //if let Some(pLineSink) = pLineSink 
+            {
+                let mut v0: PointXYA = Default::default(); let mut v1: PointXYA = Default::default();
+                v0.x = rPixelXBegin;
+                v0.y = rPixelY;
+                v0.a = rCoverage;
+
+                v1.x = rPixelXEnd;
+                v1.y = rPixelY;
+                v1.a = rCoverage;
+
+                IFC!(self.m_pVB.AddLine(&v0,&v1));
+            }
+            //else
+            {
+                /* 
+                let dwDiffuse = ReinterpretFloatAsDWORD(rCoverage);
+
+                pVertex[0].ptPt.X = rPixelXBegin;
+                pVertex[0].ptPt.Y = rPixelY;
+                pVertex[0].Diffuse = dwDiffuse;
+
+                pVertex[1].ptPt.X = rPixelXEnd;
+                pVertex[1].ptPt.Y = rPixelY;
+                pVertex[1].Diffuse = dwDiffuse;
+
+                // Advance output vertex pointer
+                pVertex += 2;*/
+            }
+        }
+
+        //
+        // Advance coverage buffer
+        //
+
+        pIntervalSpanStart = (*pIntervalSpanStart).m_pNext.get();
+    }
+
+
+//Cleanup:
+    RRETURN!(hr);
+
+}
+}
+
+//+----------------------------------------------------------------------------
+//
+//  Member:    CHwTVertexBuffer<TVertex>::Builder::AddLineAsTriangleList
+//
+//  Synopsis:  Adds a horizontal line as a triangle list to work around
+//             issue in D3D9 where horizontal lines with y = 0 may not render.
+//
+//              Line clipping in D3D9
+//             This behavior will change in D3D10 and this work-around will no
+//             longer be needed.  (Pixel center conventions will also change.)
+//              
+//-----------------------------------------------------------------------------
+impl CHwVertexBuffer<'_> {
+    fn AddLineAsTriangleList(&mut self,
+    pBegin: &CD3DVertexXYZDUV2, // Begin
+    pEnd: &CD3DVertexXYZDUV2    // End
+    ) -> HRESULT
+{
+    let hr = S_OK;
+
+    // Collect pertinent data from vertices.
+    debug_assert!(pBegin.y == pEnd.y);
+    debug_assert!(pBegin.coverage == pEnd.coverage);
+
+    // Offset begin and end X left by 0.5 because the line starts on the first
+    // pixel center and ends on the center of the pixel after the line segment.
+    let x0 = pBegin.x - 0.5;
+    let x1 = pEnd.x - 0.5;
+    let y = pBegin.y;
+    let dwDiffuse = pBegin.coverage;
+
+    //
+    // Add the vertices
+    //
+
+    // OpenGL doesn't specify how vertex positions are converted to fixed point prior to rasterization. On macOS, with AMD GPUs,
+    // the GPU appears to truncate to fixed point instead of rounding. This behaviour is controlled by PA_SU_VTX_CNTL
+    // register. To handle this we'll add a 1./512. subpixel bias to the center vertex to cause the coordinates to round instead
+    // of truncate.
+    //
+    // D3D11 requires the fixed point integer result to be within 0.6ULP which implicitly disallows the truncate behaviour above.
+    // This means that D2D doesn't need to deal with this problem.
+    let subpixel_bias = self.subpixel_bias;
+
+
+    // Use a single triangle to cover the entire line
+    self.AddTriVertices(
+        OutputVertex{ x: x0, y: y - 0.5, coverage: dwDiffuse },
+        OutputVertex{ x: x0, y: y + 0.5, coverage: dwDiffuse },
+        OutputVertex{ x: x1, y: y + subpixel_bias, coverage: dwDiffuse },
+    );
+
+    self.AddedNonLineSegment();
+
+  //Cleanup:
+    RRETURN!(hr);
+}
+}
+
+/* 
+//+----------------------------------------------------------------------------
+//
+//  Member:    CHwTVertexBuffer<TVertex>::Builder::AddParallelogram
+//
+//  Synopsis:  This function adds the coordinates of a parallelogram to the vertex strip buffer. 
+//
+//  Parameter: rgPosition contains four coordinates of the parallelogram. Coordinates should have 
+//              a winding order
+//
+//-----------------------------------------------------------------------------
+template <class TVertex>
+HRESULT
+CHwTVertexBuffer<TVertex>::Builder::AddParallelogram(
+        __in_ecount(4)  const MilPoint2F *rgPosition
+        )
+{
+    HRESULT hr = S_OK;
+
+    if (AreWaffling())
+    {
+        PointXYA rgPoints[4];
+        for (int i = 0; i < 4; ++i)
+        {
+            rgPoints[i].x = rgPosition[i].X;
+            rgPoints[i].y = rgPosition[i].Y;
+            rgPoints[i].a = 1;
+        }
+        TriangleWaffler<PointXYA> wafflers[NUM_OF_VERTEX_TEXTURE_COORDS(TVertex) * 2];
+        TriangleWaffler<PointXYA>::ISink *pWaffleSinkNoRef = BuildWafflePipeline(wafflers);
+        IFC(pWaffleSinkNoRef->AddTriangle(rgPoints[0], rgPoints[1], rgPoints[3]));
+        IFC(pWaffleSinkNoRef->AddTriangle(rgPoints[3], rgPoints[1], rgPoints[2]));
+    }
+    else
+    {
+        TVertex *pVertex;
+  
+        //
+        // Add the vertices
+        //
+
+        IFC(m_pVB->AddTriStripVertices(6, &pVertex));
+
+        //
+        // Duplicate the first vertex. This creates 2 degenerate triangles: one connecting
+        // the previous rect to this one and another between vertices 0 and 1.
+        //
+
+        pVertex[0].ptPt = rgPosition[0];
+        pVertex[0].Diffuse = FLOAT_ONE;
+
+        pVertex[1].ptPt = rgPosition[0];
+        pVertex[1].Diffuse = FLOAT_ONE;
+    
+        pVertex[2].ptPt = rgPosition[1];
+        pVertex[2].Diffuse = FLOAT_ONE;
+
+        pVertex[3].ptPt = rgPosition[3];
+        pVertex[3].Diffuse = FLOAT_ONE;
+
+        pVertex[4].ptPt = rgPosition[2];
+        pVertex[4].Diffuse = FLOAT_ONE;
+
+        //
+        // Duplicate the last vertex. This creates 2 degenerate triangles: one
+        // between vertices 4 and 5 and one connecting this Rect to the
+        // next one.
+        //
+
+        pVertex[5].ptPt = rgPosition[2];
+        pVertex[5].Diffuse = FLOAT_ONE;
+    }
+        
+  Cleanup:
+    RRETURN(hr);
+}
+    
+//+----------------------------------------------------------------------------
+//
+//  Member:    CHwTVertexBuffer<TVertex>::Builder::BuildWafflePipeline<TWaffler>
+//
+//  Synopsis:  Builds a pipeline of wafflers into the provided array of wafflers.
+//             And returns a pointer (not to be deleted) to the input sink
+//             of the waffle pipeline.
+//             the final result is sinked int m_pVB.
+//
+//-----------------------------------------------------------------------------
+
+template<class TVertex>
+template<class TWaffler>
+__out_ecount(1) typename TWaffler::ISink *
+CHwTVertexBuffer<TVertex>::Builder::BuildWafflePipeline(
+        __out_xcount(NUM_OF_VERTEX_TEXTURE_COORDS(TVertex) * 2) TWaffler *wafflers,
+        __out_ecount(1) bool &fWafflersUsed
+    ) const
+{
+    UINT count = 0;
+
+    for (int i = 0; i < NUM_OF_VERTEX_TEXTURE_COORDS(TVertex); ++i)
+    {
+        if (m_map.m_rgWaffleMode[i] != 0)
+        {
+            const MILMatrix3x2 &pMatWaffle = m_map.m_rgmatPointToUV[i];
+
+            // Each column ([a,b,c] transpose) of this matrix specifies a waffler that
+            // partitions the plane into regions between the lines:
+            //                    ax + by + c = k
+            // for every integer k.
+            //
+            // If this partition width is substantially less than a pixel we have
+            // serious problems with waffling generating too many triangles for
+            // doubtful visual effect so we don't perform a waffling with width less
+            // than c_rMinWaffleWidthPixels.  So we need to know the width of the partition
+            // regions:
+            //
+            // Changing c just translates the partition so let's assume c = 0.
+            // The line ax + by = 0 goes through the origin and the line ax + by
+            // = 1 is adjacent to it in the partition.  The distance between
+            // these lines is also the distance from ax + by = 1 to the origin.
+            // Using Lagrange multipliers we can determine that this distance
+            // is
+            //                     1/sqrt(a*a+b*b).
+            // We want to avoid waffling if this is less than c_rMinWaffleWidthPixels
+            // or equivalently:
+            //   1/sqrt(a*a+b*b) < c_rMinWaffleWidthPixels
+            //     sqrt(a*a+b*b) > 1/c_rMinWaffleWidthPixels
+            //          a*a+b*b  > 1/(c_rMinWaffleWidthPixels*c_rMinWaffleWidthPixels)
+            //          
+
+            const float c_rMaxWaffleMagnitude = 1/(c_rMinWaffleWidthPixels*c_rMinWaffleWidthPixels);
+            
+            float mag0 = pMatWaffle.m_00*pMatWaffle.m_00+pMatWaffle.m_10*pMatWaffle.m_10;
+            if (mag0 < c_rMaxWaffleMagnitude)
+            {
+                wafflers[count].Set(pMatWaffle.m_00, pMatWaffle.m_10, pMatWaffle.m_20, wafflers+count+1);
+                ++count;
+            }
+
+            float mag1 = pMatWaffle.m_01*pMatWaffle.m_01+pMatWaffle.m_11*pMatWaffle.m_11;
+            if (mag1 < c_rMaxWaffleMagnitude)
+            {
+                wafflers[count].Set(pMatWaffle.m_01, pMatWaffle.m_11, pMatWaffle.m_21, wafflers+count+1);
+                ++count;
+            }
+        }
+    }
+
+    if (count)
+    {
+        fWafflersUsed = true;
+        // As the last step in the chain we send the triangles to our vertex buffer.
+        wafflers[count-1].SetSink(m_pVB);
+        return &wafflers[0];
+    }
+    else
+    {
+        fWafflersUsed = false;
+        // If we built no wafflers then sink straight into the vertex buffer.
+        return m_pVB;
+    }
+}
+
+
+//+----------------------------------------------------------------------------
+//
+//  Member:    CHwTVertexBuffer<TVertex>::Builder::IsEmpty
+//
+//  Synopsis:  Does our VB have any triangles/lines?
+//
+//-----------------------------------------------------------------------------
+template <class TVertex>
+BOOL
+CHwTVertexBuffer<TVertex>::Builder::IsEmpty()
+{
+    return m_pVB->IsEmpty();
+}
+*/
+//+----------------------------------------------------------------------------
+//
+//  Member:    CHwTVertexBuffer<TVertex>::Builder::AddTrapezoid
+//
+//  Synopsis:  Add a trapezoid to the vertex buffer
+//
+//
+//      left edge       right edge
+//      ___+_________________+___      <<< top edge
+//     /  +  /             \  +  \
+//    /  +  /               \  +  \
+//   /  +  /                 \  +  \
+//  /__+__/___________________\__+__\  <<< bottom edge
+//    + ^^                        +
+//      delta
+//
+impl CHwVertexBufferBuilder<'_, '_> {
+
+//+----------------------------------------------------------------------------
+//
+//  Member:    CHwTVertexBuffer<TVertex>::Builder::AddTrapezoidStandard
+//
+//  Synopsis:  See AddTrapezoid.  This doesn't do waffling & uses tri strips.
+//
+
+fn AddTrapezoidStandard(&mut self,
+    rPixelYTop: f32,              // In: y coordinate of top of trapezoid
+    rPixelXTopLeft: f32,          // In: x coordinate for top left
+    rPixelXTopRight: f32,         // In: x coordinate for top right
+    rPixelYBottom: f32,           // In: y coordinate of bottom of trapezoid
+    rPixelXBottomLeft: f32,       // In: x coordinate for bottom left
+    rPixelXBottomRight: f32,      // In: x coordinate for bottom right
+    rPixelXLeftDelta: f32,        // In: trapezoid expand radius for left edge
+    rPixelXRightDelta: f32        // In: trapezoid expand radius for right edge
+    ) -> HRESULT
+{
+    type TVertex = CD3DVertexXYZDUV2;
+    let hr = S_OK;
+    //TVertex *pVertex;
+
+    IFC!(self.PrepareStratum(
+        rPixelYTop,
+        rPixelYBottom,
+        true, /* Trapezoid */
+        rPixelXTopLeft.min(rPixelXBottomLeft),
+        rPixelXTopRight.max(rPixelXBottomRight),
+        rPixelXTopLeft - rPixelXLeftDelta, rPixelXBottomLeft - rPixelXLeftDelta,
+        rPixelXTopRight + rPixelXRightDelta, rPixelXBottomRight + rPixelXRightDelta
+        ));
+    
+    //
+    // Add the vertices
+    //
+
+	let fNeedOutsideGeometry: bool; let fNeedInsideGeometry: bool;
+    fNeedOutsideGeometry = self.NeedOutsideGeometry();
+    fNeedInsideGeometry = self.NeedInsideGeometry();
+
+    //
+    // Fill in the vertices
+    //
+
+    self.m_pVB.AddTrapezoidVertices(
+        OutputVertex{
+            x: rPixelXTopLeft - rPixelXLeftDelta,
+            y: rPixelYTop,
+            coverage: FLOAT_ZERO,
+        },
+        OutputVertex{
+            x: rPixelXBottomLeft - rPixelXLeftDelta,
+            y: rPixelYBottom,
+            coverage: FLOAT_ZERO,
+        },
+        OutputVertex{
+            x: rPixelXTopLeft + rPixelXLeftDelta,
+            y: rPixelYTop,
+            coverage: FLOAT_ONE,
+        },
+        OutputVertex{
+            x: rPixelXBottomLeft + rPixelXLeftDelta,
+            y: rPixelYBottom,
+            coverage: FLOAT_ONE,
+        }
+    );
+
+
+    if (fNeedInsideGeometry)
+    {
+        self.m_pVB.AddTrapezoidVertices(
+            OutputVertex{
+                x: rPixelXTopLeft + rPixelXLeftDelta,
+                y: rPixelYTop,
+                coverage: FLOAT_ONE,
+            },
+            OutputVertex{
+                x: rPixelXBottomLeft + rPixelXLeftDelta,
+                y: rPixelYBottom,
+                coverage: FLOAT_ONE,
+            },
+            OutputVertex{
+                x: rPixelXTopRight - rPixelXRightDelta,
+                y: rPixelYTop,
+                coverage: FLOAT_ONE,
+            },
+            OutputVertex{
+                x: rPixelXBottomRight - rPixelXRightDelta,
+                y: rPixelYBottom,
+                coverage: FLOAT_ONE,
+            }
+        );
+    }
+
+    self.m_pVB.AddTrapezoidVertices(
+        OutputVertex{
+            x: rPixelXTopRight - rPixelXRightDelta,
+            y: rPixelYTop,
+            coverage: FLOAT_ONE,
+        },
+        OutputVertex{
+            x: rPixelXBottomRight - rPixelXRightDelta,
+            y: rPixelYBottom,
+            coverage: FLOAT_ONE,
+        },
+        OutputVertex{
+            x: rPixelXTopRight + rPixelXRightDelta,
+            y: rPixelYTop,
+            coverage: FLOAT_ZERO,
+        },
+        OutputVertex{
+            x: rPixelXBottomRight + rPixelXRightDelta,
+            y: rPixelYBottom,
+            coverage: FLOAT_ZERO,
+        }
+    );
+
+    if (!fNeedOutsideGeometry)
+    {
+        //
+        // Duplicate the last vertex. This creates 2 degenerate triangles: one
+        // between vertices 8 and 9 and one connecting this trapezoid to the
+        // next one.
+        //
+
+        //pVertex.push(OutputVertex{
+        //  x: rPixelXBottomRight + rPixelXRightDelta,
+        //  y: rPixelYBottom,
+        //  coverage: FLOAT_ZERO,
+        //});
+    }
+
+    self.m_pVB.AddedNonLineSegment();
+
+//Cleanup:
+    RRETURN!(hr);
+}
+}
+/* 
+//+----------------------------------------------------------------------------
+//
+//  Member:    CHwTVertexBuffer<TVertex>::Builder::AddTrapezoidWaffle
+//
+//  Synopsis:  See AddTrapezoid.  This adds a waffled trapezoid.
+//
+//-----------------------------------------------------------------------------
+template <class TVertex>
+HRESULT
+CHwTVertexBuffer<TVertex>::Builder::AddTrapezoidWaffle(
+    float rPixelYTop,              // In: y coordinate of top of trapezoid
+    float rPixelXTopLeft,          // In: x coordinate for top left
+    float rPixelXTopRight,         // In: x coordinate for top right
+    float rPixelYBottom,           // In: y coordinate of bottom of trapezoid
+    float rPixelXBottomLeft,       // In: x coordinate for bottom left
+    float rPixelXBottomRight,      // In: x coordinate for bottom right
+    float rPixelXLeftDelta,        // In: trapezoid expand radius for left edge
+    float rPixelXRightDelta        // In: trapezoid expand radius for right edge
+    )
+{
+    HRESULT hr = S_OK;
+
+    // We have 2 (u & v) wafflers per texture coordinate that need waffling.
+    TriangleWaffler<PointXYA> wafflers[NUM_OF_VERTEX_TEXTURE_COORDS(TVertex) * 2];
+    bool fWafflersUsed = false;
+
+    TriangleWaffler<PointXYA>::ISink *pWaffleSinkNoRef = BuildWafflePipeline(wafflers, OUT fWafflersUsed);
+
+    PointXYA vertices[8];
+
+    //
+    // Fill in the strip vertices
+    //
+
+    // Nonstandard coverage mapping and waffling are not supported at the same time.
+    Assert(!NeedOutsideGeometry());
+
+    vertices[0].x = rPixelXTopLeft - rPixelXLeftDelta;
+    vertices[0].y = rPixelYTop;
+    vertices[0].a = 0;
+
+    vertices[1].x = rPixelXBottomLeft - rPixelXLeftDelta;
+    vertices[1].y = rPixelYBottom;
+    vertices[1].a = 0;
+
+    vertices[2].x = rPixelXTopLeft + rPixelXLeftDelta;
+    vertices[2].y = rPixelYTop;
+    vertices[2].a = 1;
+
+    vertices[3].x = rPixelXBottomLeft + rPixelXLeftDelta;
+    vertices[3].y = rPixelYBottom;
+    vertices[3].a = 1;
+
+    vertices[4].x = rPixelXTopRight - rPixelXRightDelta;
+    vertices[4].y = rPixelYTop;
+    vertices[4].a = 1;
+
+    vertices[5].x = rPixelXBottomRight - rPixelXRightDelta;
+    vertices[5].y = rPixelYBottom;
+    vertices[5].a = 1;
+
+    vertices[6].x = rPixelXTopRight + rPixelXRightDelta;
+    vertices[6].y = rPixelYTop;
+    vertices[6].a = 0;
+
+    vertices[7].x = rPixelXBottomRight + rPixelXRightDelta;
+    vertices[7].y = rPixelYBottom;
+    vertices[7].a = 0;
+
+    // Send the triangles in the strip through the waffle pipeline.
+    for (int i = 0; i < 6; ++i)
+    {
+        IFC(pWaffleSinkNoRef->AddTriangle(vertices[i+1], vertices[i], vertices[i+2]));
+    }
+
+Cleanup:
+    RRETURN(hr);
+}
+*/
+impl CHwVertexBufferBuilder<'_, '_> {
+
+    //+----------------------------------------------------------------------------
+//
+//  Member:    CHwTVertexBuffer<TVertex>::Builder::NeedCoverageGeometry
+//
+//  Synopsis:  Returns true if the coverage value needs to be rendered
+//             based on NeedInsideGeometry() and NeedOutsideGeometry()
+//
+//             Two cases where we don't need to generate geometry:
+//              1. NeedInsideGeometry is false, and coverage is c_nShiftSizeSquared.
+//              2. NeedOutsideGeometry is false and coverage is 0
+//
+//-----------------------------------------------------------------------------
+fn NeedCoverageGeometry(&self,
+    nCoverage: INT
+    ) -> bool
+{
+    return    (self.NeedInsideGeometry()  || nCoverage != c_nShiftSizeSquared)
+           && (self.NeedOutsideGeometry() || nCoverage != 0);
+}
+
+    //+------------------------------------------------------------------------
+    //
+    //  Member:    NeedOutsideGeometry
+    //
+    //  Synopsis:  True if we should create geometry with zero alpha for
+    //             areas outside the input geometry but within a given
+    //             bounding box.
+    //
+    //-------------------------------------------------------------------------
+    fn NeedOutsideGeometry(&self) -> bool
+    {
+        return self.m_fNeedOutsideGeometry;
+    }
+
+        //+------------------------------------------------------------------------
+    //
+    //  Member:    NeedInsideGeometry
+    //
+    //  Synopsis:  True if we should create geometry for areas completely
+    //             withing the input geometry (i.e. alpha 1.)  Should only
+    //             be false if NeedOutsideGeometry is true.
+    //
+    //-------------------------------------------------------------------------
+    fn NeedInsideGeometry(&self) -> bool
+    {
+        assert!(self.m_fNeedOutsideGeometry || self.m_fNeedInsideGeometry);
+        return self.m_fNeedInsideGeometry;
+    }
+
+
+
+    // Helpers that handle extra shapes in trapezoid mode.
+    fn PrepareStratum(&mut self,
+        rStratumTop: f32,
+        rStratumBottom: f32,
+        fTrapezoid: bool,
+        rTrapezoidLeft: f32,
+        rTrapezoidRight: f32,
+        rTrapezoidTopLeft: f32, // = 0
+        rTrapezoidBottomLeft: f32, // = 0
+        rTrapezoidTopRight: f32, // = 0
+        rTrapezoidBottomRight: f32, // = 0
+
+        ) -> HRESULT
+    {
+        return if self.NeedOutsideGeometry() {
+            self.PrepareStratumSlow(
+                rStratumTop,
+                rStratumBottom,
+                fTrapezoid,
+                rTrapezoidLeft,
+                rTrapezoidRight,
+                rTrapezoidTopLeft,
+                rTrapezoidBottomLeft,
+                rTrapezoidTopRight,
+                rTrapezoidBottomRight
+                )
+     } else { S_OK };
+    }
+
+//+----------------------------------------------------------------------------
+//
+//  Member:    CHwTVertexBuffer<TVertex>::Builder::PrepareStratumSlow
+//
+//  Synopsis:  Call before producing a new stratum (complex span or trapezoid.)
+//             Handles several tasks:
+//               1. Producing between top of complement geometry & the 1st
+//                  stratum or when a gap between strata occurs (because
+//                  the geometry is not closed and has horizontal gaps.)
+//                  Passing in FLT_MAX for rStratumTop and rStratumBottom
+//                  Fills the gap between the last stratum and the bottom
+//                  of the outside.
+//               2. Begins and/or ends the triangle strip corresponding to
+//                  a trapezoid row.
+//               3. Updates status vars m_rCurStratumTop & m_rCurStratumBottom
+//
+//  Note:      Call PrepareStratum which inlines the check for NeedOutsideGeometry()
+//             If NeedOutsideGeometry is false PrepareStratum() does nothing.
+//             This (slow) version asserts NeedOutsideGeometry()
+//
+//-----------------------------------------------------------------------------
+fn PrepareStratumSlow(&mut self,
+    rStratumTop: f32,
+    rStratumBottom: f32,
+    fTrapezoid: bool,
+    rTrapezoidLeft: f32,
+    rTrapezoidRight: f32,
+    rTrapezoidTopLeft: f32,
+    rTrapezoidBottomLeft: f32,
+    rTrapezoidTopRight: f32,
+    rTrapezoidBottomRight: f32,
+    ) -> HRESULT
+{
+    type TVertex = OutputVertex;
+    let hr: HRESULT = S_OK;
+    
+    assert!(!(rStratumTop > rStratumBottom));
+    assert!(self.NeedOutsideGeometry());
+
+    // There's only once case where a stratum can go "backwards"
+    // and that's when we're done building & calling from
+    // EndBuildingOutside
+        
+    let fEndBuildingOutside: f32 = (rStratumBottom == self.OutsideBottom() &&
+                                rStratumTop == self.OutsideBottom()) as i32 as f32;
+
+    if (fEndBuildingOutside == 1.)
+    {
+        assert!(!fTrapezoid);
+    }
+    else
+    {
+        assert!(!(rStratumBottom < self.m_rCurStratumBottom));
+    }
+    
+    if (   fEndBuildingOutside == 1.
+        || rStratumBottom != self.m_rCurStratumBottom)
+    {
+        
+        // New stratum starting now.  Two things to do
+        //  1. Close out current trapezoid stratum if necessary.
+        //  2. Begin new trapezoid stratum if necessary.
+        
+        if (self.m_rCurStratumTop != f32::MAX)
+        {
+            // we do not clip trapezoids so RIGHT boundary
+            // of the stratus can be outside of m_rcOutsideBounds.
+            
+            let rOutsideRight: f32 = self.OutsideRight().max(self.m_rLastTrapezoidRight);
+
+            // End current trapezoid stratum.
+
+            self.m_pVB.AddTrapezoidVertices(
+                OutputVertex{
+                    x: self.m_rLastTrapezoidTopRight,
+                    y: self.m_rCurStratumTop,
+                    coverage: FLOAT_ZERO,
+                },
+                OutputVertex{
+                    x: self.m_rLastTrapezoidBottomRight,
+                    y: self.m_rCurStratumBottom,
+                    coverage: FLOAT_ZERO,
+                },
+                OutputVertex{
+                    x: rOutsideRight,
+                    y: self.m_rCurStratumTop,
+                    coverage: FLOAT_ZERO,
+                },
+                OutputVertex{
+                    x: rOutsideRight,
+                    y: self.m_rCurStratumBottom,
+                    coverage: FLOAT_ZERO,
+                }
+            );
+        }
+        // Compute the gap between where the last stratum ended and where
+        // this one begins.
+        let flGap: f32 = rStratumTop - self.m_rCurStratumBottom;
+
+        if (flGap > 0.)
+        {
+            // The "special" case of a gap at the beginning is caught here
+            // using the sentinel initial value of m_rCurStratumBottom.
+
+            let flRectTop: f32 = if self.m_rCurStratumBottom == -f32::MAX {
+                              self.OutsideTop() } else {
+                              self.m_rCurStratumBottom };
+            let flRectBot: f32  = (rStratumTop as f32);
+
+            // Produce rectangular for any horizontal intervals in the
+            // outside bounds that have no generated geometry.
+            assert!(self.m_rCurStratumBottom != -f32::MAX || self.m_rCurStratumTop == f32::MAX);
+
+            let outside_left = self.OutsideLeft();
+            let outside_right = self.OutsideRight();
+            
+            // Duplicate first vertex.
+            self.m_pVB.AddTrapezoidVertices(
+                OutputVertex{
+                    x: outside_left,
+                    y: flRectTop,
+                    coverage: FLOAT_ZERO,
+                },
+                OutputVertex{
+                    x: outside_left,
+                    y: flRectBot,
+                    coverage: FLOAT_ZERO,
+                },
+                OutputVertex{
+                    x: outside_right,
+                    y: flRectTop,
+                    coverage: FLOAT_ZERO,
+                },
+                OutputVertex{
+                    x: outside_right,
+                    y: flRectBot,
+                    coverage: FLOAT_ZERO,
+                }
+            );
+        }
+
+        if (fTrapezoid)
+        {
+
+            // we do not clip trapezoids so left boundary
+            // of the stratus can be outside of m_rcOutsideBounds.
+            
+            let rOutsideLeft: f32 = self.OutsideLeft().min(rTrapezoidLeft);
+
+            // Begin new trapezoid stratum.
+
+            self.m_pVB.AddTrapezoidVertices(
+                OutputVertex{
+                    x: rOutsideLeft,
+                    y: rStratumTop,
+                    coverage: FLOAT_ZERO,
+                },
+                OutputVertex{
+                    x: rOutsideLeft,
+                    y: rStratumBottom,
+                    coverage: FLOAT_ZERO,
+                },
+                OutputVertex{
+                    x: rTrapezoidTopLeft,
+                    y: rStratumTop,
+                    coverage: FLOAT_ZERO,
+                },
+                OutputVertex{
+                    x: rTrapezoidBottomLeft,
+                    y: rStratumBottom,
+                    coverage: FLOAT_ZERO,
+                }
+            );
+        }
+    }
+    
+    if (fTrapezoid)
+    {
+        self.m_rLastTrapezoidTopRight = rTrapezoidTopRight;
+        self.m_rLastTrapezoidBottomRight = rTrapezoidBottomRight;
+        self.m_rLastTrapezoidRight = rTrapezoidRight;
+    }
+
+    self.m_rCurStratumTop = if fTrapezoid { rStratumTop } else { f32::MAX };
+    self.m_rCurStratumBottom = rStratumBottom;
+
+    RRETURN!(hr);
+}
+ 
+//+----------------------------------------------------------------------------
+//
+//  Member:    CHwTVertexBuffer<TVertex>::Builder::EndBuildingOutside
+//
+//  Synopsis:  Finish creating outside geometry.
+//             1. If no geometry was created then just fill bounds.
+//             2. Otherwise:
+//                 A. End last trapezoid row
+//                 B. Produce stop stratum
+// 
+//-----------------------------------------------------------------------------
+fn EndBuildingOutside(&mut self) -> HRESULT
+{
+    return self.PrepareStratum(
+        self.OutsideBottom(),
+        self.OutsideBottom(),
+        false, /* Not a trapezoid. */
+        0., 0.,
+        0., 0.,
+        0., 0.,
+        );
+}
+
+//+----------------------------------------------------------------------------
+//
+//  Member:    CHwTVertexBuffer<TVertex>::Builder::EndBuilding
+//
+//  Synopsis:  Expand all vertices to the full required format and return
+//             vertex buffer.
+//
+//-----------------------------------------------------------------------------
+pub fn EndBuilding(&mut self) -> HRESULT
+{
+    let hr = S_OK;
+
+    IFC!(self.EndBuildingOutside());
+    
+//Cleanup:
+    RRETURN!(hr);
+}
+
+}
diff --git a/third_party/rust/wpf-gpu-raster/src/lib.rs b/third_party/rust/wpf-gpu-raster/src/lib.rs
new file mode 100644
index 0000000000..4f5a4495aa
--- /dev/null
+++ b/third_party/rust/wpf-gpu-raster/src/lib.rs
@@ -0,0 +1,691 @@
+/*!
+Converts a 2D path into a set of vertices of a triangle strip mesh that represents the antialiased fill of that path.
+
+```rust
+    use wpf_gpu_raster::PathBuilder;
+    let mut p = PathBuilder::new();
+    p.move_to(10., 10.);
+    p.line_to(40., 10.);
+    p.line_to(40., 40.);
+    let result = p.rasterize_to_tri_list(0, 0, 100, 100);
+```
+
+*/
+#![allow(unused_parens)]
+#![allow(overflowing_literals)]
+#![allow(non_snake_case)]
+#![allow(non_camel_case_types)]
+#![allow(non_upper_case_globals)]
+#![allow(dead_code)]
+#![allow(unused_macros)]
+
+#[macro_use]
+mod fix;
+#[macro_use]
+mod helpers;
+#[macro_use]
+mod real;
+mod bezier;
+#[macro_use]
+mod aarasterizer;
+mod hwrasterizer;
+mod aacoverage;
+mod hwvertexbuffer;
+
+mod types;
+mod geometry_sink;
+mod matrix;
+
+mod nullable_ref;
+
+#[cfg(feature = "c_bindings")]
+pub mod c_bindings;
+
+#[cfg(test)]
+mod tri_rasterize;
+
+use aarasterizer::CheckValidRange28_4;
+use hwrasterizer::CHwRasterizer;
+use hwvertexbuffer::{CHwVertexBuffer, CHwVertexBufferBuilder};
+use real::CFloatFPU;
+use types::{MilFillMode, PathPointTypeStart, MilPoint2F, MilPointAndSizeL, PathPointTypeLine, MilVertexFormat, MilVertexFormatAttribute, DynArray, BYTE, PathPointTypeBezier, PathPointTypeCloseSubpath, CMILSurfaceRect, POINT};
+
+#[repr(C)]
+#[derive(Clone, Debug, Default)]
+pub struct OutputVertex {
+    pub x: f32,
+    pub y: f32,
+    pub coverage: f32
+}
+
+#[repr(C)]
+#[derive(Copy, Clone)]
+pub enum FillMode {
+    EvenOdd = 0,
+    Winding = 1,
+}
+
+impl Default for FillMode {
+    fn default() -> Self {
+        FillMode::EvenOdd
+    }
+}
+
+#[derive(Clone, Default)]
+pub struct OutputPath {
+    fill_mode: FillMode,
+    points: Box<[POINT]>,
+    types: Box<[BYTE]>,
+}
+
+impl std::hash::Hash for OutputVertex {
+    fn hash<H: std::hash::Hasher>(&self, state: &mut H) {
+        self.x.to_bits().hash(state);
+        self.y.to_bits().hash(state);
+        self.coverage.to_bits().hash(state);
+    }
+}
+
+pub struct PathBuilder {
+    points: DynArray<POINT>,
+    types: DynArray<BYTE>,
+    initial_point: Option<MilPoint2F>,
+    current_point: Option<MilPoint2F>,
+    in_shape: bool,
+    fill_mode: FillMode,
+    outside_bounds: Option<CMILSurfaceRect>,
+    need_inside: bool,
+    valid_range: bool,
+    rasterization_truncates: bool,
+}
+
+impl PathBuilder {
+    pub fn new() -> Self {
+        Self {
+            points: Vec::new(),
+            types: Vec::new(),
+            initial_point: None,
+            current_point: None,
+            in_shape: false,
+            fill_mode: FillMode::EvenOdd,
+            outside_bounds: None,
+            need_inside: true,
+            valid_range: true,
+            rasterization_truncates: false,
+        }
+    }
+    fn add_point(&mut self, x: f32, y: f32) {
+        self.current_point = Some(MilPoint2F{X: x, Y: y});
+        // Transform from pixel corner at 0.0 to pixel center at 0.0. Scale into 28.4 range.
+        // Validate that the point before rounding is within expected bounds for the rasterizer.
+        let (x, y) = ((x - 0.5) * 16.0, (y - 0.5) * 16.0);
+        self.valid_range = self.valid_range && CheckValidRange28_4(x, y);
+        self.points.push(POINT {
+            x: CFloatFPU::Round(x),
+            y: CFloatFPU::Round(y),
+        });
+    }
+    pub fn line_to(&mut self, x: f32, y: f32) {
+        if let Some(initial_point) = self.initial_point {
+            if !self.in_shape {
+                self.types.push(PathPointTypeStart);
+                self.add_point(initial_point.X, initial_point.Y);
+                self.in_shape = true;
+            }
+            self.types.push(PathPointTypeLine);
+            self.add_point(x, y);
+        } else {
+            self.initial_point = Some(MilPoint2F{X: x, Y: y})
+        }
+    }
+    pub fn move_to(&mut self, x: f32, y: f32) {
+        self.in_shape = false;
+        self.initial_point = Some(MilPoint2F{X: x, Y: y});
+        self.current_point = self.initial_point;
+    }
+    pub fn curve_to(&mut self, c1x: f32, c1y: f32, c2x: f32, c2y: f32, x: f32, y: f32) {
+        let initial_point = match self.initial_point {
+            Some(initial_point) => initial_point,
+            None => MilPoint2F{X:c1x, Y:c1y}
+        };
+        if !self.in_shape {
+            self.types.push(PathPointTypeStart);
+            self.add_point(initial_point.X, initial_point.Y);
+            self.initial_point = Some(initial_point);
+            self.in_shape = true;
+        }
+        self.types.push(PathPointTypeBezier);
+        self.add_point(c1x, c1y);
+        self.add_point(c2x, c2y);
+        self.add_point(x, y);
+    }
+    pub fn quad_to(&mut self, cx: f32, cy: f32, x: f32, y: f32) {
+        // For now we just implement quad_to on top of curve_to.
+        // Long term we probably want to support quad curves
+        // directly.
+        let c0 = match self.current_point {
+            Some(current_point) => current_point,
+            None => MilPoint2F{X:cx, Y:cy}
+        };
+
+        let c1x = c0.X + (2./3.) * (cx - c0.X);
+        let c1y = c0.Y + (2./3.) * (cy - c0.Y);
+
+        let c2x = x + (2./3.) * (cx - x);
+        let c2y = y + (2./3.) * (cy - y);
+
+        self.curve_to(c1x, c1y, c2x, c2y, x, y);
+    }
+    pub fn close(&mut self) {
+        if self.in_shape {
+          // Only close the path if we are inside a shape. Otherwise, the point
+          // should be safe to elide.
+          if let Some(last) = self.types.last_mut() {
+              *last |= PathPointTypeCloseSubpath;
+          }
+          self.in_shape = false;
+        }
+        // Close must install a new initial point that is the same as the
+        // initial point of the just-closed sub-path. Thus, just leave the
+        // initial point unchanged.
+        self.current_point = self.initial_point;
+    }
+    pub fn set_fill_mode(&mut self, fill_mode: FillMode) {
+        self.fill_mode = fill_mode;
+    }
+    /// Enables rendering geometry for areas outside the shape but
+    /// within the bounds.  These areas will be created with
+    /// zero alpha.
+    ///
+    /// This is useful for creating geometry for other blend modes.
+    /// For example:
+    /// - `IN(dest, geometry)` can be done with `outside_bounds` and `need_inside = false`
+    /// - `IN(dest, geometry, alpha)` can be done with `outside_bounds` and `need_inside = true`
+    ///
+    /// Note: trapezoidal areas won't be clipped to outside_bounds
+    pub fn set_outside_bounds(&mut self, outside_bounds: Option<(i32, i32, i32, i32)>, need_inside: bool) {
+        self.outside_bounds = outside_bounds.map(|r| CMILSurfaceRect { left: r.0, top: r.1, right: r.2, bottom: r.3 });
+        self.need_inside = need_inside;
+    }
+
+    /// Set this to true if post vertex shader coordinates are converted to fixed point
+    /// via truncation. This has been observed with OpenGL on AMD GPUs on macOS. 
+    pub fn set_rasterization_truncates(&mut self, rasterization_truncates: bool) {
+        self.rasterization_truncates = rasterization_truncates;
+    }
+
+    /// Note: trapezoidal areas won't necessarily be clipped to the clip rect
+    pub fn rasterize_to_tri_list(&self, clip_x: i32, clip_y: i32, clip_width: i32, clip_height: i32) -> Box<[OutputVertex]> {
+        if !self.valid_range {
+            // If any of the points are outside of valid 28.4 range, then just return an empty triangle list.
+            return Box::new([]);
+        }
+        let (x, y, width, height, need_outside) = if let Some(CMILSurfaceRect { left, top, right, bottom }) = self.outside_bounds {
+            let x0 = clip_x.max(left);
+            let y0 = clip_y.max(top);
+            let x1 = (clip_x + clip_width).min(right);
+            let y1 = (clip_y + clip_height).min(bottom);
+            (x0, y0, x1 - x0, y1 - y0, true)
+        } else {
+            (clip_x, clip_y, clip_width, clip_height, false)
+        };
+        rasterize_to_tri_list(self.fill_mode, &self.types, &self.points, x, y, width, height, self.need_inside, need_outside, self.rasterization_truncates, None)
+            .flush_output()
+    }
+
+    pub fn get_path(&mut self) -> Option<OutputPath> {
+        if self.valid_range && !self.points.is_empty() && !self.types.is_empty() {
+            Some(OutputPath {
+                fill_mode: self.fill_mode,
+                points: std::mem::take(&mut self.points).into_boxed_slice(),
+                types: std::mem::take(&mut self.types).into_boxed_slice(),
+            })
+        } else {
+            None
+        }
+    }
+}
+
+// Converts a path that is specified as an array of edge types, each associated with a fixed number
+// of points that are serialized to the points array. Edge types are specified via PathPointType
+// masks, whereas points must be supplied in 28.4 signed fixed-point format. By default, users can
+// fill the inside of the path excluding the outside. It may alternatively be desirable to fill the
+// outside the path out to the clip boundary, optionally keeping the inside. PathBuilder may be
+// used instead as a simpler interface to this function that handles building the path arrays.
+pub fn rasterize_to_tri_list<'a>(
+    fill_mode: FillMode,
+    types: &[BYTE],
+    points: &[POINT],
+    clip_x: i32,
+    clip_y: i32,
+    clip_width: i32,
+    clip_height: i32,
+    need_inside: bool,
+    need_outside: bool,
+    rasterization_truncates: bool,
+    output_buffer: Option<&'a mut [OutputVertex]>,
+) -> CHwVertexBuffer<'a> {
+    let clipRect = MilPointAndSizeL {
+        X: clip_x,
+        Y: clip_y,
+        Width: clip_width,
+        Height: clip_height,
+    };
+
+    let mil_fill_mode = match fill_mode {
+        FillMode::EvenOdd => MilFillMode::Alternate,
+        FillMode::Winding => MilFillMode::Winding,
+    };
+
+    let m_mvfIn: MilVertexFormat = MilVertexFormatAttribute::MILVFAttrXY as MilVertexFormat;
+    let m_mvfGenerated: MilVertexFormat  = MilVertexFormatAttribute::MILVFAttrNone as MilVertexFormat;
+    //let mvfaAALocation  = MILVFAttrNone;
+    const HWPIPELINE_ANTIALIAS_LOCATION: MilVertexFormatAttribute = MilVertexFormatAttribute::MILVFAttrDiffuse;
+    let mvfaAALocation = HWPIPELINE_ANTIALIAS_LOCATION;
+
+    let outside_bounds = if need_outside {
+        Some(CMILSurfaceRect {
+            left: clip_x,
+            top: clip_y,
+            right: clip_x + clip_width,
+            bottom: clip_y + clip_height,
+        })
+    } else {
+        None
+    };
+
+    let mut vertexBuffer = CHwVertexBuffer::new(rasterization_truncates, output_buffer);
+    {
+        let mut vertexBuilder = CHwVertexBufferBuilder::Create(
+            m_mvfIn, m_mvfIn | m_mvfGenerated, mvfaAALocation, &mut vertexBuffer);
+        vertexBuilder.SetOutsideBounds(outside_bounds.as_ref(), need_inside);
+        vertexBuilder.BeginBuilding();
+        {
+            let mut rasterizer = CHwRasterizer::new(
+                &mut vertexBuilder, mil_fill_mode, None, clipRect);
+            rasterizer.SendGeometry(points, types);
+        }
+        vertexBuilder.EndBuilding();
+    }
+
+    vertexBuffer
+}
+
+#[cfg(test)]
+mod tests {
+    use std::{hash::{Hash, Hasher}, collections::hash_map::DefaultHasher};
+    use crate::{*, tri_rasterize::rasterize_to_mask};
+    fn calculate_hash<T: Hash>(t: &T) -> u64 {
+        let mut s = DefaultHasher::new();
+        t.hash(&mut s);
+        s.finish()
+    }
+    #[test]
+    fn basic() {
+        let mut p = PathBuilder::new();
+        p.move_to(10., 10.);
+        p.line_to(10., 30.);
+        p.line_to(30., 30.);
+        p.line_to(30., 10.);
+        p.close();
+        let result = p.rasterize_to_tri_list(0, 0, 100, 100);
+        assert_eq!(result.len(), 18);
+        //assert_eq!(dbg!(calculate_hash(&result)), 0x5851570566450135);
+        assert_eq!(calculate_hash(&rasterize_to_mask(&result, 100, 100)), 0xfbb7c3932059e240);
+    }
+
+    #[test]
+    fn simple() {
+        let mut p = PathBuilder::new();
+        p.move_to(10., 10.);
+        p.line_to(40., 10.);
+        p.line_to(40., 40.);
+        let result = p.rasterize_to_tri_list(0, 0, 100, 100);
+        //assert_eq!(dbg!(calculate_hash(&result)), 0x81a9af7769f88e68);
+        assert_eq!(calculate_hash(&rasterize_to_mask(&result, 100, 100)), 0x6d1595533d40ef92);
+    }
+
+    #[test]
+    fn rust() {
+        let mut p = PathBuilder::new();
+        p.move_to(10., 10.);
+        p.line_to(40., 10.);
+        p.line_to(40., 40.);
+        let result = p.rasterize_to_tri_list(0, 0, 100, 100);
+        //assert_eq!(dbg!(calculate_hash(&result)), 0x81a9af7769f88e68);
+        assert_eq!(calculate_hash(&rasterize_to_mask(&result, 100, 100)), 0x6d1595533d40ef92);
+    }
+
+    #[test]
+    fn fill_mode() {
+        let mut p = PathBuilder::new();
+        p.move_to(10., 10.);
+        p.line_to(40., 10.);
+        p.line_to(40., 40.);
+        p.line_to(10., 40.);
+        p.close();
+        p.move_to(15., 15.);
+        p.line_to(35., 15.);
+        p.line_to(35., 35.);
+        p.line_to(15., 35.);
+        p.close();
+        let result = p.rasterize_to_tri_list(0, 0, 100, 100);
+        //assert_eq!(dbg!(calculate_hash(&result)), 0xb34344234f2f75a8);
+        assert_eq!(calculate_hash(&rasterize_to_mask(&result, 100, 100)), 0xc7bf999c56ccfc34);
+
+        let mut p = PathBuilder::new();
+        p.move_to(10., 10.);
+        p.line_to(40., 10.);
+        p.line_to(40., 40.);
+        p.line_to(10., 40.);
+        p.close();
+        p.move_to(15., 15.);
+        p.line_to(35., 15.);
+        p.line_to(35., 35.);
+        p.line_to(15., 35.);
+        p.close();
+        p.set_fill_mode(FillMode::Winding);
+        let result = p.rasterize_to_tri_list(0, 0, 100, 100);
+        //assert_eq!(dbg!(calculate_hash(&result)), 0xee4ecd8a738fc42c);
+        assert_eq!(calculate_hash(&rasterize_to_mask(&result, 100, 100)), 0xfafad659db9a2efd);
+
+    }
+
+    #[test]
+    fn range() {
+        // test for a start point out of range
+        let mut p = PathBuilder::new();
+        p.curve_to(8.872974e16, 0., 0., 0., 0., 0.);
+        let result = p.rasterize_to_tri_list(0, 0, 100, 100);
+        assert_eq!(result.len(), 0);
+
+        // test for a subsequent point out of range
+        let mut p = PathBuilder::new();
+        p.curve_to(0., 0., 8.872974e16, 0., 0., 0.);
+        let result = p.rasterize_to_tri_list(0, 0, 100, 100);
+        assert_eq!(result.len(), 0);
+    }
+
+    #[test]
+    fn multiple_starts() {
+        let mut p = PathBuilder::new();
+        p.line_to(10., 10.);
+        p.move_to(0., 0.);
+        let result = p.rasterize_to_tri_list(0, 0, 100, 100);
+        assert_eq!(result.len(), 0);
+    }
+
+    #[test]
+    fn path_closing() {
+        let mut p = PathBuilder::new();
+        p.curve_to(0., 0., 0., 0., 0., 32.0);
+        p.close();
+        p.curve_to(0., 0., 0., 0., 0., 32.0);
+        let result = p.rasterize_to_tri_list(0, 0, 100, 100);
+        assert_eq!(result.len(), 0);
+    }
+
+    #[test]
+    fn curve() {
+        let mut p = PathBuilder::new();
+        p.move_to(10., 10.);
+        p.curve_to(40., 10., 40., 10., 40., 40.);
+        p.close();
+        let result = p.rasterize_to_tri_list(0, 0, 100, 100);
+        assert_eq!(calculate_hash(&rasterize_to_mask(&result, 100, 100)), 0xa92aae8dba7b8cd4);
+        assert_eq!(dbg!(calculate_hash(&result)), 0x8dbc4d23f9bba38d);
+    }
+
+    #[test]
+    fn partial_coverage_last_line() {
+        let mut p = PathBuilder::new();
+
+        p.move_to(10., 10.);
+        p.line_to(40., 10.);
+        p.line_to(40., 39.6);
+        p.line_to(10., 39.6);
+
+        let result = p.rasterize_to_tri_list(0, 0, 100, 100);
+        assert_eq!(result.len(), 21);
+        assert_eq!(calculate_hash(&rasterize_to_mask(&result, 100, 100)), 0xfa200c3bae144952);
+        assert_eq!(dbg!(calculate_hash(&result)), 0xf90cb6afaadfb559);
+    }
+
+    #[test]
+    fn delta_upper_bound() {
+        let mut p = PathBuilder::new();
+        p.move_to(-122.3 + 200.,84.285);
+        p.curve_to(-122.3 + 200., 84.285, -122.2 + 200.,86.179, -123.03 + 200., 86.16);
+        p.curve_to(-123.85 + 200., 86.141, -140.3 + 200., 38.066, -160.83 + 200., 40.309);
+        p.curve_to(-160.83 + 200., 40.309, -143.05 + 200., 32.956,  -122.3 + 200., 84.285);
+        p.close();
+
+        let result = p.rasterize_to_tri_list(0, 0, 400, 400);
+        assert_eq!(result.len(), 429);
+        assert_eq!(calculate_hash(&rasterize_to_mask(&result, 100, 100)), 0x5e82d98fdb47a796);
+        assert_eq!(dbg!(calculate_hash(&result)), 0x52d52992e249587a);
+    }
+
+
+    #[test]
+    fn self_intersect() {
+        let mut p = PathBuilder::new();
+        p.move_to(10., 10.);
+        p.line_to(40., 10.);
+        p.line_to(10., 40.);
+        p.line_to(40., 40.);
+        p.close();
+        let result = p.rasterize_to_tri_list(0, 0, 100, 100);
+        assert_eq!(calculate_hash(&rasterize_to_mask(&result, 100, 100)), 0x49ecc769e1d4ec01);
+        assert_eq!(dbg!(calculate_hash(&result)), 0xf10babef5c619d19);
+    }
+
+    #[test]
+    fn grid() {
+        let mut p = PathBuilder::new();
+
+        for i in 0..200 {
+            let offset = i as f32 * 1.3;
+            p.move_to(0. + offset, -8.);
+            p.line_to(0.5 + offset, -8.);
+            p.line_to(0.5 + offset, 40.);
+            p.line_to(0. + offset, 40.);
+            p.close();
+        }
+        let result = p.rasterize_to_tri_list(0, 0, 100, 100);
+        assert_eq!(result.len(), 12000);
+        assert_eq!(calculate_hash(&rasterize_to_mask(&result, 100, 100)), 0x5a7df39d9e9292f0);
+    }
+
+    #[test]
+    fn outside() {
+        let mut p = PathBuilder::new();
+        p.move_to(10., 10.);
+        p.line_to(40., 10.);
+        p.line_to(10., 40.);
+        p.line_to(40., 40.);
+        p.close();
+        p.set_outside_bounds(Some((0, 0, 50, 50)), false);
+        let result = p.rasterize_to_tri_list(0, 0, 100, 100);
+        assert_eq!(calculate_hash(&rasterize_to_mask(&result, 100, 100)), 0x59403ddbb7e1d09a);
+        assert_eq!(dbg!(calculate_hash(&result)), 0x805fd385e47e6f2);
+
+        // ensure that adjusting the outside bounds changes the results
+        p.set_outside_bounds(Some((5, 5, 50, 50)), false);
+        let result = p.rasterize_to_tri_list(0, 0, 100, 100);
+        assert_eq!(calculate_hash(&rasterize_to_mask(&result, 100, 100)), 0x59403ddbb7e1d09a);
+        assert_eq!(dbg!(calculate_hash(&result)), 0xcec2ed688999c966);
+    }
+
+    #[test]
+    fn outside_inside() {
+        let mut p = PathBuilder::new();
+        p.move_to(10., 10.);
+        p.line_to(40., 10.);
+        p.line_to(10., 40.);
+        p.line_to(40., 40.);
+        p.close();
+        p.set_outside_bounds(Some((0, 0, 50, 50)), true);
+        let result = p.rasterize_to_tri_list(0, 0, 100, 100);
+        assert_eq!(calculate_hash(&rasterize_to_mask(&result, 100, 100)), 0x49ecc769e1d4ec01);
+        assert_eq!(dbg!(calculate_hash(&result)), 0xaf76b42a5244d1ec);
+    }
+
+    #[test]
+    fn outside_clipped() {
+        let mut p = PathBuilder::new();
+        p.move_to(10., 10.);
+        p.line_to(10., 40.);
+        p.line_to(90., 40.);
+        p.line_to(40., 10.);
+        p.close();
+        p.set_outside_bounds(Some((0, 0, 50, 50)), false);
+        let result = p.rasterize_to_tri_list(0, 0, 50, 50);
+        assert_eq!(calculate_hash(&rasterize_to_mask(&result, 100, 100)), 0x3d2a08f5d0bac999);
+        assert_eq!(dbg!(calculate_hash(&result)), 0xbd42b934ab52be39);
+    }
+
+    #[test]
+    fn clip_edge() {
+        let mut p = PathBuilder::new();
+        // tests the bigNumerator < 0 case of aarasterizer::ClipEdge
+        p.curve_to(-24., -10., -300., 119., 0.0, 0.0);
+        let result = p.rasterize_to_tri_list(0, 0, 100, 100);
+        // The edge merging only happens between points inside the enumerate buffer. This means
+        // that the vertex output can depend on the size of the enumerate buffer because there
+        // the number of edges and positions of vertices will change depending on edge merging.
+        if ENUMERATE_BUFFER_NUMBER!() == 32 {
+            assert_eq!(result.len(), 111);
+        } else {
+            assert_eq!(result.len(), 171);
+        }
+        assert_eq!(calculate_hash(&rasterize_to_mask(&result, 100, 100)), 0x50b887b09a4c16e);
+    }
+
+    #[test]
+    fn enum_buffer_num() {
+        let mut p = PathBuilder::new();
+        p.curve_to(0.0, 0.0, 0.0, 12.0, 0.0, 44.919434);
+        p.line_to(64.0, 36.0 );
+        p.line_to(0.0, 80.0,);
+        let result = p.rasterize_to_tri_list(0, 0, 100, 100);
+        assert_eq!(result.len(), 300);
+        assert_eq!(calculate_hash(&rasterize_to_mask(&result, 100, 100)), 0x659cc742f16b42f2);
+    }
+
+    #[test]
+    fn fill_alternating_empty_interior_pairs() {
+        let mut p = PathBuilder::new();
+        p.line_to( 0., 2. );
+        p.curve_to(0.0, 0.0,1., 6., 0.0, 0.0);
+        let result = p.rasterize_to_tri_list(0, 0, 100, 100);
+        assert_eq!(result.len(), 9);
+        assert_eq!(calculate_hash(&rasterize_to_mask(&result, 100, 100)), 0x726606a662fe46a0);
+    }
+
+    #[test]
+    fn fill_winding_empty_interior_pairs() {
+        let mut p = PathBuilder::new();
+        p.curve_to(45., 61., 0.09, 0., 0., 0.);
+        p.curve_to(45., 61., 0.09, 0., 0., 0.);
+        p.curve_to(0., 0., 0., 38., 0.09, 15.);
+        p.set_fill_mode(FillMode::Winding);
+        let result = p.rasterize_to_tri_list(0, 0, 100, 100);
+        assert_eq!(result.len(), 462);
+        assert_eq!(calculate_hash(&rasterize_to_mask(&result, 100, 100)), 0x651ea4ade5543087);
+    }
+
+    #[test]
+    fn empty_fill() {
+        let mut p = PathBuilder::new();
+        p.move_to(0., 0.);
+        p.line_to(10., 100.);
+        let result = p.rasterize_to_tri_list(0, 0, 100, 100);
+        assert_eq!(result.len(), 0);
+    }
+
+    #[test]
+    fn rasterize_line() {
+        let mut p = PathBuilder::new();
+        p.move_to(1., 1.);
+        p.line_to(2., 1.);
+        p.line_to(2., 2.);
+        p.line_to(1., 2.);
+        p.close();
+        let result = p.rasterize_to_tri_list(0, 0, 100, 100);
+        let mask = rasterize_to_mask(&result, 3, 3);
+        assert_eq!(&mask[..], &[0,   0, 0,
+                                0, 255, 0,
+                                0,   0, 0][..]);
+    }
+
+    #[test]
+    fn triangle() {
+        let mut p = PathBuilder::new();
+        p.move_to(1., 10.);
+        p.line_to(100., 13.);
+        p.line_to(1., 16.);
+        p.close();
+        let result = p.rasterize_to_tri_list(0, 0, 100, 100);
+        assert_eq!(calculate_hash(&rasterize_to_mask(&result, 100, 100)), 0x4757b0c5a19b02f0);
+    }
+
+    #[test]
+    fn single_pixel() {
+        let mut p = PathBuilder::new();
+        p.move_to(1.5, 1.5);
+        p.line_to(2., 1.5);
+        p.line_to(2., 2.);
+        p.line_to(1.5, 2.);
+        p.close();
+        let result = p.rasterize_to_tri_list(0, 0, 100, 100);
+        assert_eq!(result.len(), 3);
+        assert_eq!(calculate_hash(&rasterize_to_mask(&result, 4, 4)), 0x9f481fe5588e341c);
+    }
+
+    #[test]
+    fn traps_outside_bounds() {
+        let mut p = PathBuilder::new();
+        p.move_to(10., 10.0);
+        p.line_to(30., 10.);
+        p.line_to(50., 20.);
+        p.line_to(30., 30.);
+        p.line_to(10., 30.);
+        p.close();
+        // The generated trapezoids are not necessarily clipped to the outside bounds rect
+        // and in this case the outside bounds geometry ends up drawing on top of the
+        // edge geometry which could be considered a bug.
+        p.set_outside_bounds(Some((0, 0, 50, 30)), true);
+        let result = p.rasterize_to_tri_list(0, 0, 100, 100);
+        assert_eq!(calculate_hash(&rasterize_to_mask(&result, 100, 100)), 0x6514e3d79d641f09);
+
+    }
+
+    #[test]
+    fn quad_to() {
+        let mut p = PathBuilder::new();
+        p.move_to(10., 10.0);
+        p.quad_to(30., 10., 30., 30.);
+        p.quad_to(10., 30., 30., 30.);
+        p.quad_to(60., 30., 60., 10.);
+        p.close();
+        let result = p.rasterize_to_tri_list(0, 0, 70, 40);
+        assert_eq!(result.len(), 279);
+        assert_eq!(calculate_hash(&rasterize_to_mask(&result, 70, 40)), 0xbd2eec3cfe9bd30b);
+    }
+
+    #[test]
+    fn close_after_move_to() {
+        let mut p = PathBuilder::new();
+        p.move_to(10., 0.);
+        p.close();
+        p.move_to(0., 0.);
+        p.line_to(0., 10.);
+        p.line_to(10., 10.);
+        p.move_to(10., 0.);
+        p.close();
+        let result = p.rasterize_to_tri_list(0, 0, 20, 20);
+        assert_eq!(result.len(), 27);
+        assert_eq!(dbg!(calculate_hash(&result)), 0xecfdf5bdfa25a1dd);
+    }
+}
diff --git a/third_party/rust/wpf-gpu-raster/src/matrix.rs b/third_party/rust/wpf-gpu-raster/src/matrix.rs
new file mode 100644
index 0000000000..ed873410f8
--- /dev/null
+++ b/third_party/rust/wpf-gpu-raster/src/matrix.rs
@@ -0,0 +1,37 @@
+use std::marker::PhantomData;
+
+use crate::types::CoordinateSpace;
+
+pub type CMILMatrix  = CMatrix<CoordinateSpace::Shape,CoordinateSpace::Device>;
+#[derive(Default, Clone)]
+pub struct CMatrix<InCoordSpace, OutCoordSpace> {
+    _11: f32, _12: f32, _13: f32, _14: f32,
+    _21: f32, _22: f32, _23: f32 , _24: f32,
+    _31: f32, _32: f32, _33: f32, _34: f32,
+     _41: f32, _42: f32, _43: f32, _44: f32,
+    in_coord: PhantomData<InCoordSpace>,
+    out_coord: PhantomData<OutCoordSpace>
+}
+
+impl<InCoordSpace: Default, OutCoordSpace: Default> CMatrix<InCoordSpace, OutCoordSpace> {
+    pub fn Identity() -> Self { let mut ret: Self = Default::default();
+        ret._11 = 1.;
+        ret._22 = 1.;
+        ret._33 = 1.;
+        ret._44 = 1.;
+        ret
+    }
+    pub fn GetM11(&self) -> f32 { self._11 }
+    pub fn GetM12(&self) -> f32 { self._12 }
+    pub fn GetM21(&self) -> f32 { self._21 }
+    pub fn GetM22(&self) -> f32 { self._22 }
+    pub fn GetDx(&self) -> f32 { self._41 }
+    pub fn GetDy(&self) -> f32 { self._42 }
+
+    pub fn SetM11(&mut self, r: f32) { self._11 = r}
+    pub fn SetM12(&mut self, r: f32) { self._12 = r}
+    pub fn SetM21(&mut self, r: f32) { self._21 = r}
+    pub fn SetM22(&mut self, r: f32) { self._22 = r}
+    pub fn SetDx(&mut self, dx: f32) { self._41 = dx }
+    pub fn SetDy(&mut self, dy: f32) { self._42 = dy }
+}
\ No newline at end of file
diff --git a/third_party/rust/wpf-gpu-raster/src/notes b/third_party/rust/wpf-gpu-raster/src/notes
new file mode 100644
index 0000000000..2737f1903e
--- /dev/null
+++ b/third_party/rust/wpf-gpu-raster/src/notes
@@ -0,0 +1,12 @@
+crossing goto
+./MultiSpaceRectF.inl:70:5: error: call to implicitly-deleted default constructor of 'union (anonymous union at ./MultiSpaceRectF.inl:138:5)'
+
+
+Rust conversion
+---------------
+CEdge is a singly linked list
+
+Future
+------
+When flatening curves if we try to flatten at integer values
+we can avoid the ComplexSpan code path.
diff --git a/third_party/rust/wpf-gpu-raster/src/nullable_ref.rs b/third_party/rust/wpf-gpu-raster/src/nullable_ref.rs
new file mode 100644
index 0000000000..1e8389e5b7
--- /dev/null
+++ b/third_party/rust/wpf-gpu-raster/src/nullable_ref.rs
@@ -0,0 +1,53 @@
+use std::{marker::PhantomData, ops::Deref};
+
+pub struct Ref<'a, T> {
+    ptr: *const T,
+    _phantom: PhantomData<&'a T>
+}
+
+impl<'a, T> Copy for Ref<'a, T> { }
+
+impl<'a, T> Clone for Ref<'a, T> {
+    fn clone(&self) -> Self {
+        *self
+    }
+}
+
+impl<'a, T> Ref<'a, T> {
+    pub fn new(p: &'a T) -> Self {
+        Ref { ptr: p as *const T, _phantom: PhantomData}
+    }
+    pub unsafe fn null() -> Self {
+        Ref { ptr: std::ptr::null(), _phantom: PhantomData}
+    }
+    pub fn is_null(&self) -> bool {
+        self.ptr.is_null()
+    }
+    pub fn get_ref(self) -> &'a T {
+        unsafe { &*self.ptr }
+    }
+}
+
+impl<'a, T> PartialEq for Ref<'a, T> {
+    fn eq(&self, other: &Self) -> bool {
+        self.ptr == other.ptr && self._phantom == other._phantom
+    }
+}
+
+impl<'a, T> PartialOrd for Ref<'a, T> {
+    fn partial_cmp(&self, other: &Self) -> Option<std::cmp::Ordering> {
+        match self.ptr.partial_cmp(&other.ptr) {
+            Some(core::cmp::Ordering::Equal) => {}
+            ord => return ord,
+        }
+        self._phantom.partial_cmp(&other._phantom)
+    }
+}
+
+impl<'a, T> Deref for Ref<'a, T> {
+    type Target = T;
+
+    fn deref(&self) -> &Self::Target {
+        unsafe { &*self.ptr }
+    }
+}
\ No newline at end of file
diff --git a/third_party/rust/wpf-gpu-raster/src/real.rs b/third_party/rust/wpf-gpu-raster/src/real.rs
new file mode 100644
index 0000000000..a9144ec149
--- /dev/null
+++ b/third_party/rust/wpf-gpu-raster/src/real.rs
@@ -0,0 +1,163 @@
+pub mod CFloatFPU {
+    // Maximum allowed argument for SmallRound
+    // const sc_uSmallMax: u32 = 0xFFFFF;
+
+    // Binary representation of static_cast<float>(sc_uSmallMax)
+    const sc_uBinaryFloatSmallMax: u32 = 0x497ffff0;
+
+    fn LargeRound(x: f32) -> i32 {
+        //XXX: the SSE2 version is probably slower than a naive SSE4 implementation that can use roundss
+        #[cfg(target_feature = "sse2")]
+        unsafe {
+            #[cfg(target_arch = "x86")]
+            use std::arch::x86::{__m128, _mm_set_ss, _mm_cvtss_si32, _mm_cvtsi32_ss, _mm_sub_ss, _mm_cmple_ss, _mm_store_ss, _mm_setzero_ps};
+            #[cfg(target_arch = "x86_64")]
+            use std::arch::x86_64::{__m128, _mm_set_ss, _mm_cvtss_si32, _mm_cvtsi32_ss, _mm_sub_ss, _mm_cmple_ss, _mm_store_ss, _mm_setzero_ps};
+
+            let given: __m128 = _mm_set_ss(x);                       // load given value
+            let result = _mm_cvtss_si32(given);
+            let rounded: __m128 = _mm_setzero_ps();             // convert it to integer (rounding mode doesn't matter)
+            let rounded = _mm_cvtsi32_ss(rounded, result);   // convert back to float
+            let diff = _mm_sub_ss(rounded, given);           // diff = (rounded - given)
+            let negHalf = _mm_set_ss(-0.5);                 // load -0.5f
+            let mask = _mm_cmple_ss(diff, negHalf);          // get all-ones if (rounded - given) < -0.5f
+            let mut correction: i32 = 0;                                 
+            _mm_store_ss((&mut correction) as *mut _ as *mut _, mask); // get comparison result as integer
+            return result - correction;                         // correct the result of rounding
+        }
+        #[cfg(not(target_feature = "sse2"))]
+        return (x + 0.5).floor() as i32;
+    }
+
+
+    //+------------------------------------------------------------------------
+//
+//  Function:   CFloatFPU::SmallRound
+//
+//  Synopsis:   Convert given floating point value to nearest integer.
+//              Half-integers are rounded up.
+//
+//              Important: this routine is fast but restricted:
+//              given x should be within (-(0x100000-.5) < x < (0x100000-.5))
+//
+//  Details:    Implementation has abnormal looking that use to confuse
+//              many people. However, it indeed works, being tested
+//              thoroughly on x86 and ia64 platforms for literally
+//              each possible argument values in the given range.
+//              
+//  More details:
+//      Implementation is based on the knowledge of floating point
+//      value representation. This 32-bits value consists of three parts:
+//      v & 0x80000000 = sign
+//      v & 0x7F800000 = exponent
+//      v & 0x007FFFFF - mantissa
+//
+//      Let N to be a floating point number within -0x400000 <= N <= 0x3FFFFF.
+//      The sum (S = 0xC00000 + N) thus will satisfy Ox800000 <= S <= 0xFFFFFF.
+//      All the numbers within this range (sometimes referred to as "binade")
+//      have same position of most significant bit, i.e. 0x800000.
+//      Therefore they are normalized equal way, thus
+//      providing the weights on mantissa's bits to be the same
+//      as integer numbers have. In other words, to get
+//      integer value of floating point S, when Ox800000 <= S <= 0xFFFFFF,
+//      we can just throw away the exponent and sign, and add assumed
+//      most significant bit (that is always 1 and therefore is not stored
+//      in floating point value):
+//      (int)S = (<float S as int> & 0x7FFFFF | 0x800000);
+//      To get given N in as integer, we need to subtract back
+//      the value 0xC00000 that was added in order to obtain
+//      proper normalization:
+//      N = (<float S as int> & 0x7FFFFF | 0x800000) - 0xC00000.
+//      or
+//      N = (<float S as int> & 0x7FFFFF           ) - 0x400000.
+//
+//      Hopefully, the text above explains how
+//      following routine works:
+//        int SmallRound1(float x)
+//        {
+//            union
+//            {
+//                __int32 i;
+//                float f;
+//            } u;
+//
+//            u.f = x + float(0x00C00000);
+//            return ((u.i - (int)0x00400000) << 9) >> 9;
+//        }
+//      Unfortunatelly it is imperfect, due to the way how FPU
+//      use to round intermediate calculation results.
+//      By default, rounding mode is set to "nearest".
+//      This means that when it calculates N+float(0x00C00000),
+//      the 80-bit precise result will not fit in 32-bit float,
+//      so some least significant bits will be thrown away.
+//      Rounding to nearest means that S consisting of intS + fraction,
+//      where 0 <= fraction < 1, will be converted to intS
+//      when fraction < 0.5 and to intS+1 if fraction > 0.5.
+//      What would happen with fraction exactly equal to 0.5?
+//      Smart thing: S will go to intS if intS is even and
+//      to intS+1 if intS is odd. In other words, half-integers
+//      are rounded to nearest even number.
+//      This FPU feature apparently is useful to minimize
+//      average rounding error when somebody is, say,
+//      digitally simulating electrons' behavior in plasma.
+//      However for graphics this is not desired.
+//
+//      We want to move half-integers up, therefore
+//      define SmallRound(x) as {return SmallRound1(x*2+.5) >> 1;}.
+//      This may require more comments.
+//      Let given x = i+f, where i is integer and f is fraction, 0 <= f < 1.
+//      Let's wee what is y = x*2+.5:
+//          y = i*2 + (f*2 + .5) = i*2 + g, where g = f*2 + .5;
+//      If "f" is in the range 0 <= f < .5 (so correct rounding result should be "i"),
+//      then range for "g" is .5 <= g < 1.5. The very first value, .5 will force
+//      SmallRound1 result to be "i*2", due to round-to-even rule; the remaining
+//      will lead to "i*2+1". Consequent shift will throw away extra "1" and give
+//      us desired "i".
+//      When "f" in in the range .5 <= f < 1, then 1.5 <= g < 2.5.
+//      All these values will round to 2, so SmallRound1 will return (2*i+2),
+//      and the final shift will give desired 1+1.
+//
+//      To get final routine looking we need to transform the combines
+//      expression for u.f:
+//            (x*2) + .5 + float(0x00C00000) ==
+//             (x + (.25 + double(0x00600000)) )*2
+//      Note that the ratio "2" means nothing for following operations,
+//      since it affects only exponent bits that are ignored anyway.
+//      So we can save some processor cycles avoiding this multiplication.
+//
+//      And, the very final beautification:
+//      to avoid subtracting 0x00400000 let's ignore this bit.
+//      This mean that we effectively decrease available range by 1 bit,
+//      but we're chasing for performance and found it acceptable.
+//      So 
+//         return ((u.i - (int)0x00400000) << 9) >> 9;
+//      is converted to
+//         return ((u.i                  ) << 10) >> 10;
+//      Eventually, will found that final shift by 10 bits may be combined
+//      with shift by 1 in the definition {return SmallRound1(x*2+.5) >> 1;},
+//      we'll just shift by 11 bits. That's it.
+//      
+//-------------------------------------------------------------------------
+fn SmallRound(x: f32) -> i32
+{
+    //AssertPrecisionAndRoundingMode();
+    debug_assert!(-(0x100000 as f64 -0.5) < x as f64 && (x as f64) < (0x100000 as f64 -0.5));
+
+ 
+    let fi = (x as f64 + (0x00600000 as f64 + 0.25)) as f32;
+    let result = ((fi.to_bits() as i32) << 10) >> 11;
+
+    debug_assert!(x < (result as f32) + 0.5 && x >= (result as f32) - 0.5);
+    return result;
+}
+
+pub fn Round(x: f32) -> i32
+{
+    // cut off sign
+    let xAbs: u32 = x.to_bits() & 0x7FFFFFFF;
+
+    return if xAbs <= sc_uBinaryFloatSmallMax {SmallRound(x)} else {LargeRound(x)};
+}
+}
+
+macro_rules! TOREAL { ($e: expr) => { $e as REAL } }
diff --git a/third_party/rust/wpf-gpu-raster/src/tri_rasterize.rs b/third_party/rust/wpf-gpu-raster/src/tri_rasterize.rs
new file mode 100644
index 0000000000..28fbf6a73f
--- /dev/null
+++ b/third_party/rust/wpf-gpu-raster/src/tri_rasterize.rs
@@ -0,0 +1,190 @@
+/* The rasterization code here is based off of piglit/tests/general/triangle-rasterization.cpp:
+
+    /**************************************************************************
+     *
+     * Copyright 2012 VMware, Inc.
+     * All Rights Reserved.
+     *
+     * Permission is hereby granted, free of charge, to any person obtaining a
+     * copy of this software and associated documentation files (the
+     * "Software"), to deal in the Software without restriction, including
+     * without limitation the rights to use, copy, modify, merge, publish,
+     * distribute, sub license, and/or sell copies of the Software, and to
+     * permit persons to whom the Software is furnished to do so, subject to
+     * the following conditions:
+     *
+     * The above copyright notice and this permission notice (including the
+     * next paragraph) shall be included in all copies or substantial portions
+     * of the Software.
+     *
+     * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+     * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+     * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+     * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
+     * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+     * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+     * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+     *
+     **************************************************************************/
+
+*/
+
+use std::ops::Index;
+use crate::OutputVertex;
+#[derive(Debug)]
+struct Vertex {
+    x: f32,
+    y: f32,
+    coverage: f32
+}
+#[derive(Debug)]
+struct Triangle {
+    v: [Vertex; 3],
+}
+
+impl Index<usize> for Triangle  {
+    type Output = Vertex;
+
+    fn index(&self, index: usize) -> &Self::Output {
+        &self.v[index]
+    }
+}
+
+// D3D11 mandates 8 bit subpixel precision:
+// https://microsoft.github.io/DirectX-Specs/d3d/archive/D3D11_3_FunctionalSpec.htm#CoordinateSnapping
+const FIXED_SHIFT: i32 = 8;
+const FIXED_ONE: f32 = (1 << FIXED_SHIFT) as f32;
+
+/* Proper rounding of float to integer */
+fn iround(mut v: f32) -> i64 {
+    if v > 0.0 {
+        v += 0.5;
+    }
+    if v < 0.0 {
+        v -= 0.5;
+    }
+    return v as i64
+}
+
+/* Based on http://devmaster.net/forums/topic/1145-advanced-rasterization */
+fn rast_triangle(buffer: &mut [u8], width: usize, height: usize, tri: &Triangle) {
+    let center_offset = -0.5;
+
+    let mut coverage1 = tri[0].coverage;
+    let mut coverage2 = tri[1].coverage;
+    let mut coverage3 = tri[2].coverage;
+
+    /* fixed point coordinates */
+    let mut x1 = iround(FIXED_ONE * (tri[0].x + center_offset));
+    let     x2 = iround(FIXED_ONE * (tri[1].x + center_offset));
+    let mut x3 = iround(FIXED_ONE * (tri[2].x + center_offset));
+
+    let mut y1 = iround(FIXED_ONE * (tri[0].y + center_offset));
+    let     y2 = iround(FIXED_ONE * (tri[1].y + center_offset));
+    let mut y3 = iround(FIXED_ONE * (tri[2].y + center_offset));
+
+
+    /* Force correct vertex order */
+    let cross = (x2 - x1) * (y3 - y2) - (y2 - y1) * (x3 - x2);
+    if cross > 0 {
+        std::mem::swap(&mut x1, &mut x3);
+        std::mem::swap(&mut y1, &mut y3);
+        // I don't understand why coverage 2 and 3 are swapped instead of 1 and 3
+        std::mem::swap(&mut coverage2, &mut coverage3);
+    } else {
+        std::mem::swap(&mut coverage1, &mut coverage3);
+    }
+
+    /* Deltas */
+    let dx12 = x1 - x2;
+    let dx23 = x2 - x3;
+    let dx31 = x3 - x1;
+
+    let dy12 = y1 - y2;
+    let dy23 = y2 - y3;
+    let dy31 = y3 - y1;
+
+    /* Fixed-point deltas */
+    let fdx12 = dx12 << FIXED_SHIFT;
+    let fdx23 = dx23 << FIXED_SHIFT;
+    let fdx31 = dx31 << FIXED_SHIFT;
+
+    let fdy12 = dy12 << FIXED_SHIFT;
+    let fdy23 = dy23 << FIXED_SHIFT;
+    let fdy31 = dy31 << FIXED_SHIFT;
+
+    /* Bounding rectangle */
+    let mut minx = x1.min(x2).min(x3) >> FIXED_SHIFT;
+    let mut maxx = x1.max(x2).max(x3) >> FIXED_SHIFT;
+
+    let mut miny = y1.min(y2).min(y3) >> FIXED_SHIFT;
+    let mut maxy = y1.max(y2).max(y3) >> FIXED_SHIFT;
+
+    minx = minx.max(0);
+    maxx = maxx.min(width as i64 - 1);
+
+    miny = miny.max(0);
+    maxy = maxy.min(height as i64 - 1);
+
+    /* Half-edge constants */
+    let mut c1 = dy12 * x1 - dx12 * y1;
+    let mut c2 = dy23 * x2 - dx23 * y2;
+    let mut c3 = dy31 * x3 - dx31 * y3;
+
+    /* Correct for top-left filling convention */
+    if dy12 < 0 || (dy12 == 0 && dx12 < 0) { c1 += 1 }
+    if dy23 < 0 || (dy23 == 0 && dx23 < 0) { c2 += 1 }
+    if dy31 < 0 || (dy31 == 0 && dx31 < 0) { c3 += 1 }
+
+    let mut cy1 = c1 + dx12 * (miny << FIXED_SHIFT) - dy12 * (minx << FIXED_SHIFT);
+    let mut cy2 = c2 + dx23 * (miny << FIXED_SHIFT) - dy23 * (minx << FIXED_SHIFT);
+    let mut cy3 = c3 + dx31 * (miny << FIXED_SHIFT) - dy31 * (minx << FIXED_SHIFT);
+    //dbg!(minx, maxx, tri, cross);
+    /* Perform rasterization */
+    let mut buffer = &mut buffer[miny as usize * width..];
+    for _y in miny..=maxy {
+        let mut cx1 = cy1;
+        let mut cx2 = cy2;
+        let mut cx3 = cy3;
+
+        for x in minx..=maxx {
+            if cx1 > 0 && cx2 > 0 && cx3 > 0 {
+                // cross is equal to 2*area of the triangle.
+                // we can normalize cx by 2*area to get barycentric coords.
+                let area = cross.abs() as f32;
+                let bary = (cx1 as f32 / area, cx2 as f32 / area, cx3 as f32 / area);
+                let coverages = coverage1 * bary.0 + coverage2 * bary.1 + coverage3 * bary.2;
+                let color = (coverages * 255. + 0.5) as u8;
+
+                buffer[x as usize] = color;
+            }
+
+            cx1 -= fdy12;
+            cx2 -= fdy23;
+            cx3 -= fdy31;
+        }
+
+        cy1 += fdx12;
+        cy2 += fdx23;
+        cy3 += fdx31;
+
+        buffer = &mut buffer[width..];
+    }
+}
+
+pub fn rasterize_to_mask(vertices: &[OutputVertex], width: u32, height: u32) -> Box<[u8]> {
+    let mut mask = vec![0; (width * height) as usize];
+    for n in (0..vertices.len()).step_by(3) {
+        let tri =
+            [&vertices[n], &vertices[n+1], &vertices[n+2]];
+
+        let tri = Triangle { v: [
+            Vertex { x: tri[0].x, y: tri[0].y, coverage: tri[0].coverage},
+            Vertex { x: tri[1].x, y: tri[1].y, coverage: tri[1].coverage},
+            Vertex { x: tri[2].x, y: tri[2].y, coverage: tri[2].coverage}
+            ]
+        };
+        rast_triangle(&mut mask, width as usize, height as usize, &tri);
+    }
+    mask.into_boxed_slice()
+}
diff --git a/third_party/rust/wpf-gpu-raster/src/types.rs b/third_party/rust/wpf-gpu-raster/src/types.rs
new file mode 100644
index 0000000000..696976f185
--- /dev/null
+++ b/third_party/rust/wpf-gpu-raster/src/types.rs
@@ -0,0 +1,181 @@
+pub(crate) type LONG = i32;
+pub(crate) type INT = i32;
+pub(crate) type UINT = u32;
+pub(crate) type ULONG = u32;
+pub(crate) type DWORD = ULONG;
+pub(crate) type WORD = u16;
+pub(crate) type LONGLONG = i64;
+pub(crate) type ULONGLONG = u64;
+pub(crate) type BYTE = u8;
+pub(crate) type FLOAT = f32;
+pub(crate) type REAL = FLOAT;
+pub(crate) type HRESULT = LONG;
+
+pub(crate) const S_OK: HRESULT = 0;
+pub(crate) const INTSAFE_E_ARITHMETIC_OVERFLOW: HRESULT = 0x80070216;
+pub(crate) const WGXERR_VALUEOVERFLOW: HRESULT = INTSAFE_E_ARITHMETIC_OVERFLOW;
+pub(crate) const WINCODEC_ERR_VALUEOVERFLOW: HRESULT = INTSAFE_E_ARITHMETIC_OVERFLOW;
+const fn MAKE_HRESULT(sev: LONG,fac: LONG,code: LONG) -> HRESULT {
+    ( (((sev)<<31) | ((fac)<<16) | ((code))) )
+}
+
+const FACILITY_WGX: LONG = 0x898;
+
+
+const fn MAKE_WGXHR( sev: LONG, code: LONG) -> HRESULT {
+        MAKE_HRESULT( sev, FACILITY_WGX, (code) )
+}
+    
+const fn MAKE_WGXHR_ERR( code: LONG ) -> HRESULT 
+{
+        MAKE_WGXHR( 1, code )
+}
+
+pub const WGXHR_CLIPPEDTOEMPTY: HRESULT =                MAKE_WGXHR(0, 1);
+pub const WGXHR_EMPTYFILL: HRESULT =                      MAKE_WGXHR(0, 2);
+pub const WGXHR_INTERNALTEMPORARYSUCCESS: HRESULT =      MAKE_WGXHR(0, 3);
+pub const WGXHR_RESETSHAREDHANDLEMANAGER: HRESULT =      MAKE_WGXHR(0, 4);
+
+pub const WGXERR_BADNUMBER: HRESULT =                     MAKE_WGXHR_ERR(0x00A);   //  4438
+
+pub fn FAILED(hr: HRESULT) -> bool {
+    hr != S_OK
+}
+pub trait NullPtr {
+    fn make() -> Self;
+}
+
+impl<T> NullPtr for *mut T {
+    fn make() -> Self {
+        std::ptr::null_mut()
+    }
+}
+
+impl<T> NullPtr for *const T {
+    fn make() -> Self {
+        std::ptr::null()
+    }
+}
+
+pub fn NULL<T: NullPtr>() -> T {
+    T::make()
+}
+#[derive(Default, Clone)]
+pub struct RECT {
+    pub left: LONG,
+    pub top: LONG,
+    pub right: LONG,
+    pub bottom: LONG,
+}
+#[derive(Default, Clone, Copy, PartialEq, Eq)]
+pub struct POINT {
+    pub x: LONG,
+    pub y: LONG
+}
+#[derive(Clone, Copy)]
+pub struct MilPoint2F
+{
+    pub X: FLOAT,
+    pub Y: FLOAT,
+}
+
+#[derive(Default, Clone)]
+pub struct MilPointAndSizeL
+{
+    pub X: INT,
+    pub Y: INT,
+    pub Width: INT,
+    pub Height: INT,
+}
+
+pub type CMILSurfaceRect = RECT;
+
+#[derive(PartialEq)]
+pub enum MilAntiAliasMode {
+    None = 0,
+    EightByEight = 1,
+}
+#[derive(PartialEq, Clone, Copy)]
+pub enum MilFillMode {
+    Alternate = 0,
+    Winding = 1,
+}
+
+pub const    PathPointTypeStart: u8           = 0;    // move, 1 point
+pub const    PathPointTypeLine: u8            = 1;    // line, 1 point
+pub const    PathPointTypeBezier: u8          = 3;    // default Bezier (= cubic Bezier), 3 points
+pub const    PathPointTypePathTypeMask: u8    = 0x07; // type mask (lowest 3 bits).
+pub const    PathPointTypeCloseSubpath: u8    = 0x80; // closed flag
+
+
+pub type DynArray<T> = Vec<T>;
+
+pub trait DynArrayExts<T> {
+    fn Reset(&mut self, shrink: bool);
+    fn GetCount(&self) -> usize;
+    fn SetCount(&mut self, count: usize);
+    fn GetDataBuffer(&self) -> &[T];
+}
+
+impl<T> DynArrayExts<T> for DynArray<T> {
+    fn Reset(&mut self, shrink: bool) {
+        self.clear();
+        if shrink {
+            self.shrink_to_fit();
+        }
+    }
+    fn GetCount(&self) -> usize {
+        self.len()
+    }
+    fn SetCount(&mut self, count: usize) {
+        assert!(count <= self.len());
+        self.truncate(count);
+    }
+
+    fn GetDataBuffer(&self) -> &[T] {
+        self
+    }
+}
+
+pub struct CHwPipelineBuilder;
+
+pub mod CoordinateSpace {
+    #[derive(Default, Clone)]
+    pub struct Shape;
+    #[derive(Default, Clone)]
+    pub struct Device;
+}
+
+pub trait IShapeData {
+    fn GetFillMode(&self) -> MilFillMode;
+}
+
+pub type MilVertexFormat = DWORD;
+
+pub enum MilVertexFormatAttribute {
+    MILVFAttrNone = 0x0,
+    MILVFAttrXY = 0x1,
+    MILVFAttrZ = 0x2,
+    MILVFAttrXYZ = 0x3,
+    MILVFAttrNormal = 0x4,
+    MILVFAttrDiffuse = 0x8,
+    MILVFAttrSpecular = 0x10,
+    MILVFAttrUV1 = 0x100,
+    MILVFAttrUV2 = 0x300,
+    MILVFAttrUV3 = 0x700,
+    MILVFAttrUV4 = 0xf00,
+    MILVFAttrUV5 = 0x1f00,
+    MILVFAttrUV6 = 0x3f00,
+    MILVFAttrUV7 = 0x7f00,
+    MILVFAttrUV8 = 0xff00,      // Vertex fields that are pre-generated
+
+}
+
+pub struct CHwPipeline;
+
+pub struct CBufferDispenser;
+#[derive(Default)]
+pub struct PointXYA
+{
+    pub x: f32,pub y: f32, pub a: f32,
+}