Major speedups with SIMD.

Cargo.lock

@@ -1223,7 +1223,7 @@ name = "euc"
 version = "0.5.1"
 source = "git+https://github.com/zesterer/euc.git#c9a7c17a03d45fce00caeeca09afa1e1558cd183"
 dependencies = [
- "vek",
+ "vek 0.11.2",
 ]
 
 [[package]]
@@ -4468,7 +4468,7 @@ version = "0.1.0"
 source = "git+https://gitlab.com/yusdacra/treeculler.git#efcf5283cf386117a7e654abdaa45ef664a08e42"
 dependencies = [
  "num-traits",
- "vek",
+ "vek 0.11.2",
 ]
 
 [[package]]
@@ -4636,6 +4636,19 @@ dependencies = [
  "static_assertions",
 ]
 
+[[package]]
+name = "vek"
+version = "0.12.0"
+source = "git+https://gitlab.com/veloren/vek.git?branch=fix_intrinsics#237a78528b505f34f6dde5dc77db3b642388fe4a"
+dependencies = [
+ "approx",
+ "num-integer",
+ "num-traits",
+ "rustc_version",
+ "serde",
+ "static_assertions",
+]
+
 [[package]]
 name = "veloren-chat-cli"
 version = "0.6.0"
@@ -4663,7 +4676,7 @@ dependencies = [
  "specs",
  "tracing",
  "uvth 3.1.1",
- "vek",
+ "vek 0.12.0",
  "veloren-common",
  "veloren_network",
 ]
@@ -4694,7 +4707,7 @@ dependencies = [
  "specs-idvs",
  "sum_type",
  "tracing",
- "vek",
+ "vek 0.12.0",
 ]
 
 [[package]]
@@ -4725,7 +4738,7 @@ dependencies = [
  "tiny_http",
  "tracing",
  "uvth 3.1.1",
- "vek",
+ "vek 0.12.0",
  "veloren-common",
  "veloren-world",
  "veloren_network",
@@ -4786,7 +4799,7 @@ dependencies = [
  "tracing-subscriber",
  "treeculler",
  "uvth 3.1.1",
- "vek",
+ "vek 0.12.0",
  "veloren-client",
  "veloren-common",
  "veloren-server",
@@ -4805,7 +4818,7 @@ dependencies = [
  "libloading 0.6.2",
  "notify",
  "tracing",
- "vek",
+ "vek 0.12.0",
  "veloren-common",
 ]
 
@@ -4835,7 +4848,7 @@ dependencies = [
  "serde",
  "tracing",
  "tracing-subscriber",
- "vek",
+ "vek 0.12.0",
  "veloren-common",
 ]
 

Cargo.toml

@@ -80,3 +80,4 @@ debug = 1
 [patch.crates-io]
 # cpal conflict fix isn't released yet
 winit = { git = "https://github.com/Imberflur/winit.git", branch = "macos-test" }
+vek = { git = "https://gitlab.com/veloren/vek.git", branch = "fix_intrinsics" }

client/Cargo.toml

@@ -19,6 +19,6 @@ num_cpus = "1.10.1"
 tracing = { version = "0.1", default-features = false }
 rayon = "^1.3.0"
 specs = { git = "https://github.com/amethyst/specs.git", rev = "7a2e348ab2223818bad487695c66c43db88050a5" }
-vek = { version = "0.11.2", features = ["repr_simd", "serde"] }
+vek = { version = "0.12.0", features = ["platform_intrinsics", "serde"] }
 hashbrown = { version = "0.7.2", features = ["rayon", "serde", "nightly"] }
 authc = { git = "https://gitlab.com/veloren/auth.git", rev = "b943c85e4a38f5ec60cd18c34c73097640162bfe" }

common/Cargo.toml

@@ -13,7 +13,7 @@ specs-idvs = { git = "https://gitlab.com/veloren/specs-idvs.git", branch = "spec
 
 roots = "0.0.5"
 specs = { git = "https://github.com/amethyst/specs.git", features = ["serde", "storage-event-control"], rev = "7a2e348ab2223818bad487695c66c43db88050a5" }
-vek = { version = "0.11.2", features = ["repr_simd", "serde"] }
+vek = { version = "0.12.0", features = ["platform_intrinsics", "serde"] }
 dot_vox = "4.0"
 image = { version = "0.23.8", default-features = false, features = ["png"] }
 serde = { version = "1.0.110", features = ["derive"] }

server/Cargo.toml

@@ -17,7 +17,7 @@ specs-idvs = { git = "https://gitlab.com/veloren/specs-idvs.git", branch = "spec
 
 tracing = "0.1"
 specs = { git = "https://github.com/amethyst/specs.git", features = ["shred-derive"], rev = "7a2e348ab2223818bad487695c66c43db88050a5" }
-vek = { version = "0.11.2", features = ["repr_simd", "serde"] }
+vek = { version = "0.12.0", features = ["platform_intrinsics", "serde"] }
 uvth = "3.1.1"
 futures-util = "0.3"
 futures-executor = "0.3"

voxygen/Cargo.toml

@@ -38,7 +38,7 @@ specs = { git = "https://github.com/amethyst/specs.git", rev = "7a2e348ab2223818
 specs-idvs = { git = "https://gitlab.com/veloren/specs-idvs.git", branch = "specs-git" }
 
 # Mathematics
-vek = { version = "0.11.2", features = ["repr_simd", "serde"] }
+vek = { version = "0.12.0", features = ["platform_intrinsics", "serde"] }
 
 # Controller
 gilrs = { version = "0.7", features = ["serde"] }

voxygen/src/anim/Cargo.toml

@@ -17,7 +17,7 @@ be-dyn-lib = []
 default = ["be-dyn-lib"]
 
 [dependencies]
-vek = { version = "0.11.2", features = ["repr_simd", "serde"] }
+vek = { version = "0.12.0", features = ["platform_intrinsics", "serde"] }
 common = { package = "veloren-common", path = "../../../common" }
 libloading = { version = "0.6.2", optional = true }
 notify = { version = "5.0.0-pre.2", optional = true }

voxygen/src/scene/figure/mod.rs

@@ -331,9 +331,6 @@ impl FigureMgr {
 
     #[allow(clippy::redundant_pattern_matching)]
     // TODO: Pending review in #587
-    // NOTE: All of the "useless" conversion reported here allow us to abstract over repr_c vs.
-    // simd vectors, so fixing this warning would make the code worse in this case.
-    #[allow(clippy::useless_conversion)]
     pub fn update_lighting(&mut self, scene_data: &SceneData) {
         let ecs = scene_data.state.ecs();
         for (entity, light_emitter) in (&ecs.entities(), &ecs.read_storage::<LightEmitter>()).join()
@@ -411,9 +408,6 @@ impl FigureMgr {
 
     #[allow(clippy::or_fun_call)]
     // TODO: Pending review in #587
-    // NOTE: All of the "useless" conversion reported here allow us to abstract over repr_c vs.
-    // simd vectors, so fixing this warning would make the code worse in this case.
-    #[allow(clippy::useless_conversion)]
     pub fn maintain(
         &mut self,
         renderer: &mut Renderer,
@@ -422,6 +416,7 @@ impl FigureMgr {
         visible_psr_bounds: math::Aabr<f32>,
         camera: &Camera,
     ) -> anim::vek::Aabb<f32> {
+        let visible_psr_bounds = math::Aabr::from(visible_psr_bounds);
         let state = scene_data.state;
         let time = state.get_time();
         let tick = scene_data.tick;
@@ -455,9 +450,10 @@ impl FigureMgr {
             let ray_mat = ray_mat * math::Mat4::translation_3d(-focus_off);
 
             let collides_with_aabr = |a: math::Aabr<f32>, b: math::Aabr<f32>| {
-                a.min.partial_cmple(&b.max).reduce_and() && a.max.partial_cmpge(&b.min).reduce_and()
+                let min = math::Vec4::new(a.min.x, a.min.y, b.min.x, b.min.y);
+                let max = math::Vec4::new(b.max.x, b.max.y, a.max.x, a.max.y);
+                min.partial_cmple_simd(max).reduce_and()
             };
-            // println!("Aabr: {:?}", visible_bounds);
             move |pos: (anim::vek::Vec3<f32>,), radius: f32| {
                 // Short circuit when there are no shadows to cast.
                 if !can_shadow_sun {

voxygen/src/scene/math.rs

@@ -1,9 +1,9 @@
 use core::{iter, mem};
 use hashbrown::HashMap;
 use num::traits::Float;
-// pub use vek::{geom::repr_simd::*, mat::repr_simd::column_major::Mat4, ops::*,
+pub use vek::{geom::repr_simd::*, mat::repr_simd::column_major::Mat4, ops::*, vec::repr_simd::*};
-// vec::repr_simd::*};
+// pub use vek::{geom::repr_c::*, mat::repr_c::column_major::Mat4, ops::*,
-pub use vek::{geom::repr_c::*, mat::repr_c::column_major::Mat4, ops::*, vec::repr_c::*};
+// vec::repr_c::*};
 
 pub fn aabb_to_points<T: Float>(bounds: Aabb<T>) -> [Vec3<T>; 8] {
     [
@@ -18,7 +18,7 @@ pub fn aabb_to_points<T: Float>(bounds: Aabb<T>) -> [Vec3<T>; 8] {
     ]
 }
 
-/// Each Vec4 <a, b, c, d> should be interpreted as reprenting plane
+/// Each Vec4 <a, b, c, -d> should be interpreted as reprenting plane
 /// equation
 ///
 /// a(x - x0) + b(y - y0) + c(z - z0) = 0, i.e.
@@ -28,23 +28,23 @@ pub fn aabb_to_points<T: Float>(bounds: Aabb<T>) -> [Vec3<T>; 8] {
 /// ax + by + cz = d
 ///
 /// where d is the distance of the plane from the origin.
-pub fn aabb_to_planes<T: Float>(bounds: Aabb<T>) -> [(Vec3<T>, T); 6] {
+pub fn aabb_to_planes<T: Float>(bounds: Aabb<T>) -> [Vec4<T>; 6] {
     let zero = T::zero();
     let one = T::one();
     let bounds = bounds.map(|e| e.abs());
     [
         // bottom
-        (Vec3::new(zero, -one, zero), bounds.min.y),
+        Vec4::new(zero, -one, zero, -bounds.min.y),
         // top
-        (Vec3::new(zero, one, zero), bounds.max.y),
+        Vec4::new(zero, one, zero, -bounds.max.y),
         // left
-        (Vec3::new(-one, zero, zero), bounds.min.x),
+        Vec4::new(-one, zero, zero, -bounds.min.x),
         // right
-        (Vec3::new(one, zero, zero), bounds.max.x),
+        Vec4::new(one, zero, zero, -bounds.max.x),
         // near
-        (Vec3::new(zero, zero, -one), bounds.min.z),
+        Vec4::new(zero, zero, -one, -bounds.min.z),
         // far
-        (Vec3::new(zero, zero, one), bounds.max.z),
+        Vec4::new(zero, zero, one, -bounds.max.z),
     ]
 }
 
@@ -87,11 +87,11 @@ pub fn calc_view_frustum_world_coord<T: Float + MulAdd<T, T, Output = T>>(
     world_pts
 }
 
-pub fn point_plane_distance<T: Float>(point: Vec3<T>, (norm, dist): (Vec3<T>, T)) -> T {
+pub fn point_plane_distance<T: Float>(point: Vec3<T>, norm_dist: Vec4<T>) -> T {
-    norm.dot(point) - dist
+    norm_dist.dot(Vec4::from_point(point))
 }
 
-pub fn point_before_plane<T: Float>(point: Vec3<T>, plane: (Vec3<T>, T)) -> bool {
+pub fn point_before_plane<T: Float>(point: Vec3<T>, plane: Vec4<T>) -> bool {
     point_plane_distance(point, plane) > T::zero()
 }
 
@@ -100,7 +100,7 @@ pub fn point_before_plane<T: Float>(point: Vec3<T>, plane: (Vec3<T>, T)) -> bool
 /// (this implies that the polygon must be non-degenerate).
 pub fn clip_points_by_plane<T: Float + MulAdd<T, T, Output = T> + core::fmt::Debug>(
     points: &mut Vec<Vec3<T>>,
-    plane: (Vec3<T>, T),
+    plane: Vec4<T>,
     intersection_points: &mut Vec<Vec3<T>>,
 ) -> bool {
     if points.len() < 3 {
@@ -108,18 +108,17 @@ pub fn clip_points_by_plane<T: Float + MulAdd<T, T, Output = T> + core::fmt::Deb
     }
     // NOTE: Guaranteed to succeed since points.len() > 3.
     let mut current_point = points[points.len() - 1];
-    let (norm, dist) = plane;
     let intersect_plane_edge = |a, b| {
-        let diff = b - a;
+        let diff: Vec3<_> = b - a;
-        let t = norm.dot(diff);
+        let t = plane.dot(Vec4::from_direction(diff));
         if t == T::zero() {
             None
         } else {
-            let t = (dist - norm.dot(a)) / t;
+            let t = -(plane.dot(Vec4::from_point(a)) / t);
             if t < T::zero() || T::one() < t {
                 None
             } else {
-                Some(diff.mul_add(Vec3::broadcast(t), a))
+                Some(diff * t + a)
             }
         }
     };
@@ -247,7 +246,7 @@ fn append_intersection_points<T: Float + core::fmt::Debug>(
 
 pub fn clip_object_by_plane<T: Float + MulAdd<T, T, Output = T> + core::fmt::Debug>(
     polys: &mut Vec<Vec<Vec3<T>>>,
-    plane: (Vec3<T>, T),
+    plane: Vec4<T>,
     tolerance: T,
 ) {
     let mut intersection_points = Vec::new();

voxygen/src/scene/terrain.rs

@@ -2365,9 +2365,6 @@ impl<V: RectRasterableVol> Terrain<V> {
     }
 
     /// Maintain terrain data. To be called once per tick.
-    // NOTE: All of the "useless" conversion reported here allow us to abstract over repr_c vs.
-    // simd vectors, so fixing this warning would make the code worse in this case.
-    #[allow(clippy::useless_conversion)]
     #[allow(clippy::for_loops_over_fallibles)] // TODO: Pending review in #587
     #[allow(clippy::len_zero)] // TODO: Pending review in #587
     pub fn maintain(
@@ -2689,7 +2686,7 @@ impl<V: RectRasterableVol> Terrain<V> {
         // Update chunk visibility
         let chunk_sz = V::RECT_SIZE.x as f32;
         for (pos, chunk) in &mut self.chunks {
-            let chunk_pos = pos.map(|e| e as f32 * chunk_sz);
+            let chunk_pos = pos.as_::<f32>() * chunk_sz;
 
             chunk.can_shadow_sun = false;
 
@@ -2745,33 +2742,32 @@ impl<V: RectRasterableVol> Terrain<V> {
 
         // PSCs: Potential shadow casters
         let ray_direction = scene_data.get_sun_dir();
-        let collides_with_aabr = |a: math::Aabr<f32>, b: math::Aabr<f32>| {
+        let collides_with_aabr = |a: math::Aabb<f32>, b: math::Aabr<f32>| {
-            a.min.partial_cmple(&b.max).reduce_and() && a.max.partial_cmpge(&b.min).reduce_and()
+            let min = math::Vec4::new(a.min.x, a.min.y, b.min.x, b.min.y);
+            let max = math::Vec4::new(b.max.x, b.max.y, a.max.x, a.max.y);
+            min.partial_cmple_simd(max).reduce_and()
         };
         let (visible_light_volume, visible_psr_bounds) = if ray_direction.z < 0.0
             && renderer.render_mode().shadow.is_map()
         {
-            let visible_bounding_box = Aabb {
+            let visible_bounding_box = math::Aabb::<f32> {
-                min: visible_bounding_box.min - focus_off,
+                min: math::Vec3::from(visible_bounding_box.min - focus_off),
-                max: visible_bounding_box.max - focus_off,
+                max: math::Vec3::from(visible_bounding_box.max - focus_off),
             };
             let focus_off = math::Vec3::from(focus_off);
-            let visible_bounds_fine = math::Aabb::<f64> {
+            let visible_bounds_fine = visible_bounding_box.as_::<f64>();
-                min: math::Vec3::from(visible_bounding_box.min.map(f64::from)),
-                max: math::Vec3::from(visible_bounding_box.max.map(f64::from)),
-            };
             let inv_proj_view =
                 math::Mat4::from_col_arrays((proj_mat * view_mat).into_col_arrays())
-                    .map(f64::from)
+                    .as_::<f64>()
                     .inverted();
             let ray_direction = math::Vec3::<f32>::from(ray_direction);
             let visible_light_volume = math::calc_focused_light_volume_points(
                 inv_proj_view,
-                ray_direction.map(f64::from),
+                ray_direction.as_::<f64>(),
                 visible_bounds_fine,
                 1e-6,
             )
-            .map(|v| v.map(|e| e as f32))
+            .map(|v| v.as_::<f32>())
             .collect::<Vec<_>>();
 
             let cam_pos = math::Vec4::from(view_mat.inverted() * Vec4::unit_w()).xyz();
@@ -2786,7 +2782,7 @@ impl<V: RectRasterableVol> Terrain<V> {
             let ray_mat = ray_mat * math::Mat4::translation_3d(-focus_off);
 
             let can_shadow_sun = |pos: Vec2<i32>, chunk: &TerrainChunkData| {
-                let chunk_pos = pos.map(|e| e as f32 * chunk_sz);
+                let chunk_pos = pos.as_::<f32>() * chunk_sz;
 
                 // Ensure the chunk is within the PSR set.
                 let chunk_box = math::Aabb {
@@ -2798,11 +2794,11 @@ impl<V: RectRasterableVol> Terrain<V> {
                     ),
                 };
 
-                let chunk_from_light = math::Aabr::from(math::fit_psr(
+                let chunk_from_light = math::fit_psr(
                     ray_mat,
                     math::aabb_to_points(chunk_box).iter().copied(),
                     |p| p,
-                ));
+                );
                 collides_with_aabr(chunk_from_light, visible_bounds)
             };
 

world/Cargo.toml

@@ -11,7 +11,7 @@ bitvec = "0.17.4"
 fxhash = "0.2.1"
 image = { version = "0.23.8", default-features = false, features = ["png"] }
 itertools = "0.9"
-vek = { version = "0.11.2", features = ["repr_simd", "serde"] }
+vek = { version = "0.12.0", features = ["platform_intrinsics", "serde"] }
 noise = { version = "0.6.0", default-features = false }
 num = "0.2"
 ordered-float = "1.0"