feat: remove cuda support

CHANGELOG.md

@@ -1,5 +1,31 @@
 # Change Log
 
+## Unreleased
+
+### Modified
+
+- Remove CUDA support to break free from the toolchain restriction required by cust.
+- Rework internal edges resolution using normal cones. This implies the modification of the
+  `SimdCompositeShape::map_part_at`, `TypedSimdCompositeShape::map_typed_part`, and
+  `TypedSimdCompositeShape::map_untyped_part` trait functions so that the closure argument takes
+  an extra argument for the (optional) normal constraints. This argument can be safely ignored
+  by user code unless applying the normal collection is relevant to your use-case.
+- Contact manifolds will now retain all contacts (including the ones further than the specified `prediction`
+  distance) whenever any contact is actually closer than this `prediction` distance.
+- Typo fix: renamed `TopologyError::BadAdjascentTrianglesOrientation` to `BadAdjacentTrianglesOrientation`.
+
+### Fixed
+
+- Fix contacts between convex shapes being occasionally ignored due to some rounding errors.
+- Remove crash when entering unreachable code in non-linear TOI calculation.
+- Fix accuracy issue in triangle-mesh center-of-mass calculation when the mesh isn’t manifold.
+
+### Added
+
+- Add `SdpMatrix2::inverse_and_get_determinant_unchecked`. This is useful for computing the
+  inverse in a AoSoA SIMD setting.
+- Add `Aabb::intersects_moving_aabb` to perform a swept test between two moving aabbs.
+
 ## v0.13.8
 ### Added
 

crates/parry2d-f64/Cargo.toml

@@ -28,7 +28,6 @@ bytemuck-serialize = ["bytemuck", "nalgebra/convert-bytemuck"]
 simd-stable = ["simba/wide", "simd-is-enabled"]
 simd-nightly = ["simba/packed_simd", "simd-is-enabled"]
 enhanced-determinism = ["simba/libm_force", "indexmap"]
-cuda = ["cust_core", "cust", "nalgebra/cuda"]
 parallel = ["rayon"]
 
 # Do not enable this feature directly. It is automatically
@@ -62,9 +61,6 @@ rayon = { version = "1", optional = true }
 bytemuck = { version = "1", features = ["derive"], optional = true }
 log = "0.4"
 
-[target.'cfg(not(target_os = "cuda"))'.dependencies]
-cust = { version = "0.3", optional = true }
-
 [dev-dependencies]
 simba = { version = "0.8", default-features = false, features = ["partial_fixed_point_support"] }
 oorandom = "11"

crates/parry2d/Cargo.toml

@@ -28,7 +28,6 @@ bytemuck-serialize = ["bytemuck", "nalgebra/convert-bytemuck"]
 simd-stable = ["simba/wide", "simd-is-enabled"]
 simd-nightly = ["simba/packed_simd", "simd-is-enabled"]
 enhanced-determinism = ["simba/libm_force", "indexmap"]
-cuda = ["cust_core", "cust", "nalgebra/cuda"]
 parallel = ["rayon"]
 
 # Do not enable this feature directly. It is automatically
@@ -62,9 +61,6 @@ rayon = { version = "1", optional = true }
 bytemuck = { version = "1", features = ["derive"], optional = true }
 log = "0.4"
 
-[target.'cfg(not(target_os = "cuda"))'.dependencies]
-cust = { version = "0.3", optional = true }
-
 [dev-dependencies]
 simba = { version = "0.8", default-features = false, features = ["partial_fixed_point_support"] }
 oorandom = "11"

crates/parry3d-f64/Cargo.toml

@@ -28,7 +28,6 @@ bytemuck-serialize = ["bytemuck", "nalgebra/convert-bytemuck"]
 simd-stable = ["simba/wide", "simd-is-enabled"]
 simd-nightly = ["simba/packed_simd", "simd-is-enabled"]
 enhanced-determinism = ["simba/libm_force", "indexmap"]
-cuda = ["cust_core", "cust", "nalgebra/cuda"]
 parallel = ["rayon"]
 
 # Do not enable this feature directly. It is automatically
@@ -62,9 +61,6 @@ rayon = { version = "1", optional = true }
 bytemuck = { version = "1", features = ["derive"], optional = true }
 log = "0.4"
 
-[target.'cfg(not(target_os = "cuda"))'.dependencies]
-cust = { version = "0.3", optional = true }
-
 [dev-dependencies]
 oorandom = "11"
 ptree = "0.4.0"

crates/parry3d/Cargo.toml

@@ -29,7 +29,6 @@ bytemuck-serialize = ["bytemuck", "nalgebra/convert-bytemuck"]
 simd-stable = ["simba/wide", "simd-is-enabled"]
 simd-nightly = ["simba/packed_simd", "simd-is-enabled"]
 enhanced-determinism = ["simba/libm_force", "indexmap"]
-cuda = ["cust_core", "cust", "nalgebra/cuda"]
 parallel = ["rayon"]
 
 # Do not enable this feature directly. It is automatically
@@ -63,9 +62,6 @@ rayon = { version = "1", optional = true }
 bytemuck = { version = "1", features = ["derive"], optional = true }
 log = "0.4"
 
-[target.'cfg(not(target_os = "cuda"))'.dependencies]
-cust = { version = "0.3", optional = true }
-
 [dev-dependencies]
 oorandom = "11"
 ptree = "0.4.0"

src/bounding_volume/aabb.rs

@@ -23,7 +23,6 @@ use rkyv::{bytecheck, CheckBytes};
     derive(rkyv::Archive, rkyv::Deserialize, rkyv::Serialize, CheckBytes),
     archive(as = "Self")
 )]
-#[cfg_attr(feature = "cuda", derive(cust_core::DeviceCopy))]
 #[derive(Debug, PartialEq, Copy, Clone)]
 #[repr(C)]
 pub struct Aabb {

src/bounding_volume/aabb_heightfield.rs

@@ -1,8 +1,8 @@
 use crate::bounding_volume::Aabb;
 use crate::math::{Isometry, Real};
-use crate::shape::{GenericHeightField, HeightFieldStorage};
+use crate::shape::HeightField;
 
-impl<Storage: HeightFieldStorage> GenericHeightField<Storage> {
+impl HeightField {
     /// Computes the world-space [`Aabb`] of this heightfield, transformed by `pos`.
     #[inline]
     pub fn aabb(&self, pos: &Isometry<Real>) -> Aabb {

src/bounding_volume/bounding_sphere_heightfield.rs

@@ -1,8 +1,8 @@
 use crate::bounding_volume::BoundingSphere;
 use crate::math::{Isometry, Real};
-use crate::shape::{GenericHeightField, HeightFieldStorage};
+use crate::shape::HeightField;
 
-impl<Storage: HeightFieldStorage> GenericHeightField<Storage> {
+impl HeightField {
     /// Computes the world-space bounding sphere of this height-field, transformed by `pos`.
     #[inline]
     pub fn bounding_sphere(&self, pos: &Isometry<Real>) -> BoundingSphere {

src/bounding_volume/simd_aabb.rs

@@ -12,7 +12,6 @@ use simba::simd::{SimdPartialOrd, SimdValue};
     derive(rkyv::Archive, rkyv::Deserialize, rkyv::Serialize),
     archive(check_bytes)
 )]
-#[cfg_attr(feature = "cuda", derive(cust_core::DeviceCopy))]
 pub struct SimdAabb {
     /// The min coordinates of the Aabbs.
     pub mins: Point<SimdReal>,

src/partitioning/mod.rs

@@ -2,10 +2,8 @@
 
 #[cfg(feature = "std")]
 pub use self::qbvh::{
-    CenterDataSplitter, QbvhDataGenerator, QbvhNonOverlappingDataSplitter, QbvhUpdateWorkspace,
-};
-pub use self::qbvh::{
-    GenericQbvh, IndexedData, NodeIndex, Qbvh, QbvhNode, QbvhProxy, QbvhStorage, SimdNodeIndex,
+    CenterDataSplitter, IndexedData, NodeIndex, Qbvh, QbvhDataGenerator, QbvhNode,
+    QbvhNonOverlappingDataSplitter, QbvhProxy, QbvhUpdateWorkspace, SimdNodeIndex,
 };
 #[cfg(feature = "parallel")]
 pub use self::visitor::{ParallelSimdSimultaneousVisitor, ParallelSimdVisitor};
@@ -18,5 +16,6 @@ pub use self::visitor::{
 #[deprecated(note = "Renamed to Qbvh")]
 pub type SimdQbvh<T> = Qbvh<T>;
 
+#[cfg(feature = "std")]
 mod qbvh;
 mod visitor;

src/partitioning/qbvh/mod.rs

@@ -1,23 +1,15 @@
-#[cfg(feature = "std")]
 pub use self::{
     build::{CenterDataSplitter, QbvhDataGenerator, QbvhNonOverlappingDataSplitter},
     update::QbvhUpdateWorkspace,
 };
 
 pub use self::qbvh::{
-    GenericQbvh, IndexedData, NodeIndex, Qbvh, QbvhNode, QbvhNodeFlags, QbvhProxy, SimdNodeIndex,
+    IndexedData, NodeIndex, Qbvh, QbvhNode, QbvhNodeFlags, QbvhProxy, SimdNodeIndex,
 };
-pub use self::storage::QbvhStorage;
 
 mod qbvh;
-mod storage;
 mod utils;
 
-#[cfg(feature = "std")]
 mod build;
-#[cfg(feature = "std")]
 mod traversal;
-#[cfg(not(feature = "std"))]
-mod traversal_no_std;
-#[cfg(feature = "std")]
 mod update;

src/partitioning/qbvh/qbvh.rs

@@ -1,20 +1,11 @@
 use crate::bounding_volume::{Aabb, SimdAabb};
 use crate::math::{Real, Vector};
-use crate::partitioning::qbvh::storage::QbvhStorage;
-use crate::utils::DefaultStorage;
 use bitflags::bitflags;
 
 use na::SimdValue;
 
 #[cfg(feature = "rkyv")]
 use rkyv::{bytecheck, CheckBytes};
-#[cfg(all(feature = "std", feature = "cuda"))]
-use {crate::utils::CudaArray1, cust::error::CudaResult};
-#[cfg(feature = "cuda")]
-use {
-    crate::utils::{CudaStorage, CudaStoragePtr},
-    cust_core::DeviceCopy,
-};
 
 /// A data to which an index is associated.
 pub trait IndexedData: Copy {
@@ -64,7 +55,6 @@ pub type SimdNodeIndex = u32;
     derive(rkyv::Archive, rkyv::Deserialize, rkyv::Serialize, CheckBytes),
     archive(as = "Self")
 )]
-#[cfg_attr(feature = "cuda", derive(cust_core::DeviceCopy))]
 /// The index of one specific node of a Qbvh.
 pub struct NodeIndex {
     /// The index of the SIMD node containing the addressed node.
@@ -97,7 +87,6 @@ bitflags! {
         derive(rkyv::Archive, rkyv::Deserialize, rkyv::Serialize),
         archive(as = "Self")
     )]
-    #[cfg_attr(feature = "cuda", derive(cust_core::DeviceCopy))]
     #[derive(Default)]
     /// The status of a QBVH node.
     pub struct QbvhNodeFlags: u8 {
@@ -120,7 +109,6 @@ bitflags! {
     derive(rkyv::Archive, rkyv::Deserialize, rkyv::Serialize),
     archive(check_bytes)
 )]
-#[cfg_attr(feature = "cuda", derive(cust_core::DeviceCopy))]
 pub struct QbvhNode {
     /// The Aabbs of the qbvh nodes represented by this node.
     pub simd_aabb: SimdAabb,
@@ -194,7 +182,6 @@ impl QbvhNode {
     derive(rkyv::Archive, rkyv::Deserialize, rkyv::Serialize),
     archive(check_bytes)
 )]
-#[cfg_attr(feature = "cuda", derive(cust_core::DeviceCopy))]
 /// Combination of a leaf data and its associated node’s index.
 pub struct QbvhProxy<LeafData> {
     /// Index of the leaf node the leaf data is associated to.
@@ -235,86 +222,22 @@ impl<LeafData> QbvhProxy<LeafData> {
     derive(rkyv::Archive, rkyv::Deserialize, rkyv::Serialize),
     archive(check_bytes)
 )]
-#[repr(C)] // Needed for Cuda.
-#[derive(Debug)]
-pub struct GenericQbvh<LeafData, Storage: QbvhStorage<LeafData>> {
+#[repr(C)]
+#[derive(Debug, Clone)]
+pub struct Qbvh<LeafData> {
     pub(super) root_aabb: Aabb,
-    pub(super) nodes: Storage::Nodes,
-    pub(super) dirty_nodes: Storage::ArrayU32,
-    pub(super) free_list: Storage::ArrayU32,
-    pub(super) proxies: Storage::ArrayProxies,
+    pub(super) nodes: Vec<QbvhNode>,
+    pub(super) dirty_nodes: Vec<u32>,
+    pub(super) free_list: Vec<u32>,
+    pub(super) proxies: Vec<QbvhProxy<LeafData>>,
 }
 
-/// A quaternary bounding-volume-hierarchy.
-///
-/// This is a bounding-volume-hierarchy where each node has either four children or none.
-pub type Qbvh<LeafData> = GenericQbvh<LeafData, DefaultStorage>;
-#[cfg(feature = "cuda")]
-/// A Qbvh stored in CUDA memory.
-pub type CudaQbvh<LeafData> = GenericQbvh<LeafData, CudaStorage>;
-#[cfg(feature = "cuda")]
-/// A Qbvh accessible from CUDA kernels.
-pub type CudaQbvhPtr<LeafData> = GenericQbvh<LeafData, CudaStoragePtr>;
-
-impl<LeafData, Storage> Clone for GenericQbvh<LeafData, Storage>
-where
-    Storage: QbvhStorage<LeafData>,
-    Storage::Nodes: Clone,
-    Storage::ArrayU32: Clone,
-    Storage::ArrayProxies: Clone,
-{
-    fn clone(&self) -> Self {
-        Self {
-            root_aabb: self.root_aabb,
-            nodes: self.nodes.clone(),
-            dirty_nodes: self.dirty_nodes.clone(),
-            free_list: self.free_list.clone(),
-            proxies: self.proxies.clone(),
-        }
-    }
-}
-
-impl<LeafData, Storage> Copy for GenericQbvh<LeafData, Storage>
-where
-    Storage: QbvhStorage<LeafData>,
-    Storage::Nodes: Copy,
-    Storage::ArrayU32: Copy,
-    Storage::ArrayProxies: Copy,
-{
-}
-
-#[cfg(feature = "cuda")]
-unsafe impl<LeafData, Storage> DeviceCopy for GenericQbvh<LeafData, Storage>
-where
-    Storage: QbvhStorage<LeafData>,
-    Storage::Nodes: DeviceCopy,
-    Storage::ArrayU32: DeviceCopy,
-    Storage::ArrayProxies: DeviceCopy,
-{
-}
-
-#[cfg(all(feature = "std", feature = "cuda"))]
-impl<LeafData: DeviceCopy> CudaQbvh<LeafData> {
-    /// Returns the qbvh usable from within a CUDA kernel.
-    pub fn as_device_ptr(&self) -> CudaQbvhPtr<LeafData> {
-        GenericQbvh {
-            root_aabb: self.root_aabb,
-            nodes: self.nodes.as_device_ptr(),
-            dirty_nodes: self.dirty_nodes.as_device_ptr(),
-            free_list: self.free_list.as_device_ptr(),
-            proxies: self.proxies.as_device_ptr(),
-        }
-    }
-}
-
-#[cfg(feature = "std")]
 impl<LeafData: IndexedData> Default for Qbvh<LeafData> {
     fn default() -> Self {
         Self::new()
     }
 }
 
-#[cfg(feature = "std")]
 impl<LeafData: IndexedData> Qbvh<LeafData> {
     /// Initialize an empty Qbvh.
     pub fn new() -> Self {
@@ -327,21 +250,6 @@ impl<LeafData: IndexedData> Qbvh<LeafData> {
         }
     }
 
-    /// Converts this RAM-based qbvh to an qbvh based on CUDA memory.
-    #[cfg(feature = "cuda")]
-    pub fn to_cuda(&self) -> CudaResult<CudaQbvh<LeafData>>
-    where
-        LeafData: DeviceCopy,
-    {
-        Ok(CudaQbvh {
-            root_aabb: self.root_aabb,
-            nodes: CudaArray1::new(&self.nodes)?,
-            dirty_nodes: CudaArray1::new(&self.dirty_nodes)?,
-            free_list: CudaArray1::new(&self.free_list)?,
-            proxies: CudaArray1::new(&self.proxies)?,
-        })
-    }
-
     /// Iterates mutably through all the leaf data in this Qbvh.
     pub fn iter_data_mut(&mut self) -> impl Iterator<Item = (NodeIndex, &mut LeafData)> {
         self.proxies.iter_mut().map(|p| (p.node, &mut p.data))
@@ -405,7 +313,7 @@ impl<LeafData: IndexedData> Qbvh<LeafData> {
     }
 }
 
-impl<LeafData: IndexedData, Storage: QbvhStorage<LeafData>> GenericQbvh<LeafData, Storage> {
+impl<LeafData: IndexedData> Qbvh<LeafData> {
     /// The Aabb of the root of this tree.
     pub fn root_aabb(&self) -> &Aabb {
         &self.root_aabb