Add support for rust-cuda

Cargo.toml

@@ -32,6 +32,7 @@ compare = [ "matrixcompare-core" ]
 libm    = [ "simba/libm" ]
 libm-force = [ "simba/libm_force" ]
 macros = [ "nalgebra-macros" ]
+cuda   = [ "cust" ]
 
 # Conversion
 convert-mint = [ "mint" ]
@@ -86,7 +87,7 @@ pest           = { version = "2", optional = true }
 pest_derive    = { version = "2", optional = true }
 bytemuck       = { version = "1.5", optional = true }
 matrixcompare-core = { version = "0.1", optional = true }
-proptest           = { version = "1", optional = true, default-features = false, features = ["std"] }
+proptest       = { version = "1", optional = true, default-features = false, features = ["std"] }
 glam013        = { package = "glam", version = "0.13", optional = true }
 glam014        = { package = "glam", version = "0.14", optional = true }
 glam015        = { package = "glam", version = "0.15", optional = true }
@@ -94,6 +95,9 @@ glam016        = { package = "glam", version = "0.16", optional = true }
 glam017        = { package = "glam", version = "0.17", optional = true }
 glam018        = { package = "glam", version = "0.18", optional = true }
 
+[target.'cfg(not(target_os = "cuda"))'.dependencies]
+cust           = { version = "0.1", optional = true }
+
 
 [dev-dependencies]
 serde_json = "1.0"

src/base/array_storage.rs

@@ -32,6 +32,10 @@ use std::mem;
 /// A array-based statically sized matrix data storage.
 #[repr(transparent)]
 #[derive(Copy, Clone, PartialEq, Eq, Hash)]
+#[cfg_attr(
+    all(not(target_os = "cuda"), feature = "cuda"),
+    derive(cust::DeviceCopy)
+)]
 pub struct ArrayStorage<T, const R: usize, const C: usize>(pub [[T; R]; C]);
 
 impl<T, const R: usize, const C: usize> ArrayStorage<T, R, C> {

src/base/dimension.rs

@@ -13,6 +13,10 @@ use serde::{Deserialize, Deserializer, Serialize, Serializer};
 
 /// Dim of dynamically-sized algebraic entities.
 #[derive(Clone, Copy, Eq, PartialEq, Debug)]
+#[cfg_attr(
+    all(not(target_os = "cuda"), feature = "cuda"),
+    derive(cust::DeviceCopy)
+)]
 pub struct Dynamic {
     value: usize,
 }
@@ -197,6 +201,10 @@ dim_ops!(
 );
 
 #[derive(Debug, Copy, Clone, PartialEq, Eq, Hash)]
+#[cfg_attr(
+    all(not(target_os = "cuda"), feature = "cuda"),
+    derive(cust::DeviceCopy)
+)]
 pub struct Const<const R: usize>;
 
 /// Trait implemented exclusively by type-level integers.

src/base/matrix.rs

@@ -154,6 +154,10 @@ pub type MatrixCross<T, R1, C1, R2, C2> =
 /// some concrete types for `T` and a compatible data storage type `S`).
 #[repr(C)]
 #[derive(Clone, Copy)]
+#[cfg_attr(
+    all(not(target_os = "cuda"), feature = "cuda"),
+    derive(cust::DeviceCopy)
+)]
 pub struct Matrix<T, R, C, S> {
     /// The data storage that contains all the matrix components. Disappointed?
     ///

src/base/unit.rs

@@ -26,6 +26,10 @@ use crate::{Dim, Matrix, OMatrix, RealField, Scalar, SimdComplexField, SimdRealF
 /// in their documentation, read their dedicated pages directly.
 #[repr(transparent)]
 #[derive(Clone, Hash, Copy)]
+#[cfg_attr(
+    all(not(target_os = "cuda"), feature = "cuda"),
+    derive(cust::DeviceCopy)
+)]
 pub struct Unit<T> {
     pub(crate) value: T,
 }

src/geometry/dual_quaternion.rs

@@ -39,6 +39,10 @@ use simba::scalar::{ClosedNeg, RealField};
 ///  See <https://github.com/dimforge/nalgebra/issues/487>
 #[repr(C)]
 #[derive(Debug, Copy, Clone)]
+#[cfg_attr(
+    all(not(target_os = "cuda"), feature = "cuda"),
+    derive(cust::DeviceCopy)
+)]
 pub struct DualQuaternion<T> {
     /// The real component of the quaternion
     pub real: Quaternion<T>,

src/geometry/isometry.rs

@@ -54,7 +54,11 @@ use crate::geometry::{AbstractRotation, Point, Translation};
 /// * [Conversion to a matrix <span style="float:right;">`to_matrix`…</span>](#conversion-to-a-matrix)
 ///
 #[repr(C)]
-#[derive(Debug)]
+#[derive(Debug, Copy, Clone)]
+#[cfg_attr(
+    all(not(target_os = "cuda"), feature = "cuda"),
+    derive(cust::DeviceCopy)
+)]
 #[cfg_attr(feature = "serde-serialize-no-std", derive(Serialize, Deserialize))]
 #[cfg_attr(
     feature = "serde-serialize-no-std",
@@ -170,20 +174,6 @@ where
     }
 }
 
-impl<T: Scalar + Copy, R: Copy, const D: usize> Copy for Isometry<T, R, D> where
-    Owned<T, Const<D>>: Copy
-{
-}
-
-impl<T: Scalar, R: Clone, const D: usize> Clone for Isometry<T, R, D> {
-    #[inline]
-    fn clone(&self) -> Self {
-        Self {
-            rotation: self.rotation.clone(),
-            translation: self.translation.clone(),
-        }
-    }
-}
 /// # From the translation and rotation parts
 impl<T: Scalar, R: AbstractRotation<T, D>, const D: usize> Isometry<T, R, D> {
     /// Creates a new isometry from its rotational and translational parts.

src/geometry/orthographic.rs

@@ -19,19 +19,15 @@ use crate::geometry::{Point3, Projective3};
 
 /// A 3D orthographic projection stored as a homogeneous 4x4 matrix.
 #[repr(C)]
+#[cfg_attr(
+    all(not(target_os = "cuda"), feature = "cuda"),
+    derive(cust::DeviceCopy)
+)]
+#[derive(Copy, Clone)]
 pub struct Orthographic3<T> {
     matrix: Matrix4<T>,
 }
 
-impl<T: RealField + Copy> Copy for Orthographic3<T> {}
-
-impl<T: RealField> Clone for Orthographic3<T> {
-    #[inline]
-    fn clone(&self) -> Self {
-        Self::from_matrix_unchecked(self.matrix.clone())
-    }
-}
-
 impl<T: RealField> fmt::Debug for Orthographic3<T> {
     fn fmt(&self, f: &mut fmt::Formatter<'_>) -> Result<(), fmt::Error> {
         self.matrix.fmt(f)

src/geometry/perspective.rs

@@ -20,19 +20,15 @@ use crate::geometry::{Point3, Projective3};
 
 /// A 3D perspective projection stored as a homogeneous 4x4 matrix.
 #[repr(C)]
+#[cfg_attr(
+    all(not(target_os = "cuda"), feature = "cuda"),
+    derive(cust::DeviceCopy)
+)]
+#[derive(Copy, Clone)]
 pub struct Perspective3<T> {
     matrix: Matrix4<T>,
 }
 
-impl<T: RealField + Copy> Copy for Perspective3<T> {}
-
-impl<T: RealField> Clone for Perspective3<T> {
-    #[inline]
-    fn clone(&self) -> Self {
-        Self::from_matrix_unchecked(self.matrix.clone())
-    }
-}
-
 impl<T: RealField> fmt::Debug for Perspective3<T> {
     fn fmt(&self, f: &mut fmt::Formatter<'_>) -> Result<(), fmt::Error> {
         self.matrix.fmt(f)

src/geometry/point.rs

@@ -74,6 +74,15 @@ where
 {
 }
 
+#[cfg(all(not(target_os = "cuda"), feature = "cuda"))]
+unsafe impl<T: Scalar + cust::memory::DeviceCopy, D: DimName> cust::memory::DeviceCopy
+    for OPoint<T, D>
+where
+    DefaultAllocator: Allocator<T, D>,
+    OVector<T, D>: cust::memory::DeviceCopy,
+{
+}
+
 #[cfg(feature = "bytemuck")]
 unsafe impl<T: Scalar, D: DimName> bytemuck::Zeroable for OPoint<T, D>
 where

src/geometry/quaternion.rs

@@ -28,6 +28,10 @@ use crate::geometry::{Point3, Rotation};
 /// that may be used as a rotation.
 #[repr(C)]
 #[derive(Copy, Clone)]
+#[cfg_attr(
+    all(not(target_os = "cuda"), feature = "cuda"),
+    derive(cust::DeviceCopy)
+)]
 pub struct Quaternion<T> {
     /// This quaternion as a 4D vector of coordinates in the `[ x, y, z, w ]` storage order.
     pub coords: Vector4<T>,

src/geometry/rotation.rs

@@ -54,6 +54,11 @@ use crate::geometry::Point;
 /// * [Conversion to a matrix <span style="float:right;">`matrix`, `to_homogeneous`…</span>](#conversion-to-a-matrix)
 ///
 #[repr(C)]
+#[cfg_attr(
+    all(not(target_os = "cuda"), feature = "cuda"),
+    derive(cust::DeviceCopy)
+)]
+#[derive(Copy, Clone)]
 pub struct Rotation<T, const D: usize> {
     matrix: SMatrix<T, D, D>,
 }
@@ -73,21 +78,6 @@ where
     }
 }
 
-impl<T: Scalar + Copy, const D: usize> Copy for Rotation<T, D> where
-    <DefaultAllocator as Allocator<T, Const<D>, Const<D>>>::Buffer: Copy
-{
-}
-
-impl<T: Scalar, const D: usize> Clone for Rotation<T, D>
-where
-    <DefaultAllocator as Allocator<T, Const<D>, Const<D>>>::Buffer: Clone,
-{
-    #[inline]
-    fn clone(&self) -> Self {
-        Self::from_matrix_unchecked(self.matrix.clone())
-    }
-}
-
 #[cfg(feature = "bytemuck")]
 unsafe impl<T, const D: usize> bytemuck::Zeroable for Rotation<T, D>
 where

src/geometry/scale.rs

@@ -22,6 +22,11 @@ use crate::geometry::Point;
 
 /// A scale which supports non-uniform scaling.
 #[repr(C)]
+#[cfg_attr(
+    all(not(target_os = "cuda"), feature = "cuda"),
+    derive(cust::DeviceCopy)
+)]
+#[derive(Copy, Clone)]
 pub struct Scale<T, const D: usize> {
     /// The scale coordinates, i.e., how much is multiplied to a point's coordinates when it is
     /// scaled.
@@ -43,18 +48,6 @@ where
     }
 }
 
-impl<T: Scalar + Copy, const D: usize> Copy for Scale<T, D> {}
-
-impl<T: Scalar, const D: usize> Clone for Scale<T, D>
-where
-    Owned<T, Const<D>>: Clone,
-{
-    #[inline]
-    fn clone(&self) -> Self {
-        Scale::from(self.vector.clone())
-    }
-}
-
 #[cfg(feature = "bytemuck")]
 unsafe impl<T, const D: usize> bytemuck::Zeroable for Scale<T, D>
 where

src/geometry/similarity.rs

@@ -23,7 +23,11 @@ use crate::geometry::{AbstractRotation, Isometry, Point, Translation};
 
 /// A similarity, i.e., an uniform scaling, followed by a rotation, followed by a translation.
 #[repr(C)]
-#[derive(Debug)]
+#[derive(Debug, Copy, Clone)]
+#[cfg_attr(
+    all(not(target_os = "cuda"), feature = "cuda"),
+    derive(cust::DeviceCopy)
+)]
 #[cfg_attr(feature = "serde-serialize-no-std", derive(Serialize, Deserialize))]
 #[cfg_attr(
     feature = "serde-serialize-no-std",
@@ -73,22 +77,6 @@ where
     }
 }
 
-impl<T: Scalar + Copy + Zero, R: AbstractRotation<T, D> + Copy, const D: usize> Copy
-    for Similarity<T, R, D>
-where
-    Owned<T, Const<D>>: Copy,
-{
-}
-
-impl<T: Scalar + Zero, R: AbstractRotation<T, D> + Clone, const D: usize> Clone
-    for Similarity<T, R, D>
-{
-    #[inline]
-    fn clone(&self) -> Self {
-        Similarity::from_isometry(self.isometry.clone(), self.scaling.clone())
-    }
-}
-
 impl<T: Scalar + Zero, R, const D: usize> Similarity<T, R, D>
 where
     R: AbstractRotation<T, D>,

src/geometry/transform.rs

@@ -60,14 +60,26 @@ where
 
 /// Tag representing the most general (not necessarily inversible) `Transform` type.
 #[derive(Debug, Copy, Clone, Hash, PartialEq, Eq)]
+#[cfg_attr(
+    all(not(target_os = "cuda"), feature = "cuda"),
+    derive(cust::DeviceCopy)
+)]
 pub enum TGeneral {}
 
 /// Tag representing the most general inversible `Transform` type.
 #[derive(Debug, Copy, Clone, Hash, PartialEq, Eq)]
+#[cfg_attr(
+    all(not(target_os = "cuda"), feature = "cuda"),
+    derive(cust::DeviceCopy)
+)]
 pub enum TProjective {}
 
 /// Tag representing an affine `Transform`. Its bottom-row is equal to `(0, 0 ... 0, 1)`.
 #[derive(Debug, Copy, Clone, Hash, PartialEq, Eq)]
+#[cfg_attr(
+    all(not(target_os = "cuda"), feature = "cuda"),
+    derive(cust::DeviceCopy)
+)]
 pub enum TAffine {}
 
 impl TCategory for TGeneral {
@@ -195,6 +207,16 @@ where
 {
 }
 
+#[cfg(all(not(target_os = "cuda"), feature = "cuda"))]
+unsafe impl<T: RealField + cust::memory::DeviceCopy, C: TCategory, const D: usize>
+    cust::memory::DeviceCopy for Transform<T, C, D>
+where
+    Const<D>: DimNameAdd<U1>,
+    DefaultAllocator: Allocator<T, DimNameSum<Const<D>, U1>, DimNameSum<Const<D>, U1>>,
+    Owned<T, DimNameSum<Const<D>, U1>, DimNameSum<Const<D>, U1>>: cust::memory::DeviceCopy,
+{
+}
+
 impl<T: RealField, C: TCategory, const D: usize> Clone for Transform<T, C, D>
 where
     Const<D>: DimNameAdd<U1>,

src/geometry/translation.rs

@@ -22,6 +22,11 @@ use crate::geometry::Point;
 
 /// A translation.
 #[repr(C)]
+#[cfg_attr(
+    all(not(target_os = "cuda"), feature = "cuda"),
+    derive(cust::DeviceCopy)
+)]
+#[derive(Copy, Clone)]
 pub struct Translation<T, const D: usize> {
     /// The translation coordinates, i.e., how much is added to a point's coordinates when it is
     /// translated.
@@ -43,18 +48,6 @@ where
     }
 }
 
-impl<T: Scalar + Copy, const D: usize> Copy for Translation<T, D> {}
-
-impl<T: Scalar, const D: usize> Clone for Translation<T, D>
-where
-    Owned<T, Const<D>>: Clone,
-{
-    #[inline]
-    fn clone(&self) -> Self {
-        Translation::from(self.vector.clone())
-    }
-}
-
 #[cfg(feature = "bytemuck")]
 unsafe impl<T, const D: usize> bytemuck::Zeroable for Translation<T, D>
 where