From 6d89e2aca6b53ba2271756dfe7547855867183a0 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?S=C3=A9bastien=20Crozet?= Date: Fri, 11 Mar 2022 17:17:34 +0100 Subject: [PATCH 1/4] Update to cust 0.3 --- Cargo.toml | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/Cargo.toml b/Cargo.toml index 15a24d41..0adc5729 100644 --- a/Cargo.toml +++ b/Cargo.toml @@ -98,7 +98,7 @@ glam019 = { package = "glam", version = "0.19", optional = true } glam020 = { package = "glam", version = "0.20", optional = true } [target.'cfg(not(target_os = "cuda"))'.dependencies] -cust = { version = "0.2", optional = true } +cust = { version = "0.3", optional = true } [dev-dependencies] From 1c4c3de92e0900bbd32d6f6e3d3a64a45dcd94e4 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?S=C3=A9bastien=20Crozet?= Date: Fri, 11 Mar 2022 17:43:01 +0100 Subject: [PATCH 2/4] CI: pin the version of Cuda --- .github/workflows/nalgebra-ci-build.yml | 2 ++ 1 file changed, 2 insertions(+) diff --git a/.github/workflows/nalgebra-ci-build.yml b/.github/workflows/nalgebra-ci-build.yml index bc2f9ca6..707c4143 100644 --- a/.github/workflows/nalgebra-ci-build.yml +++ b/.github/workflows/nalgebra-ci-build.yml @@ -124,6 +124,8 @@ jobs: runs-on: ubuntu-latest steps: - uses: Jimver/cuda-toolkit@v0.2.4 + with: + cuda: '11.2.2' - name: Install nightly-2021-12-04 uses: actions-rs/toolchain@v1 with: From 30c0450075290653ef14a9b6b6a4f387f9ef57f7 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?S=C3=A9bastien=20Crozet?= Date: Fri, 11 Mar 2022 18:06:40 +0100 Subject: [PATCH 3/4] CI: set the CUDA_ARCH env var when targetting nvptx --- .github/workflows/nalgebra-ci-build.yml | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/.github/workflows/nalgebra-ci-build.yml b/.github/workflows/nalgebra-ci-build.yml index 707c4143..c00b6cbc 100644 --- a/.github/workflows/nalgebra-ci-build.yml +++ b/.github/workflows/nalgebra-ci-build.yml @@ -134,4 +134,6 @@ jobs: - uses: actions/checkout@v2 - run: rustup target add nvptx64-nvidia-cuda - run: cargo build --no-default-features --features cuda - - run: cargo build --no-default-features --features cuda --target=nvptx64-nvidia-cuda \ No newline at end of file + - run: cargo build --no-default-features --features cuda --target=nvptx64-nvidia-cuda + env: + CUDA_ARCH: "350" \ No newline at end of file From d7117e228a20aa93920ddd3c98c919710c9745aa Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?S=C3=A9bastien=20Crozet?= Date: Wed, 16 Mar 2022 18:07:29 +0100 Subject: [PATCH 4/4] Use cust_core instead of cust --- Cargo.toml | 6 ++---- src/base/array_storage.rs | 5 +---- src/base/dimension.rs | 10 ++-------- src/base/matrix.rs | 5 +---- src/base/unit.rs | 10 +++------- src/geometry/dual_quaternion.rs | 5 +---- src/geometry/isometry.rs | 5 +---- src/geometry/orthographic.rs | 5 +---- src/geometry/perspective.rs | 5 +---- src/geometry/point.rs | 7 +++---- src/geometry/quaternion.rs | 9 +++------ src/geometry/rotation.rs | 5 +---- src/geometry/scale.rs | 5 +---- src/geometry/similarity.rs | 5 +---- src/geometry/transform.rs | 23 +++++++---------------- src/geometry/translation.rs | 5 +---- src/geometry/unit_complex.rs | 4 ++-- 17 files changed, 32 insertions(+), 87 deletions(-) diff --git a/Cargo.toml b/Cargo.toml index 0adc5729..8a3fea5c 100644 --- a/Cargo.toml +++ b/Cargo.toml @@ -32,7 +32,7 @@ compare = [ "matrixcompare-core" ] libm = [ "simba/libm" ] libm-force = [ "simba/libm_force" ] macros = [ "nalgebra-macros" ] -cuda = [ "cust", "simba/cuda" ] +cuda = [ "cust_core", "simba/cuda" ] # Conversion convert-mint = [ "mint" ] @@ -96,9 +96,7 @@ glam017 = { package = "glam", version = "0.17", optional = true } glam018 = { package = "glam", version = "0.18", optional = true } glam019 = { package = "glam", version = "0.19", optional = true } glam020 = { package = "glam", version = "0.20", optional = true } - -[target.'cfg(not(target_os = "cuda"))'.dependencies] -cust = { version = "0.3", optional = true } +cust_core = { version = "0.1", optional = true } [dev-dependencies] diff --git a/src/base/array_storage.rs b/src/base/array_storage.rs index 6851c381..b6bd236a 100644 --- a/src/base/array_storage.rs +++ b/src/base/array_storage.rs @@ -27,10 +27,7 @@ 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) -)] +#[cfg_attr(feature = "cuda", derive(cust_core::DeviceCopy))] pub struct ArrayStorage(pub [[T; R]; C]); impl ArrayStorage { diff --git a/src/base/dimension.rs b/src/base/dimension.rs index 86006f3d..e43cb734 100644 --- a/src/base/dimension.rs +++ b/src/base/dimension.rs @@ -13,10 +13,7 @@ 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) -)] +#[cfg_attr(feature = "cuda", derive(cust_core::DeviceCopy))] pub struct Dynamic { value: usize, } @@ -201,10 +198,7 @@ dim_ops!( ); #[derive(Debug, Copy, Clone, PartialEq, Eq, Hash)] -#[cfg_attr( - all(not(target_os = "cuda"), feature = "cuda"), - derive(cust::DeviceCopy) -)] +#[cfg_attr(feature = "cuda", derive(cust_core::DeviceCopy))] pub struct Const; /// Trait implemented exclusively by type-level integers. diff --git a/src/base/matrix.rs b/src/base/matrix.rs index bdf3a8c7..cc69c9a1 100644 --- a/src/base/matrix.rs +++ b/src/base/matrix.rs @@ -150,10 +150,7 @@ pub type MatrixCross = /// 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) -)] +#[cfg_attr(feature = "cuda", derive(cust_core::DeviceCopy))] pub struct Matrix { /// The data storage that contains all the matrix components. Disappointed? /// diff --git a/src/base/unit.rs b/src/base/unit.rs index 9336a5e5..bb8b56a1 100644 --- a/src/base/unit.rs +++ b/src/base/unit.rs @@ -21,10 +21,7 @@ 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) -// )] +// #[cfg_attr(feature = "cuda", derive(cust_core::DeviceCopy))] pub struct Unit { pub(crate) value: T, } @@ -102,9 +99,8 @@ mod rkyv_impl { } } -#[cfg(all(not(target_os = "cuda"), feature = "cuda"))] -unsafe impl cust::memory::DeviceCopy - for Unit> +#[cfg(feature = "cuda")] +unsafe impl cust_core::DeviceCopy for Unit> where T: Scalar, R: Dim, diff --git a/src/geometry/dual_quaternion.rs b/src/geometry/dual_quaternion.rs index 4280668a..509b359a 100644 --- a/src/geometry/dual_quaternion.rs +++ b/src/geometry/dual_quaternion.rs @@ -39,10 +39,7 @@ use simba::scalar::{ClosedNeg, RealField}; /// See #[repr(C)] #[derive(Debug, Copy, Clone)] -#[cfg_attr( - all(not(target_os = "cuda"), feature = "cuda"), - derive(cust::DeviceCopy) -)] +#[cfg_attr(feature = "cuda", derive(cust_core::DeviceCopy))] pub struct DualQuaternion { /// The real component of the quaternion pub real: Quaternion, diff --git a/src/geometry/isometry.rs b/src/geometry/isometry.rs index 8cdd1bfc..0179f1ff 100755 --- a/src/geometry/isometry.rs +++ b/src/geometry/isometry.rs @@ -50,10 +50,7 @@ use crate::geometry::{AbstractRotation, Point, Translation}; /// #[repr(C)] #[derive(Debug, Copy, Clone)] -#[cfg_attr( - all(not(target_os = "cuda"), feature = "cuda"), - derive(cust::DeviceCopy) -)] +#[cfg_attr(feature = "cuda", derive(cust_core::DeviceCopy))] #[cfg_attr(feature = "serde-serialize-no-std", derive(Serialize, Deserialize))] #[cfg_attr( feature = "serde-serialize-no-std", diff --git a/src/geometry/orthographic.rs b/src/geometry/orthographic.rs index 18a7852d..085ba61b 100644 --- a/src/geometry/orthographic.rs +++ b/src/geometry/orthographic.rs @@ -19,10 +19,7 @@ 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) -)] +#[cfg_attr(feature = "cuda", derive(cust_core::DeviceCopy))] #[derive(Copy, Clone)] pub struct Orthographic3 { matrix: Matrix4, diff --git a/src/geometry/perspective.rs b/src/geometry/perspective.rs index 59b7f9f2..8ebab3e4 100644 --- a/src/geometry/perspective.rs +++ b/src/geometry/perspective.rs @@ -20,10 +20,7 @@ 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) -)] +#[cfg_attr(feature = "cuda", derive(cust_core::DeviceCopy))] #[derive(Copy, Clone)] pub struct Perspective3 { matrix: Matrix4, diff --git a/src/geometry/point.rs b/src/geometry/point.rs index b62998c3..a8d7684b 100644 --- a/src/geometry/point.rs +++ b/src/geometry/point.rs @@ -69,12 +69,11 @@ where { } -#[cfg(all(not(target_os = "cuda"), feature = "cuda"))] -unsafe impl cust::memory::DeviceCopy - for OPoint +#[cfg(feature = "cuda")] +unsafe impl cust_core::DeviceCopy for OPoint where DefaultAllocator: Allocator, - OVector: cust::memory::DeviceCopy, + OVector: cust_core::DeviceCopy, { } diff --git a/src/geometry/quaternion.rs b/src/geometry/quaternion.rs index 0aa7f3d3..6d29f34f 100755 --- a/src/geometry/quaternion.rs +++ b/src/geometry/quaternion.rs @@ -23,10 +23,7 @@ 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) -)] +#[cfg_attr(feature = "cuda", derive(cust_core::DeviceCopy))] pub struct Quaternion { /// This quaternion as a 4D vector of coordinates in the `[ x, y, z, w ]` storage order. pub coords: Vector4, @@ -1045,8 +1042,8 @@ impl fmt::Display for Quaternion { /// A unit quaternions. May be used to represent a rotation. pub type UnitQuaternion = Unit>; -#[cfg(all(not(target_os = "cuda"), feature = "cuda"))] -unsafe impl cust::memory::DeviceCopy for UnitQuaternion {} +#[cfg(feature = "cuda")] +unsafe impl cust_core::DeviceCopy for UnitQuaternion {} impl PartialEq for UnitQuaternion { #[inline] diff --git a/src/geometry/rotation.rs b/src/geometry/rotation.rs index 69c4a355..4dbcfb43 100755 --- a/src/geometry/rotation.rs +++ b/src/geometry/rotation.rs @@ -49,10 +49,7 @@ use crate::geometry::Point; /// * [Conversion to a matrix `matrix`, `to_homogeneous`…](#conversion-to-a-matrix) /// #[repr(C)] -#[cfg_attr( - all(not(target_os = "cuda"), feature = "cuda"), - derive(cust::DeviceCopy) -)] +#[cfg_attr(feature = "cuda", derive(cust_core::DeviceCopy))] #[derive(Copy, Clone)] pub struct Rotation { matrix: SMatrix, diff --git a/src/geometry/scale.rs b/src/geometry/scale.rs index 064e0075..abaeeccc 100755 --- a/src/geometry/scale.rs +++ b/src/geometry/scale.rs @@ -17,10 +17,7 @@ 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) -)] +#[cfg_attr(feature = "cuda", derive(cust_core::DeviceCopy))] #[derive(Copy, Clone)] pub struct Scale { /// The scale coordinates, i.e., how much is multiplied to a point's coordinates when it is diff --git a/src/geometry/similarity.rs b/src/geometry/similarity.rs index 46c86f5d..9658685e 100755 --- a/src/geometry/similarity.rs +++ b/src/geometry/similarity.rs @@ -18,10 +18,7 @@ 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, Copy, Clone)] -#[cfg_attr( - all(not(target_os = "cuda"), feature = "cuda"), - derive(cust::DeviceCopy) -)] +#[cfg_attr(feature = "cuda", derive(cust_core::DeviceCopy))] #[cfg_attr(feature = "serde-serialize-no-std", derive(Serialize, Deserialize))] #[cfg_attr( feature = "serde-serialize-no-std", diff --git a/src/geometry/transform.rs b/src/geometry/transform.rs index b0b5cced..2a7ca112 100755 --- a/src/geometry/transform.rs +++ b/src/geometry/transform.rs @@ -60,26 +60,17 @@ 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) -)] +#[cfg_attr(feature = "cuda", derive(cust_core::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) -)] +#[cfg_attr(feature = "cuda", derive(cust_core::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) -)] +#[cfg_attr(feature = "cuda", derive(cust_core::DeviceCopy))] pub enum TAffine {} impl TCategory for TGeneral { @@ -207,13 +198,13 @@ where { } -#[cfg(all(not(target_os = "cuda"), feature = "cuda"))] -unsafe impl - cust::memory::DeviceCopy for Transform +#[cfg(feature = "cuda")] +unsafe impl + cust_core::DeviceCopy for Transform where Const: DimNameAdd, DefaultAllocator: Allocator, U1>, DimNameSum, U1>>, - Owned, U1>, DimNameSum, U1>>: cust::memory::DeviceCopy, + Owned, U1>, DimNameSum, U1>>: cust_core::DeviceCopy, { } diff --git a/src/geometry/translation.rs b/src/geometry/translation.rs index b07cce20..5db46e82 100755 --- a/src/geometry/translation.rs +++ b/src/geometry/translation.rs @@ -17,10 +17,7 @@ use crate::geometry::Point; /// A translation. #[repr(C)] -#[cfg_attr( - all(not(target_os = "cuda"), feature = "cuda"), - derive(cust::DeviceCopy) -)] +#[cfg_attr(feature = "cuda", derive(cust_core::DeviceCopy))] #[derive(Copy, Clone)] pub struct Translation { /// The translation coordinates, i.e., how much is added to a point's coordinates when it is diff --git a/src/geometry/unit_complex.rs b/src/geometry/unit_complex.rs index 48405dd4..caf25493 100755 --- a/src/geometry/unit_complex.rs +++ b/src/geometry/unit_complex.rs @@ -31,8 +31,8 @@ use std::cmp::{Eq, PartialEq}; /// * [Conversion to a matrix `to_rotation_matrix`, `to_homogeneous`…](#conversion-to-a-matrix) pub type UnitComplex = Unit>; -#[cfg(all(not(target_os = "cuda"), feature = "cuda"))] -unsafe impl cust::memory::DeviceCopy for UnitComplex {} +#[cfg(feature = "cuda")] +unsafe impl cust_core::DeviceCopy for UnitComplex {} impl PartialEq for UnitComplex { #[inline]