Skip to content

Add Aqua tests; organize Project toml a bit better #14

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
mateuszbaran merged 6 commits into from
Apr 21, 2026
Merged

Add Aqua tests; organize Project toml a bit better #14

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
71ef67e
Add Aqua tests; organize Project toml a bit better
mateuszbaran Apr 21, 2026
82cca20
Update benchmarks, add missing dependency
mateuszbaran Apr 21, 2026
ffbd5b2
even more deps
mateuszbaran Apr 21, 2026
0094316
also make sphere less slow
mateuszbaran Apr 21, 2026
bf960f5
formatting
mateuszbaran Apr 21, 2026
16c94b8
delete a stray line
mateuszbaran Apr 21, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
24 changes: 6 additions & 18 deletions Project.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,16 +1,16 @@
name = "ManifoldsGPU"
uuid = "007d1224-8888-47ee-87d0-87e096ff9b5b"
version = "0.1.0-DEV"
version = "0.1.0"
authors = ["Mateusz Baran <mateuszbaran89@gmail.com>", "Shiwen An <sweynan@icloud.com>", "and contributors"]

[workspace]
projects = ["test", "docs"]

[deps]
GPUArrays = "0c68f7d7-f131-5f86-a1c3-88cf8149b2d7"
JLArrays = "27aeb0d3-9eb9-45fb-866b-73c2ecf80fcb"
LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
ManifoldDiff = "af67fdf4-a580-4b9f-bbec-742ef357defd"
Manifolds = "1cead3c2-87b3-11e9-0ccd-23c62b72b94e"
ManifoldsBase = "3362f125-f0bb-47a3-aa74-596ffd7ef2fb"
Runic = "62bfec6d-59d7-401d-8490-b29ee721c001"

[weakdeps]
CUDA = "052768ef-5323-5732-b1bb-66c8b64840ba"
Expand All @@ -20,20 +20,8 @@ ManifoldsGPUCUDAExt = "CUDA"

[compat]
CUDA = "5.9.6"
GPUArrays = "11"
JLArrays = "0.3"
LinearAlgebra = "1.10"
ManifoldDiff = "0.4.5"
Manifolds = "0.11.12"
ManifoldsBase = "2.3.1"
Runic = "1.5.1"
julia = "1.10.10"

[extras]
CUDA = "052768ef-5323-5732-b1bb-66c8b64840ba"
GPUArrays = "0c68f7d7-f131-5f86-a1c3-88cf8149b2d7"
JLArrays = "27aeb0d3-9eb9-45fb-866b-73c2ecf80fcb"
Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"

[targets]
test = ["Test", "CUDA", "GPUArrays", "JLArrays", "Random"]
julia = "1.10"
2 changes: 2 additions & 0 deletions README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,6 +1,8 @@
# ManifoldsGPU

[![](https://img.shields.io/badge/docs-dev-blue.svg)](https://juliamanifolds.github.io/ManifoldsGPU.jl/dev/)
[![Aqua QA](https://raw.githubusercontent.com/JuliaTesting/Aqua.jl/master/badge.svg)](https://github.com/JuliaTesting/Aqua.jl)
[![code style: runic](https://img.shields.io/badge/code_style-%E1%9A%B1%E1%9A%A2%E1%9A%BE%E1%9B%81%E1%9A%B2-black)](https://github.com/fredrikekre/Runic.jl)

General GPU/CUDA support for the JuliaManifolds ecosystem.

Expand Down
63 changes: 33 additions & 30 deletions docs/src/index.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,34 +8,37 @@ ManifoldsGPU.ManifoldsGPU

Device: NVIDIA GeForce RTX 5070 Ti, eltype: Float32/ComplexF32

| Manifold | Operation | CPU median [ms] | GPU median [ms] | Speedup CPU/GPU | Relative error |
| Manifold | Operation | CPU median [ms] | GPU median [ms] | Speedup CPU/GPU | Error |
| --- | --- | ---: | ---: | ---: | ---: |
| Euclidean(32, 16, 2048) | exp | 0.36 | 0.17 | 2.17 | 0.0 |
| Euclidean(32, 16, 2048) | log! | 0.36 | 0.16 | 2.19 | 0.0 |
| Euclidean(32, 16, 2048) | inner | 0.19 | 0.14 | 1.36 | 9.357e-8 |
| Euclidean(32, 16, 2048) | norm | 0.12 | 0.16 | 0.76 | 8.423e-8 |
| Euclidean(32, 16, 2048) | project! | 0.25 | 0.13 | 1.93 | 0.0 |
| PowerManifold(Sphere(31), 2048) | exp | 0.05 | 46.57 | 0.0 | 6.877e-8 |
| PowerManifold(Sphere(31), 2048) | log! | 0.16 | 82.22 | 0.0 | 4.262e-8 |
| PowerManifold(Sphere(31), 2048) | inner | 0.02 | 0.12 | 0.13 | 5.86e-7 |
| PowerManifold(Sphere(31), 2048) | norm | 0.02 | 0.12 | 0.15 | 1.064e-7 |
| PowerManifold(Sphere(31), 2048) | project! | 0.03 | 41.71 | 0.0 | 2.813e-8 |
| PowerManifold(Rotations(32), 2048) | exp | 36.47 | 2.36 | 15.47 | 2.594e-6 |
| PowerManifold(Rotations(32), 2048) | log! | 570.16 | 78.98 | 7.22 | 9.157e-5 |
| PowerManifold(Rotations(32), 2048) | inner | 0.38 | 0.15 | 2.58 | 4.708e-6 |
| PowerManifold(Rotations(32), 2048) | norm | 1.2 | 0.15 | 7.82 | 9.132e-7 |
| PowerManifold(Rotations(32), 2048) | project! | 21.69 | 0.23 | 94.53 | 3.644e-7 |
| PowerManifold(Rotations(32), 2048) | retract_fused!(PolarRetraction) | 116.78 | 5.21 | 22.41 | 2.555e-6 |
| PowerManifold(UnitaryMatrices(32), 2048) | exp | 85.95 | 7.66 | 11.21 | 1.957e-6 |
| PowerManifold(UnitaryMatrices(32), 2048) | log! | 739.96 | 69.31 | 10.68 | 0.0001844 |
| PowerManifold(UnitaryMatrices(32), 2048) | inner | 0.76 | 58.76 | 0.01 | 5.979e-5 |
| PowerManifold(UnitaryMatrices(32), 2048) | norm | 1.73 | 45.7 | 0.04 | 1.516e-6 |
| PowerManifold(UnitaryMatrices(32), 2048) | project! | 31.55 | 0.36 | 88.48 | 5.512e-7 |
| PowerManifold(Grassmann(32, 16), 2048) | exp | 69.77 | 5.36 | 13.01 | 7.023e-5 |
| PowerManifold(Grassmann(32, 16), 2048) | log! | 57.62 | 3.3 | 17.48 | 2.332e-5 |
| PowerManifold(Grassmann(32, 16), 2048) | inner | 0.19 | 0.14 | 1.41 | 8.675e-7 |
| PowerManifold(Grassmann(32, 16), 2048) | norm | 0.86 | 0.13 | 6.58 | 2.772e-7 |
| PowerManifold(Grassmann(32, 16), 2048) | project! | 0.96 | 0.19 | 4.97 | 1.303e-7 |
| PowerManifold(Grassmann(32, 16), 2048) | retract_fused!(PolarRetraction) | 40.89 | 2.69 | 15.18 | 1.338e-6 |
| PowerManifold(Stiefel(32, 16), 2048) | exp(ExponentialRetraction) | 70.67 | 3.52 | 20.07 | 1.164e-6 |
| PowerManifold(Stiefel(32, 16), 2048) | retract_fused!(PolarRetraction) | 43.1 | 2.85 | 15.13 | 1.37e-6 |
| Euclidean(32, 16, 2048) | exp | 0.35 | 0.17 | 2.06 | 0.0 |
| Euclidean(32, 16, 2048) | log! | 0.35 | 0.17 | 2.05 | 0.0 |
| Euclidean(32, 16, 2048) | inner | 0.19 | 0.14 | 1.34 | 9.357e-8 |
| Euclidean(32, 16, 2048) | norm | 0.13 | 0.16 | 0.83 | 8.423e-8 |
| Euclidean(32, 16, 2048) | project! | 0.23 | 0.12 | 1.89 | 0.0 |
| PowerManifold(Sphere(31), 2048) | exp | 0.05 | 0.14 | 0.35 | 7.092e-8 |
| PowerManifold(Sphere(31), 2048) | log! | 0.08 | 0.37 | 0.23 | 5.125e-8 |
| PowerManifold(Sphere(31), 2048) | inner | 0.02 | 0.13 | 0.14 | 5.86e-7 |
| PowerManifold(Sphere(31), 2048) | norm | 0.02 | 0.13 | 0.15 | 1.064e-7 |
| PowerManifold(Sphere(31), 2048) | project! | 0.03 | 0.15 | 0.18 | 2.819e-8 |
| PowerManifold(Rotations(32), 2048) | exp | 36.19 | 2.35 | 15.38 | 2.594e-6 |
| PowerManifold(Rotations(32), 2048) | log! | 565.72 | 74.44 | 7.6 | 9.157e-5 |
| PowerManifold(Rotations(32), 2048) | inner | 0.41 | 0.25 | 1.65 | 4.708e-6 |
| PowerManifold(Rotations(32), 2048) | norm | 1.36 | 0.14 | 9.53 | 1.109e-6 |
| PowerManifold(Rotations(32), 2048) | project! | 20.46 | 0.22 | 91.22 | 3.644e-7 |
| PowerManifold(Rotations(32), 2048) | retract_fused!(PolarRetraction) | 115.51 | 4.89 | 23.62 | 2.555e-6 |
| PowerManifold(Rotations(32), 2048) | retract_fused!(QRRetraction) | 90.11 | 0.83 | 108.95 | 3.204e-7 |
| PowerManifold(UnitaryMatrices(32), 2048) | exp | 85.89 | 8.07 | 10.64 | 1.957e-6 |
| PowerManifold(UnitaryMatrices(32), 2048) | log! | 729.57 | 69.89 | 10.44 | 0.0001844 |
| PowerManifold(UnitaryMatrices(32), 2048) | inner | 0.83 | 56.26 | 0.01 | 5.979e-5 |
| PowerManifold(UnitaryMatrices(32), 2048) | norm | 1.74 | 44.17 | 0.04 | 1.516e-6 |
| PowerManifold(UnitaryMatrices(32), 2048) | project! | 31.42 | 0.35 | 90.36 | 5.512e-7 |
| PowerManifold(Grassmann(32, 16), 2048) | exp | 69.72 | 5.27 | 13.22 | 7.023e-5 |
| PowerManifold(Grassmann(32, 16), 2048) | log! | 57.99 | 3.36 | 17.27 | 2.332e-5 |
| PowerManifold(Grassmann(32, 16), 2048) | inner | 0.2 | 0.13 | 1.54 | 8.056e-7 |
| PowerManifold(Grassmann(32, 16), 2048) | norm | 0.81 | 0.14 | 5.95 | 3.696e-7 |
| PowerManifold(Grassmann(32, 16), 2048) | project! | 1.0 | 0.23 | 4.45 | 1.303e-7 |
| PowerManifold(Grassmann(32, 16), 2048) | retract_fused!(PolarRetraction) | 40.67 | 2.99 | 13.59 | 0.0001873 |
| PowerManifold(Grassmann(32, 16), 2048) | retract_fused!(QRRetraction) | 17.84 | 0.72 | 24.63 | 4.623e-5 |
| PowerManifold(Stiefel(32, 16), 2048) | exp(ExponentialRetraction) | 70.99 | 3.56 | 19.91 | 1.164e-6 |
| PowerManifold(Stiefel(32, 16), 2048) | retract_fused!(PolarRetraction) | 43.66 | 2.87 | 15.21 | 1.37e-6 |
| PowerManifold(Stiefel(32, 16), 2048) | retract_fused!(QRRetraction) | 18.28 | 0.71 | 25.89 | 1.885e-7 |
58 changes: 58 additions & 0 deletions ext/ManifoldsGPUCUDAExt/Sphere.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,3 +14,61 @@ function ManifoldsBase.norm(
) where {T <: Real}
return sqrt(dot(X, X))
end

function ManifoldsBase.exp!(
::PowerManifold{ℝ, <:Sphere{ℝ}, <:Tuple, ArrayPowerRepresentation},
q::CuArray{T},
p::CuArray{T},
X::CuArray{T},
) where {T <: Real}
θ = sqrt.(sum(abs2, X; dims = 1))
q .= cos.(θ) .* p .+ Manifolds.usinc.(θ) .* X
return q
end

function ManifoldsBase.log!(
M::PowerManifold{ℝ, <:Sphere{ℝ}, <:Tuple, ArrayPowerRepresentation},
X::CuArray{T},
p::CuArray{T},
q::CuArray{T},
) where {T <: Real}
cosθ = clamp.(sum(p .* q; dims = 1), -one(T), one(T))
θ = acos.(cosθ)

X_regular = (q .- cosθ .* p) ./ Manifolds.usinc.(θ)

antipodal = abs.(cosθ .+ one(T)) .<= sqrt(eps(T))
basis = CUDA.zeros(T, size(p))
basis[1, :] .= one(T)
if size(p, 1) > 1
p1_is_one = abs.(p[1:1, :] .- one(T)) .<= sqrt(eps(T))
basis[1, :] .-= T.(p1_is_one[1, :])
basis[2, :] .= T.(p1_is_one[1, :])
end

X_antipodal = basis .- p .* sum(p .* basis; dims = 1)
X_antipodal .*= T(π) ./ sqrt.(sum(abs2, X_antipodal; dims = 1))

X .= ifelse.(antipodal, X_antipodal, X_regular)
return project!(M, X, p, X)
end

function ManifoldsBase.project!(
::PowerManifold{ℝ, <:Sphere{ℝ}, <:Tuple, ArrayPowerRepresentation},
Y::CuArray{T},
p::CuArray{T},
X::CuArray{T},
) where {T <: Real}
Y .= X .- p .* sum(p .* X; dims = 1)
return Y
end

function ManifoldsBase.project!(
::PowerManifold{ℝ, <:Sphere{ℝ}, <:Tuple, ArrayPowerRepresentation},
q::CuArray{T},
p::CuArray{T},
) where {T <: Real}
norms_p = sqrt.(sum(abs2, p; dims = 1))
q .= p ./ norms_p
return q
end
27 changes: 27 additions & 0 deletions test/Project.toml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,27 @@
[deps]
Aqua = "4c88cf16-eb10-579e-8560-4a9242c79595"
CUDA = "052768ef-5323-5732-b1bb-66c8b64840ba"
GPUArrays = "0c68f7d7-f131-5f86-a1c3-88cf8149b2d7"
JLArrays = "27aeb0d3-9eb9-45fb-866b-73c2ecf80fcb"
LinearAlgebra = "37e2e46d-f89d-539d-b4ee-838fcccc9c8e"
ManifoldDiff = "af67fdf4-a580-4b9f-bbec-742ef357defd"
Manifolds = "1cead3c2-87b3-11e9-0ccd-23c62b72b94e"
ManifoldsBase = "3362f125-f0bb-47a3-aa74-596ffd7ef2fb"
ManifoldsGPU = "007d1224-8888-47ee-87d0-87e096ff9b5b"
Random = "9a3f8284-a2c9-5f02-9a11-845980a1fd5c"
Test = "8dfed614-e22c-5e08-85e1-65c5234f0b40"

[sources]
ManifoldsGPU = {path = ".."}

[compat]
Aqua = "0.8"
CUDA = "5.9.6"
GPUArrays = "11"
JLArrays = "0.3"
LinearAlgebra = "1.10"
ManifoldDiff = "0.4.5"
Manifolds = "0.11.12"
ManifoldsBase = "2.3.1"
Test = "1.10"
julia = "1.10"
100 changes: 100 additions & 0 deletions test/cuda/test_sphere.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,3 +1,5 @@
using Manifolds, ManifoldsGPU, Test, Random, CUDA

@testset "Sphere CUDA" begin
@testset "inner and norm" begin
Random.seed!(70)
Expand Down Expand Up @@ -46,4 +48,102 @@
@test isapprox(i_gpu, i_cpu; atol = 1.0f-4, rtol = 1.0f-4)
@test isapprox(n_gpu, n_cpu; atol = 1.0f-4, rtol = 1.0f-4)
end

@testset "project!" begin
for T in [Float32, Float64]
Random.seed!(72)

M = Sphere(7)
MP = PowerManifold(M, 32)

p = T.(rand(MP))
q_cpu = similar(p)
project!(MP, q_cpu, p)

p_cu = CuArray(p)
q_cu = similar(p_cu)
project!(MP, q_cu, p_cu)

if T === Float32
@test isapprox(Array(q_cu), q_cpu; atol = 1.0f-5, rtol = 1.0f-5)
else
@test isapprox(Array(q_cu), q_cpu; atol = 1.0e-12, rtol = 1.0e-12)
end
end
end

@testset "project! tangent" begin
for (seed, T, atol, rtol) in (
(721, Float32, 1.0f-5, 1.0f-5),
(722, Float64, 1.0e-12, 1.0e-12),
)
Random.seed!(seed)

M = Sphere(7)
MP = PowerManifold(M, 32)

p = T.(rand(MP))
X = T.(randn(size(p)))

Y_cpu = similar(p)
project!(MP, Y_cpu, p, X)

p_cu = CuArray(p)
X_cu = CuArray(X)
Y_cu = similar(p_cu)
project!(MP, Y_cu, p_cu, X_cu)

@test isapprox(Array(Y_cu), Y_cpu; atol = atol, rtol = rtol)
end
end

@testset "exp!" begin
for (seed, T, atol, rtol) in (
(73, Float32, 1.0f-4, 1.0f-4),
(74, Float64, 1.0e-10, 1.0e-10),
)
Random.seed!(seed)

M = Sphere(7)
MP = PowerManifold(M, 32)

p = T.(rand(MP))
X = T.(rand(MP; vector_at = p))

q_cpu = similar(p)
exp!(MP, q_cpu, p, X)

p_cu = CuArray(p)
X_cu = CuArray(X)
q_cu = similar(p_cu)
exp!(MP, q_cu, p_cu, X_cu)

@test isapprox(Array(q_cu), q_cpu; atol = atol, rtol = rtol)
end
end

@testset "log!" begin
for (seed, T, atol, rtol) in (
(75, Float32, 5.0f-4, 5.0f-4),
(76, Float64, 1.0e-9, 1.0e-9),
)
Random.seed!(seed)

M = Sphere(7)
MP = PowerManifold(M, 32)

p = T.(rand(MP))
q = T.(rand(MP))

X_cpu = similar(p)
log!(MP, X_cpu, p, q)

p_cu = CuArray(p)
q_cu = CuArray(q)
X_cu = similar(p_cu)
log!(MP, X_cu, p_cu, q_cu)

@test isapprox(Array(X_cu), X_cpu; atol = atol, rtol = rtol)
end
end
end
2 changes: 2 additions & 0 deletions test/runtests.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -26,4 +26,6 @@ using GPUArrays
else
@info "CUDA not available, skipping CUDA tests"
end

include(joinpath(@__DIR__, "test_aqua.jl"))
end
5 changes: 5 additions & 0 deletions test/test_aqua.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,5 @@
using Aqua, ManifoldsGPU, Test

@testset "Aqua.jl" begin
Aqua.test_all(ManifoldsGPU)
end