Skip to content

Add GPU profiles #74

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
blegat merged 2 commits into from
May 28, 2026
Merged

Add GPU profiles #74

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
b7e3118
Add GPU profiles
blegat May 27, 2026
8bc418d
Fix format
blegat May 28, 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
18 changes: 18 additions & 0 deletions perf/arraydiff.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -106,4 +106,22 @@ function neural(
)
end

function profile_gpu(; T = Float32, h = 4096, d = 13, n = 178)
state = _build(T, h, d, n, true)
x = CUDA.CuVector{T}(vec(state.W1))
g = CUDA.zeros(T, h * d)
fill!(state.evaluator.backend.last_x, NaN)
CUDA.@sync CUDA.@allowscalar MOI.eval_objective_gradient(
state.evaluator,
g,
x,
)
fill!(state.evaluator.backend.last_x, NaN)
return CUDA.@profile CUDA.@sync CUDA.@allowscalar MOI.eval_objective_gradient(
state.evaluator,
g,
x,
)
end

end # module
26 changes: 26 additions & 0 deletions perf/hand_cuda.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -111,4 +111,30 @@ function neural(
end
end

function profile_gpu(;
T = Float32,
h = 4096,
d = 13,
n = 178,
prealloc::Bool = true,
)
Random.seed!(0)
W1 = randn(T, h, d)
W2 = randn(T, OUT_DIM, h)
X = randn(T, d, n)
y = randn(T, OUT_DIM, n)
W1g, W2g, Xg, yg = CuArray(W1), CuArray(W2), CuArray(X), CuArray(y)
CUDA.synchronize()
CUDA.@sync if prealloc
gradient!(Buffers{(typeof(W1g))}(h, d, n), W1g, W2g, Xg, yg)
else
gradient_alloc(W1g, W2g, Xg, yg)
end
return CUDA.@profile CUDA.@sync if prealloc
gradient!(Buffers{(typeof(W1g))}(h, d, n), W1g, W2g, Xg, yg)
else
gradient_alloc(W1g, W2g, Xg, yg)
end
end

end # module
Loading