Skip to content

[Nonlinear.ReverseAD] fix performance bug in Hessian computation #2783

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
merged 1 commit into from
Jul 10, 2025
Merged

[Nonlinear.ReverseAD] fix performance bug in Hessian computation #2783

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
6 changes: 4 additions & 2 deletions src/Nonlinear/ReverseAD/forward_over_reverse.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -92,16 +92,19 @@ function _eval_hessian_chunk(
for s in 1:chunk
# If `chunk < chunk_size`, leaves junk in the unused components
d.input_ϵ[(idx-1)*chunk_size+s] = ex.seed_matrix[r, offset+s-1]
# Ensure the output is clear in preparation for the chunk
d.output_ϵ[(idx-1)*chunk_size+s] = 0.0
end
end
_hessian_slice_inner(d, ex, chunk_size)
fill!(d.input_ϵ, 0.0)
# collect directional derivatives
for r in eachindex(ex.rinfo.local_indices)
@inbounds idx = ex.rinfo.local_indices[r]
# load output_ϵ into ex.seed_matrix[r,k,k+1,...,k+remaining-1]
for s in 1:chunk
ex.seed_matrix[r, offset+s-1] = d.output_ϵ[(idx-1)*chunk_size+s]
# Reset the input in preparation for the next chunk
d.input_ϵ[(idx-1)*chunk_size+s] = 0.0
end
end
return
Expand All @@ -122,7 +125,6 @@ end
end

function _hessian_slice_inner(d, ex, ::Type{T}) where {T}
fill!(d.output_ϵ, 0.0)
output_ϵ = _reinterpret_unsafe(T, d.output_ϵ)
subexpr_forward_values_ϵ =
_reinterpret_unsafe(T, d.subexpression_forward_values_ϵ)
Expand Down
Loading