Skip to content

[Nonlinear] Merge forward_storage_ϵ with reverse_storage_ϵ #2732

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
odow merged 6 commits into from
Apr 24, 2025
Merged

[Nonlinear] Merge forward_storage_ϵ with reverse_storage_ϵ #2732

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
adceb0f
[Nonlinear] Merge forward_storage_ϵ with reverse_storage_ϵ
blegat Apr 21, 2025
84f6ded
Fixes
blegat Apr 23, 2025
707ab59
Fix
blegat Apr 23, 2025
7e9c6cb
Update mathoptinterface_api.jl
odow Apr 23, 2025
0ad6caa
Update mathoptinterface_api.jl
odow Apr 23, 2025
6da0650
Fix format
blegat Apr 24, 2025
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
8 changes: 4 additions & 4 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 @@ -135,7 +135,7 @@ function _hessian_slice_inner(d, ex, input_ϵ, output_ϵ, ::Type{T}) where {T}
subexpr_forward_values_ϵ[i] = _forward_eval_ϵ(
d,
subexpr,
_reinterpret_unsafe(T, subexpr.forward_storage_ϵ),
_reinterpret_unsafe(T, d.storage_ϵ),
_reinterpret_unsafe(T, subexpr.partials_storage_ϵ),
input_ϵ,
subexpr_forward_values_ϵ,
Expand All @@ -145,7 +145,7 @@ function _hessian_slice_inner(d, ex, input_ϵ, output_ϵ, ::Type{T}) where {T}
_forward_eval_ϵ(
d,
ex,
_reinterpret_unsafe(T, d.forward_storage_ϵ),
_reinterpret_unsafe(T, d.storage_ϵ),
_reinterpret_unsafe(T, d.partials_storage_ϵ),
input_ϵ,
subexpr_forward_values_ϵ,
Expand All @@ -161,7 +161,7 @@ function _hessian_slice_inner(d, ex, input_ϵ, output_ϵ, ::Type{T}) where {T}
_reverse_eval_ϵ(
output_ϵ,
ex,
_reinterpret_unsafe(T, d.reverse_storage_ϵ),
_reinterpret_unsafe(T, d.storage_ϵ),
_reinterpret_unsafe(T, d.partials_storage_ϵ),
d.subexpression_reverse_values,
subexpr_reverse_values_ϵ,
Expand All @@ -174,7 +174,7 @@ function _hessian_slice_inner(d, ex, input_ϵ, output_ϵ, ::Type{T}) where {T}
_reverse_eval_ϵ(
output_ϵ,
subexpr,
_reinterpret_unsafe(T, subexpr.reverse_storage_ϵ),
_reinterpret_unsafe(T, d.storage_ϵ),
_reinterpret_unsafe(T, subexpr.partials_storage_ϵ),
d.subexpression_reverse_values,
subexpr_reverse_values_ϵ,
Expand Down
28 changes: 13 additions & 15 deletions src/Nonlinear/ReverseAD/mathoptinterface_api.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -44,6 +44,7 @@ function MOI.initialize(d::NLPEvaluator, requested_features::Vector{Symbol})
want_hess_storage = (:HessVec in requested_features) || d.want_hess
coloring_storage = Coloring.IndexedSet(N)
max_expr_length = 0
max_expr_with_sub_length = 0
#
main_expressions = [c.expression.nodes for (_, c) in d.data.constraints]
if d.data.objective !== nothing
Expand Down Expand Up @@ -71,6 +72,8 @@ function MOI.initialize(d::NLPEvaluator, requested_features::Vector{Symbol})
)
d.subexpressions[k] = subex
d.subexpression_linearity[k] = subex.linearity
max_expr_with_sub_length =
max(max_expr_with_sub_length, length(subex.nodes))
if d.want_hess
empty!(coloring_storage)
_compute_gradient_sparsity!(coloring_storage, subex.nodes)
Expand Down Expand Up @@ -138,27 +141,22 @@ function MOI.initialize(d::NLPEvaluator, requested_features::Vector{Symbol})
end
# 10 is hardcoded upper bound to avoid excess memory allocation
max_chunk = min(max_chunk, 10)
max_expr_with_sub_length = max(max_expr_with_sub_length, max_expr_length)
if d.want_hess || want_hess_storage
d.input_ϵ = zeros(max_chunk * N)
d.output_ϵ = zeros(max_chunk * N)
#
len = max_chunk * max_expr_length
d.forward_storage_ϵ = zeros(len)
d.partials_storage_ϵ = zeros(len)
d.reverse_storage_ϵ = zeros(len)
d.partials_storage_ϵ = zeros(max_chunk * max_expr_length)
d.storage_ϵ = zeros(max_chunk * max_expr_with_sub_length)
#
len = max_chunk * length(d.subexpressions)
d.subexpression_forward_values_ϵ = zeros(len)
d.subexpression_reverse_values_ϵ = zeros(len)
#
for k in d.subexpression_order
len = max_chunk * length(d.subexpressions[k].nodes)
resize!(d.subexpressions[k].forward_storage_ϵ, len)
fill!(d.subexpressions[k].forward_storage_ϵ, 0.0)
resize!(d.subexpressions[k].partials_storage_ϵ, len)
fill!(d.subexpressions[k].partials_storage_ϵ, 0.0)
resize!(d.subexpressions[k].reverse_storage_ϵ, len)
fill!(d.subexpressions[k].reverse_storage_ϵ, 0.0)
end
d.max_chunk = max_chunk
if d.want_hess
Expand Down Expand Up @@ -350,7 +348,7 @@ function MOI.eval_hessian_lagrangian_product(d::NLPEvaluator, h, x, v, σ, μ)
subexpr_forward_values_ϵ[i] = _forward_eval_ϵ(
d,
subexpr,
reinterpret(T, subexpr.forward_storage_ϵ),
reinterpret(T, d.storage_ϵ),
reinterpret(T, subexpr.partials_storage_ϵ),
input_ϵ,
subexpr_forward_values_ϵ,
Expand All @@ -361,13 +359,13 @@ function MOI.eval_hessian_lagrangian_product(d::NLPEvaluator, h, x, v, σ, μ)
subexpr_reverse_values_ϵ = reinterpret(T, d.subexpression_reverse_values_ϵ)
fill!(subexpr_reverse_values_ϵ, zero(T))
fill!(d.subexpression_reverse_values, 0.0)
fill!(d.reverse_storage_ϵ, 0.0)
fill!(d.storage_ϵ, 0.0)
fill!(output_ϵ, zero(T))
if d.objective !== nothing
_forward_eval_ϵ(
d,
something(d.objective),
reinterpret(T, d.forward_storage_ϵ),
reinterpret(T, d.storage_ϵ),
reinterpret(T, d.partials_storage_ϵ),
input_ϵ,
subexpr_forward_values_ϵ,
Expand All @@ -376,7 +374,7 @@ function MOI.eval_hessian_lagrangian_product(d::NLPEvaluator, h, x, v, σ, μ)
_reverse_eval_ϵ(
output_ϵ,
something(d.objective),
reinterpret(T, d.reverse_storage_ϵ),
reinterpret(T, d.storage_ϵ),
reinterpret(T, d.partials_storage_ϵ),
d.subexpression_reverse_values,
subexpr_reverse_values_ϵ,
Expand All @@ -388,7 +386,7 @@ function MOI.eval_hessian_lagrangian_product(d::NLPEvaluator, h, x, v, σ, μ)
_forward_eval_ϵ(
d,
con,
reinterpret(T, d.forward_storage_ϵ),
reinterpret(T, d.storage_ϵ),
reinterpret(T, d.partials_storage_ϵ),
input_ϵ,
subexpr_forward_values_ϵ,
Expand All @@ -397,7 +395,7 @@ function MOI.eval_hessian_lagrangian_product(d::NLPEvaluator, h, x, v, σ, μ)
_reverse_eval_ϵ(
output_ϵ,
con,
reinterpret(T, d.reverse_storage_ϵ),
reinterpret(T, d.storage_ϵ),
reinterpret(T, d.partials_storage_ϵ),
d.subexpression_reverse_values,
subexpr_reverse_values_ϵ,
Expand All @@ -411,7 +409,7 @@ function MOI.eval_hessian_lagrangian_product(d::NLPEvaluator, h, x, v, σ, μ)
_reverse_eval_ϵ(
output_ϵ,
subexpr,
reinterpret(T, subexpr.reverse_storage_ϵ),
reinterpret(T, d.storage_ϵ),
reinterpret(T, subexpr.partials_storage_ϵ),
d.subexpression_reverse_values,
subexpr_reverse_values_ϵ,
Expand Down
11 changes: 3 additions & 8 deletions src/Nonlinear/ReverseAD/types.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -11,9 +11,7 @@ struct _SubexpressionStorage
forward_storage::Vector{Float64}
partials_storage::Vector{Float64}
reverse_storage::Vector{Float64}
forward_storage_ϵ::Vector{Float64}
partials_storage_ϵ::Vector{Float64}
reverse_storage_ϵ::Vector{Float64}
linearity::Linearity

function _SubexpressionStorage(
Expand All @@ -34,8 +32,6 @@ struct _SubexpressionStorage
zeros(N), # partials_storage,
zeros(N), # reverse_storage,
Float64[],
Float64[],
Float64[],
linearity[1],
)
end
Expand Down Expand Up @@ -175,11 +171,10 @@ mutable struct NLPEvaluator <: MOI.AbstractNLPEvaluator
# so the length should be multiplied by the maximum number of epsilon components
disable_2ndorder::Bool # don't offer Hess or HessVec
want_hess::Bool
forward_storage_ϵ::Vector{Float64} # (longest expression)
partials_storage_ϵ::Vector{Float64} # (longest expression)
reverse_storage_ϵ::Vector{Float64} # (longest expression)
partials_storage_ϵ::Vector{Float64} # (longest expression excluding subexpressions)
storage_ϵ::Vector{Float64} # (longest expression including subexpressions)
input_ϵ::Vector{Float64} # (number of variables)
output_ϵ::Vector{Float64}# (number of variables)
output_ϵ::Vector{Float64} # (number of variables)
subexpression_forward_values_ϵ::Vector{Float64} # (number of subexpressions)
subexpression_reverse_values_ϵ::Vector{Float64} # (number of subexpressions)
hessian_sparsity::Vector{Tuple{Int64,Int64}}
Expand Down
Loading