Skip to content

Commit

Permalink
avoid unnecessary allocs and rework nln part
Browse files Browse the repository at this point in the history
  • Loading branch information
@tmigot
tmigot committed Jul 25, 2022
1 parent 78e6aa5 commit e89575d
Show file tree
Hide file tree
Showing 2 changed files with 69 additions and 113 deletions.
18 changes: 10 additions & 8 deletions src/CUTEst.jl
View file
Original file line number Diff line number Diff line change
Expand Up @@ -22,6 +22,8 @@ mutable struct CUTEstModel <: AbstractNLPModel{Float64, Vector{Float64}}
counters::Counters
hrows::Vector{Int32}
hcols::Vector{Int32}

jac_structure_reliable::Bool
jrows::Vector{Int32}
jcols::Vector{Int32}

Expand All @@ -30,11 +32,9 @@ mutable struct CUTEstModel <: AbstractNLPModel{Float64, Vector{Float64}}
clincols::Vector{Int32}
clinvals::Vector{Float64}

nln_structure_reliable::Bool
cnlnrows::Vector{Int32}
cnlncols::Vector{Int32}

work::Vector{Float64}
Jval::Vector{Cdouble}
Jvar::Vector{Cint}
end

const funit = convert(Int32, 42)
Expand Down Expand Up @@ -367,6 +367,8 @@ function CUTEstModel(

hrows = Vector{Int32}(undef, nnzh)
hcols = Vector{Int32}(undef, nnzh)

jac_structure_reliable = false
jrows = Vector{Int32}(undef, nnzj)
jcols = Vector{Int32}(undef, nnzj)
work = Vector{Int32}(undef, ncon)
Expand All @@ -375,11 +377,11 @@ function CUTEstModel(
clinrows = Vector{Int32}(undef, lin_nnzj)
clincols = Vector{Int32}(undef, lin_nnzj)
clinvals = Vector{Float64}(undef, lin_nnzj)

Jval = Array{Cdouble}(undef, nvar)
Jvar = Array{Cint}(undef, nvar)

nln_structure_reliable = false
cnlnrows = Vector{Int32}(undef, nln_nnzj)
cnlncols = Vector{Int32}(undef, nln_nnzj)
nlp = CUTEstModel(meta, Counters(), hrows, hcols, jrows, jcols, lin_structure_reliable, clinrows, clincols, clinvals, nln_structure_reliable, cnlnrows, cnlncols, work)
nlp = CUTEstModel(meta, Counters(), hrows, hcols, jac_structure_reliable, jrows, jcols, lin_structure_reliable, clinrows, clincols, clinvals, work, Jval, Jvar)

cutest_instances += 1
finalizer(cutest_finalize, nlp)
Expand Down
164 changes: 59 additions & 105 deletions src/julia_interface.jl
View file
Original file line number Diff line number Diff line change
Expand Up @@ -334,22 +334,25 @@ function NLPModels.jac_structure!(
end

function NLPModels.jac_structure!(nlp::CUTEstModel)
nnzj = nlp.meta.nnzj
io_err = Cint[0]
this_nnzj = Cint[0]
if !nlp.jac_structure_reliable
nnzj = nlp.meta.nnzj
io_err = Cint[0]
this_nnzj = Cint[0]

ccall(
dlsym(cutest_lib, :cutest_csjp_),
Nothing,
(Ptr{Int32}, Ptr{Int32}, Ref{Int32}, Ptr{Int32}, Ptr{Int32}),
io_err,
this_nnzj,
nnzj,
nlp.jcols,
nlp.jrows,
)
@cutest_error
ccall(
dlsym(cutest_lib, :cutest_csjp_),
Nothing,
(Ptr{Int32}, Ptr{Int32}, Ref{Int32}, Ptr{Int32}, Ptr{Int32}),
io_err,
this_nnzj,
nnzj,
nlp.jcols,
nlp.jrows,
)
@cutest_error

nlp.jac_structure_reliable = true
end
return nlp
end

Expand All @@ -367,25 +370,25 @@ function NLPModels.jac_structure!(
end

function eval_lin_structure!(nlp::CUTEstModel)
nvar = Cint[nlp.meta.nvar]
ci = [0.0]
nnzj = Cint[0]
Jval = @view(Array{Cdouble}(undef, nlp.meta.nvar + 1)[2:end])
Jvar = @view(Array{Cint}(undef, nlp.meta.nvar + 1)[2:end])
i = 1
for j in nlp.meta.lin
ccifsg(Cint[0], nvar, Cint[j], nlp.meta.x0, ci, nnzj, nvar, Jval, Jvar, Cint[true])
for k = 1:nnzj[1]
nlp.clinrows[i] = findfirst(x -> x == j, nlp.meta.lin)
nlp.clincols[i] = Jvar[k]
nlp.clinvals[i] = Jval[k]
i += 1
if !nlp.lin_structure_reliable
nvar = Cint[nlp.meta.nvar]
ci = [0.0]
nnzj = Cint[0]
i = 1
for j in nlp.meta.lin
ccifsg(Cint[0], nvar, Cint[j], nlp.meta.x0, ci, nnzj, nvar, nlp.Jval, nlp.Jvar, Cint[true])
for k = 1:nnzj[1]
nlp.clinrows[i] = findfirst(x -> x == j, nlp.meta.lin)
nlp.clincols[i] = nlp.Jvar[k]
nlp.clinvals[i] = nlp.Jval[k]
i += 1
end
end
nlp.clinrows[i:end] .= Int32(1)
nlp.clincols[i:end] .= Int32(1)
nlp.clinvals[i:end] .= 0.0
nlp.lin_structure_reliable = true
end
nlp.clinrows[i:end] .= Int32(1)
nlp.clincols[i:end] .= Int32(1)
nlp.clinvals[i:end] .= 0.0
nlp.lin_structure_reliable = true
return nlp
end

Expand All @@ -394,9 +397,7 @@ function NLPModels.jac_lin_structure!(
rows::AbstractVector{<:Integer},
cols::AbstractVector{<:Integer},
)
if !nlp.lin_structure_reliable
eval_lin_structure!(nlp)
end
eval_lin_structure!(nlp)
rows .= nlp.clinrows
cols .= nlp.clincols
return rows, cols
Expand All @@ -407,30 +408,16 @@ function NLPModels.jac_nln_structure!(
rows::AbstractVector{<:Integer},
cols::AbstractVector{<:Integer},
)
if !nlp.nln_structure_reliable
st = Cint[0]
nvar = Cint[nlp.meta.nvar]
ci = view([0.0], 1:1)
nnzj = view(Cint[0], 1:1)
lj = view(Cint[nlp.meta.nvar], 1:1)
Jval = @view(Array{Cdouble}(undef, nlp.meta.nvar + 1)[2:end])
Jvar = @view(Array{Cint}(undef, nlp.meta.nvar + 1)[2:end])
True = view(Cint[true], 1:1)
i = 1
for j in nlp.meta.nln
ccifsg(st, nvar, Cint[j], nlp.meta.x0, ci, nnzj, lj, Jval, Jvar, True)
for k = 1:nnzj[1]
nlp.cnlnrows[i] = findfirst(x -> x == j, nlp.meta.nln)
nlp.cnlncols[i] = Jvar[k]
i += 1
end
end
nlp.cnlnrows[i:end] .= one(eltype(rows))
nlp.cnlncols[i:end] .= one(eltype(rows))
nlp.nln_structure_reliable = true
jac_structure!(nlp)
k = 1
for i in findall(j -> j in nlp.meta.nln, nlp.jrows)
rows[k] = nlp.jrows[i] - count(x < nlp.jrows[i] for x in nlp.meta.lin)
cols[k] = nlp.jcols[i]
k += 1
end
rows .= nlp.cnlnrows
cols .= nlp.cnlncols
rows[k:end] .= zero(Int32)
cols[k:end] .= zero(Int32)

return rows, cols
end

Expand All @@ -442,31 +429,24 @@ end

function NLPModels.jac_lin_coord!(nlp::CUTEstModel, x::AbstractVector, vals::AbstractVector)
nlp.counters.neval_jac_lin += 1
if !nlp.lin_structure_reliable
eval_lin_structure!(nlp)
end
eval_lin_structure!(nlp)
vals .= nlp.clinvals
return vals
end

function NLPModels.jac_nln_coord!(nlp::CUTEstModel, x::AbstractVector, vals::AbstractVector)
st = Cint[0]
nvar = Cint[nlp.meta.nvar]
ci = view([0.0], 1:1)
nnzj = view(Cint[0], 1:1)
lj = view(Cint[nlp.meta.nvar], 1:1)
Jval = @view(Array{Cdouble}(undef, nlp.meta.nvar + 1)[2:end])
Jvar = @view(Array{Cint}(undef, nlp.meta.nvar + 1)[2:end])
True = view(Cint[true], 1:1)
ci = [0.0]
nnzj = Cint[0]
i = 1
for j in nlp.meta.nln
ccifsg(st, nvar, Cint[j], x, ci, nnzj, lj, Jval, Jvar, True)
ccifsg(Cint[0], nvar, Cint[j], x, ci, nnzj, nvar, nlp.Jval, nlp.Jvar, Cint[true])
for k = 1:nnzj[1]
vals[i] = Jval[k]
vals[i] = nlp.Jval[k]
i += 1
end
end
vals[i:end] .= zero(eltype(vals))
vals[i:end] .= 0.0
nlp.counters.neval_jac_nln += 1
return vals
end
Expand Down Expand Up @@ -528,7 +508,7 @@ function NLPModels.jprod!(
v::AbstractVector,
jv::AbstractVector,
)
jvc = zeros(nlp.meta.ncon)
jvc = nlp.work
jprod!(nlp, Vector{Float64}(x), Vector{Float64}(v), jvc)
jv[1:(nlp.meta.ncon)] .= jvc
end
Expand All @@ -539,8 +519,8 @@ function NLPModels.jprod_nln!(
v::AbstractVector,
jv::AbstractVector,
)
jvc = zeros(nlp.meta.ncon)
jprod!(nlp, Vector{Float64}(x), Vector{Float64}(v), jvc)
jvc = nlp.work
jprod!(nlp, x, v, jvc)
nlp.counters.neval_jprod -= 1
nlp.counters.neval_jprod_nln += 1
jv[1:(nlp.meta.nnln)] .= jvc[nlp.meta.nln]
Expand All @@ -553,9 +533,7 @@ function NLPModels.jprod_lin!(
jv::AbstractVector,
)
nlp.counters.neval_jprod_lin += 1
if !nlp.lin_structure_reliable
eval_lin_structure!(nlp)
end
eval_lin_structure!(nlp)
jprod_lin!(nlp, nlp.clinrows, nlp.clincols, nlp.clinvals, v, jv)
return jv
end
Expand Down Expand Up @@ -626,53 +604,29 @@ function NLPModels.jtprod_nln!(
nlp::CUTEstModel,
x::AbstractVector,
v::AbstractVector,
jtv::StrideOneVector{Float64},
jtv::AbstractVector,
)
_v = Vector{Float64}(undef, nlp.meta.ncon)
_v = nlp.work
_v[nlp.meta.lin] .= 0.0
_v[nlp.meta.nln] = v
jtprod!(nlp, Vector{Float64}(x), _v, jtv)
jtprod!(nlp, x, _v, jtv)
nlp.counters.neval_jtprod -= 1
nlp.counters.neval_jtprod_nln += 1
return jtv
end

function NLPModels.jtprod_nln!(
nlp::CUTEstModel,
x::AbstractVector,
v::AbstractVector,
jtv::AbstractVector,
)
jtvc = zeros(nlp.meta.nvar)
jtprod_nln!(nlp, x, _v, jtvc)
jtv[1:(nlp.meta.nvar)] .= jtvc
end

function NLPModels.jtprod_lin!(
nlp::CUTEstModel,
x::AbstractVector,
v::AbstractVector,
jtv::StrideOneVector{Float64},
jtv::AbstractVector,
)
nlp.counters.neval_jtprod_lin += 1
if !nlp.lin_structure_reliable
eval_lin_structure!(nlp)
end
eval_lin_structure!(nlp)
jtprod_lin!(nlp, nlp.clinrows, nlp.clincols, nlp.clinvals, v, jtv)
return jtv
end

function NLPModels.jtprod_lin!(
nlp::CUTEstModel,
x::AbstractVector,
v::AbstractVector,
jtv::AbstractVector,
)
jtvc = zeros(nlp.meta.nvar)
jtprod_lin!(nlp, x, v, jtvc)
jtv[1:(nlp.meta.nvar)] .= jtvc
end

function NLPModels.hess_structure!(
nlp::CUTEstModel,
rows::StrideOneVector{Int32},
Expand Down

0 comments on commit e89575d

Please sign in to comment.