Skip to content
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

GPU support for CompressedNLPModels #88

Merged
merged 1 commit into from
May 2, 2024
Merged

GPU support for CompressedNLPModels #88

Show file tree
Hide file tree
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
55 changes: 45 additions & 10 deletions ext/ExaModelsKernelAbstractions.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -14,7 +14,7 @@ ExaModels.ExaCore(T, backend::KernelAbstractions.CPU) =
ExaModels.ExaCore(x0 = zeros(T, 0), backend = backend)
ExaModels.ExaCore(backend::KernelAbstractions.CPU) = ExaModels.ExaCore(backend = backend)

function getptr(backend, array; cmp = isequal)
function ExaModels.getptr(backend, array; cmp = (x,y) -> x != y)

bitarray = similar(array, Bool, length(array) + 1)
kergetptr(backend)(cmp, bitarray, array; ndrange = length(array) + 1)
Expand Down Expand Up @@ -48,14 +48,14 @@ function ExaModels.ExaModel(
if !isempty(gsparsity)
ExaModels.sort!(gsparsity; lt = ((i, j), (k, l)) -> i < k)
end
gptr = getptr(c.backend, gsparsity)
gptr = ExaModels.getptr(c.backend, gsparsity; cmp = (x,y) -> x[1] != y[1])

conaugsparsity = similar(c.x0, Tuple{Int,Int}, c.nconaug)
_conaug_structure!(c.backend, c.con, conaugsparsity)
if !isempty(conaugsparsity)
ExaModels.sort!(conaugsparsity; lt = ((i, j), (k, l)) -> i < k)
end
conaugptr = getptr(c.backend, conaugsparsity)
conaugptr = ExaModels.getptr(c.backend, conaugsparsity; cmp = (x,y) -> x[1] != y[1])


if prod
Expand All @@ -73,22 +73,22 @@ function ExaModels.ExaModel(
if !isempty(jacsparsityi)
ExaModels.sort!(jacsparsityi; lt = (((i, j), k), ((n, m), l)) -> i < n)
end
jacptri = getptr(c.backend, jacsparsityi; cmp = (x, y) -> x[1] == y[1])
jacptri = ExaModels.getptr(c.backend, jacsparsityi; cmp = (x, y) -> x[1][1] != y[1][1])

if !isempty(jacsparsityj)
ExaModels.sort!(jacsparsityj; lt = (((i, j), k), ((n, m), l)) -> j < m)
end
jacptrj = getptr(c.backend, jacsparsityj; cmp = (x, y) -> x[2] == y[2])
jacptrj = ExaModels.getptr(c.backend, jacsparsityj; cmp = (x, y) -> x[1][2] != y[1][2])


if !isempty(hesssparsityi)
ExaModels.sort!(hesssparsityi; lt = (((i, j), k), ((n, m), l)) -> i < n)
end
hessptri = getptr(c.backend, hesssparsityi; cmp = (x, y) -> x[1] == y[1])
hessptri = ExaModels.getptr(c.backend, hesssparsityi; cmp = (x, y) -> x[1][1] != y[1][1])
if !isempty(hesssparsityj)
ExaModels.sort!(hesssparsityj; lt = (((i, j), k), ((n, m), l)) -> j < m)
end
hessptrj = getptr(c.backend, hesssparsityj; cmp = (x, y) -> x[2] == y[2])
hessptrj = ExaModels.getptr(c.backend, hesssparsityj; cmp = (x, y) -> x[1][2] != y[1][2])

prodhelper = (
jacbuffer = jacbuffer,
Expand Down Expand Up @@ -620,15 +620,50 @@ end
elseif I == length(array) + 1
bitarray[I] = true
else
i0, j0 = array[I-1]
i1, j1 = array[I]
i0 = array[I-1]
i1 = array[I]

if !cmp(i0, i1)
if cmp(i0, i1)
bitarray[I] = true
else
bitarray[I] = false
end
end
end

ExaModels.getbackend(m::ExaModels.ExaModel{T,VT,E}) where {T,VT,E<:KAExtension} = m.ext.backend
function ExaModels._compress!(V, buffer, ptr, sparsity, backend)
fill!(V, zero(eltype(V)))
ker_compress!(backend)(V, buffer, ptr, sparsity; ndrange= length(ptr)-1)
synchronize(backend)
end

@kernel function ker_compress!(V, @Const(buffer), @Const(ptr), @Const(sparsity))
i = @index(Global)
@inbounds for j = ptr[i]:ptr[i+1]-1
V[i] += buffer[sparsity[j][2]]
end
end

function ExaModels._structure!(I, J, ptr, sparsity, backend)
ker_structure!(backend)(I, J, ptr, sparsity, ndrange = length(ptr)-1)
synchronize(backend)
end

@kernel function ker_structure!(I, J, @Const(ptr), @Const(sparsity))
i = @index(Global)
@inbounds J[i], I[i] = sparsity[ptr[i]][1]
end

function ExaModels.get_compressed_sparsity(nnz, Ibuffer, Jbuffer, backend)
sparsity = similar(Ibuffer, Tuple{Tuple{Int,Int},Int},nnz)
ker_get_compressed_sparsity(backend)(sparsity, Ibuffer, Jbuffer; ndrange = nnz)
synchronize(backend)
return sparsity
end
@kernel function ker_get_compressed_sparsity(sparsity, @Const(I), @Const(J))
i = @index(Global)
@inbounds sparsity[i] = ((J[i],I[i]), i)
end

end # module ExaModelsKernelAbstractions
38 changes: 22 additions & 16 deletions src/utils.jl
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -382,6 +382,7 @@ Base.show(io::IO, ::MIME"text/plain", e::TimedNLPModel) = Base.print(io, e);
struct CompressedNLPModel{
T,
VT<:AbstractVector{T},
B,
VI<:AbstractVector{Int},
VI2<:AbstractVector{Tuple{Tuple{Int,Int},Int}},
M<:NLPModels.AbstractNLPModel{T,VT},
Expand All @@ -394,41 +395,43 @@ struct CompressedNLPModel{
hsparsity::VI2
buffer::VT

backend::B
meta::NLPModels.NLPModelMeta{T,VT}
counters::NLPModels.Counters
end

function getptr(array)
function getptr(backend::Nothing, array; cmp = (x,y) -> x != y)
return push!(
pushfirst!(
findall(_is_sparsity_not_equal.(@view(array[1:end-1]), @view(array[2:end]))) .+=
findall(cmp.(@view(array[1:end-1]), @view(array[2:end]))) .+=
1,
1,
),
length(array) + 1,
)
end
_is_sparsity_not_equal(a, b) = first(a) != first(b)

function CompressedNLPModel(m)

nnzj = NLPModels.get_nnzj(m)
Ibuffer = Vector{Int}(undef, nnzj)
Jbuffer = Vector{Int}(undef, nnzj)
Ibuffer = similar(m.meta.x0, Int, nnzj)
Jbuffer = similar(m.meta.x0, Int, nnzj)
NLPModels.jac_structure!(m, Ibuffer, Jbuffer)

jsparsity = map((k, i, j) -> ((j, i), k), 1:nnzj, Ibuffer, Jbuffer)
backend = getbackend(m)

jsparsity = get_compressed_sparsity(nnzj, Ibuffer, Jbuffer, backend)
sort!(jsparsity; lt = (a, b) -> a[1] < b[1])
jptr = getptr(jsparsity)
jptr = getptr(backend, jsparsity; cmp = (a, b) -> first(a) != first(b))

nnzh = NLPModels.get_nnzh(m)
resize!(Ibuffer, nnzh)
resize!(Jbuffer, nnzh)
NLPModels.hess_structure!(m, Ibuffer, Jbuffer)

hsparsity = map((k, i, j) -> ((j, i), k), 1:nnzh, Ibuffer, Jbuffer)
hsparsity = get_compressed_sparsity(nnzh, Ibuffer, Jbuffer, backend)
sort!(hsparsity; lt = (a, b) -> a[1] < b[1])
hptr = getptr(hsparsity)
hptr = getptr(backend, hsparsity; cmp = (a, b) -> first(a) != first(b))

buffer = similar(m.meta.x0, max(nnzj, nnzh))

Expand All @@ -447,9 +450,12 @@ function CompressedNLPModel(m)

counters = NLPModels.Counters()

return CompressedNLPModel(m, jptr, jsparsity, hptr, hsparsity, buffer, meta, counters)
return CompressedNLPModel(m, jptr, jsparsity, hptr, hsparsity, buffer, backend, meta, counters)
end

getbackend(m) = nothing
get_compressed_sparsity(nnz, Ibuffer, Jbuffer, backend::Nothing) = map((k, i, j) -> ((j, i), k), 1:nnz, Ibuffer, Jbuffer)

function NLPModels.obj(m::CompressedNLPModel, x::AbstractVector)
NLPModels.obj(m.inner, x)
end
Expand All @@ -464,7 +470,7 @@ end

function NLPModels.jac_coord!(m::CompressedNLPModel, x::AbstractVector, j::AbstractVector)
NLPModels.jac_coord!(m.inner, x, m.buffer)
_compress!(j, m.buffer, m.jptr, m.jsparsity)
_compress!(j, m.buffer, m.jptr, m.jsparsity, m.backend)
end

function NLPModels.hess_coord!(
Expand All @@ -475,26 +481,26 @@ function NLPModels.hess_coord!(
obj_weight = 1.0,
)
NLPModels.hess_coord!(m.inner, x, y, m.buffer; obj_weight = obj_weight)
_compress!(h, m.buffer, m.hptr, m.hsparsity)
_compress!(h, m.buffer, m.hptr, m.hsparsity, m.backend)
end

function NLPModels.jac_structure!(
m::CompressedNLPModel,
I::AbstractVector,
J::AbstractVector,
)
_structure!(I, J, m.jptr, m.jsparsity)
_structure!(I, J, m.jptr, m.jsparsity, m.backend)
end

function NLPModels.hess_structure!(
m::CompressedNLPModel,
I::AbstractVector,
J::AbstractVector,
)
_structure!(I, J, m.hptr, m.hsparsity)
_structure!(I, J, m.hptr, m.hsparsity, m.backend)
end

function _compress!(V, buffer, ptr, sparsity)
function _compress!(V, buffer, ptr, sparsity, backend::Nothing)
fill!(V, zero(eltype(V)))
@simd for i = 1:length(ptr)-1
for j = ptr[i]:ptr[i+1]-1
Expand All @@ -503,7 +509,7 @@ function _compress!(V, buffer, ptr, sparsity)
end
end

function _structure!(I, J, ptr, sparsity)
function _structure!(I, J, ptr, sparsity, backend::Nothing)
@simd for i = 1:length(ptr)-1
J[i], I[i] = sparsity[ptr[i]][1]
end
Expand Down
Loading