Add preallocation for obj

src/qpmodel.jl

@@ -8,10 +8,12 @@ mutable struct QPData{
 }
   c0::T         # constant term in objective
   c::S          # linear term
+  v::S          # vector that stores products with the hessian v = H*
   H::M1
   A::M2
 end
 
+@inline QPData(c0, c, H, A) = QPData(c0, c, similar(c), H, A)
 isdense(data::QPData{T, S, M1, M2}) where {T, S, M1, M2} = M1 <: DenseMatrix || M2 <: DenseMatrix
 
 function Base.convert(
@@ -20,7 +22,7 @@ function Base.convert(
 ) where {T, S, M1 <: AbstractMatrix, M2 <: AbstractMatrix, MCOO <: SparseMatrixCOO{T}}
   HCOO = (M1 <: SparseMatrixCOO) ? data.H : SparseMatrixCOO(data.H)
   ACOO = (M2 <: SparseMatrixCOO) ? data.A : SparseMatrixCOO(data.A)
-  return QPData(data.c0, data.c, HCOO, ACOO)
+  return QPData(data.c0, data.c, data.v, HCOO, ACOO)
 end
 Base.convert(
   ::Type{QPData{T, S, MCOO, MCOO}},
@@ -256,9 +258,8 @@ end
 
 function NLPModels.obj(qp::AbstractQuadraticModel{T, S}, x::AbstractVector) where {T, S}
   NLPModels.increment!(qp, :neval_obj)
-  Hx = fill!(S(undef, qp.meta.nvar), zero(T))
-  mul!(Hx, Symmetric(qp.data.H, :L), x)
-  return qp.data.c0 + dot(qp.data.c, x) + dot(Hx, x) / 2
+  mul!(qp.data.v, Symmetric(qp.data.H, :L), x)
+  return qp.data.c0 + dot(qp.data.c, x) + dot(qp.data.v, x) / 2
 end
 
 function NLPModels.grad!(qp::AbstractQuadraticModel, x::AbstractVector, g::AbstractVector)

test/runtests.jl

@@ -259,3 +259,4 @@ end
 end
 
 include("test_presolve.jl")
+include("test_allocations.jl")

test/test_allocations.jl

@@ -0,0 +1,47 @@
+function test_only_zeros(table)
+  for (key, vals) in table
+    if !isnan(vals)
+      @test vals == 0
+      vals != 0 && println(key)
+    end
+  end
+end
+
+@testset "allocations" begin
+  @testset "allocs QPSData" begin
+    for problem in qp_problems_Matrix
+      nlp_qps = eval(Symbol(problem * "_QPSData"))()
+      test_only_zeros(test_allocs_nlpmodels(nlp_qps))
+    end
+  end
+
+  @testset "allocs QP_dense" begin
+    for problem in qp_problems_Matrix
+      nlp_qm_dense = eval(Symbol(problem * "_QP_dense"))()
+      test_only_zeros(test_allocs_nlpmodels(nlp_qm_dense))
+    end
+  end
+
+  @testset "allocs COO QPSData" begin
+    for problem in qp_problems_COO
+      nlp_qps = eval(Symbol(problem * "_QPSData"))()
+      test_only_zeros(test_allocs_nlpmodels(nlp_qps))
+    end
+  end
+
+  @testset "allocs COO QP" begin
+    for problem in qp_problems_COO
+      nlp_qm_dense = eval(Symbol(problem * "_QP"))()
+      test_only_zeros(test_allocs_nlpmodels(nlp_qm_dense))
+    end
+  end
+
+  @testset "allocs quadratic approximation" begin
+    for problem in NLPModelsTest.nlp_problems
+      nlp = eval(Symbol(problem))()
+      x = nlp.meta.x0
+      nlp_qm = QuadraticModel(nlp, x)
+      test_only_zeros(test_allocs_nlpmodels(nlp_qm))
+    end
+  end
+end