Update for KNITRO.jl v0.14.0

Project.toml

@@ -10,10 +10,13 @@ NLPModelsModifiers = "e01155f1-5c6f-4375-a9d8-616dd036575f"
 SolverCore = "ff4d7338-4cf1-434d-91df-b86cb86fb843"
 
 [compat]
-KNITRO = "0.13"
+ADNLPModels = "0.7"
+KNITRO = "0.14"
+LinearAlgebra = "<0.0.1, 1.6"
 NLPModels = "0.18, 0.19, 0.20"
 NLPModelsModifiers = "0.6"
 SolverCore = "0.3"
+Test = "<0.0.1, 1.6"
 julia = "^1.6.0"
 
 [extras]

src/NLPModelsKnitro.jl

@@ -34,7 +34,7 @@ function knitro end
 
 Returns a `KnitroSolver` structure to solve the problem `nlp` with `knitro!`.
 
-Knitro does not accept least-squares problems with constraints other than bounds. 
+Knitro does not accept least-squares problems with constraints other than bounds.
 If an NLSModel has constraints other than bounds, we convert it to a FeasibilityFormNLS.
 The first argument is `Val(false)` if the problem has been converted, and `Val(true)` otherwise.
 
@@ -44,8 +44,7 @@ mutable struct KnitroSolver <: AbstractOptimizationSolver
   kc
 end
 
-const KNITRO_VERSION = KNITRO.KNITRO_VERSION
-if KNITRO_VERSION == v"0.0.0"
+if KNITRO.knitro_version() == v"0.0.0"
   @error "KNITRO is not installed correctly"
 else
   include("api.jl")

src/api.jl

@@ -64,11 +64,18 @@ function setparams!(solver::KnitroSolver; kwargs...)
   end
 
   # specify that we are able to provide the Hessian without including the objective
-  KNITRO.KN_set_param(kc, KNITRO.KN_PARAM_HESSIAN_NO_F, KNITRO.KN_HESSIAN_NO_F_ALLOW)
+  KNITRO.KN_set_int_param(kc, KNITRO.KN_PARAM_HESSIAN_NO_F, KNITRO.KN_HESSIAN_NO_F_ALLOW)
 
   # pass options to KNITRO
   for (k, v) in kwargs
-    KNITRO.KN_set_param(kc, string(k), v)
+    if v isa Integer
+      KNITRO.KN_set_int_param_by_name(kc, string(k), v)
+    elseif v isa Cdouble
+      KNITRO.KN_set_double_param_by_name(kc, string(k), v)
+    else
+      @assert v isa AbstractString
+      KNITRO.KN_set_char_param_by_name(kc, string(k), v)
+    end
   end
 
   return solver
@@ -153,12 +160,19 @@ function SolverCore.solve!(
   nStatus, obj_val, x, lambda_ = KNITRO.KN_get_solution(kc)
   n = length(x)
   m = length(lambda_) - n
-  primal_feas = KNITRO.KN_get_abs_feas_error(kc)
-  dual_feas = KNITRO.KN_get_abs_opt_error(kc)
-  iter = KNITRO.KN_get_number_iters(kc)
-  if KNITRO_VERSION ≥ v"12.0"
-    Δt = KNITRO.KN_get_solve_time_cpu(kc)
-    real_time = KNITRO.KN_get_solve_time_real(kc)
+  pCdouble = Ref{Cdouble}()
+  KNITRO.KN_get_abs_feas_error(kc, pCdouble)
+  primal_feas = pCdouble[]
+  KNITRO.KN_get_abs_opt_error(kc, pCdouble)
+  dual_feas = pCdouble[]
+  pCint = Ref{Cint}()
+  KNITRO.KN_get_number_iters(kc, pCint)
+  iter = pCint[]
+  if KNITRO.knitro_version() ≥ v"12.0"
+    KNITRO.KN_get_solve_time_cpu(kc, pCdouble)
+    Δt = pCdouble[]
+    KNITRO.KN_get_solve_time_real(kc, pCdouble)
+    real_time = pCdouble[]
   else
     Δt = real_time = t[2]
   end

src/nlp.jl

@@ -15,7 +15,7 @@ function KnitroSolver(
   end
 
   # add variables and bound constraints
-  KNITRO.KN_add_vars(kc, n)
+  KNITRO.KN_add_vars(kc, n, C_NULL)
 
   lvarinf = isinf.(nlp.meta.lvar)
   if !all(lvarinf)
@@ -38,7 +38,7 @@ function KnitroSolver(
   end
 
   # add constraints
-  KNITRO.KN_add_cons(kc, m)
+  KNITRO.KN_add_cons(kc, m, C_NULL)
   lcon = nlp.meta.lcon
   lconinf = isinf.(lcon)
   if any(lconinf)
@@ -58,7 +58,7 @@ function KnitroSolver(
     jlrows, jlcols = jac_lin_structure(nlp)
     for klin = 1:(nlp.meta.lin_nnzj)
       row = nlp.meta.lin[jlrows[klin]]
-      KNITRO.KN_add_con_linear_struct(kc, Int32(row - 1), Int32(jlcols[klin] - 1), jlvals[klin])
+      KNITRO.KN_add_con_linear_struct_one(kc, 1, row - 1, Cint[jlcols[klin] - 1], Cdouble[jlvals[klin]])
     end
   end
 
@@ -157,11 +157,18 @@ function KnitroSolver(
   )
 
   # specify that we are able to provide the Hessian without including the objective
-  KNITRO.KN_set_param(kc, KNITRO.KN_PARAM_HESSIAN_NO_F, KNITRO.KN_HESSIAN_NO_F_ALLOW)
+  KNITRO.KN_set_int_param(kc, KNITRO.KN_PARAM_HESSIAN_NO_F, KNITRO.KN_HESSIAN_NO_F_ALLOW)
 
   # pass options to KNITRO
   for (k, v) in kwargs
-    KNITRO.KN_set_param(kc, string(k), v)
+    if v isa Integer
+      KNITRO.KN_set_int_param_by_name(kc, string(k), v)
+    elseif v isa Cdouble
+      KNITRO.KN_set_double_param_by_name(kc, string(k), v)
+    else
+      @assert v isa AbstractString
+      KNITRO.KN_set_char_param_by_name(kc, string(k), v)
+    end
   end
 
   # set user-defined callback called after each iteration

src/nls.jl

@@ -16,7 +16,7 @@ function KnitroSolver(
   end
 
   # add variables and bound constraints
-  KNITRO.KN_add_vars(kc, n)
+  KNITRO.KN_add_vars(kc, n, C_NULL)
 
   lvarinf = isinf.(nls.meta.lvar)
   if !all(lvarinf)
@@ -52,7 +52,7 @@ function KnitroSolver(
   end
 
   # add number of residual functions
-  KNITRO.KN_add_rsds(kc, ne)
+  KNITRO.KN_add_rsds(kc, ne, C_NULL)
 
   jrows, jcols = jac_structure_residual(nls)
 
@@ -89,7 +89,14 @@ function KnitroSolver(
 
   # pass options to KNITRO
   for (k, v) in kwargs
-    KNITRO.KN_set_param(kc, string(k), v)
+    if v isa Integer
+      KNITRO.KN_set_int_param_by_name(kc, string(k), v)
+    elseif v isa Cdouble
+      KNITRO.KN_set_double_param_by_name(kc, string(k), v)
+    else
+      @assert v isa AbstractString
+      KNITRO.KN_set_char_param_by_name(kc, string(k), v)
+    end
   end
 
   # set user-defined callback called after each iteration

test/runtests.jl

@@ -30,14 +30,20 @@ function test_qp_with_solver_and_evals()
   @test stats.iter == 1
   @test stats.status == :first_order
 
-  gx = KNITRO.KN_get_objgrad_values(solver.kc)[2]
+  gx = zeros(2)
+  KNITRO.KN_get_objgrad_values_all(solver.kc, gx)
   @test isapprox(gx, [-4.8; -4.8], rtol = 1e-6)
-  cx = KNITRO.KN_get_con_values(solver.kc)
+  cx = zeros(1)
+  KNITRO.KN_get_con_values_all(solver.kc, cx)
   @test isapprox(norm(cx), 0, atol = 1e-6)
-  Jx = KNITRO.KN_get_jacobian_values(solver.kc)
-  @test Jx[1] == [0; 0]
-  @test Jx[2] == [0; 1]
-  @test Jx[3] == [1; 1]
+  nnz = Ref{Clonglong}()
+  KN_get_jacobian_nnz(solver.kc, nnz)
+  @test nnz[] == 2
+  vars, cons, coef = zeros(Cint, 2), zeros(Cint, 2), zeros(2)
+  KNITRO.KN_get_jacobian_values(solver.kc, vars, cons, coef)
+  @test vars == Cint[0, 0]
+  @test cons == Cint[0, 1]
+  @test coef == Cdouble[1, 1]
   finalize(solver)
 end
 
@@ -66,7 +72,9 @@ end
 
 function test_with_callback()
   function callback(kc, x, lambda_, userParams)
-    if KNITRO.KN_get_number_iters(kc) > 1
+    pCint = Ref{Cint}()
+    KNITRO.KN_get_number_iters(kc, pCint)
+    if pCint[] > 1
       return KNITRO.KN_RC_USER_TERMINATION
     end
     return 0
@@ -177,10 +185,17 @@ function test_linear_constraints()
   @test isapprox(stats.solution, [-1; 1], rtol = 1e-6)
   @test isapprox(stats.objective, 0.0, atol = 1e-6)
   @test stats.status == :first_order
-  cx = KNITRO.KN_get_con_values(solver.kc)
+  cx = zeros(2)
+  KNITRO.KN_get_con_values_all(solver.kc, cx)
   @test isapprox(cx, [1, 1], rtol = 1e-6)
-  Jx = KNITRO.KN_get_jacobian_values(solver.kc)
-  @test Jx == (Int32[0, 0, 1, 1], Int32[0, 1, 0, 1], [1.0, 2.0, 3.0, 4.0])
+  nnz = Ref{Clonglong}()
+  KN_get_jacobian_nnz(solver.kc, nnz)
+  @test nnz[] == 4
+  vars, cons, coef = zeros(Cint, 4), zeros(Cint, 4), zeros(4)
+  KNITRO.KN_get_jacobian_values(solver.kc, vars, cons, coef)
+  @test vars == Int32[0, 0, 1, 1]
+  @test cons == Int32[0, 1, 0, 1]
+  @test coef == [1.0, 2.0, 3.0, 4.0]
   finalize(solver)
 end
 
@@ -200,10 +215,17 @@ function test_mixed_linear_constraints()
   @test isapprox(stats.solution, [-1; 1], rtol = 1e-6)
   @test isapprox(stats.objective, 0.0, atol = 1e-6)
   @test stats.status == :first_order
-  cx = KNITRO.KN_get_con_values(solver.kc)
+  cx = zeros(2)
+  KNITRO.KN_get_con_values_all(solver.kc, cx)
   @test isapprox(cx, [1, 1], rtol = 1e-6)
-  Jx = KNITRO.KN_get_jacobian_values(solver.kc)
-  @test Jx == (Int32[0, 0, 1, 1], Int32[0, 1, 0, 1], [3.0, 4.0, 1.0, 2.0])
+  nnz = Ref{Clonglong}()
+  KN_get_jacobian_nnz(solver.kc, nnz)
+  @test nnz[] == 4
+  vars, cons, coef = zeros(Cint, 4), zeros(Cint, 4), zeros(4)
+  KNITRO.KN_get_jacobian_values(solver.kc, vars, cons, coef)
+  @test vars == Int32[0, 0, 1, 1]
+  @test cons == Int32[0, 1, 0, 1]
+  @test coef == [3.0, 4.0, 1.0, 2.0]
   finalize(solver)
 end