Skip to content

Commit

Permalink
restore handlers from julia
Browse files Browse the repository at this point in the history
  • Loading branch information
@SobhanMP @amontoison
SobhanMP authored and amontoison committed Jul 20, 2022
1 parent d72c5aa commit dface8f
Show file tree
Hide file tree
Showing 2 changed files with 152 additions and 120 deletions.
35 changes: 27 additions & 8 deletions src/c_wrappers.jl
View file
Original file line number Diff line number Diff line change
Expand Up @@ -7,6 +7,18 @@ else
using NOMAD_jll
end

const SIGNALS = [2, 11]

function sigsegv_handler(new=C_NULL, old=zeros(UInt8, 256); signal)
@static if Sys.isunix()
@ccall sigaction(
signal::Cint,
new::Ptr{Cvoid},
old::Ptr{Cvoid})::Cint
end
return old
end

######################################################
# NOMAD C low level types #
######################################################
Expand All @@ -17,12 +29,13 @@ mutable struct C_NomadProblem
nb_outputs::Int

eval_bb::Function # callback function

handlers::Vector{Vector{UInt8}}
function C_NomadProblem(ref::Ptr{Cvoid},
nb_inputs::Int,
nb_outputs::Int,
eval_bb::Function)
p = new(ref, nb_inputs, nb_outputs, eval_bb)
eval_bb::Function,
handlers)
p = new(ref, nb_inputs, nb_outputs, eval_bb, handlers)
finalizer(free_c_nomad_problem, p)
return p
end
Expand All @@ -34,7 +47,9 @@ function eval_bb_wrapper(nb_inputs::Cint, inputs_ptr::Ptr{Float64},
count_eval_ptr::Ptr{Cint}, user_data_ptr::Ptr{Cvoid})
# extract the Julia problem from pointer
prob = unsafe_pointer_to_objref(user_data_ptr)::C_NomadProblem

nomad_handlers = map(zip(SIGNALS, prob.handlers)) do (signal, handler)
sigsegv_handler(handler; signal)
end
# get the new outputs from the black box
new_bb_outputs = unsafe_wrap(Array, outputs_ptr, Int(nb_outputs))

Expand All @@ -48,6 +63,10 @@ function eval_bb_wrapper(nb_inputs::Cint, inputs_ptr::Ptr{Float64},

# affect count eval to count_eval flag
unsafe_store!(count_eval_ptr, count_eval)

foreach(zip(SIGNALS, nomad_handlers)) do (signal, old_handler)
sigsegv_handler(old_handler, C_NULL; signal)
end

return Int32(success_flag)
end
Expand All @@ -58,7 +77,8 @@ function create_c_nomad_problem(eval_bb::Function,
lower_bound::Vector{Float64},
upper_bound::Vector{Float64},
type_inputs::String,
type_outputs::String)
type_outputs::String,
handlers)
# wrap callback function
eval_bb_cb = @cfunction(eval_bb_wrapper, Cint, (Cint, Ptr{Float64}, Cint, Ptr{Float64}, Ptr{Cint}, Ptr{Cvoid}))

Expand Down Expand Up @@ -95,13 +115,13 @@ function create_c_nomad_problem(eval_bb::Function,
# Output types
ccall((:addNomadStringParam, libnomadCInterface), Cint, (Ptr{Cvoid}, Ptr{UInt8}, Ptr{UInt8}), internal_ref, "BB_OUTPUT_TYPE", type_outputs)

return C_NomadProblem(internal_ref, nb_inputs, nb_outputs, eval_bb)
return C_NomadProblem(internal_ref, nb_inputs, nb_outputs, eval_bb, handlers)
end

end

function free_c_nomad_problem(prob::C_NomadProblem)
if prob.ref != C_NULL
if prob.ref != C_NULL
ccall((:freeNomadProblem, libnomadCInterface), Cvoid, (Ptr{Cvoid},), prob.ref)
prob.ref = C_NULL
end
Expand Down Expand Up @@ -142,7 +162,6 @@ function solve_nomad_problem(prob::C_NomadProblem, x0s::Vector{Float64}, nb_star
x_inf_sol = zeros(Float64, prob.nb_inputs)
outputs_feas_sol = zeros(Float64, prob.nb_outputs)
outputs_inf_sol = zeros(Float64, prob.nb_outputs)

statusflag = ccall((:solveNomadProblem, libnomadCInterface), Cint,
(Ptr{Cvoid}, Cint, Ptr{Float64}, # internal data, number of starting points, starting points,
Ptr{Cint}, Ptr{Float64}, Ptr{Float64}, # feasible solution flag, x_feas, outputs feas
Expand Down
237 changes: 125 additions & 112 deletions src/core.jl
View file
Original file line number Diff line number Diff line change
Expand Up @@ -586,133 +586,146 @@ result = solve(p, [5.0;5.0])
"""
function solve(p::NomadProblem, x0::Vector{Float64})
# 1- make a first check before manipulating the c library
check_problem(p)
check_options(p.options)
@assert p.nb_inputs == length(x0) "NOMAD.jl error : wrong parameters, starting point size is not consistent with bb inputs"

# verify starting point satisfies approximately the linear constraints
if p.A !== nothing
isapprox(p.A * x0, p.b, atol=p.options.linear_constraints_atol) || error("NOMAD.jl error : starting point x0 does does not satisfy the linear constraints")
handlers = map(SIGNALS) do signal
sigsegv_handler(; signal)
end
sols = try
# 1- make a first check before manipulating the c library
check_problem(p)
check_options(p.options)
@assert p.nb_inputs == length(x0) "NOMAD.jl error : wrong parameters, starting point size is not consistent with bb inputs"

# verify starting point satisfies approximately the linear constraints
if p.A !== nothing
isapprox(p.A * x0, p.b, atol=p.options.linear_constraints_atol) || error("NOMAD.jl error : starting point x0 does does not satisfy the linear constraints")
end

# 2- create c_nomad_problem
# identity converter in case there are no linear constraints
converter = identity

# TODO refactorize
c_nomad_problem = begin
if p.A === nothing # no linear constraints
input_types_wrapper = "( " * join(p.input_types, " ") * " )"
output_types_wrapper = join(p.output_types, " ")
create_c_nomad_problem(p.eval_bb ∘ converter, p.nb_inputs, p.nb_outputs,
p.lower_bound, p.upper_bound,
input_types_wrapper, output_types_wrapper)
else # linear constraints
# TODO to change with new converters
converter = p.options.linear_converter == "SVD" ? SVDConverter(p.A, p.b) : QRConverter(p.A, p.b)

nz = get_nz_dimension(converter)

# fix blackbox new dimensions
input_types_wrapper = "( " * join(["R" for i in 1:nz], " ") * " )"
output_types_wrapper = join(p.output_types, " ")

# fix new problem bounds
print("Computing variable bounds for the reduced problem ...")
lower_bound = get_lower_bound(converter, p.lower_bound, p.upper_bound)
upper_bound = get_upper_bound(converter, p.lower_bound, p.upper_bound)
println("Done")

# x0 satisfies the equality constraints and the bounds constraints
is_x0_evaluated = false

new_eval_bb = z -> begin
if !is_x0_evaluated
is_x0_evaluated = true
return p.eval_bb(x0)
# 2- create c_nomad_problem
# identity converter in case there are no linear constraints
converter = identity

# TODO refactorize
c_nomad_problem = begin
if p.A === nothing # no linear constraints
input_types_wrapper = "( " * join(p.input_types, " ") * " )"
output_types_wrapper = join(p.output_types, " ")
create_c_nomad_problem(p.eval_bb ∘ converter, p.nb_inputs, p.nb_outputs,
p.lower_bound, p.upper_bound,
input_types_wrapper, output_types_wrapper, handlers)
else # linear constraints
# TODO to change with new converters
converter = p.options.linear_converter == "SVD" ? SVDConverter(p.A, p.b) : QRConverter(p.A, p.b)

nz = get_nz_dimension(converter)

# fix blackbox new dimensions
input_types_wrapper = "( " * join(["R" for i in 1:nz], " ") * " )"
output_types_wrapper = join(p.output_types, " ")

# fix new problem bounds
print("Computing variable bounds for the reduced problem ...")
lower_bound = get_lower_bound(converter, p.lower_bound, p.upper_bound)
upper_bound = get_upper_bound(converter, p.lower_bound, p.upper_bound)
println("Done")

# x0 satisfies the equality constraints and the bounds constraints
is_x0_evaluated = false


new_eval_bb = z -> begin
if !is_x0_evaluated
is_x0_evaluated = true
return p.eval_bb(x0)
end
φ_x = convert_to_x(converter, z)
if all(p.lower_bound .<= φ_x .<= p.upper_bound)
return p.eval_bb(φ_x)
else # in this case, the function is not evaluated
return (false, false, [Inf for i in 1:p.nb_outputs])
end
end
φ_x = convert_to_x(converter, z)
if all(p.lower_bound .<= φ_x .<= p.upper_bound)
return p.eval_bb(φ_x)
else # in this case, the function is not evaluated
return (false, false, [Inf for i in 1:p.nb_outputs])
end
end

create_c_nomad_problem(new_eval_bb, nz, p.nb_outputs,
lower_bound, upper_bound,
input_types_wrapper, output_types_wrapper)
create_c_nomad_problem(new_eval_bb, nz, p.nb_outputs,
lower_bound, upper_bound,
input_types_wrapper, output_types_wrapper, handlers)
end
end
end

# 3- set options
if p.A === nothing
add_nomad_array_of_double_param!(c_nomad_problem, "GRANULARITY", p.granularity)
if any(p.min_mesh_size .> 0)
# conversion to Nomad tolerance to avoid warnings.
add_nomad_array_of_double_param!(c_nomad_problem, "MIN_MESH_SIZE", max.(1e-13, p.min_mesh_size))
end
if !isempty(p.initial_mesh_size)
add_nomad_array_of_double_param!(c_nomad_problem, "INITIAL_MESH_SIZE", p.initial_mesh_size)
# 3- set options
if p.A === nothing
add_nomad_array_of_double_param!(c_nomad_problem, "GRANULARITY", p.granularity)
if any(p.min_mesh_size .> 0)
# conversion to Nomad tolerance to avoid warnings.
add_nomad_array_of_double_param!(c_nomad_problem, "MIN_MESH_SIZE", max.(1e-13, p.min_mesh_size))
end
if !isempty(p.initial_mesh_size)
add_nomad_array_of_double_param!(c_nomad_problem, "INITIAL_MESH_SIZE", p.initial_mesh_size)
end
end
end

if p.options.max_cache_size != typemax(Int64)
add_nomad_val_param!(c_nomad_problem, "MAX_CACHE_SIZE", p.options.max_cache_size)
end
if p.options.max_cache_size != typemax(Int64)
add_nomad_val_param!(c_nomad_problem, "MAX_CACHE_SIZE", p.options.max_cache_size)
end

add_nomad_val_param!(c_nomad_problem, "DISPLAY_DEGREE", p.options.display_degree)
add_nomad_bool_param!(c_nomad_problem, "DISPLAY_ALL_EVAL", p.options.display_all_eval)
add_nomad_bool_param!(c_nomad_problem, "DISPLAY_INFEASIBLE", p.options.display_infeasible)
add_nomad_bool_param!(c_nomad_problem, "DISPLAY_UNSUCCESSFUL", p.options.display_unsuccessful)
add_nomad_val_param!(c_nomad_problem, "DISPLAY_DEGREE", p.options.display_degree)
add_nomad_bool_param!(c_nomad_problem, "DISPLAY_ALL_EVAL", p.options.display_all_eval)
add_nomad_bool_param!(c_nomad_problem, "DISPLAY_INFEASIBLE", p.options.display_infeasible)
add_nomad_bool_param!(c_nomad_problem, "DISPLAY_UNSUCCESSFUL", p.options.display_unsuccessful)

if !isempty(p.options.display_stats)
display_stats_wrapper = join(map(elt -> elt == "SOL" ? "( " * elt * " )" : elt, p.options.display_stats), " ")
add_nomad_string_param!(c_nomad_problem, "DISPLAY_STATS", display_stats_wrapper)
end
if !isempty(p.options.display_stats)
display_stats_wrapper = join(map(elt -> elt == "SOL" ? "( " * elt * " )" : elt, p.options.display_stats), " ")
add_nomad_string_param!(c_nomad_problem, "DISPLAY_STATS", display_stats_wrapper)
end

add_nomad_val_param!(c_nomad_problem, "MAX_BB_EVAL", p.options.max_bb_eval)
add_nomad_val_param!(c_nomad_problem, "MAX_SGTE_EVAL", p.options.max_sgte_eval)
add_nomad_bool_param!(c_nomad_problem, "OPPORTUNISTIC_EVAL", p.options.opportunistic_eval)
add_nomad_bool_param!(c_nomad_problem, "USE_CACHE", p.options.use_cache)
add_nomad_bool_param!(c_nomad_problem, "RANDOM_EVAL_SORT", p.options.random_eval_sort)

add_nomad_string_param!(c_nomad_problem, "LH_SEARCH", string(p.options.lh_search[1]) * " " * string(p.options.lh_search[2]))
add_nomad_bool_param!(c_nomad_problem, "QUAD_MODEL_SEARCH", p.options.quad_model_search)
add_nomad_bool_param!(c_nomad_problem, "SGTELIB_SEARCH", p.options.sgtelib_search)
add_nomad_val_param!(c_nomad_problem, "SGTELIB_MODEL_TRIALS", p.options.sgtelib_model_trials)
add_nomad_bool_param!(c_nomad_problem, "SPECULATIVE_SEARCH", p.options.speculative_search)
add_nomad_val_param!(c_nomad_problem, "SPECULATIVE_SEARCH_MAX", p.options.speculative_search_max)
add_nomad_bool_param!(c_nomad_problem, "NM_SEARCH", p.options.nm_search)
add_nomad_bool_param!(c_nomad_problem, "NM_SEARCH_STOP_ON_SUCCESS", p.options.nm_search_stop_on_success)

if p.options.max_time !== nothing
add_nomad_val_param!(c_nomad_problem, "MAX_TIME", p.options.max_time)
end
add_nomad_val_param!(c_nomad_problem, "MAX_BB_EVAL", p.options.max_bb_eval)
add_nomad_val_param!(c_nomad_problem, "MAX_SGTE_EVAL", p.options.max_sgte_eval)
add_nomad_bool_param!(c_nomad_problem, "OPPORTUNISTIC_EVAL", p.options.opportunistic_eval)
add_nomad_bool_param!(c_nomad_problem, "USE_CACHE", p.options.use_cache)
add_nomad_bool_param!(c_nomad_problem, "RANDOM_EVAL_SORT", p.options.random_eval_sort)

add_nomad_string_param!(c_nomad_problem, "LH_SEARCH", string(p.options.lh_search[1]) * " " * string(p.options.lh_search[2]))
add_nomad_bool_param!(c_nomad_problem, "QUAD_MODEL_SEARCH", p.options.quad_model_search)
add_nomad_bool_param!(c_nomad_problem, "SGTELIB_SEARCH", p.options.sgtelib_search)
add_nomad_val_param!(c_nomad_problem, "SGTELIB_MODEL_TRIALS", p.options.sgtelib_model_trials)
add_nomad_bool_param!(c_nomad_problem, "SPECULATIVE_SEARCH", p.options.speculative_search)
add_nomad_val_param!(c_nomad_problem, "SPECULATIVE_SEARCH_MAX", p.options.speculative_search_max)
add_nomad_bool_param!(c_nomad_problem, "NM_SEARCH", p.options.nm_search)
add_nomad_bool_param!(c_nomad_problem, "NM_SEARCH_STOP_ON_SUCCESS", p.options.nm_search_stop_on_success)

if p.options.max_time !== nothing
add_nomad_val_param!(c_nomad_problem, "MAX_TIME", p.options.max_time)
end

# 4- solve problem
result = begin
if p.A === nothing
solve_nomad_problem(c_nomad_problem, x0, 1)
else
z0 = convert_to_z(converter, x0)
solve_nomad_problem(c_nomad_problem, z0, 1)
# 4- solve problem
result = begin
if p.A === nothing
solve_nomad_problem(c_nomad_problem, x0, 1)
else
z0 = convert_to_z(converter, x0)
solve_nomad_problem(c_nomad_problem, z0, 1)
end
end
end

sols = begin
if p.A === nothing
(x_best_feas = result[1] ? result[2] : nothing,
bbo_best_feas = result[1] ? result[3] : nothing,
x_best_inf = result[4] ? result[5] : nothing,
bbo_best_inf = result[4] ? result[6] : nothing)
else
(x_best_feas = result[1] ? convert_to_x(converter, result[2]) : nothing,
bbo_best_feas = result[1] ? result[3] : nothing,
x_best_inf = result[4] ? convert_to_x(converter, result[5]) : nothing,
bbo_best_inf = result[4] ? result[6] : nothing)
sols = begin
if p.A === nothing
(x_best_feas = result[1] ? result[2] : nothing,
bbo_best_feas = result[1] ? result[3] : nothing,
x_best_inf = result[4] ? result[5] : nothing,
bbo_best_inf = result[4] ? result[6] : nothing)
else
(x_best_feas = result[1] ? convert_to_x(converter, result[2]) : nothing,
bbo_best_feas = result[1] ? result[3] : nothing,
x_best_inf = result[4] ? convert_to_x(converter, result[5]) : nothing,
bbo_best_inf = result[4] ? result[6] : nothing)
end
end
finalize(c_nomad_problem)
sols
finally
foreach(zip(SIGNALS, handlers)) do (signal, handler)
sigsegv_handler(handler, C_NULL; signal)
end
end

return sols
end

0 comments on commit dface8f

Please sign in to comment.