mirror of
https://github.com/Z3Prover/z3
synced 2025-04-06 17:44:08 +00:00
1837 lines
61 KiB
C++
1837 lines
61 KiB
C++
/*++
|
|
Copyright (c) 2013 Microsoft Corporation
|
|
|
|
Module Name:
|
|
|
|
opt_context.cpp
|
|
|
|
Abstract:
|
|
|
|
Facility for running optimization problem.
|
|
|
|
Author:
|
|
|
|
Anh-Dung Phan (t-anphan) 2013-10-16
|
|
|
|
Notes:
|
|
|
|
--*/
|
|
|
|
#include "util/gparams.h"
|
|
#include "ast/for_each_expr.h"
|
|
#include "ast/ast_pp.h"
|
|
#include "ast/bv_decl_plugin.h"
|
|
#include "ast/pb_decl_plugin.h"
|
|
#include "ast/ast_smt_pp.h"
|
|
#include "ast/ast_pp_util.h"
|
|
#include "ast/ast_ll_pp.h"
|
|
#include "ast/display_dimacs.h"
|
|
#include "model/model_smt2_pp.h"
|
|
#include "tactic/goal.h"
|
|
#include "tactic/tactic.h"
|
|
#include "tactic/arith/lia2card_tactic.h"
|
|
#include "tactic/core/solve_eqs_tactic.h"
|
|
#include "tactic/core/simplify_tactic.h"
|
|
#include "tactic/core/propagate_values_tactic.h"
|
|
#include "tactic/core/solve_eqs_tactic.h"
|
|
#include "tactic/core/elim_uncnstr_tactic.h"
|
|
#include "tactic/tactical.h"
|
|
#include "tactic/arith/card2bv_tactic.h"
|
|
#include "tactic/arith/eq2bv_tactic.h"
|
|
#include "tactic/bv/dt2bv_tactic.h"
|
|
#include "ast/converters/generic_model_converter.h"
|
|
#include "ackermannization/ackermannize_bv_tactic.h"
|
|
#include "sat/sat_solver/inc_sat_solver.h"
|
|
#include "sat/sat_params.hpp"
|
|
#include "opt/opt_context.h"
|
|
#include "opt/opt_solver.h"
|
|
#include "opt/opt_params.hpp"
|
|
|
|
|
|
namespace opt {
|
|
|
|
void context::scoped_state::push() {
|
|
m_asms_lim.push_back(m_asms.size());
|
|
m_hard_lim.push_back(m_hard.size());
|
|
m_values_lim.push_back(m_values.size());
|
|
m_objectives_lim.push_back(m_objectives.size());
|
|
m_objectives_term_trail_lim.push_back(m_objectives_term_trail.size());
|
|
}
|
|
|
|
void context::scoped_state::pop() {
|
|
m_hard.shrink(m_hard_lim.back());
|
|
m_asms.shrink(m_asms_lim.back());
|
|
m_values.shrink(m_values_lim.back());
|
|
unsigned k = m_objectives_term_trail_lim.back();
|
|
while (m_objectives_term_trail.size() > k) {
|
|
unsigned idx = m_objectives_term_trail.back();
|
|
m_objectives[idx].m_terms.pop_back();
|
|
m_objectives[idx].m_weights.pop_back();
|
|
m_objectives_term_trail.pop_back();
|
|
}
|
|
m_objectives_term_trail_lim.pop_back();
|
|
k = m_objectives_lim.back();
|
|
while (m_objectives.size() > k) {
|
|
objective& obj = m_objectives.back();
|
|
if (obj.m_type == O_MAXSMT) {
|
|
m_indices.erase(obj.m_id);
|
|
}
|
|
m_objectives.pop_back();
|
|
}
|
|
m_objectives_lim.pop_back();
|
|
m_hard_lim.pop_back();
|
|
m_asms_lim.pop_back();
|
|
m_values_lim.pop_back();
|
|
}
|
|
|
|
void context::scoped_state::add(expr* hard) {
|
|
m_hard.push_back(hard);
|
|
}
|
|
|
|
bool context::scoped_state::set(expr_ref_vector const & hard) {
|
|
bool eq = hard.size() == m_hard.size();
|
|
for (unsigned i = 0; eq && i < hard.size(); ++i) {
|
|
eq = hard.get(i) == m_hard.get(i);
|
|
}
|
|
m_hard.reset();
|
|
m_hard.append(hard);
|
|
return !eq;
|
|
}
|
|
|
|
unsigned context::scoped_state::add(expr* f, rational const& w, symbol const& id) {
|
|
if (!m.is_bool(f)) {
|
|
throw default_exception("Soft constraint should be Boolean");
|
|
}
|
|
if (!m_indices.contains(id)) {
|
|
m_objectives.push_back(objective(m, id));
|
|
m_indices.insert(id, m_objectives.size() - 1);
|
|
}
|
|
SASSERT(m_indices.contains(id));
|
|
unsigned idx = m_indices[id];
|
|
if (!w.is_zero()) {
|
|
m_objectives[idx].m_terms.push_back(f);
|
|
m_objectives[idx].m_weights.push_back(w);
|
|
m_objectives_term_trail.push_back(idx);
|
|
}
|
|
return idx;
|
|
}
|
|
|
|
unsigned context::scoped_state::add(app* t, bool is_max) {
|
|
app_ref tr(t, m);
|
|
if (!m_bv.is_bv(t) && !m_arith.is_int_real(t)) {
|
|
throw default_exception("Objective must be bit-vector, integer or real");
|
|
}
|
|
unsigned index = m_objectives.size();
|
|
m_objectives.push_back(objective(is_max, tr, index));
|
|
return index;
|
|
}
|
|
|
|
context::context(ast_manager& m):
|
|
opt_wrapper(m),
|
|
m_arith(m),
|
|
m_bv(m),
|
|
m_hard_constraints(m),
|
|
m_solver(nullptr),
|
|
m_pareto1(false),
|
|
m_box_index(UINT_MAX),
|
|
m_optsmt(m, *this),
|
|
m_scoped_state(m),
|
|
m_fm(alloc(generic_model_converter, m, "opt")),
|
|
m_model_fixed(),
|
|
m_objective_refs(m),
|
|
m_core(m),
|
|
m_unknown("unknown")
|
|
{
|
|
params_ref p;
|
|
p.set_bool("model", true);
|
|
p.set_bool("unsat_core", true);
|
|
p.set_bool("elim_to_real", true);
|
|
updt_params(p);
|
|
m_model_counter = 0;
|
|
}
|
|
|
|
context::~context() {
|
|
reset_maxsmts();
|
|
}
|
|
|
|
void context::reset_maxsmts() {
|
|
for (auto& kv : m_maxsmts) {
|
|
dealloc(kv.m_value);
|
|
}
|
|
m_maxsmts.reset();
|
|
}
|
|
|
|
void context::push() {
|
|
m_scoped_state.push();
|
|
}
|
|
|
|
void context::pop(unsigned n) {
|
|
n = std::min(n, m_scoped_state.num_scopes());
|
|
for (unsigned i = 0; i < n; ++i) {
|
|
m_scoped_state.pop();
|
|
}
|
|
clear_state();
|
|
reset_maxsmts();
|
|
m_optsmt.reset();
|
|
m_hard_constraints.reset();
|
|
}
|
|
|
|
void context::get_labels(svector<symbol> & r) {
|
|
r.append(m_labels);
|
|
}
|
|
|
|
void context::get_unsat_core(expr_ref_vector & r) {
|
|
r.append(m_core);
|
|
}
|
|
|
|
void context::set_hard_constraints(expr_ref_vector const& fmls) {
|
|
if (m_calling_on_model) {
|
|
for (expr* f : fmls)
|
|
add_hard_constraint(f);
|
|
return;
|
|
}
|
|
if (m_scoped_state.set(fmls))
|
|
clear_state();
|
|
}
|
|
|
|
void context::add_hard_constraint(expr* f) {
|
|
if (m_calling_on_model) {
|
|
if (!m_incremental)
|
|
throw default_exception("Set opt.incremental = true to allow adding constraints during search");
|
|
get_solver().assert_expr(f);
|
|
for (auto const& [k, v] : m_maxsmts)
|
|
v->reset_upper();
|
|
for (unsigned i = 0; i < num_objectives(); ++i) {
|
|
auto const& o = m_scoped_state.m_objectives[i];
|
|
if (o.m_type != O_MAXSMT)
|
|
m_optsmt.update_upper(o.m_index, inf_eps::infinity());
|
|
}
|
|
}
|
|
else {
|
|
m_scoped_state.add(f);
|
|
clear_state();
|
|
}
|
|
}
|
|
|
|
|
|
void context::add_hard_constraint(expr* f, expr* t) {
|
|
if (m_calling_on_model)
|
|
throw default_exception("adding hard constraints is not supported during callbacks");
|
|
m_scoped_state.m_asms.push_back(t);
|
|
m_scoped_state.add(m.mk_implies(t, f));
|
|
clear_state();
|
|
}
|
|
|
|
void context::get_hard_constraints(expr_ref_vector& hard) {
|
|
hard.append(m_scoped_state.m_hard);
|
|
}
|
|
|
|
expr_ref context::get_objective(unsigned i) {
|
|
SASSERT(i < num_objectives());
|
|
objective const& o = m_scoped_state.m_objectives[i];
|
|
expr_ref result(m), zero(m);
|
|
expr_ref_vector args(m);
|
|
switch (o.m_type) {
|
|
case O_MAXSMT:
|
|
zero = m_arith.mk_numeral(rational(0), false);
|
|
for (unsigned i = 0; i < o.m_terms.size(); ++i) {
|
|
args.push_back(m.mk_ite(o.m_terms[i], zero, m_arith.mk_numeral(o.m_weights[i], false)));
|
|
}
|
|
result = m_arith.mk_add(args.size(), args.data());
|
|
break;
|
|
case O_MAXIMIZE:
|
|
result = o.m_term;
|
|
if (m_arith.is_int_real(result)) {
|
|
result = m_arith.mk_uminus(result);
|
|
}
|
|
else if (m_bv.is_bv(result)) {
|
|
result = m_bv.mk_bv_neg(result);
|
|
}
|
|
else {
|
|
UNREACHABLE();
|
|
}
|
|
break;
|
|
case O_MINIMIZE:
|
|
result = o.m_term;
|
|
break;
|
|
}
|
|
return result;
|
|
}
|
|
|
|
|
|
unsigned context::add_soft_constraint(expr* f, rational const& w, symbol const& id) {
|
|
clear_state();
|
|
return m_scoped_state.add(f, w, id);
|
|
}
|
|
|
|
unsigned context::add_objective(app* t, bool is_max) {
|
|
clear_state();
|
|
return m_scoped_state.add(t, is_max);
|
|
}
|
|
|
|
|
|
|
|
void context::import_scoped_state() {
|
|
m_optsmt.reset();
|
|
reset_maxsmts();
|
|
m_objectives.reset();
|
|
m_hard_constraints.reset();
|
|
scoped_state& s = m_scoped_state;
|
|
for (unsigned i = 0; i < s.m_objectives.size(); ++i) {
|
|
objective& obj = s.m_objectives[i];
|
|
m_objectives.push_back(obj);
|
|
if (obj.m_type == O_MAXSMT) {
|
|
add_maxsmt(obj.m_id, i);
|
|
}
|
|
}
|
|
m_hard_constraints.append(s.m_hard);
|
|
}
|
|
|
|
lbool context::optimize(expr_ref_vector const& _asms) {
|
|
scoped_time _st(*this);
|
|
if (m_pareto) {
|
|
return execute_pareto();
|
|
}
|
|
if (m_box_index != UINT_MAX) {
|
|
return execute_box();
|
|
}
|
|
clear_state();
|
|
init_solver();
|
|
import_scoped_state();
|
|
expr_ref_vector asms(_asms);
|
|
asms.append(m_scoped_state.m_asms);
|
|
normalize(asms);
|
|
if (m_hard_constraints.size() == 1 && m.is_false(m_hard_constraints.get(0))) {
|
|
return l_false;
|
|
}
|
|
internalize();
|
|
update_solver();
|
|
if (contains_quantifiers()) {
|
|
warning_msg("optimization with quantified constraints is not supported");
|
|
}
|
|
solver& s = get_solver();
|
|
s.assert_expr(m_hard_constraints);
|
|
if (m_model_converter)
|
|
m_model_converter->convert_initialize_value(m_scoped_state.m_values);
|
|
for (auto & [var, value] : m_scoped_state.m_values)
|
|
s.user_propagate_initialize_value(var, value);
|
|
|
|
opt_params optp(m_params);
|
|
symbol pri = optp.priority();
|
|
|
|
IF_VERBOSE(1, verbose_stream() << "(optimize:check-sat)\n");
|
|
|
|
lbool is_sat = s.check_sat(asms.size(), asms.data());
|
|
|
|
TRACE("opt", s.display(tout << "initial search result: " << is_sat << "\n"););
|
|
if (is_sat != l_false) {
|
|
s.get_model(m_model);
|
|
s.get_labels(m_labels);
|
|
model_updated(m_model.get());
|
|
if (!m_model) {
|
|
is_sat = l_undef;
|
|
}
|
|
}
|
|
if (is_sat != l_true) {
|
|
TRACE("opt", tout << m_hard_constraints << " " << asms << "\n";);
|
|
if (!asms.empty()) {
|
|
s.get_unsat_core(m_core);
|
|
}
|
|
return is_sat;
|
|
}
|
|
s.assert_expr(asms);
|
|
IF_VERBOSE(1, verbose_stream() << "(optimize:sat)\n");
|
|
TRACE("opt", model_smt2_pp(tout, m, *m_model, 0););
|
|
m_optsmt.setup(*m_opt_solver.get());
|
|
update_lower();
|
|
|
|
switch (m_objectives.size()) {
|
|
case 0:
|
|
break;
|
|
case 1:
|
|
if (m_pareto1) {
|
|
is_sat = l_false;
|
|
m_pareto1 = false;
|
|
}
|
|
else {
|
|
m_pareto1 = (pri == symbol("pareto"));
|
|
is_sat = execute(m_objectives[0], true, false);
|
|
}
|
|
break;
|
|
default: {
|
|
opt_params optp(m_params);
|
|
symbol pri = optp.priority();
|
|
if (pri == symbol("pareto")) {
|
|
is_sat = execute_pareto();
|
|
}
|
|
else if (pri == symbol("box")) {
|
|
is_sat = execute_box();
|
|
}
|
|
else {
|
|
is_sat = execute_lex();
|
|
}
|
|
}
|
|
}
|
|
if (is_sat == l_true) validate_model();
|
|
return adjust_unknown(is_sat);
|
|
}
|
|
|
|
lbool context::adjust_unknown(lbool r) {
|
|
if (r == l_true && m_opt_solver.get() && m_opt_solver->was_unknown()) {
|
|
r = l_undef;
|
|
}
|
|
return r;
|
|
}
|
|
|
|
void context::get_base_model(model_ref& mdl) {
|
|
mdl = m_model;
|
|
}
|
|
|
|
void context::fix_model(model_ref& mdl) {
|
|
if (mdl && !m_model_fixed.contains(mdl.get())) {
|
|
TRACE("opt", m_fm->display(tout << "fix-model\n");
|
|
tout << *mdl << "\n";
|
|
if (m_model_converter) m_model_converter->display(tout););
|
|
(*m_fm)(mdl);
|
|
apply(m_model_converter, mdl);
|
|
m_model_fixed.push_back(mdl.get());
|
|
}
|
|
}
|
|
|
|
void context::set_model(model_ref& m) {
|
|
m_model = m;
|
|
opt_params optp(m_params);
|
|
symbol prefix = optp.solution_prefix();
|
|
bool model2console = optp.dump_models();
|
|
bool model2file = prefix != symbol::null && prefix != symbol("");
|
|
|
|
if ((model2console || model2file) && m) {
|
|
model_ref md = m->copy();
|
|
fix_model(md);
|
|
if (model2file) {
|
|
std::ostringstream buffer;
|
|
buffer << prefix << (m_model_counter++) << ".smt2";
|
|
std::ofstream out(buffer.str());
|
|
if (out) {
|
|
out << *md;
|
|
out.close();
|
|
}
|
|
}
|
|
if (model2console)
|
|
std::cout << *md;
|
|
}
|
|
if (m_on_model_eh && m) {
|
|
model_ref md = m->copy();
|
|
if (!m_model_fixed.contains(md.get()))
|
|
fix_model(md);
|
|
flet<bool> _calling(m_calling_on_model, true);
|
|
m_on_model_eh(m_on_model_ctx, md);
|
|
m_model_fixed.pop_back();
|
|
}
|
|
}
|
|
|
|
|
|
void context::get_model_core(model_ref& mdl) {
|
|
mdl = m_model;
|
|
CTRACE("opt", mdl, tout << *mdl;);
|
|
fix_model(mdl);
|
|
if (mdl) mdl->set_model_completion(true);
|
|
CTRACE("opt", mdl, tout << *mdl;);
|
|
}
|
|
|
|
void context::get_box_model(model_ref& mdl, unsigned index) {
|
|
if (index >= m_box_models.size()) {
|
|
throw default_exception("index into models is out of bounds");
|
|
}
|
|
mdl = m_box_models[index];
|
|
fix_model(mdl);
|
|
}
|
|
|
|
bool context::contains_quantifiers() const {
|
|
for (expr* f : m_hard_constraints) {
|
|
if (has_quantifiers(f)) return true;
|
|
}
|
|
return false;
|
|
}
|
|
|
|
|
|
lbool context::execute_min_max(unsigned index, bool committed, bool scoped, bool is_max) {
|
|
if (scoped) get_solver().push();
|
|
lbool result = m_optsmt.lex(index, is_max);
|
|
if (result == l_true) { m_optsmt.get_model(m_model, m_labels); SASSERT(m_model); }
|
|
if (scoped) get_solver().pop(1);
|
|
if (result == l_true && committed) m_optsmt.commit_assignment(index);
|
|
if (result == l_true && m_optsmt.is_unbounded(index, is_max) && contains_quantifiers()) {
|
|
throw default_exception("unbounded objectives on quantified constraints is not supported");
|
|
}
|
|
return result;
|
|
}
|
|
|
|
lbool context::execute_maxsat(symbol const& id, bool committed, bool scoped) {
|
|
model_ref tmp;
|
|
maxsmt& ms = *m_maxsmts.find(id);
|
|
if (scoped) get_solver().push();
|
|
lbool result = ms(committed);
|
|
if (result != l_false && (ms.get_model(tmp, m_labels), tmp.get())) {
|
|
ms.get_model(m_model, m_labels);
|
|
}
|
|
if (scoped) get_solver().pop(1);
|
|
if (result == l_true && committed) ms.commit_assignment();
|
|
DEBUG_CODE(if (result == l_true) validate_maxsat(id););
|
|
return result;
|
|
}
|
|
|
|
lbool context::execute(objective const& obj, bool committed, bool scoped) {
|
|
switch(obj.m_type) {
|
|
case O_MAXIMIZE: return execute_min_max(obj.m_index, committed, scoped, true);
|
|
case O_MINIMIZE: return execute_min_max(obj.m_index, committed, scoped, false);
|
|
case O_MAXSMT: return execute_maxsat(obj.m_id, committed, scoped);
|
|
default: UNREACHABLE(); return l_undef;
|
|
}
|
|
}
|
|
|
|
/**
|
|
\brief there is no need to use push/pop when all objectives are maxsat and engine
|
|
is maxres.
|
|
*/
|
|
bool context::scoped_lex() {
|
|
if (m_maxsat_engine == symbol("maxres")) {
|
|
for (auto const& o : m_objectives) {
|
|
if (o.m_type != O_MAXSMT) return true;
|
|
}
|
|
return false;
|
|
}
|
|
return true;
|
|
}
|
|
|
|
lbool context::execute_lex() {
|
|
lbool r = l_true;
|
|
bool sc = scoped_lex();
|
|
IF_VERBOSE(1, verbose_stream() << "(opt :lex)\n";);
|
|
unsigned sz = m_objectives.size();
|
|
for (unsigned i = 0; r == l_true && i < sz; ++i) {
|
|
objective const& o = m_objectives[i];
|
|
bool is_last = i + 1 == sz;
|
|
r = execute(o, i + 1 < sz, sc && !is_last);
|
|
if (r == l_true && o.m_type == O_MINIMIZE && !get_lower_as_num(i).is_finite()) {
|
|
return r;
|
|
}
|
|
if (r == l_true && o.m_type == O_MAXIMIZE && !get_upper_as_num(i).is_finite()) {
|
|
return r;
|
|
}
|
|
if (r == l_true && i + 1 < sz) {
|
|
update_lower();
|
|
}
|
|
}
|
|
DEBUG_CODE(if (r == l_true) validate_lex(););
|
|
return r;
|
|
}
|
|
|
|
lbool context::execute_box() {
|
|
if (m_box_index < m_box_models.size()) {
|
|
m_model = m_box_models[m_box_index];
|
|
CTRACE("opt", m_model, tout << *m_model << "\n";);
|
|
++m_box_index;
|
|
return l_true;
|
|
}
|
|
if (m_box_index < m_objectives.size()) {
|
|
m_model = nullptr;
|
|
++m_box_index;
|
|
return l_undef;
|
|
}
|
|
if (m_box_index != UINT_MAX && m_box_index >= m_objectives.size()) {
|
|
m_box_index = UINT_MAX;
|
|
return l_false;
|
|
}
|
|
m_box_index = 1;
|
|
m_box_models.reset();
|
|
lbool r = m_optsmt.box();
|
|
for (unsigned i = 0, j = 0; r == l_true && i < m_objectives.size(); ++i) {
|
|
objective const& obj = m_objectives[i];
|
|
if (obj.m_type == O_MAXSMT) {
|
|
solver::scoped_push _sp(get_solver());
|
|
r = execute(obj, false, false);
|
|
m_box_models.push_back(m_model.get());
|
|
}
|
|
else {
|
|
model* mdl = m_optsmt.get_model(j);
|
|
if (!mdl) mdl = m_model.get();
|
|
m_box_models.push_back(mdl);
|
|
++j;
|
|
}
|
|
}
|
|
if (r == l_true && !m_box_models.empty()) {
|
|
m_model = m_box_models[0];
|
|
CTRACE("opt", m_model, tout << *m_model << "\n";);
|
|
}
|
|
return r;
|
|
}
|
|
|
|
expr_ref context::mk_le(unsigned i, model_ref& mdl) {
|
|
objective const& obj = m_objectives[i];
|
|
return mk_cmp(false, mdl, obj);
|
|
}
|
|
|
|
expr_ref context::mk_ge(unsigned i, model_ref& mdl) {
|
|
objective const& obj = m_objectives[i];
|
|
return mk_cmp(true, mdl, obj);
|
|
}
|
|
|
|
expr_ref context::mk_gt(unsigned i, model_ref& mdl) {
|
|
expr_ref result = mk_le(i, mdl);
|
|
result = mk_not(m, result);
|
|
return result;
|
|
}
|
|
|
|
expr_ref context::mk_cmp(bool is_ge, model_ref& mdl, objective const& obj) {
|
|
rational k(0);
|
|
expr_ref val(m), result(m);
|
|
switch (obj.m_type) {
|
|
case O_MINIMIZE:
|
|
is_ge = !is_ge;
|
|
case O_MAXIMIZE:
|
|
val = (*mdl)(obj.m_term);
|
|
if (is_numeral(val, k)) {
|
|
if (is_ge) {
|
|
result = mk_ge(obj.m_term, val);
|
|
}
|
|
else {
|
|
result = mk_ge(val, obj.m_term);
|
|
}
|
|
}
|
|
else {
|
|
result = m.mk_true();
|
|
}
|
|
break;
|
|
case O_MAXSMT: {
|
|
pb_util pb(m);
|
|
unsigned sz = obj.m_terms.size();
|
|
ptr_vector<expr> terms;
|
|
vector<rational> coeffs;
|
|
for (unsigned i = 0; i < sz; ++i) {
|
|
terms.push_back(obj.m_terms[i]);
|
|
coeffs.push_back(obj.m_weights[i]);
|
|
if (mdl->is_true(obj.m_terms[i])) {
|
|
k += obj.m_weights[i];
|
|
}
|
|
else {
|
|
TRACE("opt", tout << (*mdl)(obj.m_terms[i]) << "\n";);
|
|
}
|
|
}
|
|
if (is_ge) {
|
|
result = pb.mk_ge(sz, coeffs.data(), terms.data(), k);
|
|
}
|
|
else {
|
|
result = pb.mk_le(sz, coeffs.data(), terms.data(), k);
|
|
}
|
|
break;
|
|
}
|
|
}
|
|
TRACE("opt",
|
|
tout << (is_ge?">= ":"<= ") << k << "\n";
|
|
display_objective(tout, obj);
|
|
tout << "\n";
|
|
tout << result << "\n";);
|
|
return result;
|
|
}
|
|
|
|
expr_ref context::mk_ge(expr* t, expr* s) {
|
|
expr_ref result(m);
|
|
if (m_bv.is_bv(t)) {
|
|
result = m_bv.mk_ule(s, t);
|
|
}
|
|
else {
|
|
result = m_arith.mk_ge(t, s);
|
|
}
|
|
return result;
|
|
}
|
|
|
|
void context::yield() {
|
|
SASSERT (m_pareto);
|
|
m_pareto->get_model(m_model, m_labels);
|
|
update_bound(true);
|
|
update_bound(false);
|
|
TRACE("opt", model_smt2_pp(tout, m, *m_model.get(), 0););
|
|
}
|
|
|
|
lbool context::execute_pareto() {
|
|
if (!m_pareto) {
|
|
set_pareto(alloc(gia_pareto, m, *this, m_solver.get(), m_params));
|
|
}
|
|
lbool is_sat = (*(m_pareto.get()))();
|
|
if (is_sat != l_true) {
|
|
set_pareto(nullptr);
|
|
}
|
|
if (is_sat == l_true) {
|
|
yield();
|
|
}
|
|
return is_sat;
|
|
}
|
|
|
|
|
|
std::string context::reason_unknown() const {
|
|
if (!m.inc()) {
|
|
return Z3_CANCELED_MSG;
|
|
}
|
|
if (m_solver.get()) {
|
|
return m_solver->reason_unknown();
|
|
}
|
|
return m_unknown;
|
|
}
|
|
|
|
void context::display_bounds(std::ostream& out, bounds_t const& b) const {
|
|
for (unsigned i = 0; i < m_objectives.size(); ++i) {
|
|
objective const& obj = m_objectives[i];
|
|
display_objective(out, obj);
|
|
if (obj.m_type == O_MAXIMIZE) {
|
|
out << " |-> [" << b[i].first << ":" << b[i].second << "]\n";
|
|
}
|
|
else {
|
|
out << " |-> [" << -b[i].second << ":" << -b[i].first << "]\n";
|
|
}
|
|
}
|
|
}
|
|
|
|
solver& context::get_solver() {
|
|
return *m_solver.get();
|
|
}
|
|
|
|
void context::init_solver() {
|
|
setup_arith_solver();
|
|
m_sat_solver = nullptr;
|
|
m_opt_solver = alloc(opt_solver, m, m_params, *m_fm);
|
|
m_opt_solver->set_logic(m_logic);
|
|
m_solver = m_opt_solver.get();
|
|
m_opt_solver->ensure_pb();
|
|
}
|
|
|
|
void context::setup_arith_solver() {
|
|
opt_params p(m_params);
|
|
if (p.optsmt_engine() == symbol("symba") ||
|
|
p.optsmt_engine() == symbol("farkas")) {
|
|
auto str = std::to_string((unsigned)(arith_solver_id::AS_OPTINF));
|
|
gparams::set("smt.arith.solver", str.c_str());
|
|
}
|
|
}
|
|
|
|
void context::initialize_value(expr* var, expr* value) {
|
|
m_scoped_state.m_values.push_back({expr_ref(var, m), expr_ref(value, m)});
|
|
}
|
|
|
|
|
|
/**
|
|
* Set the solver to the SAT core.
|
|
* It requres:
|
|
* - either EUF is enabled or the query is finite domain.
|
|
* - it is a MaxSAT query because linear optimiation is not exposed over the EUF core.
|
|
* - opt_solver relies on features from the legacy core.
|
|
* - the MaxSAT engine does not depend on old core features (branch and bound solver for MaxSAT)
|
|
* - proofs are not enabled
|
|
* Relaxation of these filters are possible by adding functionality to the new core.
|
|
* - Pareto optimizaiton might already be possible with EUF = true
|
|
* - optsmt needs to be disetangled from the legacy core
|
|
*/
|
|
void context::update_solver() {
|
|
sat_params p(m_params);
|
|
if (!p.euf() && (!m_enable_sat || !probe_fd()))
|
|
return;
|
|
|
|
if (!is_maxsat_query())
|
|
return;
|
|
|
|
if (m_maxsat_engine != symbol("maxres") &&
|
|
m_maxsat_engine != symbol("rc2") &&
|
|
m_maxsat_engine != symbol("rc2tot") &&
|
|
m_maxsat_engine != symbol("rc2bin") &&
|
|
m_maxsat_engine != symbol("maxres-bin") &&
|
|
m_maxsat_engine != symbol("maxres-bin-delay") &&
|
|
m_maxsat_engine != symbol("pd-maxres") &&
|
|
m_maxsat_engine != symbol("bcd2") &&
|
|
m_maxsat_engine != symbol("sls")) {
|
|
return;
|
|
}
|
|
|
|
if (opt_params(m_params).priority() == symbol("pareto"))
|
|
return;
|
|
|
|
if (m.proofs_enabled())
|
|
return;
|
|
|
|
m_params.set_bool("minimize_core_partial", true);
|
|
m_params.set_bool("minimize_core", true);
|
|
m_sat_solver = mk_inc_sat_solver(m, m_params);
|
|
expr_ref_vector fmls(m);
|
|
get_solver().get_assertions(fmls);
|
|
m_sat_solver->assert_expr(fmls);
|
|
m_solver = m_sat_solver.get();
|
|
}
|
|
|
|
void context::enable_sls(bool force) {
|
|
if ((force || m_enable_sls) && m_sat_solver.get()) {
|
|
m_params.set_bool("optimize_model", true);
|
|
m_sat_solver->updt_params(m_params);
|
|
}
|
|
}
|
|
|
|
struct context::is_fd {
|
|
struct found_fd {};
|
|
ast_manager& m;
|
|
pb_util pb;
|
|
bv_util bv;
|
|
is_fd(ast_manager& m): m(m), pb(m), bv(m) {}
|
|
void operator()(var *) { throw found_fd(); }
|
|
void operator()(quantifier *) { throw found_fd(); }
|
|
void operator()(app *n) {
|
|
family_id fid = n->get_family_id();
|
|
if (fid != m.get_basic_family_id() &&
|
|
fid != pb.get_family_id() &&
|
|
fid != bv.get_family_id() &&
|
|
(!is_uninterp_const(n) || (!m.is_bool(n) && !bv.is_bv(n)))) {
|
|
throw found_fd();
|
|
}
|
|
}
|
|
};
|
|
|
|
bool context::is_maxsat_query() {
|
|
for (objective& obj : m_objectives)
|
|
if (obj.m_type != O_MAXSMT)
|
|
return false;
|
|
return true;
|
|
}
|
|
|
|
bool context::probe_fd() {
|
|
expr_fast_mark1 visited;
|
|
is_fd proc(m);
|
|
if (!is_maxsat_query())
|
|
return false;
|
|
try {
|
|
for (objective& obj : m_objectives) {
|
|
maxsmt& ms = *m_maxsmts.find(obj.m_id);
|
|
for (unsigned j = 0; j < ms.size(); ++j)
|
|
quick_for_each_expr(proc, visited, ms[j]);
|
|
}
|
|
unsigned sz = get_solver().get_num_assertions();
|
|
for (unsigned i = 0; i < sz; i++)
|
|
quick_for_each_expr(proc, visited, get_solver().get_assertion(i));
|
|
for (expr* f : m_hard_constraints)
|
|
quick_for_each_expr(proc, visited, f);
|
|
}
|
|
catch (const is_fd::found_fd &) {
|
|
return false;
|
|
}
|
|
return true;
|
|
}
|
|
|
|
struct context::is_propositional_fn {
|
|
struct found {};
|
|
ast_manager& m;
|
|
is_propositional_fn(ast_manager& m): m(m) {}
|
|
void operator()(var *) { throw found(); }
|
|
void operator()(quantifier *) { throw found(); }
|
|
void operator()(app *n) {
|
|
family_id fid = n->get_family_id();
|
|
if (fid != m.get_basic_family_id() &&
|
|
!is_uninterp_const(n)) {
|
|
throw found();
|
|
}
|
|
}
|
|
};
|
|
|
|
bool context::is_propositional(expr* p) {
|
|
expr* np;
|
|
if (is_uninterp_const(p) || (m.is_not(p, np) && is_uninterp_const(np))) {
|
|
return true;
|
|
}
|
|
is_propositional_fn proc(m);
|
|
expr_fast_mark1 visited;
|
|
try {
|
|
quick_for_each_expr(proc, visited, p);
|
|
}
|
|
catch (const is_propositional_fn::found &) {
|
|
return false;
|
|
}
|
|
return true;
|
|
}
|
|
|
|
|
|
void context::add_maxsmt(symbol const& id, unsigned index) {
|
|
maxsmt* ms = alloc(maxsmt, *this, index);
|
|
ms->updt_params(m_params);
|
|
m_maxsmts.insert(id, ms);
|
|
}
|
|
|
|
bool context::is_numeral(expr* e, rational & n) const {
|
|
unsigned sz;
|
|
return m_arith.is_numeral(e, n) || m_bv.is_numeral(e, n, sz);
|
|
}
|
|
|
|
void context::normalize(expr_ref_vector const& asms) {
|
|
expr_ref_vector fmls(m);
|
|
m_model_converter = nullptr;
|
|
to_fmls(fmls);
|
|
simplify_fmls(fmls, asms);
|
|
from_fmls(fmls);
|
|
}
|
|
|
|
void context::simplify_fmls(expr_ref_vector& fmls, expr_ref_vector const& asms) {
|
|
if (m_is_clausal) {
|
|
return;
|
|
}
|
|
|
|
goal_ref g(alloc(goal, m, true, !asms.empty()));
|
|
for (expr* fml : fmls)
|
|
g->assert_expr(fml);
|
|
for (expr * a : asms)
|
|
g->assert_expr(a, a);
|
|
tactic_ref tac0 =
|
|
and_then(mk_simplify_tactic(m, m_params),
|
|
mk_propagate_values_tactic(m),
|
|
m_incremental ? mk_skip_tactic() : mk_solve_eqs_tactic(m),
|
|
mk_simplify_tactic(m));
|
|
opt_params optp(m_params);
|
|
tactic_ref tac1, tac2, tac3, tac4;
|
|
bool has_dep = false;
|
|
for (unsigned i = 0; !has_dep && i < g->size(); ++i) {
|
|
ptr_vector<expr> deps;
|
|
expr_dependency_ref core(g->dep(i), m);
|
|
m.linearize(core, deps);
|
|
has_dep |= !deps.empty();
|
|
}
|
|
if (optp.elim_01() && m_logic.is_null() && !has_dep && !m_incremental) {
|
|
tac1 = mk_dt2bv_tactic(m);
|
|
tac2 = mk_lia2card_tactic(m);
|
|
tac3 = mk_eq2bv_tactic(m);
|
|
params_ref lia_p;
|
|
lia_p.set_bool("compile_equality", optp.pb_compile_equality());
|
|
tac2->updt_params(lia_p);
|
|
set_simplify(and_then(tac0.get(), tac1.get(), tac2.get(), tac3.get(), mk_simplify_tactic(m)));
|
|
}
|
|
else {
|
|
set_simplify(tac0.get());
|
|
}
|
|
goal_ref_buffer result;
|
|
TRACE("opt", g->display(tout););
|
|
(*m_simplify)(g, result);
|
|
SASSERT(result.size() == 1);
|
|
goal* r = result[0];
|
|
m_model_converter = r->mc();
|
|
CTRACE("opt", r->mc(), r->mc()->display(tout););
|
|
fmls.reset();
|
|
expr_ref tmp(m);
|
|
for (unsigned i = 0; i < r->size(); ++i) {
|
|
if (asms.empty()) {
|
|
fmls.push_back(r->form(i));
|
|
continue;
|
|
}
|
|
|
|
ptr_vector<expr> deps;
|
|
expr_dependency_ref core(r->dep(i), m);
|
|
m.linearize(core, deps);
|
|
if (deps.empty())
|
|
fmls.push_back(r->form(i));
|
|
else if (deps.size() == 1 && deps[0] == r->form(i))
|
|
continue;
|
|
else if (is_objective(r->form(i)))
|
|
fmls.push_back(r->form(i));
|
|
else
|
|
fmls.push_back(m.mk_implies(mk_and(m, deps.size(), deps.data()), r->form(i)));
|
|
}
|
|
if (r->inconsistent()) {
|
|
ptr_vector<expr> core_elems;
|
|
expr_dependency_ref core(r->dep(0), m);
|
|
m.linearize(core, core_elems);
|
|
m_core.append(core_elems.size(), core_elems.data());
|
|
}
|
|
}
|
|
|
|
bool context::is_objective(expr* fml) {
|
|
return is_app(fml) && m_objective_fns.contains(to_app(fml)->get_decl());
|
|
}
|
|
|
|
bool context::is_maximize(expr* fml, app_ref& term, expr_ref& orig_term, unsigned& index) {
|
|
if (is_app(fml) && m_objective_fns.find(to_app(fml)->get_decl(), index) &&
|
|
m_objectives[index].m_type == O_MAXIMIZE) {
|
|
term = to_app(to_app(fml)->get_arg(0));
|
|
orig_term = m_objective_orig.find(to_app(fml)->get_decl());
|
|
return true;
|
|
}
|
|
return false;
|
|
}
|
|
|
|
bool context::is_minimize(expr* fml, app_ref& term, expr_ref& orig_term, unsigned& index) {
|
|
if (is_app(fml) && m_objective_fns.find(to_app(fml)->get_decl(), index) &&
|
|
m_objectives[index].m_type == O_MINIMIZE) {
|
|
term = to_app(to_app(fml)->get_arg(0));
|
|
orig_term = m_objective_orig.find(to_app(fml)->get_decl());
|
|
return true;
|
|
}
|
|
return false;
|
|
}
|
|
|
|
bool context::is_maxsat(expr* fml, expr_ref_vector& terms,
|
|
vector<rational>& weights, rational& offset,
|
|
bool& neg, symbol& id, expr_ref& orig_term, unsigned& index) {
|
|
if (!is_app(fml))
|
|
return false;
|
|
neg = false;
|
|
orig_term = nullptr;
|
|
index = 0;
|
|
app* a = to_app(fml);
|
|
|
|
if (m_objective_fns.find(a->get_decl(), index) && m_objectives[index].m_type == O_MAXSMT) {
|
|
for (unsigned i = 0; i < a->get_num_args(); ++i) {
|
|
expr_ref arg(a->get_arg(i), m);
|
|
rational weight = m_objectives[index].m_weights[i];
|
|
if (weight.is_neg()) {
|
|
weight.neg();
|
|
arg = mk_not(m, arg);
|
|
offset -= weight;
|
|
}
|
|
if (m.is_true(arg)) {
|
|
IF_VERBOSE(5, verbose_stream() << weight << ": " << mk_pp(m_objectives[index].m_terms[i].get(), m) << " |-> true\n";);
|
|
}
|
|
else if (weight.is_zero()) {
|
|
// skip
|
|
}
|
|
else if (m.is_false(arg)) {
|
|
IF_VERBOSE(5, verbose_stream() << weight << ": " << mk_pp(m_objectives[index].m_terms[i].get(), m) << " |-> false\n";);
|
|
offset += weight;
|
|
}
|
|
else {
|
|
terms.push_back(arg);
|
|
weights.push_back(weight);
|
|
}
|
|
}
|
|
id = m_objectives[index].m_id;
|
|
return true;
|
|
}
|
|
app_ref term(m);
|
|
offset = rational::zero();
|
|
bool is_max = is_maximize(fml, term, orig_term, index);
|
|
bool is_min = !is_max && is_minimize(fml, term, orig_term, index);
|
|
if (is_min && get_pb_sum(term, terms, weights, offset)) {
|
|
TRACE("opt", tout << "try to convert minimization\n" << mk_pp(term, m) << "\n";);
|
|
// minimize 2*x + 3*y
|
|
// <=>
|
|
// (assert-soft (not x) 2)
|
|
// (assert-soft (not y) 3)
|
|
//
|
|
for (unsigned i = 0; i < weights.size(); ++i) {
|
|
if (weights[i].is_neg()) {
|
|
offset += weights[i];
|
|
weights[i].neg();
|
|
}
|
|
else {
|
|
terms[i] = mk_not(m, terms[i].get());
|
|
}
|
|
}
|
|
TRACE("opt",
|
|
tout << "Convert minimization " << orig_term << "\n";
|
|
tout << "to maxsat: " << term << "\n";
|
|
for (unsigned i = 0; i < weights.size(); ++i) {
|
|
tout << mk_pp(terms.get(i), m) << ": " << weights[i] << "\n";
|
|
}
|
|
tout << "offset: " << offset << "\n";
|
|
);
|
|
std::ostringstream out;
|
|
out << mk_bounded_pp(orig_term, m, 2) << ':' << index;
|
|
id = symbol(out.str());
|
|
return true;
|
|
}
|
|
if (is_max && get_pb_sum(term, terms, weights, offset)) {
|
|
TRACE("opt", tout << "try to convert maximization " << mk_pp(term, m) << "\n";);
|
|
// maximize 2*x + 3*y - z
|
|
// <=>
|
|
// (assert-soft x 2)
|
|
// (assert-soft y 3)
|
|
// (assert-soft (not z) 1)
|
|
// offset := 6
|
|
// maximize = offset - penalty
|
|
//
|
|
for (unsigned i = 0; i < weights.size(); ++i) {
|
|
if (weights[i].is_neg()) {
|
|
weights[i].neg();
|
|
terms[i] = mk_not(m, terms[i].get());
|
|
}
|
|
offset += weights[i];
|
|
}
|
|
neg = true;
|
|
std::ostringstream out;
|
|
out << mk_bounded_pp(orig_term, m) << ':' << index;
|
|
id = symbol(out.str());
|
|
return true;
|
|
}
|
|
if ((is_max || is_min) && m_bv.is_bv(term)) {
|
|
offset.reset();
|
|
unsigned bv_size = m_bv.get_bv_size(term);
|
|
expr_ref val(m);
|
|
val = m_bv.mk_numeral(is_max, 1);
|
|
for (unsigned i = 0; i < bv_size; ++i) {
|
|
rational w = power(rational(2),i);
|
|
weights.push_back(w);
|
|
terms.push_back(m.mk_eq(val, m_bv.mk_extract(i, i, term)));
|
|
if (is_max) {
|
|
offset += w;
|
|
}
|
|
}
|
|
neg = is_max;
|
|
std::ostringstream out;
|
|
out << mk_bounded_pp(orig_term, m, 2) << ':' << index;
|
|
id = symbol(out.str());
|
|
return true;
|
|
}
|
|
return false;
|
|
}
|
|
|
|
expr* context::mk_objective_fn(unsigned index, objective_t ty, unsigned sz, expr*const* args) {
|
|
ptr_vector<sort> domain;
|
|
for (unsigned i = 0; i < sz; ++i) {
|
|
domain.push_back(args[i]->get_sort());
|
|
}
|
|
char const* name = "";
|
|
switch(ty) {
|
|
case O_MAXIMIZE: name = "maximize"; break;
|
|
case O_MINIMIZE: name = "minimize"; break;
|
|
case O_MAXSMT: name = "maxsat"; break;
|
|
default: break;
|
|
}
|
|
func_decl* f = m.mk_fresh_func_decl(name,"", domain.size(), domain.data(), m.mk_bool_sort());
|
|
m_objective_fns.insert(f, index);
|
|
m_objective_refs.push_back(f);
|
|
m_objective_orig.insert(f, sz > 0 ? args[0] : nullptr);
|
|
return m.mk_app(f, sz, args);
|
|
}
|
|
|
|
expr* context::mk_maximize(unsigned index, app* t) {
|
|
expr* t_ = t;
|
|
return mk_objective_fn(index, O_MAXIMIZE, 1, &t_);
|
|
}
|
|
|
|
expr* context::mk_minimize(unsigned index, app* t) {
|
|
expr* t_ = t;
|
|
return mk_objective_fn(index, O_MINIMIZE, 1, &t_);
|
|
}
|
|
|
|
expr* context::mk_maxsat(unsigned index, unsigned num_fmls, expr* const* fmls) {
|
|
return mk_objective_fn(index, O_MAXSMT, num_fmls, fmls);
|
|
}
|
|
|
|
void context::from_fmls(expr_ref_vector const& fmls) {
|
|
TRACE("opt", tout << fmls << "\n";);
|
|
m_hard_constraints.reset();
|
|
for (expr * fml : fmls) {
|
|
app_ref tr(m);
|
|
expr_ref orig_term(m);
|
|
expr_ref_vector terms(m);
|
|
vector<rational> weights;
|
|
rational offset(0);
|
|
unsigned index = 0;
|
|
symbol id;
|
|
bool neg;
|
|
if (is_maxsat(fml, terms, weights, offset, neg, id, orig_term, index)) {
|
|
objective& obj = m_objectives[index];
|
|
|
|
if (obj.m_type != O_MAXSMT) {
|
|
// change from maximize/minimize.
|
|
obj.m_id = id;
|
|
obj.m_type = O_MAXSMT;
|
|
SASSERT(!m_maxsmts.contains(id));
|
|
add_maxsmt(id, index);
|
|
}
|
|
mk_atomic(terms);
|
|
SASSERT(obj.m_id == id);
|
|
obj.m_term = orig_term?to_app(orig_term):nullptr;
|
|
obj.m_terms.reset();
|
|
obj.m_terms.append(terms);
|
|
obj.m_weights.reset();
|
|
obj.m_weights.append(weights);
|
|
obj.m_adjust_value.set_offset(offset);
|
|
obj.m_adjust_value.set_negate(neg);
|
|
TRACE("opt", tout << "maxsat: " << neg << " " << id << " offset: " << offset << "\n";
|
|
tout << terms << "\n";);
|
|
}
|
|
else if (is_maximize(fml, tr, orig_term, index)) {
|
|
purify(tr);
|
|
m_objectives[index].m_term = tr;
|
|
}
|
|
else if (is_minimize(fml, tr, orig_term, index)) {
|
|
purify(tr);
|
|
m_objectives[index].m_term = tr;
|
|
m_objectives[index].m_adjust_value.set_negate(true);
|
|
}
|
|
else {
|
|
m_hard_constraints.push_back(fml);
|
|
}
|
|
}
|
|
// fix types of objectives:
|
|
for (objective & obj : m_objectives) {
|
|
expr* t = obj.m_term;
|
|
switch(obj.m_type) {
|
|
case O_MINIMIZE:
|
|
case O_MAXIMIZE:
|
|
if (!m_arith.is_int(t) && !m_arith.is_real(t)) {
|
|
obj.m_term = m_arith.mk_numeral(rational(0), true);
|
|
}
|
|
break;
|
|
default:
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
|
|
|
|
void context::model_updated(model* md) {
|
|
model_ref mdl = md;
|
|
set_model(mdl);
|
|
}
|
|
|
|
rational context::adjust(unsigned id, rational const& v) {
|
|
return m_objectives[id].m_adjust_value(v);
|
|
}
|
|
|
|
void context::add_offset(unsigned id, rational const& o) {
|
|
m_objectives[id].m_adjust_value.add_offset(o);
|
|
}
|
|
|
|
bool context::verify_model(unsigned index, model* md, rational const& _v) {
|
|
rational r;
|
|
app_ref term = m_objectives[index].m_term;
|
|
if (!term) {
|
|
return true;
|
|
}
|
|
rational v = m_objectives[index].m_adjust_value(_v);
|
|
expr_ref val(m);
|
|
model_ref mdl = md->copy();
|
|
fix_model(mdl);
|
|
val = (*mdl)(term);
|
|
unsigned bvsz;
|
|
if (!m_arith.is_numeral(val, r) && !m_bv.is_numeral(val, r, bvsz)) {
|
|
TRACE("opt", tout << "model does not evaluate objective to a value but instead " << val << "\n";
|
|
tout << *mdl << "\n";
|
|
);
|
|
return false;
|
|
}
|
|
if (r != v) {
|
|
TRACE("opt", tout << "Out of bounds: " << term << " " << val << " != " << v << "\n";);
|
|
return false;
|
|
}
|
|
else {
|
|
TRACE("opt", tout << "validated: " << term << " = " << val << "\n";);
|
|
}
|
|
return true;
|
|
}
|
|
|
|
void context::purify(app_ref& term) {
|
|
generic_model_converter_ref fm;
|
|
if (m_arith.is_add(term)) {
|
|
expr_ref_vector args(m);
|
|
for (expr* arg : *term) {
|
|
if (is_mul_const(arg)) {
|
|
args.push_back(arg);
|
|
}
|
|
else {
|
|
args.push_back(purify(fm, arg));
|
|
}
|
|
}
|
|
term = m_arith.mk_add(args.size(), args.data());
|
|
}
|
|
else if (m.is_ite(term) || !is_mul_const(term)) {
|
|
TRACE("opt", tout << "Purifying " << term << "\n";);
|
|
term = purify(fm, term);
|
|
}
|
|
if (fm) {
|
|
m_model_converter = concat(m_model_converter.get(), fm.get());
|
|
}
|
|
}
|
|
|
|
bool context::is_mul_const(expr* e) {
|
|
expr* e1, *e2;
|
|
return
|
|
is_uninterp_const(e) ||
|
|
m_arith.is_numeral(e) ||
|
|
(m_arith.is_mul(e, e1, e2) && m_arith.is_numeral(e1) && is_uninterp_const(e2)) ||
|
|
(m_arith.is_mul(e, e2, e1) && m_arith.is_numeral(e1) && is_uninterp_const(e2));
|
|
}
|
|
|
|
app* context::purify(generic_model_converter_ref& fm, expr* term) {
|
|
std::ostringstream out;
|
|
out << mk_bounded_pp(term, m, 3);
|
|
app* q = m.mk_fresh_const(out.str(), term->get_sort());
|
|
if (!fm) fm = alloc(generic_model_converter, m, "opt");
|
|
if (m_arith.is_int_real(term)) {
|
|
m_hard_constraints.push_back(m_arith.mk_ge(q, term));
|
|
m_hard_constraints.push_back(m_arith.mk_le(q, term));
|
|
}
|
|
else {
|
|
m_hard_constraints.push_back(m.mk_eq(q, term));
|
|
}
|
|
fm->hide(q);
|
|
return q;
|
|
}
|
|
|
|
/**
|
|
To select the proper theory solver we have to ensure that all theory
|
|
symbols from soft constraints are reflected in the hard constraints.
|
|
|
|
- filter "obj" from generated model.
|
|
*/
|
|
void context::mk_atomic(expr_ref_vector& terms) {
|
|
generic_model_converter_ref fm;
|
|
for (unsigned i = 0; i < terms.size(); ++i) {
|
|
expr_ref p(terms[i].get(), m);
|
|
app_ref q(m);
|
|
if (is_propositional(p)) {
|
|
terms[i] = p;
|
|
}
|
|
else {
|
|
terms[i] = purify(fm, p);
|
|
}
|
|
}
|
|
if (fm) {
|
|
m_model_converter = concat(m_model_converter.get(), fm.get());
|
|
}
|
|
}
|
|
|
|
void context::to_fmls(expr_ref_vector& fmls) {
|
|
m_objective_fns.reset();
|
|
fmls.append(m_hard_constraints);
|
|
for (unsigned i = 0; i < m_objectives.size(); ++i) {
|
|
objective const& obj = m_objectives[i];
|
|
switch(obj.m_type) {
|
|
case O_MINIMIZE:
|
|
fmls.push_back(mk_minimize(i, obj.m_term));
|
|
break;
|
|
case O_MAXIMIZE:
|
|
fmls.push_back(mk_maximize(i, obj.m_term));
|
|
break;
|
|
case O_MAXSMT:
|
|
fmls.push_back(mk_maxsat(i, obj.m_terms.size(), obj.m_terms.data()));
|
|
break;
|
|
}
|
|
}
|
|
TRACE("opt", tout << fmls << "\n";);
|
|
}
|
|
|
|
void context::internalize() {
|
|
for (objective & obj : m_objectives) {
|
|
switch(obj.m_type) {
|
|
case O_MINIMIZE: {
|
|
app_ref tmp(m);
|
|
tmp = obj.m_term;
|
|
if (m_arith.is_int(tmp) || m_arith.is_real(tmp)) {
|
|
tmp = m_arith.mk_uminus(obj.m_term);
|
|
}
|
|
obj.m_index = m_optsmt.add(tmp);
|
|
break;
|
|
}
|
|
case O_MAXIMIZE:
|
|
obj.m_index = m_optsmt.add(obj.m_term);
|
|
break;
|
|
case O_MAXSMT: {
|
|
maxsmt& ms = *m_maxsmts.find(obj.m_id);
|
|
for (unsigned j = 0; j < obj.m_terms.size(); ++j) {
|
|
ms.add(obj.m_terms.get(j), obj.m_weights[j]);
|
|
}
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
void context::update_bound(bool is_lower) {
|
|
expr_ref val(m);
|
|
if (!m_model.get()) return;
|
|
for (objective const& obj : m_objectives) {
|
|
rational r;
|
|
switch(obj.m_type) {
|
|
case O_MINIMIZE: {
|
|
val = (*m_model)(obj.m_term);
|
|
TRACE("opt", tout << obj.m_term << " " << val << "\n";);
|
|
if (is_numeral(val, r)) {
|
|
inf_eps val = inf_eps(obj.m_adjust_value(r));
|
|
TRACE("opt", tout << "adjusted value: " << val << "\n";);
|
|
if (is_lower) {
|
|
m_optsmt.update_lower(obj.m_index, val);
|
|
}
|
|
else {
|
|
m_optsmt.update_upper(obj.m_index, val);
|
|
}
|
|
}
|
|
break;
|
|
}
|
|
case O_MAXIMIZE: {
|
|
val = (*m_model)(obj.m_term);
|
|
TRACE("opt", tout << obj.m_term << " " << val << "\n";);
|
|
if (is_numeral(val, r)) {
|
|
inf_eps val = inf_eps(obj.m_adjust_value(r));
|
|
TRACE("opt", tout << "adjusted value: " << val << "\n";);
|
|
if (is_lower) {
|
|
m_optsmt.update_lower(obj.m_index, val);
|
|
}
|
|
else {
|
|
m_optsmt.update_upper(obj.m_index, val);
|
|
}
|
|
}
|
|
break;
|
|
}
|
|
case O_MAXSMT: {
|
|
for (unsigned j = 0; j < obj.m_terms.size(); ++j) {
|
|
val = (*m_model)(obj.m_terms[j]);
|
|
TRACE("opt", tout << mk_pp(obj.m_terms[j], m) << " " << val << "\n";);
|
|
if (!m.is_true(val))
|
|
r += obj.m_weights[j];
|
|
}
|
|
|
|
maxsmt& ms = *m_maxsmts.find(obj.m_id);
|
|
if (is_lower) {
|
|
ms.update_upper(r);
|
|
TRACE("opt", tout << "update upper from " << r << " to " << ms.get_upper() << "\n";);
|
|
}
|
|
else {
|
|
ms.update_lower(r);
|
|
TRACE("opt", tout << "update lower from " << r << " to " << ms.get_lower() << "\n";);
|
|
}
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
void context::display_benchmark() {
|
|
display(verbose_stream());
|
|
return;
|
|
|
|
if (opt_params(m_params).dump_benchmarks() &&
|
|
sat_enabled() &&
|
|
m_objectives.size() == 1 &&
|
|
m_objectives[0].m_type == O_MAXSMT
|
|
) {
|
|
objective& o = m_objectives[0];
|
|
unsigned sz = o.m_terms.size();
|
|
inc_sat_display(verbose_stream(), get_solver(), sz, o.m_terms.data(), o.m_weights.data());
|
|
}
|
|
|
|
|
|
}
|
|
|
|
void context::display(std::ostream& out) {
|
|
display_assignment(out);
|
|
}
|
|
|
|
void context::display_assignment(std::ostream& out) {
|
|
if (m_scoped_state.m_objectives.size() != m_objectives.size()) {
|
|
throw default_exception("check-sat has not been called with latest objectives");
|
|
}
|
|
out << "(objectives\n";
|
|
for (unsigned i = 0; i < m_scoped_state.m_objectives.size(); ++i) {
|
|
objective const& obj = m_scoped_state.m_objectives[i];
|
|
out << " (";
|
|
display_objective(out, obj);
|
|
if (get_lower_as_num(i) != get_upper_as_num(i)) {
|
|
out << " (interval " << get_lower(i) << " " << get_upper(i) << ")";
|
|
}
|
|
else {
|
|
out << " " << get_lower(i);
|
|
}
|
|
out << ")\n";
|
|
}
|
|
out << ")\n";
|
|
}
|
|
|
|
void context::display_objective(std::ostream& out, objective const& obj) const {
|
|
switch(obj.m_type) {
|
|
case O_MAXSMT: {
|
|
symbol s = obj.m_id;
|
|
if (s != symbol::null) {
|
|
out << s;
|
|
}
|
|
break;
|
|
}
|
|
default:
|
|
out << obj.m_term;
|
|
break;
|
|
}
|
|
}
|
|
|
|
inf_eps context::get_lower_as_num(unsigned idx) {
|
|
if (idx >= m_objectives.size()) {
|
|
throw default_exception("index out of bounds");
|
|
}
|
|
objective const& obj = m_objectives[idx];
|
|
switch(obj.m_type) {
|
|
case O_MAXSMT:
|
|
return inf_eps(m_maxsmts.find(obj.m_id)->get_lower());
|
|
case O_MINIMIZE:
|
|
return obj.m_adjust_value(m_optsmt.get_upper(obj.m_index));
|
|
case O_MAXIMIZE:
|
|
return obj.m_adjust_value(m_optsmt.get_lower(obj.m_index));
|
|
default:
|
|
UNREACHABLE();
|
|
return inf_eps();
|
|
}
|
|
}
|
|
|
|
inf_eps context::get_upper_as_num(unsigned idx) {
|
|
if (idx >= m_objectives.size()) {
|
|
throw default_exception("index out of bounds");
|
|
}
|
|
objective const& obj = m_objectives[idx];
|
|
switch(obj.m_type) {
|
|
case O_MAXSMT:
|
|
return inf_eps(m_maxsmts.find(obj.m_id)->get_upper());
|
|
case O_MINIMIZE:
|
|
return obj.m_adjust_value(m_optsmt.get_lower(obj.m_index));
|
|
case O_MAXIMIZE:
|
|
return obj.m_adjust_value(m_optsmt.get_upper(obj.m_index));
|
|
default:
|
|
UNREACHABLE();
|
|
return inf_eps();
|
|
}
|
|
}
|
|
|
|
expr_ref context::get_lower(unsigned idx) {
|
|
return to_expr(get_lower_as_num(idx));
|
|
}
|
|
|
|
expr_ref context::get_upper(unsigned idx) {
|
|
return to_expr(get_upper_as_num(idx));
|
|
}
|
|
|
|
void context::to_exprs(inf_eps const& n, expr_ref_vector& es) {
|
|
rational inf = n.get_infinity();
|
|
rational r = n.get_rational();
|
|
rational eps = n.get_infinitesimal();
|
|
es.push_back(m_arith.mk_numeral(inf, inf.is_int()));
|
|
es.push_back(m_arith.mk_numeral(r, r.is_int()));
|
|
es.push_back(m_arith.mk_numeral(eps, eps.is_int()));
|
|
}
|
|
|
|
expr_ref context::to_expr(inf_eps const& n) {
|
|
rational inf = n.get_infinity();
|
|
rational r = n.get_rational();
|
|
rational eps = n.get_infinitesimal();
|
|
expr_ref_vector args(m);
|
|
bool is_int = eps.is_zero() && r.is_int();
|
|
if (!inf.is_zero()) {
|
|
expr* oo = m.mk_const(symbol("oo"), is_int ? m_arith.mk_int() : m_arith.mk_real());
|
|
if (inf.is_one()) {
|
|
args.push_back(oo);
|
|
}
|
|
else {
|
|
args.push_back(m_arith.mk_mul(m_arith.mk_numeral(inf, is_int), oo));
|
|
}
|
|
}
|
|
if (!r.is_zero()) {
|
|
args.push_back(m_arith.mk_numeral(r, is_int));
|
|
}
|
|
if (!eps.is_zero()) {
|
|
expr* ep = m.mk_const(symbol("epsilon"), m_arith.mk_real());
|
|
if (eps.is_one()) {
|
|
args.push_back(ep);
|
|
}
|
|
else {
|
|
args.push_back(m_arith.mk_mul(m_arith.mk_numeral(eps, is_int), ep));
|
|
}
|
|
}
|
|
switch(args.size()) {
|
|
case 0: return expr_ref(m_arith.mk_numeral(rational(0), true), m);
|
|
case 1: return expr_ref(args[0].get(), m);
|
|
default: return expr_ref(m_arith.mk_add(args.size(), args.data()), m);
|
|
}
|
|
}
|
|
|
|
void context::set_simplify(tactic* tac) {
|
|
m_simplify = tac;
|
|
}
|
|
|
|
void context::clear_state() {
|
|
m_pareto = nullptr;
|
|
m_pareto1 = false;
|
|
m_box_index = UINT_MAX;
|
|
m_box_models.reset();
|
|
m_model.reset();
|
|
m_model_fixed.reset();
|
|
m_core.reset();
|
|
}
|
|
|
|
void context::set_pareto(pareto_base* p) {
|
|
m_pareto = p;
|
|
m_pareto1 = p != nullptr;
|
|
}
|
|
|
|
void context::collect_statistics(statistics& stats) const {
|
|
if (m_solver)
|
|
m_solver->collect_statistics(stats);
|
|
if (m_simplify)
|
|
m_simplify->collect_statistics(stats);
|
|
for (auto const& kv : m_maxsmts)
|
|
kv.m_value->collect_statistics(stats);
|
|
get_memory_statistics(stats);
|
|
get_rlimit_statistics(m.limit(), stats);
|
|
if (m_qmax)
|
|
m_qmax->collect_statistics(stats);
|
|
}
|
|
|
|
void context::collect_param_descrs(param_descrs & r) {
|
|
opt_params::collect_param_descrs(r);
|
|
insert_timeout(r);
|
|
insert_ctrl_c(r);
|
|
}
|
|
|
|
void context::updt_params(params_ref const& p) {
|
|
m_params.append(p);
|
|
if (m_solver) {
|
|
m_solver->updt_params(m_params);
|
|
}
|
|
if (m_sat_solver) {
|
|
m_sat_solver->updt_params(m_params);
|
|
}
|
|
m_optsmt.updt_params(m_params);
|
|
for (auto & kv : m_maxsmts) {
|
|
kv.m_value->updt_params(m_params);
|
|
}
|
|
opt_params _p(p);
|
|
m_enable_sat = _p.enable_sat();
|
|
m_enable_sls = _p.enable_sls();
|
|
m_maxsat_engine = _p.maxsat_engine();
|
|
m_pp_neat = _p.pp_neat();
|
|
m_pp_wcnf = _p.pp_wcnf();
|
|
m_incremental = _p.incremental();
|
|
}
|
|
|
|
std::string context::to_string() {
|
|
if (m_pp_wcnf)
|
|
return to_wcnf();
|
|
return to_string(false, m_scoped_state.m_hard, m_scoped_state.m_objectives);
|
|
}
|
|
|
|
std::string context::to_string_internal() const {
|
|
return to_string(true, m_hard_constraints, m_objectives);
|
|
}
|
|
|
|
std::string context::to_wcnf() {
|
|
import_scoped_state();
|
|
expr_ref_vector asms(m);
|
|
normalize(asms);
|
|
auto const& objectives = m_objectives;
|
|
if (objectives.size() > 1)
|
|
throw default_exception("only single objective weighted MaxSAT wcnf output is supported");
|
|
ptr_vector<expr> soft_f;
|
|
vector<rational> soft_w;
|
|
svector<std::pair<expr*, unsigned>> soft;
|
|
if (objectives.size() == 1) {
|
|
auto const& obj = objectives[0];
|
|
if (obj.m_type != O_MAXSMT)
|
|
throw default_exception("only single objective weighted MaxSAT wcnf output is supported");
|
|
for (unsigned j = 0; j < obj.m_terms.size(); ++j) {
|
|
rational w = obj.m_weights[j];
|
|
if (!w.is_unsigned())
|
|
throw default_exception("only single objective weighted MaxSAT wcnf output is supported");
|
|
soft_f.push_back(obj.m_terms[j]);
|
|
soft_w.push_back(w);
|
|
}
|
|
}
|
|
std::ostringstream strm;
|
|
m_sat_solver = mk_inc_sat_solver(m, m_params);
|
|
m_sat_solver->assert_expr(m_hard_constraints);
|
|
inc_sat_display(strm, *m_sat_solver.get(), soft_f.size(), soft_f.data(), soft_w.data());
|
|
return strm.str();
|
|
}
|
|
|
|
std::string context::to_string(bool is_internal, expr_ref_vector const& hard, vector<objective> const& objectives) const {
|
|
smt2_pp_environment_dbg env(m);
|
|
ast_pp_util visitor(m);
|
|
std::ostringstream out;
|
|
visitor.collect(hard);
|
|
model_converter_ref mc = concat(m_model_converter.get(), m_fm.get());
|
|
|
|
for (objective const& obj : objectives) {
|
|
switch(obj.m_type) {
|
|
case O_MAXIMIZE:
|
|
case O_MINIMIZE:
|
|
visitor.collect(obj.m_term);
|
|
break;
|
|
case O_MAXSMT:
|
|
visitor.collect(obj.m_terms);
|
|
break;
|
|
default:
|
|
UNREACHABLE();
|
|
break;
|
|
}
|
|
}
|
|
|
|
if (is_internal && mc) {
|
|
mc->set_env(&visitor);
|
|
}
|
|
|
|
param_descrs descrs;
|
|
collect_param_descrs(descrs);
|
|
m_params.display_smt2(out, "opt", descrs);
|
|
visitor.display_decls(out);
|
|
visitor.display_asserts(out, hard, m_pp_neat);
|
|
for (objective const& obj : objectives) {
|
|
switch(obj.m_type) {
|
|
case O_MAXIMIZE:
|
|
out << "(maximize ";
|
|
ast_smt2_pp(out, obj.m_term, env);
|
|
out << ")\n";
|
|
break;
|
|
case O_MINIMIZE:
|
|
out << "(minimize ";
|
|
ast_smt2_pp(out, obj.m_term, env);
|
|
out << ")\n";
|
|
break;
|
|
case O_MAXSMT:
|
|
for (unsigned j = 0; j < obj.m_terms.size(); ++j) {
|
|
out << "(assert-soft ";
|
|
ast_smt2_pp(out, obj.m_terms[j], env);
|
|
rational w = obj.m_weights[j];
|
|
|
|
w.display_decimal(out << " :weight ", 3, true);
|
|
if (obj.m_id != symbol::null) {
|
|
if (is_smt2_quoted_symbol(obj.m_id)) {
|
|
out << " :id " << mk_smt2_quoted_symbol(obj.m_id);
|
|
}
|
|
else {
|
|
out << " :id " << obj.m_id;
|
|
}
|
|
}
|
|
out << ")\n";
|
|
}
|
|
break;
|
|
default:
|
|
UNREACHABLE();
|
|
break;
|
|
}
|
|
}
|
|
if (is_internal && mc) {
|
|
mc->display(out);
|
|
}
|
|
if (is_internal && mc) {
|
|
mc->set_env(nullptr);
|
|
}
|
|
out << "(check-sat)\n";
|
|
return out.str();
|
|
}
|
|
|
|
void context::validate_model() {
|
|
return;
|
|
if (!gparams::get_ref().get_bool("model_validate", false)) return;
|
|
expr_ref_vector fmls(m);
|
|
get_hard_constraints(fmls);
|
|
expr_ref tmp(m);
|
|
model_ref mdl;
|
|
get_model(mdl);
|
|
mdl->set_model_completion(true);
|
|
for (expr * f : fmls) {
|
|
if (!mdl->is_true(f)) {
|
|
IF_VERBOSE(0,
|
|
verbose_stream() << "Failed to validate " << mk_pp(f, m) << "\n" << tmp << "\n";
|
|
m_fm->display(verbose_stream() << "fm\n");
|
|
m_model_converter->display(verbose_stream() << "mc\n");
|
|
model_smt2_pp(verbose_stream(), m, *mdl, 0);
|
|
verbose_stream() << to_string_internal() << "\n");
|
|
}
|
|
}
|
|
}
|
|
|
|
void context::validate_maxsat(symbol const& id) {
|
|
maxsmt& ms = *m_maxsmts.find(id);
|
|
TRACE("opt", tout << "Validate: " << id << "\n";);
|
|
for (objective const& obj : m_objectives) {
|
|
if (obj.m_id == id && obj.m_type == O_MAXSMT) {
|
|
SASSERT(obj.m_type == O_MAXSMT);
|
|
rational value(0);
|
|
expr_ref val(m);
|
|
for (unsigned i = 0; i < obj.m_terms.size(); ++i) {
|
|
auto const& t = obj.m_terms[i];
|
|
if (!m_model->is_true(t)) {
|
|
value += obj.m_weights[i];
|
|
}
|
|
// TBD: check that optimal was not changed.
|
|
}
|
|
value = obj.m_adjust_value(value);
|
|
rational value0 = ms.get_lower();
|
|
TRACE("opt", tout << "value " << value << " " << value0 << "\n";);
|
|
// TBD is this correct? SASSERT(value == value0);
|
|
}
|
|
}
|
|
}
|
|
|
|
void context::validate_lex() {
|
|
rational r1;
|
|
expr_ref val(m);
|
|
SASSERT(m_model);
|
|
for (unsigned i = 0; i < m_objectives.size(); ++i) {
|
|
objective const& obj = m_objectives[i];
|
|
switch(obj.m_type) {
|
|
case O_MINIMIZE:
|
|
case O_MAXIMIZE:
|
|
break;
|
|
case O_MAXSMT: {
|
|
rational value(0);
|
|
for (unsigned i = 0; i < obj.m_terms.size(); ++i) {
|
|
if (!m_model->is_true(obj.m_terms[i])) {
|
|
value += obj.m_weights[i];
|
|
}
|
|
// TBD: check that optimal was not changed.
|
|
}
|
|
maxsmt& ms = *m_maxsmts.find(obj.m_id);
|
|
rational value0 = ms.get_lower();
|
|
TRACE("opt", tout << "value " << value << " other " << value0 << "\n";);
|
|
// TBD SASSERT(value0 == value);
|
|
break;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
bool context::is_qsat_opt() {
|
|
if (m_objectives.size() != 1) {
|
|
return false;
|
|
}
|
|
if (m_objectives[0].m_type != O_MAXIMIZE &&
|
|
m_objectives[0].m_type != O_MINIMIZE) {
|
|
return false;
|
|
}
|
|
if (!m_arith.is_real(m_objectives[0].m_term)) {
|
|
return false;
|
|
}
|
|
for (expr* fml : m_hard_constraints) {
|
|
if (has_quantifiers(fml)) {
|
|
return true;
|
|
}
|
|
}
|
|
return false;
|
|
}
|
|
|
|
lbool context::run_qsat_opt() {
|
|
SASSERT(is_qsat_opt());
|
|
objective const& obj = m_objectives[0];
|
|
app_ref term(obj.m_term);
|
|
if (obj.m_type == O_MINIMIZE) {
|
|
term = m_arith.mk_uminus(term);
|
|
}
|
|
inf_eps value;
|
|
m_qmax = alloc(qe::qmax, m, m_params);
|
|
lbool result = (*m_qmax)(m_hard_constraints, term, value, m_model);
|
|
if (result != l_undef && obj.m_type == O_MINIMIZE) {
|
|
value.neg();
|
|
}
|
|
m_optsmt.setup(*m_opt_solver.get());
|
|
if (result == l_undef) {
|
|
if (obj.m_type == O_MINIMIZE) {
|
|
m_optsmt.update_upper(obj.m_index, value);
|
|
}
|
|
else {
|
|
m_optsmt.update_lower(obj.m_index, value);
|
|
}
|
|
}
|
|
else {
|
|
m_optsmt.update_lower(obj.m_index, value);
|
|
m_optsmt.update_upper(obj.m_index, value);
|
|
}
|
|
return result;
|
|
}
|
|
}
|