3
0
Fork 0
mirror of https://github.com/Z3Prover/z3 synced 2025-06-22 22:03:39 +00:00

Pseudo-inverse op_constraint

This commit is contained in:
Clemens Eisenhofer 2023-01-03 17:47:54 +01:00
parent 84a5ec221f
commit 79e7380ffc
10 changed files with 144 additions and 47 deletions

View file

@ -509,4 +509,10 @@ namespace polysat {
pdd constraint_manager::bnor(pdd const& p, pdd const& q) { pdd constraint_manager::bnor(pdd const& p, pdd const& q) {
return bnot(bor(p, q)); return bnot(bor(p, q));
} }
pdd constraint_manager::pseudo_inv(pdd const& p) {
if (p.is_val())
return p.manager().mk_val(p.val().pseudo_inverse(p.power_of_2()));
return mk_op_term(op_constraint::code::inv_op, p, p.manager().zero());
}
} }

View file

@ -126,6 +126,7 @@ namespace polysat {
pdd bxor(pdd const& p, pdd const& q); pdd bxor(pdd const& p, pdd const& q);
pdd bnand(pdd const& p, pdd const& q); pdd bnand(pdd const& p, pdd const& q);
pdd bnor(pdd const& p, pdd const& q); pdd bnor(pdd const& p, pdd const& q);
pdd pseudo_inv(pdd const& p);
constraint* const* begin() const { return m_constraints.data(); } constraint* const* begin() const { return m_constraints.data(); }
constraint* const* end() const { return m_constraints.data() + m_constraints.size(); } constraint* const* end() const { return m_constraints.data() + m_constraints.size(); }

View file

@ -40,6 +40,8 @@ namespace polysat {
if (p.index() > q.index()) if (p.index() > q.index())
std::swap(m_p, m_q); std::swap(m_p, m_q);
break; break;
case code::inv_op:
SASSERT(q.is_zero());
default: default:
break; break;
} }
@ -61,6 +63,8 @@ namespace polysat {
return eval_shl(p, q, r); return eval_shl(p, q, r);
case code::and_op: case code::and_op:
return eval_and(p, q, r); return eval_and(p, q, r);
case code::inv_op:
return eval_inv(p, r);
default: default:
return l_undef; return l_undef;
} }
@ -84,6 +88,8 @@ namespace polysat {
return out << "<<"; return out << "<<";
case op_constraint::code::and_op: case op_constraint::code::and_op:
return out << "&"; return out << "&";
case op_constraint::code::inv_op:
return out << "inv";
default: default:
UNREACHABLE(); UNREACHABLE();
return out; return out;
@ -96,6 +102,9 @@ namespace polysat {
} }
std::ostream& op_constraint::display(std::ostream& out, char const* eq) const { std::ostream& op_constraint::display(std::ostream& out, char const* eq) const {
if (m_op == code::inv_op)
return out << r() << " " << eq << " " << m_op << " " << p();
return out << r() << " " << eq << " " << p() << " " << m_op << " " << q(); return out << r() << " " << eq << " " << p() << " " << m_op << " " << q();
} }
@ -161,6 +170,8 @@ namespace polysat {
return lemma_shl(s, a); return lemma_shl(s, a);
case code::and_op: case code::and_op:
return lemma_and(s, a); return lemma_and(s, a);
case code::inv_op:
return lemma_inv(s, a);
default: default:
NOT_IMPLEMENTED_YET(); NOT_IMPLEMENTED_YET();
return {}; return {};
@ -178,6 +189,8 @@ namespace polysat {
// handle masking of high order bits // handle masking of high order bits
activate_and(s); activate_and(s);
break; break;
case code::inv_op:
break;
default: default:
break; break;
} }
@ -571,6 +584,73 @@ namespace polysat {
return true; return true;
} }
/**
* Produce lemmas for constraint: r == inv p
* p = 0 => r = 0
* r = 0 => p = 0
* odd(r) -- for now we are looking for the smallest pseudo-inverse (there are 2^parity(p) of them)
* parity(p) >= k && p * r < 2^k => p * r >= 2^k
* parity(p) < k && p * r >= 2^k => p * r < 2^k
*/
clause_ref op_constraint::lemma_inv(solver& s, assignment const& a) {
auto& m = p().manager();
auto pv = a.apply_to(p());
auto rv = a.apply_to(r());
if (!pv.is_val() || !rv.is_val() || eval_inv(pv, rv) == l_true)
return {};
unsigned parity_pv = pv.val().trailing_zeros();
unsigned parity_rv = rv.val().trailing_zeros();
signed_constraint const invc(this, true);
// p = 0 => r = 0
if (pv.is_zero())
return s.mk_clause(~invc, ~s.eq(p()), s.eq(r()), true);
// r = 0 => p = 0
if (rv.is_zero())
return s.mk_clause(~invc, ~s.eq(r()), s.eq(p()), true);
// odd(r)
if (parity_rv != 0)
return s.mk_clause(~invc, s.odd(r()), true);
// parity(p) >= k && p * r < 2^k => p * r >= 2^k
// parity(p) < k && p * r >= 2^k => p * r < 2^k
rational prod = (p() * r()).val();
SASSERT(prod != rational::power_of_two(parity_pv)); // Why did it evaluate to false in this case?
unsigned lower = 0, upper = p().power_of_2();
// binary search for the parity
while (lower + 1 < upper) {
unsigned middle = (upper + lower) / 2;
LOG("Splitting on " << middle);
if (parity_pv >= middle) {
lower = middle;
LOG("Its in [" << lower << "; " << upper << ")");
if (prod < rational::power_of_two(middle))
return s.mk_clause(~invc, ~s.parity_at_least(p(), middle), s.uge(p() * r(), rational::power_of_two(middle)), false);
}
else {
upper = middle;
LOG("Its in [" << lower << "; " << upper << ")");
if (prod >= rational::power_of_two(middle))
return s.mk_clause(~invc, s.parity_at_least(p(), middle), s.ult(p() * r(), rational::power_of_two(middle)), false);
}
}
UNREACHABLE();
return {};
}
/** Evaluate constraint: r == inv p */
lbool op_constraint::eval_inv(pdd const& p, pdd const& r) {
if (!p.is_val() || !r.is_val())
return l_undef;
if (p.is_zero() || r.is_zero()) // the inverse of 0 is 0 (by arbitrary definition). Just to have some unique value
return p.is_zero() && r.is_zero() ? l_true : l_false;
return p.val().pseudo_inverse(p.power_of_2()) == r.val() ? l_true : l_false;
}
void op_constraint::add_to_univariate_solver(pvar v, solver& s, univariate_solver& us, unsigned dep, bool is_positive) const { void op_constraint::add_to_univariate_solver(pvar v, solver& s, univariate_solver& us, unsigned dep, bool is_positive) const {
pdd pv = s.subst(p()); pdd pv = s.subst(p());
if (!pv.is_univariate_in(v)) if (!pv.is_univariate_in(v))

View file

@ -26,7 +26,7 @@ namespace polysat {
class op_constraint final : public constraint { class op_constraint final : public constraint {
public: public:
enum class code { lshr_op, ashr_op, shl_op, and_op }; enum class code { lshr_op, ashr_op, shl_op, and_op, inv_op };
protected: protected:
friend class constraint_manager; friend class constraint_manager;
@ -51,6 +51,9 @@ namespace polysat {
static lbool eval_and(pdd const& p, pdd const& q, pdd const& r); static lbool eval_and(pdd const& p, pdd const& q, pdd const& r);
bool propagate_bits_and(solver& s, bool is_positive); bool propagate_bits_and(solver& s, bool is_positive);
clause_ref lemma_inv(solver& s, assignment const& a);
static lbool eval_inv(pdd const& p, pdd const& r);
std::ostream& display(std::ostream& out, char const* eq) const; std::ostream& display(std::ostream& out, char const* eq) const;
void activate(solver& s); void activate(solver& s);

View file

@ -1245,34 +1245,26 @@ namespace polysat {
if (c->is_ule()) { if (c->is_ule()) {
// If both are equalities this boils down to polynomial superposition => Might generate the same lemma twice // If both are equalities this boils down to polynomial superposition => Might generate the same lemma twice
auto const& ule = c->to_ule(); auto const& ule = c->to_ule();
auto [lhs_new, changed_lhs, side_condition_lhs] = m_parity_tracker.eliminate_variable(*this, x, a, b, ule.lhs()); m_lemma.reset();
auto [rhs_new, changed_rhs, side_condition_rhs] = m_parity_tracker.eliminate_variable(*this, x, a, b, ule.rhs()); auto [lhs_new, changed_lhs] = m_parity_tracker.eliminate_variable(*this, x, a, b, ule.lhs(), m_lemma);
auto [rhs_new, changed_rhs] = m_parity_tracker.eliminate_variable(*this, x, a, b, ule.rhs(), m_lemma);
if (!changed_lhs && !changed_rhs) if (!changed_lhs && !changed_rhs)
continue; // nothing changed - no reason for propagating lemmas continue; // nothing changed - no reason for propagating lemmas
m_lemma.reset();
m_lemma.insert(~c); m_lemma.insert(~c);
m_lemma.insert_eval(~s.eq(y)); m_lemma.insert_eval(~s.eq(y));
for (auto& sc_lhs : side_condition_lhs) // the "path to get the parities"
m_lemma.insert(sc_lhs);
for (auto& sc_rhs : side_condition_rhs)
m_lemma.insert(sc_rhs);
if (propagate(x, core, a_l_b, c.is_positive() ? s.ule(lhs_new, rhs_new) : ~s.ule(lhs_new, rhs_new))) if (propagate(x, core, a_l_b, c.is_positive() ? s.ule(lhs_new, rhs_new) : ~s.ule(lhs_new, rhs_new)))
prop = true; prop = true;
} }
else if (c->is_umul_ovfl()) { else if (c->is_umul_ovfl()) {
auto const& ovf = c->to_umul_ovfl(); auto const& ovf = c->to_umul_ovfl();
auto [lhs_new, changed_lhs, side_condition_lhs] = m_parity_tracker.eliminate_variable(*this, x, a, b, ovf.p()); auto [lhs_new, changed_lhs] = m_parity_tracker.eliminate_variable(*this, x, a, b, ovf.p(), m_lemma);
auto [rhs_new, changed_rhs, side_condition_rhs] = m_parity_tracker.eliminate_variable(*this, x, a, b, ovf.q()); auto [rhs_new, changed_rhs] = m_parity_tracker.eliminate_variable(*this, x, a, b, ovf.q(), m_lemma);
if (!changed_lhs && !changed_rhs) if (!changed_lhs && !changed_rhs)
continue; continue;
m_lemma.reset(); m_lemma.reset();
m_lemma.insert(~c); m_lemma.insert(~c);
m_lemma.insert_eval(~s.eq(y)); m_lemma.insert_eval(~s.eq(y));
for (auto& sc_lhs : side_condition_lhs)
m_lemma.insert(sc_lhs);
for (auto& sc_rhs : side_condition_rhs)
m_lemma.insert(sc_rhs);
if (propagate(x, core, a_l_b, c.is_positive() ? s.umul_ovfl(lhs_new, rhs_new) : ~s.umul_ovfl(lhs_new, rhs_new))) if (propagate(x, core, a_l_b, c.is_positive() ? s.umul_ovfl(lhs_new, rhs_new) : ~s.umul_ovfl(lhs_new, rhs_new)))
prop = true; prop = true;

View file

@ -132,7 +132,7 @@ namespace polysat {
auto const eq_it = std::find(cl.begin(), cl.end(), eq.blit()); auto const eq_it = std::find(cl.begin(), cl.end(), eq.blit());
if (eq_it == cl.end()) if (eq_it == cl.end())
continue; continue;
unsigned const eq_idx = std::distance(cl.begin(), eq_it); unsigned eq_idx = (unsigned)std::distance(cl.begin(), eq_it);
any_removed = true; any_removed = true;
should_remove[eq_idx] = true; should_remove[eq_idx] = true;
if (c.is_positive()) { if (c.is_positive()) {

View file

@ -405,6 +405,9 @@ namespace polysat {
/** Create expression for bit-wise nor of p, q. */ /** Create expression for bit-wise nor of p, q. */
pdd bnor(pdd const& p, pdd const& q) { return m_constraints.bnor(p, q); } pdd bnor(pdd const& p, pdd const& q) { return m_constraints.bnor(p, q); }
/** Create expression for the smallest pseudo-inverse of p. */
pdd pseudo_inv(pdd const& p) { return m_constraints.pseudo_inv(p); }
/** /**
* Create polynomial constant. * Create polynomial constant.
*/ */

View file

@ -583,7 +583,7 @@ namespace polysat {
return inv_pdd; return inv_pdd;
} }
pdd parity_tracker::get_odd(const pdd& p, unsigned parity, svector<signed_constraint>& path) { pdd parity_tracker::get_odd(const pdd& p, unsigned parity, clause_builder& precondition) {
LOG("Getting odd part of " << p); LOG("Getting odd part of " << p);
if (p.is_val()) { if (p.is_val()) {
SASSERT(!p.val().is_zero()); SASSERT(!p.val().is_zero());
@ -618,14 +618,14 @@ namespace polysat {
LOG("Splitting on " << middle << " with " << parity); LOG("Splitting on " << middle << " with " << parity);
if (parity >= middle) { if (parity >= middle) {
lower = middle; lower = middle;
path.push_back(~c); precondition.insert(~c);
if (needs_propagate) if (needs_propagate)
m_builder.insert(~c); m_builder.insert(~c);
verbose_stream() << "Side-condition: " << ~c << "\n"; verbose_stream() << "Side-condition: " << ~c << "\n";
} }
else { else {
upper = middle; upper = middle;
path.push_back(c); precondition.insert(c);
if (needs_propagate) if (needs_propagate)
m_builder.insert(c); m_builder.insert(c);
verbose_stream() << "Side-condition: " << c << "\n"; verbose_stream() << "Side-condition: " << c << "\n";
@ -643,40 +643,40 @@ namespace polysat {
} }
// a * x + b = 0 (x not in a or b; i.e., the equation is linear in x) // a * x + b = 0 (x not in a or b; i.e., the equation is linear in x)
// C[p, ...] resp., C[..., p] // C[x, ...] resp., C[..., x]
std::tuple<pdd, bool, svector<signed_constraint>> parity_tracker::eliminate_variable(saturation& saturation, pvar x, const pdd& a, const pdd& b, const pdd& p) { std::tuple<pdd, bool> parity_tracker::eliminate_variable(saturation& saturation, pvar x, const pdd& a, const pdd& b, const pdd& p, clause_builder& precondition) {
unsigned p_degree = p.degree(x); unsigned p_degree = p.degree(x);
if (p_degree == 0) if (p_degree == 0)
return { p, false, {} }; return { p, false };
if (a.is_val() && a.val().is_odd()) { // just invert and plug it in if (a.is_val() && a.val().is_odd()) { // just invert and plug it in
rational a_inv; rational a_inv;
VERIFY(a.val().mult_inverse(a.power_of_2(), a_inv)); VERIFY(a.val().mult_inverse(a.power_of_2(), a_inv));
// this works as well if the degree of "p" is not 1: 3 x = a (mod 4) && x^2 <= b => (3a)^2 <= b // this works as well if the degree of "p" is not 1: 3 x = a (mod 4) && x^2 <= b => (3a)^2 <= b
return { p.subst_pdd(x, -b * a_inv), true, {} }; return { p.subst_pdd(x, -b * a_inv), true, };
} }
// from now on we require linear factors // from now on we require linear factors
if (p_degree != 1) if (p_degree != 1)
return { p, false, {} }; // TODO: Maybe fallback to brute-force return { p, false }; // TODO: Maybe fallback to brute-force
pdd a1 = a.manager().zero(), b1 = a1, mul_fac = a1; pdd a1 = a.manager().zero(), b1 = a1, mul_fac = a1;
p.factor(x, 1, a1, b1); p.factor(x, 1, a1, b1);
lbool is_multiple = saturation.get_multiple(a1, a, mul_fac); lbool is_multiple = saturation.get_multiple(a1, a, mul_fac);
if (is_multiple == l_false) if (is_multiple == l_false)
return { p, false, {} }; // there is no chance to invert return { p, false }; // there is no chance to invert
if (is_multiple == l_true) // we multiply with a factor to make them equal if (is_multiple == l_true) // we multiply with a factor to make them equal
return { b1 - b * mul_fac, true, {} }; return { b1 - b * mul_fac, true };
#if 1 #if 0
return { p, false, {} }; return { p, false };
#else #else
if (!a.is_monomial() || !a1.is_monomial()) if (!a.is_monomial() || !a1.is_monomial())
return { p , false, {} }; return { p , false };
if (!a1.is_var() && !a1.is_val()) { if (!a1.is_var() && !a1.is_val()) {
// TODO: Compromise: Maybe only monomials...? Does this make sense? // TODO: Compromise: Maybe only monomials...?
//return { p, false, {} }; //return { p, false, {} };
LOG("Warning: Inverting " << a1 << " although it is not a single variable - might not be a good idea"); LOG("Warning: Inverting " << a1 << " although it is not a single variable - might not be a good idea");
} }
@ -685,37 +685,49 @@ namespace polysat {
LOG("Warning: Inverting " << a << " although it is not a single variable - might not be a good idea"); LOG("Warning: Inverting " << a << " although it is not a single variable - might not be a good idea");
} }
// We don't know whether it will work. Use the parity of the assignment
#if 1 #if 1
unsigned a_parity; unsigned a_parity;
if ((a_parity = saturation.min_parity(a)) != saturation.max_parity(a) || saturation.min_parity(a1) < a_parity) if ((a_parity = saturation.min_parity(a)) != saturation.max_parity(a) || saturation.min_parity(a1) < a_parity)
return { p, false, {} }; // We need the parity of a and this has to be for sure less than the parity of a1 return { p, false }; // We need the parity of a and this has to be for sure less than the parity of a1
if (b.is_zero())
return { b1, true };
svector<signed_constraint> precondition; #if 0
pdd a_pi = get_pseudo_inverse(a, a_parity); pdd a_pi = get_pseudo_inverse(a, a_parity);
#else
pdd a_pi = s.pseudo_inv(a);
//precondition.insert(~s.eq(a_pi * a, rational::power_of_two(a_parity))); // TODO: This is unfortunately not a justification as the inverse might not be set yet (Can we make it to one?)
precondition.insert(~s.parity_at_most(a, a_parity));
#endif
pdd shift = a;
if (a_parity > 0) { if (a_parity > 0) {
pdd shift = s.lshr(a1, a1.manager().mk_val(a_parity)); shift = s.lshr(a1, a1.manager().mk_val(a_parity));
precondition.push_back(s.eq(rational::power_of_two(a_parity) * shift, a1)); // TODO: Or s.parity_at_least(a1, a_parity) but we want to reuse the variable introduced by the shift precondition.insert(~s.eq(rational::power_of_two(a_parity) * shift, a1)); // TODO: Or s.parity_at_least(a1, a_parity) but we want to reuse the variable introduced by the shift
return { a_pi * (-b) * shift + b1, true, {std::move(precondition)} };
} }
// Special case: If it is already odd we can directly use the pseudo inverse (as it is the inverse in this case!) LOG("Forced elimination: " << a_pi * (-b) * shift + b1);
return { a_pi * (-b) * a + b1, true, {std::move(precondition)} }; LOG("a: " << a);
LOG("a1: " << a1);
LOG("parity of a: " << a_parity);
LOG("pseudo inverse: " << a_pi);
LOG("-b: " << (-b));
LOG("shifted a" << shift);
LOG("Forced elimination: " << a_pi * (-b) * shift + b1);
return { a_pi * (-b) * shift + b1, true };
#else #else
unsigned a_parity; unsigned a_parity;
unsigned a1_parity; unsigned a1_parity;
if ((a_parity = saturation.min_parity(a)) != saturation.max_parity(a) || (a1_parity = saturation.min_parity(a1)) != saturation.max_parity(a1)) if ((a_parity = saturation.min_parity(a)) != saturation.max_parity(a) || (a1_parity = saturation.min_parity(a1)) != saturation.max_parity(a1))
return { p, false, {} }; // We need the parity, but we failed to get it precisely return { p, false }; // We need the parity, but we failed to get it precisely
if (a_parity > a1_parity) { if (a_parity > a1_parity) {
SASSERT(false); // get_multiple should have excluded this case already SASSERT(false); // get_multiple should have excluded this case already
return { p, false, {} }; return { p, false };
} }
svector<signed_constraint> precondition;
auto odd_a = get_odd(a, a_parity, precondition); auto odd_a = get_odd(a, a_parity, precondition);
auto odd_a1 = get_odd(a1, a1_parity, precondition); auto odd_a1 = get_odd(a1, a1_parity, precondition);
pdd inv_odd_a = get_inverse(odd_a); pdd inv_odd_a = get_inverse(odd_a);
@ -723,7 +735,7 @@ namespace polysat {
LOG("Forced elimination: " << odd_a1 * inv_odd_a * rational::power_of_two(a1_parity - a_parity) * b + b1); LOG("Forced elimination: " << odd_a1 * inv_odd_a * rational::power_of_two(a1_parity - a_parity) * b + b1);
verbose_stream() << "Forced elimination: " << odd_a1 * inv_odd_a * rational::power_of_two(a1_parity - a_parity) * (-b) + b1 << "\n"; verbose_stream() << "Forced elimination: " << odd_a1 * inv_odd_a * rational::power_of_two(a1_parity - a_parity) * (-b) + b1 << "\n";
verbose_stream() << "From: " << "eliminated v" << x << " with a = " << a << "; -b = " << -b << "; p = " << p << "\n"; verbose_stream() << "From: " << "eliminated v" << x << " with a = " << a << "; -b = " << -b << "; p = " << p << "\n";
return { odd_a1 * inv_odd_a * rational::power_of_two(a1_parity - a_parity) * (-b) + b1, true, {std::move(precondition)} }; return { odd_a1 * inv_odd_a * rational::power_of_two(a1_parity - a_parity) * (-b) + b1, true };
#endif #endif
#endif #endif
} }

View file

@ -87,8 +87,8 @@ namespace polysat {
pdd get_pseudo_inverse(const pdd& p, unsigned parity); pdd get_pseudo_inverse(const pdd& p, unsigned parity);
pdd get_inverse(const pdd& p); pdd get_inverse(const pdd& p);
pdd get_odd(const pdd& p, unsigned parity, svector<signed_constraint>& pat); pdd get_odd(const pdd& p, unsigned parity, clause_builder& precondition);
std::tuple<pdd, bool, svector<signed_constraint>> eliminate_variable(saturation& saturation, pvar x, const pdd& a, const pdd& b, const pdd& p); std::tuple<pdd, bool> eliminate_variable(saturation& saturation, pvar x, const pdd& a, const pdd& b, const pdd& p, clause_builder& precondition);
}; };
} }

View file

@ -154,7 +154,7 @@ bool rational::mult_inverse(unsigned num_bits, rational & result) const {
} }
/** /**
* Compute multiplicative pseudo-inverse modulo 2^num_bits: * Compute the smallest multiplicative pseudo-inverse modulo 2^num_bits:
* *
* mod(n * n.pseudo_inverse(bits), 2^bits) == 2^k, * mod(n * n.pseudo_inverse(bits), 2^bits) == 2^k,
* where k is maximal such that 2^k divides n. * where k is maximal such that 2^k divides n.
@ -167,7 +167,7 @@ rational rational::pseudo_inverse(unsigned num_bits) const {
SASSERT(!n.is_zero()); // TODO: or we define pseudo-inverse of 0 as 0. SASSERT(!n.is_zero()); // TODO: or we define pseudo-inverse of 0 as 0.
unsigned const k = n.trailing_zeros(); unsigned const k = n.trailing_zeros();
rational const odd = machine_div2k(n, k); rational const odd = machine_div2k(n, k);
VERIFY(odd.mult_inverse(num_bits, result)); VERIFY(odd.mult_inverse(num_bits - k, result));
SASSERT_EQ(mod(n * result, rational::power_of_two(num_bits)), rational::power_of_two(k)); SASSERT_EQ(mod(n * result, rational::power_of_two(num_bits)), rational::power_of_two(k));
return result; return result;
} }