/*++ Copyright (c) 2021 Microsoft Corporation Module Name: maintain viable domains Author: Nikolaj Bjorner (nbjorner) 2021-03-19 Jakob Rath 2021-04-06 Notes: --*/ #include "util/debug.h" #include "util/log.h" #include "sat/smt/polysat/viable.h" #include "sat/smt/polysat/core.h" #include "sat/smt/polysat/number.h" #include "sat/smt/polysat/refine.h" #include "sat/smt/polysat/project_interval.h" #include "sat/smt/polysat/ule_constraint.h" namespace polysat { using dd::val_pp; viable::viable(core& c) : c(c), cs(c.cs()), m_forbidden_intervals(c), m_projection(c), m_fixed_bits(c) {} viable::~viable() { for (auto* e : m_alloc) dealloc(e); } std::ostream& operator<<(std::ostream& out, find_t f) { switch (f) { case find_t::empty: return out << "empty"; case find_t::singleton: return out << "singleton"; case find_t::multiple: return out << "multiple"; case find_t::resource_out: return out << "resource-out"; default: return out << ""; } } struct viable::pop_viable_trail : public trail { viable& m_s; entry* e; entry_kind k; public: pop_viable_trail(viable& s, entry* e, entry_kind k) : m_s(s), e(e), k(k) {} void undo() override { m_s.pop_viable(e, k); } }; struct viable::push_viable_trail : public trail { viable& m_s; entry* e; public: push_viable_trail(viable& s, entry* e) : m_s(s), e(e) {} void undo() override { m_s.push_viable(e); } }; viable::entry* viable::alloc_entry(pvar var, constraint_id constraint_index) { entry* e = nullptr; if (m_alloc.empty()) e = alloc(entry); else { e = m_alloc.back(); m_alloc.pop_back(); } e->reset(); e->var = var; e->constraint_index = constraint_index; return e; } bool viable::assign(pvar v, rational const& value, dependency dep) { m_var = v; m_explain_kind = explain_t::none; m_num_bits = c.size(v); m_fixed_bits.init(v); m_explain.reset(); m_assign_dep = null_dependency; init_overlaps(v); bool first = true; while (true) { for (auto const& [w, offset] : m_overlaps) { for (auto& layer : m_units[w].get_layers()) { entry* e = find_overlap(w, layer, value); if (!e) continue; m_explain.push_back({ e, value }); m_explain_kind = explain_t::assignment; m_assign_dep = dep; return false; } } if (!first) return true; first = false; if (!check_fixed_bits(v, value)) continue; if (!check_disequal_lin(v, value)) continue; if (!check_equal_lin(v, value)) continue; break; } return true; } find_t viable::find_viable(pvar v, rational& lo) { m_explain.reset(); m_var = v; m_num_bits = c.size(v); m_fixed_bits.init(v); init_overlaps(v); m_explain_kind = explain_t::none; m_assign_dep = null_dependency; for (unsigned rounds = 0; rounds < 10; ) { entry* n = find_overlap(lo); if (m_explain_kind == explain_t::conflict) return find_t::empty; if (n) continue; if (!check_fixed_bits(v, lo)) continue; ++rounds; if (!check_disequal_lin(v, lo)) continue; if (!check_equal_lin(v, lo)) continue; if (is_propagation(lo)) { m_explain_kind = explain_t::propagation; return find_t::singleton; } return find_t::multiple; } TRACE("bv", display(tout << "resource-out v" << v << "\n")); return find_t::resource_out; } void viable::init_overlaps(pvar v) { m_overlaps.reset(); c.get_bitvector_suffixes(v, m_overlaps); std::sort(m_overlaps.begin(), m_overlaps.end(), [&](auto const& x, auto const& y) { return c.size(x.child) < c.size(y.child); }); } // // // from smallest size(w) overlap [w] to largest // from smallest bit_width layer [bit_width, entries] to largest // check if val is allowed by entries or advance val to next allowed value // viable::entry* viable::find_overlap(rational& val) { entry* last = nullptr; for (auto const& [w, offset] : m_overlaps) { for (auto& layer : m_units[w].get_layers()) { entry* e = find_overlap(w, layer, val); if (!e) continue; last = e; if (e->interval.is_proper()) update_value_to_high(val, e); m_explain.push_back({ e, val }); if (is_conflict()) { m_explain_kind = explain_t::conflict; return nullptr; } } } return last; } viable::entry* viable::find_overlap(pvar w, layer& l, rational const& val) { if (!l.entries) return nullptr; unsigned v_width = m_num_bits; unsigned b_width = l.bit_width; if (v_width == b_width) return find_overlap(val, l.entries); rational val1 = mod(val, rational::power_of_two(b_width)); return find_overlap(val1, l.entries); } void viable::update_value_to_high(rational& val, entry* e) { unsigned v_width = m_num_bits; unsigned b_width = e->bit_width; SASSERT(b_width <= v_width); auto const& hi = e->interval.hi_val(); auto const& lo = e->interval.lo_val(); if (b_width == v_width) { val = hi; SASSERT(0 <= val && val <= c.var2pdd(m_var).max_value()); return; } auto p2b = rational::power_of_two(b_width); rational val2 = clear_lower_bits(val, b_width); if (lo <= mod(val, p2b) && hi < lo) { val2 += p2b; if (val2 == rational::power_of_two(v_width)) val2 = 0; } val = val2 + hi; SASSERT(0 <= hi && hi < p2b); SASSERT(0 <= val && val <= c.var2pdd(m_var).max_value()); } /* * In either case we are checking a constraint $v[u-1:0]\not\in[lo, hi[$ * where $u := w'-k-1$ and using it to compute $\forward(v)$. * Thus, suppose $v[u-1:0] \in [lo, hi[$. * - $lo < hi$: $\forward(v) := \forward(2^u v[w-1:w-u] + hi)$. * - $lo > hi$, $v[w-1:w-u]+1 = 2^{w-u}$: $\forward(v) := \bot$ * - $lo > hi$, $v[w-1:w-u]+1 < 2^{w-u}$: $\forward(v) := \forward(2^u (v[w-1:w-u]+1) + hi)$ */ // Find interval that contains 'val', or, if no such interval exists, null. viable::entry* viable::find_overlap(rational const& val, entry* entries) { SASSERT(entries); entry* const first = entries; entry* e = entries; do { auto const& i = e->interval; if (i.currently_contains(val)) return e; entry* const n = e->next(); // there is only one interval, and it does not contain 'val' if (e == n) return nullptr; // check whether 'val' is contained in the gap between e and n bool const overlapping = e->interval.currently_contains(n->interval.lo_val()); if (!overlapping && r_interval::contains(e->interval.hi_val(), n->interval.lo_val(), val)) return nullptr; e = n; } while (e != first); UNREACHABLE(); return nullptr; } bool viable::check_equal_lin(pvar v, rational const& val) { // LOG_H2("refine-equal-lin with v" << v << ", val = " << val); entry const* e = m_equal_lin[v]; if (!e) return true; entry const* first = e; auto& m = c.var2pdd(v); unsigned const N = m.power_of_2(); rational const& max_value = m.max_value(); rational const& mod_value = m.two_to_N(); SASSERT(0 <= val && val <= max_value); // Rotate the 'first' entry, to prevent getting stuck in a refinement loop // with an early entry when a later entry could give a better interval. m_equal_lin[v] = m_equal_lin[v]->next(); do { rational coeff_val = mod(e->coeff * val, mod_value); if (e->interval.currently_contains(coeff_val)) { // IF_LOGGING( // verbose_stream() << "refine-equal-lin for v" << v << " in src: "; // for (const auto& src : e->src) // verbose_stream() << lit_pp(s, src) << "\n"; // ); // LOG("forbidden interval v" << v << " " << num_pp(s, v, val) << " " << num_pp(s, v, e->coeff, true) << " * " << e->interval); if (mod(e->interval.hi_val() + 1, mod_value) == e->interval.lo_val()) { // We have an equation: a * v == b rational const a = e->coeff; rational const b = e->interval.hi_val(); LOG("refine-equal-lin: equation detected: " << dd::val_pp(m, a, true) << " * v" << v << " == " << dd::val_pp(m, b, false)); unsigned const parity_a = get_parity(a, N); unsigned const parity_b = get_parity(b, N); if (parity_a > parity_b) { // No solution LOG("refined: no solution due to parity"); entry* ne = alloc_entry(v, e->constraint_index); ne->refined = true; ne->src = e->src; ne->side_cond = e->side_cond; ne->coeff = 1; ne->bit_width = e->bit_width; ne->interval = eval_interval::full(); intersect(v, ne); return false; } if (parity_a == 0) { // "fast path" for odd a rational a_inv; VERIFY(a.mult_inverse(N, a_inv)); rational const hi = mod(a_inv * b, mod_value); rational const lo = mod(hi + 1, mod_value); // LOG("refined to [" << num_pp(c, v, lo) << ", " << num_pp(c, v, hi) << "["); SASSERT_EQ(mod(a * hi, mod_value), b); // hi is the solution entry* ne = alloc_entry(v, e->constraint_index); ne->refined = true; ne->src = e->src; ne->side_cond = e->side_cond; ne->coeff = 1; ne->bit_width = e->bit_width; ne->interval = eval_interval::proper(m.mk_val(lo), lo, m.mk_val(hi), hi); SASSERT(ne->interval.currently_contains(val)); intersect(v, ne); return false; } // 2^k * v == a_inv * b // 2^k solutions because only the lower N-k bits of v are fixed. // // Smallest solution is v0 == a_inv * (b >> k) // Solutions are of the form v_i = v0 + 2^(N-k) * i for i in { 0, 1, ..., 2^k - 1 }. // Forbidden intervals: [v_i + 1; v_{i+1}[ == [ v_i + 1; v_i + 2^(N-k) [ // We need the interval that covers val: // v_i + 1 <= val < v_i + 2^(N-k) // // TODO: create one interval for v[N-k:] instead of 2^k intervals for v. unsigned const k = parity_a; rational const a_inv = a.pseudo_inverse(N); unsigned const N_minus_k = N - k; rational const two_to_N_minus_k = rational::power_of_two(N_minus_k); rational const v0 = mod(a_inv * machine_div2k(b, k), two_to_N_minus_k); SASSERT(mod(val, two_to_N_minus_k) != v0); // val is not a solution rational const vi = v0 + clear_lower_bits(mod(val - v0, mod_value), N_minus_k); rational const lo = mod(vi + 1, mod_value); rational const hi = mod(vi + two_to_N_minus_k, mod_value); // LOG("refined to [" << num_pp(c, v, lo) << ", " << num_pp(c, v, hi) << "["); SASSERT_EQ(mod(a * (lo - 1), mod_value), b); // lo-1 is a solution SASSERT_EQ(mod(a * hi, mod_value), b); // hi is a solution entry* ne = alloc_entry(v, e->constraint_index); ne->refined = true; ne->src = e->src; ne->side_cond = e->side_cond; ne->coeff = 1; ne->bit_width = e->bit_width; ne->interval = eval_interval::proper(m.mk_val(lo), lo, m.mk_val(hi), hi); SASSERT(ne->interval.currently_contains(val)); intersect(v, ne); return false; } // TODO: special handling for the even factors of e->coeff = 2^k * a', a' odd // (create one interval for v[N-k:] instead of 2^k intervals for v) // TODO: often, the intervals alternate between short forbidden and short allowed intervals. // we should be able to handle this similarly to compute_y_bounds, // and be able to represent such periodic intervals (inside safe bounds). // // compute_y_bounds calculates with inclusive upper bound, so we need to adjust argument and result accordingly. rational const hi_val_incl = e->interval.hi_val().is_zero() ? max_value : (e->interval.hi_val() - 1); auto [lo, hi] = refine_equal::compute_y_bounds(val, e->coeff, e->interval.lo_val(), hi_val_incl, mod_value); hi += 1; //LOG("refined to [" << num_pp(c, v, lo) << ", " << num_pp(c, v, hi) << "["); // verbose_stream() << "lo=" << lo << " val=" << val << " hi=" << hi << "\n"; if (lo <= hi) { SASSERT(0 <= lo && lo <= val && val < hi && hi <= mod_value); } else { SASSERT(0 < hi && hi < lo && lo < mod_value && (val < hi || lo <= val)); } bool full = (lo == 0 && hi == mod_value); if (hi == mod_value) hi = 0; entry* ne = alloc_entry(v, e->constraint_index); ne->refined = true; ne->src = e->src; ne->side_cond = e->side_cond; ne->coeff = 1; ne->bit_width = e->bit_width; if (full) ne->interval = eval_interval::full(); else ne->interval = eval_interval::proper(m.mk_val(lo), lo, m.mk_val(hi), hi); SASSERT(ne->interval.currently_contains(val)); intersect(v, ne); return false; } e = e->next(); } while (e != first); return true; } bool viable::check_fixed_bits(pvar v, rational const& val) { auto e = alloc_entry(v, constraint_id::null()); if (m_fixed_bits.check(val, *e)) { m_alloc.push_back(e); return true; } else { TRACE("bv", tout << "fixed " << val << " " << *e << "\n"); if (!intersect(v, e)) { display(verbose_stream()); display_explain(verbose_stream() << "explain\n"); UNREACHABLE(); SASSERT(false); } return false; } } bool viable::check_disequal_lin(pvar v, rational const& val) { // LOG_H2("refine-disequal-lin with v" << v << ", val = " << val); entry const* e = m_diseq_lin[v]; if (!e) return true; entry const* first = e; auto& m = c.var2pdd(v); rational const& max_value = m.max_value(); rational const& mod_value = m.two_to_N(); SASSERT(0 <= val && val <= max_value); // Rotate the 'first' entry, to prevent getting stuck in a refinement loop // with an early entry when a later entry could give a better interval. m_diseq_lin[v] = m_diseq_lin[v]->next(); do { // We compute an interval if the concrete value 'val' violates the constraint: // p*val + q > r*val + s if e->src.is_positive() // p*val + q >= r*val + s if e->src.is_negative() // Note that e->interval is meaningless in this case, // we just use it to transport the values p,q,r,s rational const& p = e->interval.lo_val(); rational const& q_ = e->interval.lo().val(); rational const& r = e->interval.hi_val(); rational const& s_ = e->interval.hi().val(); SASSERT(p != r && p != 0 && r != 0); SASSERT(e->src.size() == 1); rational const a = mod(p * val + q_, mod_value); rational const b = mod(r * val + s_, mod_value); rational const np = mod_value - p; rational const nr = mod_value - r; int const corr = e->src[0].is_negative() ? 1 : 0; auto delta_l = [&](rational const& val) { rational num = a - b + corr; rational l1 = floor(b / r); rational l2 = val; if (p > r) l2 = ceil(num / (p - r)) - 1; rational l3 = ceil(num / (p + nr)) - 1; rational l4 = ceil((mod_value - a) / np) - 1; rational d1 = l3; rational d2 = std::min(l1, l2); rational d3 = std::min(l1, l4); rational d4 = std::min(l2, l4); rational dmax = std::max(std::max(d1, d2), std::max(d3, d4)); return std::min(val, dmax); }; auto delta_u = [&](rational const& val) { rational num = a - b + corr; rational h1 = floor(b / nr); rational h2 = max_value - val; if (r > p) h2 = ceil(num / (r - p)) - 1; rational h3 = ceil(num / (np + r)) - 1; rational h4 = ceil((mod_value - a) / p) - 1; rational d1 = h3; rational d2 = std::min(h1, h2); rational d3 = std::min(h1, h4); rational d4 = std::min(h2, h4); rational dmax = std::max(std::max(d1, d2), std::max(d3, d4)); return std::min(max_value - val, dmax); }; if (a > b || (e->src[0].is_negative() && a == b)) { rational lo = val - delta_l(val); rational hi = val + delta_u(val) + 1; LOG("refine-disequal-lin: " << " [" << lo << ", " << hi << "["); SASSERT(0 <= lo && lo <= val); SASSERT(val <= hi && hi <= mod_value); if (hi == mod_value) hi = 0; pdd lop = c.var2pdd(v).mk_val(lo); pdd hip = c.var2pdd(v).mk_val(hi); entry* ne = alloc_entry(v, e->constraint_index); ne->refined = true; ne->src = e->src; ne->side_cond = e->side_cond; ne->coeff = 1; ne->bit_width = e->bit_width; ne->interval = eval_interval::proper(lop, lo, hip, hi); intersect(v, ne); return false; } e = e->next(); } while (e != first); return true; } /* * The current explanation trail is a conflict if the top-most entry * is repeated below and there is no entry with higher bit-width between. */ bool viable::is_conflict() { auto last = m_explain.back(); unsigned bw = last.e->bit_width; if (last.e->interval.is_full()) { m_explain.reset(); m_explain.push_back(last); return true; } for (unsigned i = m_explain.size() - 1; i-- > 0; ) { auto e = m_explain[i]; if (bw < e.e->bit_width) return false; if (last.e == e.e) return true; } return false; } bool viable::is_propagation(rational const& val) { // disable for now return false; if (m_explain.empty()) return false; auto last = m_explain.back(); auto first = m_explain[0]; if (first.e->interval.lo_val() == val + 1 && last.e->interval.hi_val() == val && first.e->bit_width == last.e->bit_width && first.e->bit_width == c.size(m_var)) { return true; } return false; } std::ostream& operator<<(std::ostream& out, viable::explain_t e) { switch(e) { case viable::explain_t::conflict: return out << "conflict"; case viable::explain_t::propagation: return out << "propagation"; case viable::explain_t::assignment: return out << "assignment"; case viable::explain_t::none: return out << "none"; default: UNREACHABLE(); } return out; } /* * Explain why the current variable is not viable or * or why it can only have a single value. */ dependency_vector viable::explain() { dependency_vector result; auto last = m_explain.back(); auto after = last; verbose_stream() << "viable::explain: " << m_explain_kind << " v" << m_var << "\n"; if (c.inconsistent()) verbose_stream() << "inconsistent explain\n"; TRACE("bv", display_explain(tout)); auto unmark = [&]() { for (auto e : m_explain) e.e->marked = false; }; auto explain_entry = [&](entry* e) { auto index = e->constraint_index; if (c.inconsistent()) return; if (e->marked) return; e->marked = true; if (m_var != e->var) result.push_back(offset_claim(m_var, { e->var, 0 })); for (auto const& sc : e->side_cond) { auto d = c.propagate(sc, c.explain_weak_eval(sc), "entry weak eval"); if (c.inconsistent()) { verbose_stream() << "inconsistent " << d << " " << sc << "\n"; } result.push_back(d); } result.append(e->deps); if (!index.is_null()) { VERIFY_EQ(e->src.size(), 1); VERIFY_EQ(c.get_constraint(index), e->src[0]); result.push_back(c.get_dependency(index)); } }; if (last.e->interval.is_full()) { SASSERT(m_explain_kind == explain_t::none); explain_entry(last.e); SASSERT(m_explain.size() == 1); unmark(); return result; } SASSERT(m_explain_kind != explain_t::none); for (unsigned i = m_explain.size() - 1; i-- > 0; ) { auto e = m_explain[i]; explain_overlap(e, after, result); after = e; explain_entry(e.e); if (e.e == last.e) break; } if (m_explain_kind == explain_t::propagation) { // assume first and last have same bit-width auto first = m_explain[0]; SASSERT(first.e->bit_width == last.e->bit_width); explain_entry(first.e); // add constraint that there is only a single viable value. auto sc = cs.eq(last.e->interval.hi() + 1, first.e->interval.lo()); auto exp = c.propagate(sc, c.explain_weak_eval(sc)); if (c.inconsistent()) { verbose_stream() << "inconsistent " << sc << " " << exp << "\n"; } result.push_back(exp); } if (m_explain_kind == explain_t::assignment) { // there is just one entry SASSERT(m_explain.size() == 1); SASSERT(!m_assign_dep.is_null()); entry* e = last.e; explain_entry(e); // assignment conflict and propagation from containing slice depends on concrete values, // so we also need the evaluations of linear terms SASSERT(!c.is_assigned(m_var)); // assignment of m_var is justified by m_assign_dep auto index = last.e->constraint_index; if (!index.is_null()) result.append(c.explain_weak_eval(c.get_constraint(index))); // 'result' so far contains explanation for e and its weak evaluation m_projection.init(e->var, e->interval, e->bit_width, result); switch (m_projection()) { case l_true: // propagated interval onto subslice m_projection.explain(result); break; case l_false: // conflict (projected interval is full) m_projection.explain(result); break; case l_undef: // unable to propagate interval to containing slice // fallback explanation uses assignment of m_var result.push_back(m_assign_dep); break; }; } unmark(); if (c.inconsistent()) verbose_stream() << "inconsistent after explain\n"; return result; } /* * For two consecutive intervals * * - 2^kx \not\in [lo, hi[, * - 2^k'y \not\in [lo', hi'[ * - a value v such that * - 2^kv \not\in [lo, hi[ * - 2^k'v \in [lo', hi'[ * - hi \in ] (v - 1) * 2^{bw - ebw} ; v * 2^{bw - ebw} ] * * Where: * - bw is the width of x, aw the width of y * - ebw is the bit-width of x, abw the bit-width of y * - k = bw - ebw, k' = aw - abw * * We want to encode the constraint that (2^k' hi)[w'] in [lo', hi'[ * * Note that x in [lo, hi[ <=> x - lo < hi - lo * If bw = aw, ebw = abw there is nothing else to do. * - hi \in [lo', hi'[ * * If bw != aw or aw < bw: * - hi \in ] (v - 1) * 2^{bw - ebw} ; v * 2^{bw - ebw} ] * - hi := v mod aw * * - 2^k'hi \in [lo', hi'[ * */ void viable::explain_overlap(explanation const& e, explanation const& after, dependency_vector& deps) { auto bw = c.size(e.e->var); auto ebw = e.e->bit_width; auto aw = c.size(after.e->var); auto abw = after.e->bit_width; auto t = e.e->interval.hi(); auto lo = after.e->interval.lo(); auto hi = after.e->interval.hi(); SASSERT(abw <= aw); SASSERT(ebw <= bw); if (ebw < bw || aw != bw) { auto const& p2b = rational::power_of_two(bw); auto const& p2eb = rational::power_of_two(bw - ebw); // let coeff = 2^{bw - ebw} // let assume coeff * x \not\in [lo, t[ // Then value is chosen, min x . coeff * x >= t. // In other words: // // x >= t div coeff // => t <= coeff * x // (x - 1) * coeff < t <= x * coeff // a < x <= b <=> // a + 1 <= x < b + 1 // x - a - 1 < b - a auto vlo = c.value(mod((e.value - 1) * p2eb + 1, p2b), bw); auto vhi = c.value(mod(e.value * p2eb + 1, p2b), bw); #if 0 verbose_stream() << "value " << e.value << "\n"; verbose_stream() << "t " << t << "\n"; verbose_stream() << "[" << vlo << " " << vhi << "[\n"; verbose_stream() << "before bw " << ebw << " " << bw << " " << *e.e << "\nafter bw " << abw << " " << aw << " " << *after.e << "\n"; if (!t.is_val()) IF_VERBOSE(0, verbose_stream() << "symbolic t : " << t << "\n"); verbose_stream() << t - vlo << " " << vhi - vlo << "\n"; #endif auto sc = cs.ult(t - vlo, vhi - vlo); CTRACE("bv", sc.is_always_false(), c.display(tout)); VERIFY(!sc.is_always_false()); if (!sc.is_always_true()) deps.push_back(c.propagate(sc, c.explain_weak_eval(sc))); t.reset(lo.manager()); t = c.value(mod(e.value, rational::power_of_two(aw)), aw); // verbose_stream() << "after " << t << "\n"; if (c.inconsistent()) verbose_stream() << "inconsistent overlap " << sc << " " << "\n"; } if (abw < aw) t *= rational::power_of_two(aw - abw); auto sc = cs.ult(t - lo, hi - lo); SASSERT(!sc.is_always_false()); if (!sc.is_always_true()) deps.push_back(c.propagate(sc, c.explain_weak_eval(sc))); if (c.inconsistent()) { verbose_stream() << "inconsistent ult " << sc << " " << "\n"; verbose_stream() << "before bw " << ebw << " " << bw << " " << *e.e << "\nafter bw " << abw << " " << aw << " " << *after.e << "\n"; display(verbose_stream()); // UNREACHABLE(); } } /* * Register constraint at index 'idx' as unitary in v. * Returns 'multiple' when either intervals are unchanged or there really are multiple values left. */ find_t viable::add_unitary(pvar v, constraint_id idx, rational& var_value) { if (c.is_assigned(v)) return find_t::multiple; auto [sc, d, truth_value] = c.m_constraint_index[idx.id]; SASSERT(truth_value != l_undef); if (truth_value == l_false) sc = ~sc; if (!sc.is_linear()) return find_t::multiple; entry* ne = alloc_entry(v, idx); if (!m_forbidden_intervals.get_interval(sc, v, *ne)) { m_alloc.push_back(ne); return find_t::multiple; } // verbose_stream() << "v" << v << " " << sc << " " << ne->interval << "\n"; TRACE("bv", tout << "v" << v << " " << sc << " " << ne->interval << "\n"; display_one(tout, ne) << "\n"); if (ne->interval.is_currently_empty()) { m_alloc.push_back(ne); return find_t::multiple; } if (ne->coeff == 1) intersect(v, ne); else if (ne->coeff == -1) insert(ne, v, m_diseq_lin, entry_kind::diseq_e); else if (!ne->coeff.is_power_of_two()) insert(ne, v, m_equal_lin, entry_kind::equal_e); else if (ne->interval.is_full()) insert(ne, v, m_equal_lin, entry_kind::equal_e); else { unsigned const w = c.size(v); unsigned const k = ne->coeff.parity(w); SASSERT(k > 0); IF_VERBOSE(3, display_one(verbose_stream() << "try to reduce entry: ", ne) << "\n"); // reduction of coeff gives us a unit entry // // 2^k x \not\in [ lo ; hi [ // // new_lo = lo[w-1:k] if lo[k-1:0] = 0 // lo[w-1:k] + 1 otherwise // // new_hi = hi[w-1:k] if hi[k-1:0] = 0 // hi[w-1:k] + 1 otherwise // // Reference: Fig. 1 (dtrim) in BitvectorsMCSAT TRACE("bv", display_one(tout << "try to reduce entry: ", ne) << "\n"); pdd const& pdd_lo = ne->interval.lo(); pdd const& pdd_hi = ne->interval.hi(); rational const& lo = ne->interval.lo_val(); rational const& hi = ne->interval.hi_val(); rational twoK = rational::power_of_two(k); rational new_lo = machine_div2k(mod2k(lo + twoK - 1, w), k); pdd lo_eq = pdd_lo * rational::power_of_two(w - k); if (mod2k(lo, k).is_zero()) { if (!lo_eq.is_zero()) ne->side_cond.push_back(cs.eq(lo_eq)); } else { SASSERT(!lo_eq.is_val() || !lo_eq.is_zero()); if (!lo_eq.is_val()) ne->side_cond.push_back(~cs.eq(lo_eq)); } rational new_hi = machine_div2k(mod2k(hi + twoK - 1, w), k); pdd hi_eq = pdd_hi * rational::power_of_two(w - k); if (mod2k(hi, k).is_zero()) { if (!hi_eq.is_zero()) ne->side_cond.push_back(cs.eq(hi_eq)); } else { SASSERT(!hi_eq.is_val() || !hi_eq.is_zero()); if (!hi_eq.is_val()) ne->side_cond.push_back(~cs.eq(hi_eq)); } // // If new_lo = new_hi it means that // mod(ceil(lo / 2^k), 2^(w-k)) = mod(ceil(hi / 2^k), 2^(w-k)) // or // div(mod(lo + 2^k - 1, 2^w), 2^k) = div(mod(hi + 2^k - 1, 2^w), 2^k) // but we also have lo != hi. // // Cases: // 0 < lo < hi: empty (2^k does not divide any of [lo, hi[) // 0 == lo < hi: full // 0 < hi < lo: full // 0 == hi < lo: empty // if (new_lo == new_hi) { bool is_empty_ivl; if (lo < hi) { ne->side_cond.push_back(cs.ult(pdd_lo, pdd_hi)); if (lo == 0) { ne->side_cond.push_back(cs.eq(pdd_lo)); is_empty_ivl = false; } else { ne->side_cond.push_back(~cs.eq(pdd_lo)); is_empty_ivl = true; } } else { SASSERT(hi < lo); ne->side_cond.push_back(cs.ult(pdd_hi, pdd_lo)); if (hi == 0) { ne->side_cond.push_back(cs.eq(pdd_hi)); is_empty_ivl = true; } else { ne->side_cond.push_back(~cs.eq(pdd_hi)); is_empty_ivl = false; } } if (is_empty_ivl) { m_alloc.push_back(ne); return find_t::multiple; } else { ne->interval = eval_interval::full(); ne->coeff = 1; m_explain.reset(); m_explain.push_back({ ne, rational::zero() }); m_fixed_bits.reset(); m_var = v; return find_t::empty; } } ne->coeff = 1; ne->interval = eval_interval::proper(pdd_lo, new_lo, pdd_hi, new_hi); ne->bit_width -= k; TRACE("bv", display_one(tout << "reduced: ", ne) << "\n"); intersect(v, ne); } if (ne->interval.is_full()) { m_explain.reset(); m_explain.push_back({ ne, rational::zero() }); m_fixed_bits.reset(); m_var = v; return find_t::empty; } return find_viable(v, var_value); } void viable::ensure_var(pvar v) { while (v >= m_units.size()) { m_units.push_back(layers()); m_equal_lin.push_back(nullptr); m_diseq_lin.push_back(nullptr); } } bool viable::intersect(pvar v, entry* ne) { SASSERT(!c.is_assigned(v)); entry*& entries = m_units[v].ensure_layer(ne->bit_width).entries; entry* e = entries; if (e && e->interval.is_full()) { m_alloc.push_back(ne); return false; } if (ne->interval.is_currently_empty()) { m_alloc.push_back(ne); return false; } auto create_entry = [&]() { c.trail().push(pop_viable_trail(*this, ne, entry_kind::unit_e)); ne->init(ne); return ne; }; auto remove_entry = [&](entry* e) { c.trail().push(push_viable_trail(*this, e)); e->remove_from(entries, e); e->active = false; }; if (ne->interval.is_full()) { while (entries) remove_entry(entries); entries = create_entry(); return true; } if (!e) entries = create_entry(); else { entry* first = e; do { if (e->interval.currently_contains(ne->interval)) { m_alloc.push_back(ne); return false; } while (ne->interval.currently_contains(e->interval)) { entry* n = e->next(); remove_entry(e); if (!entries) { entries = create_entry(); return true; } if (e == first) first = n; e = n; } SASSERT(e->interval.lo_val() != ne->interval.lo_val()); if (e->interval.lo_val() > ne->interval.lo_val()) { if (first->prev()->interval.currently_contains(ne->interval)) { m_alloc.push_back(ne); return false; } e->insert_before(create_entry()); if (e == first) entries = e->prev(); SASSERT(well_formed(m_units[v])); return true; } e = e->next(); } while (e != first); // otherwise, append to end of list first->insert_before(create_entry()); } SASSERT(well_formed(m_units[v])); return true; } viable::layer& viable::layers::ensure_layer(unsigned bit_width) { for (unsigned i = 0; i < m_layers.size(); ++i) { layer& l = m_layers[i]; if (l.bit_width == bit_width) return l; else if (l.bit_width > bit_width) { m_layers.push_back(layer(0)); for (unsigned j = m_layers.size(); --j > i; ) m_layers[j] = m_layers[j - 1]; m_layers[i] = layer(bit_width); return m_layers[i]; } } m_layers.push_back(layer(bit_width)); return m_layers.back(); } viable::layer* viable::layers::get_layer(unsigned bit_width) { return const_cast(std::as_const(*this).get_layer(bit_width)); } viable::layer const* viable::layers::get_layer(unsigned bit_width) const { for (layer const& l : m_layers) if (l.bit_width == bit_width) return &l; return nullptr; } void viable::pop_viable(entry* e, entry_kind k) { unsigned v = e->var; SASSERT(well_formed(m_units[v])); SASSERT(e->active); e->active = false; switch (k) { case entry_kind::unit_e: entry::remove_from(m_units[v].get_layer(e)->entries, e); break; case entry_kind::equal_e: entry::remove_from(m_equal_lin[v], e); break; case entry_kind::diseq_e: entry::remove_from(m_diseq_lin[v], e); break; default: UNREACHABLE(); break; } SASSERT(well_formed(m_units[v])); m_alloc.push_back(e); } void viable::push_viable(entry* e) { // display_one(verbose_stream() << "Push entry: ", e) << "\n"; auto v = e->var; entry*& entries = m_units[v].get_layer(e)->entries; SASSERT(e->prev() != e || !entries); SASSERT(e->prev() != e || e->next() == e); SASSERT(!e->active); e->active = true; SASSERT(well_formed(m_units[v])); if (e->prev() != e) { entry* pos = e->prev(); e->init(e); pos->insert_after(e); if (!entries || e->interval.lo_val() < entries->interval.lo_val()) entries = e; } else entries = e; SASSERT(well_formed(m_units[v])); } void viable::insert(entry* e, pvar v, ptr_vector& entries, entry_kind k) { SASSERT(well_formed(m_units[v])); c.trail().push(pop_viable_trail(*this, e, k)); e->init(e); if (!entries[v]) entries[v] = e; else e->insert_after(entries[v]); SASSERT(entries[v]->invariant()); SASSERT(well_formed(m_units[v])); } std::ostream& viable::display_one(std::ostream& out, entry const* e) const { auto& m = c.var2pdd(e->var); if (e->coeff == -1) { // p*val + q > r*val + s if e->src.is_positive() // p*val + q >= r*val + s if e->src.is_negative() // Note that e->interval is meaningless in this case, // we just use it to transport the values p,q,r,s rational const& p = e->interval.lo_val(); rational const& q_ = e->interval.lo().val(); rational const& r = e->interval.hi_val(); rational const& s_ = e->interval.hi().val(); out << "[ "; out << val_pp(m, p, true) << "*v" << e->var << " + " << val_pp(m, q_); out << (e->src[0].is_positive() ? " > " : " >= "); out << val_pp(m, r, true) << "*v" << e->var << " + " << val_pp(m, s_); out << " ] "; } else { if (e->coeff != 1) out << e->coeff << " * "; out << "v" << e->var; if (e->bit_width != c.size(e->var)) out << "[" << e->bit_width << "]"; out << " " << e->interval << " "; } if (!e->deps.empty()) out << " deps " << e->deps; if (e->side_cond.empty()) ; else if (e->side_cond.size() <= 5) out << " side-conds " << e->side_cond; else out << " side-conds " << e->side_cond.size() << " side-conditions"; out << " src "; unsigned count = 0; for (const auto& src : e->src) { ++count; out << src << "; "; if (count > 10) { out << " ..."; break; } } return out; } std::ostream& viable::display_all(std::ostream& out, entry const* e, char const* delimiter) const { if (!e) return out; entry const* first = e; unsigned count = 0; do { display_one(out, e) << delimiter; e = e->next(); ++count; if (count > 10) { out << " ... (total: " << count << " entries)"; break; } } while (e != first); return out; } std::ostream& viable::display(std::ostream& out) const { for (unsigned v = 0; v < m_units.size(); ++v) { for (auto const& layer : m_units[v].get_layers()) { if (!layer.entries) continue; out << "v" << v << "[" << layer.bit_width << "]: "; display_all(out, layer.entries, " "); out << "\n"; } } return out; } std::ostream& viable::display_state(std::ostream& out) const { out << "v" << m_var << ":"; for (auto const& slice : m_overlaps) { out << " v" << slice.child << ":" << c.size(slice.child) << "@" << slice.offset; if (c.is_assigned(slice.child)) out << " value=" << c.get_assignment().value(slice.child); } out << "\n"; return out; } std::ostream& viable::display_explain(std::ostream& out) const { out << "explain_kind " << m_explain_kind << "\n"; display_state(out); for (auto const& e : m_explain) display_one(out << "v" << m_var << "[" << e.e->bit_width << "] := " << e.value << " ", e.e) << "\n"; return out; } /* * Lower bounds are strictly ascending. * Intervals don't contain each-other (since lower bounds are ascending, it suffices to check containment in one direction). */ bool viable::well_formed(entry* e) { if (!e) return true; entry* first = e; while (true) { CTRACE("bv", !e->active, tout << "inactive entry v" << e->var << " " << e->interval << "\n"; display(tout)); if (!e->active) return false; if (e->interval.is_full()) return e->next() == e; if (e->interval.is_currently_empty()) return false; auto* n = e->next(); if (n != e && e->interval.currently_contains(n->interval)) { TRACE("bv", tout << "currently contains\n"); return false; } if (n == first) break; if (e->interval.lo_val() >= n->interval.lo_val()) { TRACE("bv", tout << "lo-val >= n->lo_val\n"); return false; } e = n; } return true; } /* * Layers are ordered in strictly ascending bit-width. * Entries in each layer are well-formed. */ bool viable::well_formed(layers const& ls) { bool first = true; unsigned prev_width = 0; for (layer const& l : ls.get_layers()) { if (!well_formed(l.entries)) { TRACE("bv", tout << "entries are not well-formed\n"); return false; } if (!all_of(dll_elements(l.entries), [&l](entry const& e) { return e.bit_width == l.bit_width; })) { TRACE("bv", tout << "elements don't have same bit-width\n"); return false; } if (!first && prev_width >= l.bit_width) { TRACE("bv", tout << "previous width is " << prev_width << " vs " << l.bit_width << "\n"); return false; } first = false; prev_width = l.bit_width; } return true; } }