z3-src 503.0.0

Source distribution of the Z3 SMT solver, for use as a build dependency
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207

/*++
Copyright (c) 2026 Romain Soulat

Module Name:

    ff_polynomial.h

Abstract:

    Sparse multivariate polynomial representation (monomial -> rational
    coefficient maps) over Q, with input-constraint provenance tracking,
    used by the Groebner-basis elimination engine for prime fields.

Author:

    Romain Soulat

--*/
#pragma once
#include "util/rational.h"
#include "util/rlimit.h"
#include "util/lbool.h"
#include "util/statistics.h"
#include <map>
#include <vector>
#include <set>

namespace ff {
    using monomial = std::vector<unsigned>;
    struct monomial_order {
        bool operator()(monomial const &a, monomial const &b) const {
            return a.size() != b.size() ? a.size() > b.size() : a > b;
        }
    };
    struct polynomial : public std::map<monomial, rational, monomial_order> {
        // Input-constraint provenance for conflict clauses. This is not a proof:
        // v2 additionally needs the polynomial multipliers witnessing each step.
        std::set<unsigned> dependencies;
        // Degree of the homogenized computation, including cancelled terms.
        // This is scheduling metadata only, never an algebraic premise.
        unsigned sugar = 0;
    };
    struct exhausted {};
    // Exact, bounded memoization of basis computations. Entries contain only
    // polynomial data and numeric premise indices, never context-owned ASTs.
    struct basis_cache {
        struct entry {
            rational prime;
            std::vector<polynomial> input, output;
        };
        std::vector<entry> entries;
        unsigned hits = 0, misses = 0;
        void clear() {
            entries.clear();
        }
    };

    // Backend-independent modular polynomial arithmetic. All transformations below
    // are ideal operations or invertible variable eliminations. A future certificate
    // recorder can attach polynomial-combination witnesses at add_scaled/reduce.
    struct f4_config;
    struct f4_stats;

    class engine {
        // Unit tests exercise bounded matrix admission directly, without an
        // unrelated Groebner pair schedule deciding whether the cap is reached.
        friend struct test_engine;
        friend class certificate_builder;
        rational p;
        reslimit &limit;
        unsigned work = 0, max_work, max_terms;
        unsigned random_state = 17;
        enum work_phase { other_work, scalar_work, insertion_work, pair_work, symbolic_work, matrix_work, num_work_phases };
        unsigned m_phase_work[num_work_phases] = {};
        work_phase m_work_phase = other_work;
        unsigned m_phase_start = 0;
        // Exclusive work attribution: nested scopes charge only their own
        // ticks, including the failing tick on resource exhaustion. Scope
        // transitions do not consume work or alter shared cancellation.
        class scoped_work {
            engine &e;
            work_phase previous;
            void enter(work_phase phase) {
                e.m_phase_work[e.m_work_phase] += e.work - e.m_phase_start;
                e.m_phase_start = e.work;
                e.m_work_phase = phase;
            }
        public:
            scoped_work(engine &e, work_phase phase) : e(e), previous(e.m_work_phase) { enter(phase); }
            ~scoped_work() { enter(previous); }
            scoped_work(scoped_work const &) = delete;
            scoped_work &operator=(scoped_work const &) = delete;
        };
        bool bit_propagation, batch_enabled, sparse_enabled;
        basis_cache *memo = nullptr;
        unsigned m_bit_facts = 0, m_bit_rounds = 0;
        unsigned m_eliminations = 0, m_substitutions = 0, m_substituted_terms = 0;
        unsigned m_basis_calls = 0, m_basis_pairs = 0, m_root_calls = 0;
        unsigned m_chain_skips = 0, m_fused_reductions = 0;
        unsigned m_batches = 0, m_matrix_rows = 0;
        unsigned m_extra_matrix_reducers = 0, m_lazy_matrix_reducers = 0, m_sparse_matrix_reducers = 0;
        unsigned m_sparse_trials = 0, m_sparse_witnesses = 0;
        unsigned m_step_exhaustions = 0, m_term_exhaustions = 0, m_basis_exhaustions = 0, m_matrix_exhaustions = 0;
        // Sub-counters identify the unchanged guard that ended an attempt.
        // Peaks are per engine (not process RSS); wrapper statistics may sum
        // them across independent engine calls rather than take a global max.
        unsigned m_local_work_exhaustions = 0, m_shared_limit_exhaustions = 0;
        unsigned m_polynomial_terms_exhaustions = 0, m_monomial_degree_exhaustions = 0;
        unsigned m_active_basis_exhaustions = 0, m_basis_slots_exhaustions = 0;
        unsigned m_basis_storage_exhaustions = 0, m_pair_storage_exhaustions = 0;
        unsigned m_pair_compactions = 0, m_discarded_pairs = 0;
        size_t m_peak_basis_bytes = 0, m_peak_pair_bytes = 0, m_peak_active_basis = 0;
        size_t m_basis_storage_limit = 16 * 1024 * 1024, m_pair_storage_limit = 16 * 1024 * 1024;
        unsigned m_matrix_symbolic_exhaustions = 0, m_matrix_reducer_exhaustions = 0;
        unsigned m_matrix_column_exhaustions = 0, m_matrix_row_exhaustions = 0, m_matrix_pivot_exhaustions = 0;
        size_t m_peak_matrix_columns = 0, m_peak_matrix_reducers = 0, m_peak_matrix_row_terms = 0;
        size_t m_peak_matrix_symbolic_bytes = 0, m_peak_matrix_coefficient_bytes = 0, m_peak_matrix_dependency_bytes = 0;
        std::vector<polynomial> batch_reduce(std::vector<polynomial> const &rows,
                                             std::vector<polynomial> const &basis);
        std::set<unsigned> m_conflict;
        unsigned m_deferred_eliminations = 0, m_scalar_fallbacks = 0, m_gm_skips = 0, m_mask_skips = 0, m_bucket_reductions = 0, m_small_products = 0;
        rational coefficient_product(rational const &a, rational const &b);
        rational coefficient_residue(rational const &a);
        void configure_probe(engine &probe) const;
        void tick();
        rational random_value();
        polynomial substitute(polynomial const &f, unsigned v, polynomial const &value);
        polynomial remainder(polynomial f, polynomial const &divisor);
        polynomial gcd(polynomial a, polynomial b);
        polynomial power_mod(polynomial a, rational exponent, polynomial const &modulus);
        void split_roots(polynomial const &f, unsigned v, std::vector<rational> &out);
        lbool solve_core(std::vector<polynomial> eqs, std::vector<polynomial> neqs, std::vector<rational> &values,
                         unsigned depth);

        void split_consequences(std::vector<polynomial> &eqs);
        lbool tiny_solve(std::vector<polynomial> const &eqs, std::vector<polynomial> const &neqs,
                         std::vector<rational> &values);
        polynomial minimal_polynomial(unsigned v, std::vector<polynomial> const &bs);
        bool small_quotient(std::vector<polynomial> const &bs, std::set<unsigned> &vars);
        bool quotient_field_basis(std::vector<polynomial> &bs);
        unsigned m_split_facts = 0, m_minpolys = 0, m_model_probes = 0, m_bound_facts = 0;
        unsigned m_quotient_probes = 0, m_quotient_facts = 0, m_quotient_exhaustions = 0;
    public:
        std::set<unsigned> definition_variables;
        bool bounded_elimination = false;
        bool adaptive_reduction = false;
        bool adaptive_matrix = false;
        bool adaptive_basis = false;
        bool lazy_matrix = false;
        bool sparse_matrix_reducers = false;
        bool fused_reduction = false;
        bool sugar_pairs = false, gm_pairs = false, div_masks = false, geobucket = false, small_coefficients = false, compact_encoding = false;
        bool linear_split = false, basis_bits = false, compact_matrix = false, model_search = false, bit_bounds = false;
        bool root_completion = false, quotient_field = false;
        // Fixed-width F4 backend with zero-dimensional model construction.
        bool f4 = true;
        unsigned f4_max_quotient = 1024;
        unsigned f4_budget_factor = 40;  // F4 work units allowed per legacy step (algebraic systems)
        unsigned f4_short_budget = 1;     // same, for circuit-like systems with many Boolean variables
        unsigned f4_bit_heavy = 0;
        bool m_bit_heavy_input = false;
        bool f4_value_split = false;
        unsigned f4_slice = 4;  // random slices per level for positive-dimensional ideals (SAT only)
        // Complete finite-domain search over tiny fields (p < 64).
        bool tiny_search = true;
        unsigned tiny_budget = 200;  // search work units per legacy step
        unsigned tiny_calls = 0, tiny_sat = 0, tiny_unsat = 0, tiny_undef = 0;
        uint64_t tiny_nodes = 0;
        bool f4_handback = false;  // also hand inconclusive F4 bases to the legacy heuristics over small fields
        uint64_t f4_steps = 0;
        unsigned f4_calls = 0, f4_sat = 0, f4_unsat = 0, f4_undef = 0;
        std::vector<unsigned> f4_counters = std::vector<unsigned>(11, 0);
        engine(rational const &p, reslimit &limit, unsigned max_work = 200000, unsigned max_terms = 4096,
               bool bit_propagation = true, bool batch_enabled = true, bool sparse_enabled = true)
            : p(p), limit(limit), max_work(max_work), max_terms(max_terms), bit_propagation(bit_propagation),
              batch_enabled(batch_enabled), sparse_enabled(sparse_enabled) {}
        void add_term(polynomial &f, monomial const &mon, rational const &c);
        polynomial constant(rational const &c);
        polynomial variable(unsigned v);
        polynomial add(polynomial a, polynomial const &b, rational const &scale = rational(1));
        polynomial mul(polynomial const &a, polynomial const &b);
        polynomial scale(polynomial a, rational const &c);
        rational inverse(rational a);
        rational evaluate(polynomial const &a, std::vector<rational> const &values);
        polynomial reduce(polynomial f, std::vector<polynomial> const &basis);
        void basis(std::vector<polynomial> &eqs);
        unsigned propagate_bits(std::vector<polynomial> &eqs);
        void set_basis_storage_limit(unsigned bytes) {
            // Keep accumulation safe on 32-bit hosts, and preserve an absolute
            // bound even when a caller requests the largest unsigned value.
            m_basis_storage_limit = std::min<size_t>(bytes, 1024u * 1024u * 1024u);
        }
        void set_basis_cache(basis_cache *c) {
            memo = c;
        }
        lbool solve(std::vector<polynomial> const &eqs, std::vector<polynomial> const &neqs,
                    std::vector<rational> &values);
        std::set<unsigned> const &conflict() const {
            return m_conflict;
        }
        bool polynomial_limit_hit() const { return m_term_exhaustions != 0; }
        unsigned steps() const {
            return work;
        }
        void collect_statistics(statistics &st) const;
    };
}  // namespace ff