結果
| 問題 |
No.5022 XOR Printer
|
| ユーザー |
 Ang1077
|
| 提出日時 | 2025-07-26 16:36:58 |
| 言語 | C++23 (gcc 13.3.0 + boost 1.87.0) |
| 結果 |
AC
|
| 実行時間 | 1,983 ms / 2,000 ms |
| コード長 | 13,491 bytes |
| コンパイル時間 | 4,982 ms |
| コンパイル使用メモリ | 343,360 KB |
| 実行使用メモリ | 7,720 KB |
| スコア | 5,193,786,499 |
| 最終ジャッジ日時 | 2025-07-26 16:39:07 |
| 合計ジャッジ時間 | 107,122 ms |
|
ジャッジサーバーID （参考情報） |
judge4 / judge5 |
| 純コード判定しない問題か言語 |
（要ログイン）
| ファイルパターン | 結果 |
|---|---|
| other | AC * 50 |
ソースコード
// #pragma GCC optimize
// "-O3,omit-frame-pointer,inline,unroll-all-loops,fast-math" #pragma GCC target
// "tune=native"
#include <bits/stdc++.h>
#include <ext/pb_ds/assoc_container.hpp>
using namespace std;
// ==== Typedefs / utils ====
using u8 = uint8_t;
using u32 = uint32_t;
using u64 = uint64_t;
using i64 = long long;
struct RNG {
uint64_t s[2];
RNG() {
reset((uint64_t)chrono::steady_clock::now().time_since_epoch().count());
}
static inline uint64_t rotl(uint64_t x, int k) {
return (x << k) | (x >> (64 - k));
}
void reset(uint64_t seed) {
auto sm = [s = seed]() mutable {
s += 0x9E3779B97f4A7C15ull;
u64 z = s;
z = (z ^ (z >> 30)) * 0xBF58476D1CE4E5B9ull;
z = (z ^ (z >> 27)) * 0x94D049BB133111EBull;
return z ^ (z >> 31);
};
s[0] = sm();
s[1] = sm();
}
uint64_t next() {
uint64_t s0 = s[0], s1 = s[1];
uint64_t r = rotl(s0 * 5, 7) * 9;
s1 ^= s0;
s[0] = rotl(s0, 24) ^ s1 ^ (s1 << 16);
s[1] = rotl(s1, 37);
return r;
}
double randomDouble() { return (double)(uint32_t)next() / 4294967296.0; }
uint32_t random32(uint32_t r) {
return (uint64_t)(uint32_t)next() * r >> 32;
}
} rng;
struct timer {
chrono::high_resolution_clock::time_point t0;
timer() { t0 = chrono::high_resolution_clock::now(); }
float elapsed() const {
return chrono::duration<float>(chrono::high_resolution_clock::now() -
t0)
.count();
}
};
struct EvalResult {
u64 sum = 0;
vector<char> ops;
vector<u32> Aflat; // final matrix (flattened by row-major)
};
// ==== Simulator (fast eval + emitting version) ====
struct Simulator {
int N, T, NN; // N=10, NN=100
vector<u32> A0; // flattened A
vector<int> xr, yc; // coordinates per id
vector<int> dist; // dist[a*NN + b] = Manhattan
// MSB-based pruning for 1C
array<vector<int>, 20>
msb_bucket; // indices of cells with given MSB (0..19)
array<array<vector<int>, 20>, 100>
near_msb; // near_msb[target][bit] = top-L closest ids
// working state (per simulation)
vector<u32> A; // working grid (flattened)
bitset<100> written; // whether W already done
int cur; // current id
u32 s; // register
int used; // used turns
inline int DID(int a, int b) const { return dist[a * NN + b]; }
void init_precompute(int N_, int T_, const vector<vector<u32>> &A2d,
int L = 12) {
N = N_;
T = T_;
NN = N * N;
A0.resize(NN);
xr.resize(NN);
yc.resize(NN);
int id = 0;
for (int i = 0; i < N; i++)
for (int j = 0; j < N; j++) {
A0[id] = A2d[i][j];
xr[id] = i;
yc[id] = j;
++id;
}
// distances
dist.assign(NN * NN, 0);
for (int a = 0; a < NN; a++)
for (int b = 0; b < NN; b++)
dist[a * NN + b] = abs(xr[a] - xr[b]) + abs(yc[a] - yc[b]);
// msb buckets
for (int b = 0; b < 20; b++)
msb_bucket[b].clear();
for (int i2 = 0; i2 < NN; i2++) {
u32 v = A0[i2];
if (v == 0)
continue;
int b = 31 - __builtin_clz(v);
if (0 <= b && b < 20)
msb_bucket[b].push_back(i2);
}
// near_msb precompute
for (int t = 0; t < NN; t++) {
for (int b = 0; b < 20; b++) {
auto &src = msb_bucket[b];
auto cmp = [&](int p, int q) { return DID(p, t) < DID(q, t); };
if (src.empty()) {
near_msb[t][b].clear();
continue;
}
vector<int> tmp = src;
if ((int)tmp.size() > L) {
nth_element(tmp.begin(), tmp.begin() + L, tmp.end(), cmp);
tmp.resize(L);
sort(tmp.begin(), tmp.end(), cmp);
} else {
sort(tmp.begin(), tmp.end(), cmp);
}
near_msb[t][b] = move(tmp);
}
}
}
// candidate action for a target id, using current (cur, s, used)
struct Cand {
long long val;
int cost;
int mode;
int pr;
}; // mode: 0=0C, 1=1C, -1=none
inline Cand best_action_for_target(int t) const {
Cand best{LLONG_MIN, INT_MAX, -1, -1};
int rem = T - used;
// 0C
{
int c0 = DID(cur, t) + 1;
if (c0 <= rem) {
u32 nv = (A[t] ^ s);
if ((long long)nv > (long long)A[t]) {
best = {(long long)nv, c0, 0, -1};
}
}
}
// 1C: focus on first highest zero-bit of base
u32 base = A[t] ^ s;
int k = -1;
for (int b = 19; b >= 0; --b) {
if (((base >> b) & 1u) == 0u) {
k = b;
break;
}
}
auto try_bucket = [&](int bit) {
if (bit < 0)
return;
auto const &cand = near_msb[t][bit];
for (int pr : cand) {
int c1 = DID(cur, pr) + 1 + DID(pr, t) + 1;
if (c1 > rem)
continue;
u32 s2 = s ^ A[pr];
u32 nv = (A[t] ^ s2);
if ((long long)nv <= (long long)A[t])
continue;
if (nv > (u32)best.val ||
(nv == (u32)best.val && c1 < best.cost)) {
best = {(long long)nv, c1, 1, pr};
}
}
};
if (k >= 0) {
try_bucket(k);
if (best.mode < 0)
try_bucket(k - 1);
}
return best;
}
// --- FAST evaluation (no ops emission) ---
u64 simulate_value(const vector<int> &order) {
A = A0;
written.reset();
cur = 0;
s = 0;
used = 0;
bool progressed = true;
while (progressed && used < T) {
progressed = false;
for (int id : order) {
if (used >= T)
break;
if (written.test(id))
continue;
Cand cand = best_action_for_target(id);
if (cand.mode < 0)
continue;
if (cand.mode == 0) {
// move -> W
if (used + cand.cost > T)
continue;
used += DID(cur, id);
cur = id;
used += 1; // W
A[id] ^= s;
written.set(id);
progressed = true;
} else {
// move pr -> C -> move t -> W
if (used + cand.cost > T)
continue;
used += DID(cur, cand.pr);
cur = cand.pr;
used += 1; // C
s ^= A[cand.pr];
used += DID(cur, id);
cur = id;
used += 1; // W
A[id] ^= s;
written.set(id);
progressed = true;
}
}
}
u64 sum = 0;
for (int i = 0; i < NN; i++)
sum += A[i];
return sum;
}
// --- Full simulation with ops emission (UDLR + C/W) ---
inline void emit_move_path(int from, int to, vector<char> &ops_out) {
int dr = xr[to] - xr[from];
int dc = yc[to] - yc[from];
if (dr >= 0)
for (int k = 0; k < dr; ++k)
ops_out.push_back('D');
else
for (int k = 0; k < -dr; ++k)
ops_out.push_back('U');
if (dc >= 0)
for (int k = 0; k < dc; ++k)
ops_out.push_back('R');
else
for (int k = 0; k < -dc; ++k)
ops_out.push_back('L');
}
EvalResult simulate_emit(const vector<int> &order) {
A = A0;
written.reset();
cur = 0;
s = 0;
used = 0;
vector<char> ops_out;
ops_out.reserve(6000);
bool progressed = true;
while (progressed && used < T) {
progressed = false;
for (int id : order) {
if (used >= T)
break;
if (written.test(id))
continue;
Cand cand = best_action_for_target(id);
if (cand.mode < 0)
continue;
if (used + cand.cost > T)
continue;
if (cand.mode == 0) {
// move to t
emit_move_path(cur, id, ops_out);
used += DID(cur, id);
cur = id;
// W
ops_out.push_back('W');
++used;
A[id] ^= s;
written.set(id);
progressed = true;
} else {
// move to pr
emit_move_path(cur, cand.pr, ops_out);
used += DID(cur, cand.pr);
cur = cand.pr;
// C
ops_out.push_back('C');
++used;
s ^= A[cand.pr];
// move to t
emit_move_path(cur, id, ops_out);
used += DID(cur, id);
cur = id;
// W
ops_out.push_back('W');
++used;
A[id] ^= s;
written.set(id);
progressed = true;
}
}
}
EvalResult res;
res.ops = move(ops_out);
res.Aflat = A;
res.sum = 0;
for (int i = 0; i < NN; i++)
res.sum += A[i];
return res;
}
};
// ==== Main with Simulated Annealing (adjacent swap / 2-point swap / 1-point
// insertion) ====
int main() {
ios::sync_with_stdio(false);
cin.tie(nullptr);
int N, T;
if (!(cin >> N >> T))
return 0;
vector<vector<u32>> A2(N, vector<u32>(N));
for (int i = 0; i < N; i++)
for (int j = 0; j < N; j++)
cin >> A2[i][j];
T = 1000;
Simulator sim;
sim.init_precompute(N, T, A2, /*L=*/12);
vector<int> order(sim.NN);
iota(order.begin(), order.end(), 0);
timer tim;
const double LIM = 1.98;
const double T0 = 1e8, T1 = 1e-3;
// initial
u64 cur_sum = sim.simulate_value(order);
u64 best_sum = cur_sum;
vector<int> best_order = order;
EvalResult best_emit =
sim.simulate_emit(order); // initial feasible sequence
// SA loop
while (tim.elapsed() < LIM) {
// temperature
double ft = tim.elapsed() / LIM;
double temp = T0 * pow(T1 / T0, ft);
// Neighborhood: 0=adjacent swap, 1=two-point swap, 2=one-point
// insertion
vector<int> cand = order;
u32 kind = rng.random32(3);
if (kind == 0) {
// adjacent swap
if ((int)cand.size() >= 2) {
int i = rng.random32((uint32_t)max(1, (int)cand.size() - 1));
swap(cand[i], cand[i + 1]);
}
} else if (kind == 1) {
// two-point swap (i != j)
int n = (int)cand.size();
if (n >= 2) {
int i = rng.random32((uint32_t)n);
int j = rng.random32((uint32_t)(n - 1));
if (j >= i)
++j;
swap(cand[i], cand[j]);
}
} else {
// one-point insertion
int n = (int)cand.size();
int i = rng.random32((uint32_t)n);
int j = rng.random32((uint32_t)(n - 1));
if (j >= i)
++j; // ensure j != i
int v = cand[i];
if (i < j) {
for (int k = i; k < j; ++k)
cand[k] = cand[k + 1];
cand[j] = v;
} else {
for (int k = i; k > j; --k)
cand[k] = cand[k - 1];
cand[j] = v;
}
}
u64 cand_sum = sim.simulate_value(cand);
long long delta = (long long)cand_sum - (long long)cur_sum;
bool accept = (delta >= 0) || (rng.randomDouble() < exp(delta / temp));
if (accept) {
order.swap(cand);
cur_sum = cand_sum;
if (cur_sum > best_sum) {
best_sum = cur_sum;
best_order = order;
best_emit = sim.simulate_emit(best_order); // only for the best
}
}
}
// --- Debug output to stderr ---
cerr << best_emit.ops.size() << '\n';
u64 out_sum = 0;
for (int i = 0; i < sim.NN; i++) {
cerr << best_emit.Aflat[i] << ((i % N == N - 1) ? '\n' : ' ');
out_sum += best_emit.Aflat[i];
}
cerr << out_sum << '\n';
// --- Answer (ops to stdout, one per line) ---
for (char ch : best_emit.ops)
cout << ch << '\n';
return 0;
}