void testPow()
{
#ifdef NDEBUG
const size_t n = 1000;
#else
const size_t n = 1;
#endif
Vuint m = 2, d = 2;
for (size_t i = 0; i < 160 - 1; i++) {
m *= d;
}
m += 7;
ZmZ<>::setModulo(m);
TEST_EQUAL(m, Vuint("1461501637330902918203684832716283019655932542983"));
ZmZ<> x, y = ZmZ<>(m) - 1;
Xbyak::util::Clock clk;
for (size_t i = 0; i < n; i++) {
x = 3;
clk.begin();
x = power(x, y);
clk.end();
}
TEST_EQUAL(x, 1);
// clk=1428899.430clk (VariableBuffer)
// clk= 555733.590clk (FixedBuffer)
printf("clk=%.3fclk\n", clk.getClock() / double(clk.getCount()));
}
void bench()
{
Xbyak::util::Clock clk;
const size_t N = 1000;
const size_t M = 1000;
Vuint m("14615016373309029182036848327162830196559325429831461501636150163733090291820368483271628301965593254298314615016373309029182036848327162830196559325429831461501637330902918203684832716283019655932542983");
ZmZ<>::setModulo(m);
// const char *str = "123456789012342342342499924234242422333333333333333333333256789112345678901234234234242423424242233333333333333333333325678911234567890123423423424242342424223333333333333333333332567891";
const char *str = "82434016654300679721217353503190038836571781811386228921167322412819029493182";
// ZmZ<> a(str);
Vuint a(str), b(a);
for (size_t i = 0; i < N; i++) {
clk.begin();
for (size_t j = 0; j < M; j++) {
// a *= a;
Vuint::mul(b, a, a); // 4268clk
// Vuint::mul1(b, a, 123456789); // 128clk
// Vuint::add(b, a, a); // 113clk
// local::PrimitiveFunction::add1(&b[0], &a[0], a.size(), 1234566);
// local::PrimitiveFunction::mul1(&b[0], &a[0], a.size(), 1234566);
// Vuint::sub(b, a, a); // 140clk
}
clk.end();
}
printf("clk=%.3fclk\n", clk.getClock() / double(clk.getCount()) / M);
}
Ret benchmark1(const std::basic_string<C>& str, const std::basic_string<C>& key, F f)
{
const int N = 1;
int val = 0;
f.set(str, key);
Xbyak::util::Clock clk;
for (int i = 0; i < N; i++) {
typename F::type p = f.begin();
typename F::type end = f.end();
for (;;) {
clk.begin();
typename F::type q = f.find(p);
clk.end();
if (q == end) break;
val += (int)(q - p);
p = q + 1;
}
}
if (val == 0) val = (int)(str.size()) * N;
Ret ret;
ret.val = val;
ret.clk = clk.getClock() / (double)val;
return ret;
}
double getDummyLoopClock(size_t n, size_t bitLen)
{
uint64_t ret = 0;
Xbyak::util::Clock clk;
#ifdef USE_C11
g_rg.seed(0);
std::uniform_int_distribution<uint64_t> dist(0, (1ULL << bitLen) - 1);
#else
XorShift128 r;
const uint64_t mask = (1ULL << bitLen) - 1;
#endif
const int lp = 5;
for (int i = 0; i < lp; i++) {
clk.begin();
for (size_t i = 0; i < n; i++) {
#ifdef USE_C11
uint64_t v = dist(g_rg);
#else
uint64_t v = r.get64();
v += r.get() >> 5;
v &= mask;
#endif
ret += v;
}
clk.end();
}
printf("(%llx)", (long long)ret);
return clk.getClock() / double(n) / lp;
}
void test(const char *msg, void (*f)())
{
Xbyak::util::Clock clk;
counter = 0;
clk.begin();
f();
clk.end();
printf("%s %.2f clk/loop counter=%lld\n", msg, clk.getClock() / double(clk.getCount() * N), (long long)counter);
}
void test(const std::string& s, F f)
{
const int M = 100;
uint64_t x = 0;
Xbyak::util::Clock clk;
for (int i = 0; i < M; i++) {
clk.begin();
x += f((const uint8_t*)s.c_str(), s.size());
clk.end();
}
printf("x=%16lld, %.2fclk\n", (long long)x, clk.getClock() / double(M) / s.size());
}
int main()
{
const int count = 1000;
Xbyak::util::Clock clk;
Code c;
void (*f)() = (void (*)())c.getCode();
for (int i = 0; i < count; i++) {
clk.begin();
f();
clk.end();
}
printf("%.3fclk\n", clk.getClock() / double(N) / clk.getCount());
}
void test(size_t n, bool doPut, F pred)
{
IntVec iv;
PtrVec pv;
init(iv, pv, n);
if (doPut) put(pv);
Xbyak::util::Clock clk;
clk.begin();
std::sort(pv.begin(), pv.end(), pred);
clk.end();
int sum = std::accumulate(iv.begin(), iv.end(), 0);
printf("clk=%.2fclk, sum=%d\n", clk.getClock() / double(n), sum);
if (doPut) put(pv);
}
int main(int argc, char *argv[])
{
int mode = argc == 1 ? -1 : atoi(argv[1]);
for (int i = 0; i < 2; i++) {
if (mode >= 0 && mode != i) continue;
Code c(i == 0);
void (*f)() = c.getCode<void (*)()>();
Xbyak::util::Clock clk;
clk.begin();
f();
clk.end();
printf("%.2fclk\n", clk.getClock() / double(N));
}
}
void test(int mode)
{
printf("mode:%6s ", mode2str(mode));
memset(data, 0, sizeof(data));
double time = 0;
{
Xbyak::util::Clock clk;
clk.begin();
std::thread t1(write1[mode]);
std::thread t2(write2[mode]);
t1.join();
t2.join();
clk.end();
time = (double)clk.getClock();
}
int num[4] = { };
for (int i = 0; i < N; i++) {
const Data& d = data[i];
if (d.sa == 0 || d.sb == 0) {
printf("ERR %d %d\n", d.sa, d.sb);
exit(1);
}
if (d.r1 == 0 && d.r2 == 0) {
num[0]++;
} else
if (d.r1 == 1 && d.r2 == 0) {
num[1]++;
} else
if (d.r1 == 0 && d.r2 == 1) {
num[2]++;
} else
{
num[3]++;
}
}
int sum = 0;
printf("num=");
for (int i = 0; i < 4; i++) {
printf("%8d ", num[i]);
sum += num[i];
}
if (sum != N) {
fprintf(stderr, "ERR sum=%d\n", sum);
exit(1);
}
printf("clk=%7.3fMclk\n", time * 1e-6);
}
void bench(int mode)
{
const int N = 100000;
Code code;
code.makeBench(N, mode);
int (*p)(uint64_t*, const uint64_t*, const uint64_t*) = (int (*)(uint64_t*, const uint64_t*, const uint64_t*))code.getCode();
uint64_t a[4] = { uint64_t(-1), uint64_t(-2), uint64_t(-3), 544443221 };
uint64_t b[4] = { uint64_t(-123), uint64_t(-3), uint64_t(-4), 222222222 };
uint64_t c[5] = { 0, 0, 0, 0, 0 };
const int M = 100;
Xbyak::util::Clock clk;
for (int i = 0; i < M; i++) {
clk.begin();
p(c, a, b);
clk.end();
}
printf("%.2fclk\n", clk.getClock() / double(M) / double(N) / innerN);
}
uint64_t bench(const uint64_t *block, size_t blockNum, size_t n, size_t bitLen, double baseClk, bool useSelect)
{
const T sbv(block, blockNum);
uint64_t ret = 0;
Xbyak::util::Clock clk;
#ifdef USE_C11
std::uniform_int_distribution<uint64_t> dist(0, (1ULL << bitLen) - 1);
#else
XorShift128 r;
const uint64_t mask = (1ULL << bitLen) - 1;
#endif
const int lp = 5;
for (int j = 0; j < lp; j++) {
clk.begin();
for (size_t i = 0; i < n; i++) {
#ifdef USE_C11
uint64_t v = dist(g_rg);
#else
uint64_t v = r.get64();
v &= mask;
#endif
ret += sbv.rank1(v);
}
clk.end();
}
printf("%11lld ret %08x %6.2f clk(%6.2f)\n", 1LL << bitLen, (int)ret, (double)clk.getClock() / double(n) / lp - baseClk, baseClk);
if (useSelect) {
clk.clear();
const size_t maxNum = sbv.rank1(blockNum * 64 - 1);
for (int j = 0; j < lp; j++) {
clk.begin();
for (size_t i = 0; i < n; i++) {
#ifdef USE_C11
uint64_t v = dist(g_rg);
#else
uint64_t v = r.get64();
v %= maxNum;
#endif
ret += sbv.select1(v);
}
clk.end();
}
printf("%11lld ret %08x %6.2f clk(%6.2f)\n", 1LL << bitLen, (int)ret, (double)clk.getClock() / double(n) / lp - baseClk, baseClk);
}
return ret;
}