void run(size_t bitLen)
{
if (bitLen < 8) {
printf("too small bitLen=%d\n", (int)bitLen);
exit(1);
}
const size_t N = size_t(1) << bitLen;
RandomGenerator rg;
printf("%09llx\n", (long long)N);
puts("init");
Vec8 v;
v.resize(N);
for (size_t i = 0; i < 256; i++) {
v[i] = uint8_t(i);
}
for (size_t i = 256; i < N; i++) {
v[i] = uint8_t(rg());
}
puts("start");
bench<cybozu::WaveletMatrix>("wm", v, N);
#ifdef COMPARE_WAT
bench<Wat>("wat", v, N);
#endif
#ifdef COMPARE_WAVELET
bench<Wavelet>("wavelet", v, N);
#endif
#ifdef COMPARE_SHELLINFORD
bench<Shellinford>("shellinford", v, N);
#endif
}
void init(const Vec8& v8, int)
{
std::vector<uint64_t> v64;
v64.resize(v8.size());
for (size_t i = 0; i < v8.size(); i++) {
v64[i] = v8[i];
}
wm.Init(v64);
}
void bench_select(const T& wm, const Vec8& v8, RG& rg, size_t C)
{
cybozu::disable_warning_unused_variable(v8);
size_t ret = 0;
std::vector<int> maxTbl;
maxTbl.resize(256);
for (int i = 0; i < 256; i++) {
int v = (int)wm.size(i);
if (v == 0) v = 1;
maxTbl[i] = v;
}
double begin = cybozu::GetCurrentTimeSec();
for (size_t i = 0; i < C; i++) {
uint8_t c = uint8_t(rg());
size_t pos = rg() % maxTbl[c];
uint64_t a = wm.select(c, pos);
#if 0
uint64_t b = v8.select(c, pos);
if (a != b) {
printf("ERR i=%d a=%d b=%d c=%d pos=%d\n", (int)i, (int)a, (int)b, (int)c, (int)pos);
exit(1);
}
#endif
ret += a;
}
double t = cybozu::GetCurrentTimeSec() - begin;
printf("select %08x %9.2fusec\n", (int)ret, t / C * 1e6);
}
void put() const
{
putSub();
if (freqTbl.empty()) return;
const uint32_t compSize = (uint32_t)vec.size();
for (size_t i = 0; i < freqTbl.size(); i++) {
printf("freqTbl[%2d] = %8d(%5.2f%%, %5.2f%%)\n", (int)i, freqTbl[i], freqTbl[i] * 100.0 / compSize, freqTbl[i] * encTbl[i].len * 100.0 / bitSize_);
}
}
void putSub() const
{
const uint32_t inSize = bitSize_ / 8;
const uint32_t compSize = (uint32_t)vec.size();
const uint32_t idxSize = (uint32_t)(blkVec.size() * sizeof(blkVec[0]));
const double cr = compSize * 100.0 / inSize;
const double ir = idxSize * 100.0 / inSize;
if (inSize == 0) return;
printf("in Size= %9d, rank=%u\n", inSize, rk_);
printf("comp Size= %9u\n", compSize);
printf("idx Size= %9u(blkVec.size=%7u)\n", idxSize, (uint32_t)blkVec.size());
printf("totalSize= %9u\n", compSize + idxSize);
printf("rate=%5.2f%%(%5.2f%% + %5.2f%%)\n", cr + ir, cr, ir);
}
void initBlockVec()
{
blkVec.reserve(bitSize_ / skip + 16);
uint32_t orgPos = 0;
uint32_t rk = 0;
uint32_t samplingPos = 0;
for (size_t vecPos = 0, n = vec.size(); vecPos < n; vecPos++) {
uint8_t v = vec[vecPos];
uint32_t next = orgPos + encTbl[v].len;
while (samplingPos < next) {
blkVec.push_back(Block(orgPos, (uint32_t)vecPos, rk));
samplingPos += skip;
}
orgPos = next;
rk += encTbl[v].rk;
}
}
OutputStream(Vec32& freqTbl, Vec8& vec, uint32_t& rk, const uint64_t *buf, uint32_t bitSize, const EncodingTbl& encTbl)
: freqTbl(freqTbl)
, vec(vec)
, rk(rk)
, bi(encTbl)
, encTbl(encTbl)
{
csucvector_util::InputStream is(buf, bitSize);
freqTbl.clear();
freqTbl.resize(encTbl.size());
vec.clear();
rk = 0;
for (;;) {
uint32_t s = append(is);
is.consume(s);
if (is.empty()) break;
}
printf("bitSize=%u\n",bitSize);
}
void bench_rank(const T& wm, const Vec8& v8, RG& rg, size_t C, size_t N)
{
cybozu::disable_warning_unused_variable(v8);
size_t ret = 0;
double begin = cybozu::GetCurrentTimeSec();
for (size_t i = 0; i < C; i++) {
size_t pos = rg() & (N - 1);
uint8_t c = uint8_t(rg());
uint64_t a = wm.rank(c, pos);
#if 0
uint64_t b = v8.rank(c, pos);
if (a != b) {
printf("ERR i=%d a=%d b=%d c=%d pos=%d\n", (int)i, (int)a, (int)b, (int)c, (int)pos);
exit(1);
}
#endif
ret += a;
}
double t = cybozu::GetCurrentTimeSec() - begin;
printf("rank %08x %9.2fusec\n", (int)ret, t / C * 1e6);
}