kate/BadFish
1
0
Fork 0
mirror of https://github.com/sockspls/badfish synced 2025-07-11 19:49:14 +00:00
Code Issues Releases Wiki Activity

Clean up decompress_pairs

Browse source 
Also replace platform specific byte swap with
a software version. Amazingly it seems it is
even faster now!

Also removing the templatized form does not slow down.
This commit is contained in:
Marco Costalba 2016-04-23 08:17:36 +02:00
parent 5414eaba35
commit 7bc5946fc9
1 changed files with 39 additions and 50 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

73
src/syzygy/tbprobe.cpp
View file

@ -36,14 +36,6 @@
#include <windows.h>
#endif
#ifndef _MSC_VER
#define BSWAP32(v) __builtin_bswap32(v)
#define BSWAP64(v) __builtin_bswap64(v)
#else
#define BSWAP32(v) _byteswap_ulong(v)
#define BSWAP64(v) _byteswap_uint64(v)
#endif
#define WDLSUFFIX ".rtbw"
#define DTZSUFFIX ".rtbz"
#define TBPIECES 6
@ -356,6 +348,7 @@ const short KK_idx[10][64] = {
const uint8_t WDL_MAGIC[] = { 0x71, 0xE8, 0x23, 0x5D };
const uint8_t DTZ_MAGIC[] = { 0xD7, 0x66, 0x0C, 0xA5 };
const int wdl_to_dtz[] = { -1, -101, 0, 101, 1 };
const int wdl_to_map[] = { 1, 3, 0, 2, 0 };
const uint8_t pa_flags[] = { 8, 0, 0, 0, 4 };
@ -1326,36 +1319,49 @@ int init_table_dtz(TBEntry *entry)
return 1;
}
template<bool LittleEndian>
uint8_t decompress_pairs(PairsData *d, uint64_t idx)
template<typename T, int Half = sizeof(T)/2, int End = sizeof(T)-1>
inline void byteSwap(T& x)
{
char tmp, *c = (char*)(&x);
for (int i = 0; i < Half; ++i)
tmp = c[i], c[i] = c[End-i], c[End-i] = tmp;
}
int decompress_pairs(PairsData* d, uint64_t idx)
{
const union { uint32_t i; char c[4]; } LE = { 0x01020304 };
const bool LittleEndian = (LE.c[0] == 4);
if (!d->idxbits)
return uint8_t(d->min_len);
return d->min_len;
uint32_t mainidx = (uint32_t)(idx >> d->idxbits);
// idx = blockidx | litidx where litidx is a signed number of lenght d->idxbits
uint32_t blockidx = (uint32_t)(idx >> d->idxbits);
int litidx = (idx & ((1ULL << d->idxbits) - 1)) - (1ULL << (d->idxbits - 1));
uint32_t block = *(uint32_t *)(d->indextable + 6 * mainidx);
if (!LittleEndian)
block = BSWAP32(block);
// indextable points to an array of blocks of 6 bytes representing numbers in
// little endian. The low 4 bytes are the block, the high 2 bytes the idxOffset.
uint32_t block = *(uint32_t *)(d->indextable + 6 * blockidx);
uint16_t idxOffset = *(uint16_t *)(d->indextable + 6 * blockidx + 4);
uint16_t idxOffset = *(uint16_t *)(d->indextable + 6 * mainidx + 4);
if (!LittleEndian)
idxOffset = uint16_t((idxOffset << 8) | (idxOffset >> 8));
if (!LittleEndian) {
byteSwap(block);
byteSwap(idxOffset);
}
litidx += idxOffset;
if (litidx < 0) {
do {
while (litidx < 0)
litidx += d->sizetable[--block] + 1;
} while (litidx < 0);
} else {
while (litidx > d->sizetable[block])
litidx -= d->sizetable[block++] + 1;
}
uint32_t* ptr = (uint32_t*)(d->data + (block << d->blocksize));
uint64_t code = *((uint64_t*)ptr);
if (LittleEndian)
byteSwap(code);
int m = d->min_len;
uint16_t *offset = d->offset;
@ -1363,11 +1369,6 @@ uint8_t decompress_pairs(PairsData *d, uint64_t idx)
uint8_t* symlen = d->symlen;
int sym, bitcnt;
uint64_t code = *((uint64_t *)ptr);
if (LittleEndian)
code = BSWAP64(code);
ptr += 2;
bitcnt = 0; // number of "empty bits" in code
@ -1396,7 +1397,7 @@ uint8_t decompress_pairs(PairsData *d, uint64_t idx)
uint32_t tmp = *ptr++;
if (LittleEndian)
tmp = BSWAP32(tmp);
byteSwap(tmp);
code |= (uint64_t)tmp << bitcnt;
}
@ -1419,14 +1420,6 @@ uint8_t decompress_pairs(PairsData *d, uint64_t idx)
return sympat[3 * sym];
}
uint8_t decompress_pairs(PairsData *d, uint64_t idx)
{
const union { uint32_t i; char c[4]; } bint = {0x01020304};
const bool isLittleEndian = (bint.c[0] == 4);
return isLittleEndian ? decompress_pairs<true >(d, idx)
: decompress_pairs<false>(d, idx);
}
void load_dtz_table(const std::string& str, uint64_t key1, uint64_t key2)
{
DTZ_table[0].key1 = key1;
@ -2059,10 +2052,6 @@ static int probe_dtz_no_ep(Position& pos, int *success)
}
}
static int wdl_to_dtz[] = {
-1, -101, 0, 101, 1
};
// Probe the DTZ table for a particular position.
// If *success != 0, the probe was successful.
// The return value is from the point of view of the side to move: