kate/BadFish
1
0
Fork 0
mirror of https://github.com/sockspls/badfish synced 2025-05-01 09:13:08 +00:00
Code Issues Releases Wiki Activity

Tweaks to bitcount functions

Browse source 
Seems even a bit faster now (almost 1% in 32bit case).

No functional change.
This commit is contained in:
Marco Costalba 2012-09-22 09:26:25 +02:00
parent 1cb2722c95
commit 9204a60dbb
1 changed files with 20 additions and 26 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

46
src/bitcount.h
View file

@ -33,8 +33,8 @@ enum BitCountType {
};
/// Determine at compile time the best popcount<> specialization according if
/// platform is 32 or 64 bits, to the maximum number of nonzero bits to count or
/// use hardware popcnt instruction when available.
/// platform is 32 or 64 bits, to the maximum number of nonzero bits to count
/// and if hardware popcnt instruction is available.
const BitCountType Full = HasPopCnt ? CNT_HW_POPCNT : Is64Bit ? CNT_64 : CNT_32;
const BitCountType Max15 = HasPopCnt ? CNT_HW_POPCNT : Is64Bit ? CNT_64_MAX15 : CNT_32_MAX15;
@ -44,44 +44,38 @@ template<BitCountType> inline int popcount(Bitboard);
template<>
inline int popcount<CNT_64>(Bitboard b) {
b -= ((b>>1) & 0x5555555555555555ULL);
b = ((b>>2) & 0x3333333333333333ULL) + (b & 0x3333333333333333ULL);
b = ((b>>4) + b) & 0x0F0F0F0F0F0F0F0FULL;
b *= 0x0101010101010101ULL;
return int(b >> 56);
b -= (b >> 1) & 0x5555555555555555ULL;
b = ((b >> 2) & 0x3333333333333333ULL) + (b & 0x3333333333333333ULL);
b = ((b >> 4) + b) & 0x0F0F0F0F0F0F0F0FULL;
return (b * 0x0101010101010101ULL) >> 56;
}
template<>
inline int popcount<CNT_64_MAX15>(Bitboard b) {
b -= (b>>1) & 0x5555555555555555ULL;
b = ((b>>2) & 0x3333333333333333ULL) + (b & 0x3333333333333333ULL);
b *= 0x1111111111111111ULL;
return int(b >> 60);
b -= (b >> 1) & 0x5555555555555555ULL;
b = ((b >> 2) & 0x3333333333333333ULL) + (b & 0x3333333333333333ULL);
return (b * 0x1111111111111111ULL) >> 60;
}
template<>
inline int popcount<CNT_32>(Bitboard b) {
unsigned w = unsigned(b >> 32), v = unsigned(b);
v -= (v >> 1) & 0x55555555; // 0-2 in 2 bits
w -= (w >> 1) & 0x55555555;
v = ((v >> 2) & 0x33333333) + (v & 0x33333333); // 0-4 in 4 bits
w = ((w >> 2) & 0x33333333) + (w & 0x33333333);
v = ((v >> 4) + v) & 0x0F0F0F0F; // 0-8 in 8 bits
v += (((w >> 4) + w) & 0x0F0F0F0F); // 0-16 in 8 bits
v *= 0x01010101; // mul is fast on amd procs
return int(v >> 24);
v -= (v >> 1) & 0x55555555; // 0-2 in 2 bits
w -= (w >> 1) & 0x55555555;
v = ((v >> 2) & 0x33333333) + (v & 0x33333333); // 0-4 in 4 bits
w = ((w >> 2) & 0x33333333) + (w & 0x33333333);
v = ((v >> 4) + v + (w >> 4) + w) & 0x0F0F0F0F;
return (v * 0x01010101) >> 24;
}
template<>
inline int popcount<CNT_32_MAX15>(Bitboard b) {
unsigned w = unsigned(b >> 32), v = unsigned(b);
v -= (v >> 1) & 0x55555555; // 0-2 in 2 bits
w -= (w >> 1) & 0x55555555;
v = ((v >> 2) & 0x33333333) + (v & 0x33333333); // 0-4 in 4 bits
w = ((w >> 2) & 0x33333333) + (w & 0x33333333);
v += w; // 0-8 in 4 bits
v *= 0x11111111;
return int(v >> 28);
v -= (v >> 1) & 0x55555555; // 0-2 in 2 bits
w -= (w >> 1) & 0x55555555;
v = ((v >> 2) & 0x33333333) + (v & 0x33333333); // 0-4 in 4 bits
w = ((w >> 2) & 0x33333333) + (w & 0x33333333);
return ((v + w) * 0x11111111) >> 28;
}
template<>