gvsafronov/futriix
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Speedup hllRawSum() processing 8 bytes per iteration.

Browse Source 
The internal HLL raw encoding used by PFCOUNT when merging multiple keys
is aligned to 8 bits (1 byte per register) so we can exploit this to
improve performances by processing multiple bytes per iteration.

In benchmarks the new code was several times faster with HLLs with many
registers set to zero, while no slowdown was observed with populated
HLLs.
This commit is contained in:
antirez 2014-04-17 18:02:45 +02:00
parent 192a213274
commit 5eb7ac0c92
1 changed files with 15 additions and 7 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

22
src/hyperloglog.c
View File

@ -928,16 +928,24 @@ double hllSparseSum(uint8_t *sparse, int sparselen, double *PE, int *ezp, int *i
double hllRawSum(uint8_t *registers, double *PE, int *ezp) { double hllRawSum(uint8_t *registers, double *PE, int *ezp) {
double E = 0; double E = 0;
int j, ez = 0; int j, ez = 0;
unsigned long reg; uint64_t *word = (uint64_t*) registers;
uint8_t *bytes;
for (j = 0; j < HLL_REGISTERS; j++) { for (j = 0; j < HLL_REGISTERS/8; j++) {
reg = registers[j]; if (*word == 0) {
if (reg == 0) { ez += 8;
ez++;
/* Increment E at the end of the loop. */
} else { } else {
E += PE[reg]; /* Precomputed 2^(-reg[j]). */ bytes = (uint8_t*) word;
if (bytes[0]) E += PE[bytes[0]]; else ez++;
if (bytes[1]) E += PE[bytes[1]]; else ez++;
if (bytes[2]) E += PE[bytes[2]]; else ez++;
if (bytes[3]) E += PE[bytes[3]]; else ez++;
if (bytes[4]) E += PE[bytes[4]]; else ez++;
if (bytes[5]) E += PE[bytes[5]]; else ez++;
if (bytes[6]) E += PE[bytes[6]]; else ez++;
if (bytes[7]) E += PE[bytes[7]]; else ez++;
} }
word++;
} }
E += ez; /* 2^(-reg[j]) is 1 when m is 0, add it 'ez' times for every E += ez; /* 2^(-reg[j]) is 1 when m is 0, add it 'ez' times for every
zero register in the HLL. */ zero register in the HLL. */