Potential performance issues [2019-03]

[wukefe]

Overview

The source file, profile-q5-single.txt, contains the profiled information generated by callgrind/cachegrind for query 5. After looking into the result, we identify several potential issues as shown below.

List of Potential Issues

• allocMem: /* naive system */
  • bit shift: size%16 to size>>4
  • (Dr 9,491,585) if((*cur)+size < top){
• getHashValueComplete:
  • (Dr 7,037,254) caseL dominates
• find_hash_many
  • (Dr 13,097,228) while(hN(t)){
  • (Dr 20,181,981) caseL
• lib_join_index_hash
  • (Dr 65,628,988) DOP(vLen, tempK[i]=find_hash_many(hashT,hashMask,src,val,i,typ,&tempH[i]))
• getNumOfNonZero
  • (Dr 3,910,408) DOT(xn, if(vB(x,i))z[tid]++)
• isOrdered
  • (Dr 1,650,050) DOIa(vn(x), ...)
• binarySearch (<- searchOrdered)
  • (Dr 142,639,278) if(x[mid] == val) R mid;

[wukefe]

Update

Use bit shift instead of basic arithmetic operations, e.g. use size>>4 instead of size%16.