[Libreoffice-bugs] [Bug 144777] countifs() in Calc is slower than Excel's countifs()

[bugzilla-daemon at bugs.documentfoundation.org]

https://bugs.documentfoundation.org/show_bug.cgi?id=144777

Julien Nabet <serval2412 at yahoo.fr> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |erack at redhat.com,
                   |                            |noelgrandin at gmail.com

--- Comment #13 from Julien Nabet <serval2412 at yahoo.fr> ---
Eike/Noel: the Flamegraph shows that a lot of time is in ScTable::ValidQuery
See
https://opengrok.libreoffice.org/xref/core/sc/source/core/data/table3.cxx?r=b60b6bfa#2979

I noticed in the loop "for (const auto& rItem : rItems)" at line 3039 that
perhaps changing the order of test may help.
For example, I thought about putting:
3069                  else if (rParam.mbRangeLookup)
3070                  {
3071                      std::pair<bool,bool> aThisRes =
3072                          aEval.compareByRangeLookup(aCell, nCol, nRow,
rEntry, rItem);
3073                      aRes.first |= aThisRes.first;
3074                      aRes.second |= aThisRes.second;
3075                  }
3076  
3077                  if (aRes.first && aRes.second)
3078                      break;

just after:
3041                  if (rItem.meType == ScQueryEntry::ByTextColor)
3042                  {
3043                      std::pair<bool, bool> aThisRes
3044                          = aEval.compareByTextColor(nCol, nRow, nTab,
rItem);
3045                      aRes.first |= aThisRes.first;
3046                      aRes.second |= aThisRes.second;
3047                  }
3048                  else if (rItem.meType == ScQueryEntry::ByBackgroundColor)
3049                  {
3050                      std::pair<bool,bool> aThisRes =
3051                          aEval.compareByBackgroundColor(nCol, nRow, nTab,
rItem);
3052                      aRes.first |= aThisRes.first;
3053                      aRes.second |= aThisRes.second;
3054                  }

so these would be at the end:
3055                  else if (aEval.isQueryByValue(rItem, nCol, nRow, aCell))
3056                  {
3057                      std::pair<bool,bool> aThisRes =
3058                          aEval.compareByValue(aCell, nCol, nRow, rEntry,
rItem, pContext);
3059                      aRes.first |= aThisRes.first;
3060                      aRes.second |= aThisRes.second;
3061                  }
3062                  else if (aEval.isQueryByString(rEntry, rItem, nCol, nRow,
aCell))
3063                  {
3064                      std::pair<bool,bool> aThisRes =
3065                          aEval.compareByString(aCell, nRow, rEntry, rItem,
pContext);
3066                      aRes.first |= aThisRes.first;
3067                      aRes.second |= aThisRes.second;
3068                  }

So we would call "isQueryByValue" and/or "isQueryByString" only if necessary.

Now, I don't know if the order of evaluation matters and could change the final
result.

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.freedesktop.org/archives/libreoffice-bugs/attachments/20210930/d713090d/attachment.htm>