Merge pull request #126 from georges-hatem/main

compatibility with LF rule instead of CRLF, and a few optimizations

Author: gcarreno

entries/ghatem-fpc/README.md

@@ -224,3 +224,12 @@ I implemented my own Dictionary class consisting of two arrays. We compute the m
 
 edit:
 quadratic probing improved performance even further. we could probably do better with 2-level hashing, but finding such a hash function is going to take a lot of trials, this is probably acceptable results
+
+**ACTUAL TIMING (busy machine): ~4 seconds as per gcarreno**
+
+## v.4 (2024-04-24)
+
+a few performance improvements, and measurements as per gcarreno on a busy machine:
+ - using mORMot's `crc32c` function instead of the native `crc32`, time dropped to 3.8 seconds
+ - I had removed my pre-allocated records implementation. restored it in the custom dictionary class, time dropped to 3.2 seconds
+ - skipping a few chars that we don't need to bother with, no timing yet

entries/ghatem-fpc/src/OneBRCproj.lpi

@@ -45,7 +45,6 @@
           <Linking>
             <Debugging>
               <DebugInfoType Value="dsDwarf3"/>
-              <UseHeaptrc Value="True"/>
               <TrashVariables Value="True"/>
               <UseValgrind Value="True"/>
               <UseExternalDbgSyms Value="True"/>

entries/ghatem-fpc/src/onebrc.pas

@@ -318,7 +318,7 @@ procedure TOneBRC.ProcessData (aThreadNb: UInt16; aStartIdx: Int64; aEndIdx: Int
   // the given ending point might be in the middle of a line:
   // find the beginning of that line (the last block works well)
   while True do begin
-    if FData[aEndIdx] <> #13 then
+    if FData[aEndIdx] <> #10 then
       Dec (aEndIdx)
     else
       break;
@@ -328,7 +328,7 @@ procedure TOneBRC.ProcessData (aThreadNb: UInt16; aStartIdx: Int64; aEndIdx: Int
   vLineStart := i;
 
   while i < aEndIdx do begin
-    if FData[i] = #13 then begin
+    if FData[i] = #10 then begin
       // new line parsed, process its contents
       ExtractLineData (vLineStart, i - 1, vLenStationName, vTemp);
 
@@ -360,8 +360,18 @@ procedure TOneBRC.ProcessData (aThreadNb: UInt16; aStartIdx: Int64; aEndIdx: Int
         FStationsDicts[aThreadNb].Add (vHash, vData);
       end;
 
-      // next char is #10, so we can skip 2 instead of 1
-      vLineStart := i+2;
+      // we're at a #10: next line starts at the next index
+      vLineStart := i+1;
+
+      // we're at a #10:
+      // until the next #10 char, there will be:
+      // - 1 semicolon
+      // - 3 chars for the temp (min)
+      // - 2 chars for the name (min)
+      // - the usual Inc (I)
+      // so we should be able to skip 7 chars until another #10 may appear
+      Inc (i, 7);
+      continue;
     end;
 
     Inc (i);