Merge pull request #135 from georges-hatem/main

undo custom dict size, add option for different hash function

Author: gcarreno

entries/ghatem-fpc/README.md

@@ -233,3 +233,12 @@ a few performance improvements, and measurements as per gcarreno on a busy machi
  - 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
+
+## v.5 (2024-04-27)
+
+Various attempts at dictionary sizes, ranging from 45k to 95k. Even though larger dictionaries reduce collision tremendously, a dictionary of size 45k was still optimal.
+
+Another trial with various hash functions, a simple modulus vs. a slightly more complex one: modulus is slower on my PC, remains to try on the test env.
+Can be tested with the HASHMULT build option
+
+Finally, it seems choosing a dictionary size that is a prime number is also recommended: shaves 1 second out of 20 on my PC.

entries/ghatem-fpc/src/OneBRCproj.lpi

@@ -15,7 +15,7 @@
       <UseAppBundle Value="False"/>
       <ResourceType Value="res"/>
     </General>
-    <BuildModes Count="4">
+    <BuildModes Count="5">
       <Item1 Name="Default" Default="True"/>
       <Item2 Name="Debug">
         <CompilerOptions>
@@ -108,6 +108,35 @@
           </Other>
         </CompilerOptions>
       </Item4>
+      <Item5 Name="HashMult">
+        <CompilerOptions>
+          <Version Value="11"/>
+          <PathDelim Value="\"/>
+          <Target>
+            <Filename Value="..\..\..\bin\ghatem"/>
+          </Target>
+          <SearchPaths>
+            <IncludeFiles Value="$(ProjOutDir)"/>
+            <UnitOutputDirectory Value="..\..\..\bin\lib\$(TargetCPU)-$(TargetOS)"/>
+          </SearchPaths>
+          <CodeGeneration>
+            <SmartLinkUnit Value="True"/>
+            <Optimizations>
+              <OptimizationLevel Value="3"/>
+            </Optimizations>
+          </CodeGeneration>
+          <Linking>
+            <Debugging>
+              <GenerateDebugInfo Value="False"/>
+              <RunWithoutDebug Value="True"/>
+            </Debugging>
+            <LinkSmart Value="True"/>
+          </Linking>
+          <Other>
+            <CustomOptions Value="-dRELEASE -dHASHMULT"/>
+          </Other>
+        </CompilerOptions>
+      </Item5>
     </BuildModes>
     <PublishOptions>
       <Version Value="2"/>

entries/ghatem-fpc/src/OneBRCproj.lpr

@@ -18,7 +18,6 @@   TOneBRCApp = class(TCustomApplication)
   private
     FFileName: string;
     FThreadCount: Integer;
-    FDictSize: Integer;
     procedure RunOneBRC;
   protected
     procedure DoRun; override;
@@ -34,7 +33,7 @@ procedure TOneBRCApp.RunOneBRC;
 var
   vOneBRC: TOneBRC;
 begin
-  vOneBRC := TOneBRC.Create (FThreadCount, FDictSize);
+  vOneBRC := TOneBRC.Create (FThreadCount);
   try
     try
       vOneBRC.mORMotMMF(FFileName);
@@ -89,17 +88,15 @@ procedure TOneBRCApp.DoRun;
   ErrorMsg: String;
 begin
   // quick check parameters
-  ErrorMsg:= CheckOptions(Format('%s%s%s%s%s:',[
+  ErrorMsg:= CheckOptions(Format('%s%s%s%s:',[
       cShortOptHelp,
       cShortOptThread,
-      cShortOptDictSize,
       cShortOptVersion,
       cShortOptInput
     ]),
     [
       cLongOptHelp,
       cLongOptThread+':',
-      cLongOptDictSize+':',
       cLongOptVersion,
       cLongOptInput+':'
     ]
@@ -129,11 +126,6 @@ procedure TOneBRCApp.DoRun;
     FThreadCount := StrToInt (GetOptionValue(cShortOptThread, cLongOptThread));
   end;
 
-  FDictSize := 45003;
-  if HasOption(cShortOptDictSize, cLongOptDictSize) then begin
-    FDictSize := StrToInt (GetOptionValue(cShortOptDictSize, cLongOptDictSize));
-  end;
-
   if HasOption(cShortOptInput, cLongOptInput) then begin
     FFileName := GetOptionValue(
       cShortOptInput,

entries/ghatem-fpc/src/baseline.console.pas

@@ -21,8 +21,6 @@ interface
   cLongOptInput          = 'input-file';
   cShortOptThread: Char  = 't';
   cLongOptThread         = 'threads';
-  cShortOptDictSize: Char = 's';
-  cLongOptDictSize        = 'size';
   {$ELSE}
   cOptionHelp:    array of string = ['-h', '--help'];
   cOptionVersion: array of string = ['-v', '--version'];

entries/ghatem-fpc/src/onebrc.pas

@@ -10,10 +10,8 @@ interface
 
 function RoundExDouble(const ATemp: Double): Double; inline;
 
-{$WRITEABLECONST ON}
 const
-  cDictSize: Integer = 45003;
-{$WRITEABLECONST OFF}
+  cDictSize: Integer = 45007;
 
 type
 
@@ -62,10 +60,10 @@   TOneBRC = class
     FThreads: array of TThread;
     FStationsDicts: array of TMyDictionary;
 
-    procedure ExtractLineData(const aStart: Int64; const aEnd: Int64; out aLength: ShortInt; out aTemp: SmallInt); inline;
+    procedure ExtractLineData(const aStart: Int64; const aEnd: Int64; out aLength: ShortInt; out aTemp: SmallInt);
 
   public
-    constructor Create (const aThreadCount: UInt16; const aDictSize: Integer);
+    constructor Create (const aThreadCount: UInt16);
     destructor Destroy; override;
     function mORMotMMF (const afilename: string): Boolean;
     procedure DispatchThreads;
@@ -138,9 +136,13 @@ procedure TMyDictionary.InternalFind(const aKey: Cardinal; out aFound: Boolean;
     vDbl: Double;
     vOffset: Integer;
 begin
+{$IFDEF HASHMULT}
   vDbl := aKey * cHashConst;
   vDbl := vDbl - Trunc (vDbl);
   vIdx := Trunc (vDbl * cDictSize);
+{$ELSE}
+  vIdx := aKey mod cDictSize;
+{$ENDIF}
 
   aFound := False;
 
@@ -251,11 +253,9 @@ procedure TOneBRC.ExtractLineData(const aStart: Int64; const aEnd: Int64; out aL
 
 //---------------------------------------------------
 
-constructor TOneBRC.Create (const aThreadCount: UInt16; const aDictSize: Integer);
+constructor TOneBRC.Create (const aThreadCount: UInt16);
 var I: UInt16;
 begin
-  cDictSize := aDictSize;
-
   FThreadCount := aThreadCount;
   SetLength (FStationsDicts, aThreadCount);
   SetLength (FThreads, aThreadCount);