Parallel for 32 vrs 64bits

[pyscripter 579]

I revisited this thread  and tested the code below:

 

program Project1;
{$APPTYPE CONSOLE}
{$R *.res}

uses
  System.SysUtils,
  System.Threading,
  System.Diagnostics;
var
  SW:TStopWatch;
type
  TThreadPoolStatsHelper = record helper for TThreadPoolStats
    function Formatted: string;
  end;

  function TThreadPoolStatsHelper.Formatted: string;
  begin
    Result := Format('Worker: %2d, Min: %2d, Max: %2d, Idle: %2d, Retired: %2d, Suspended: %2d, CPU(Avg): %3d, CPU: %3d',
      [self.WorkerThreadCount,
       self.MinLimitWorkerThreadCount, self.MaxLimitWorkerThreadCount,
       self.IdleWorkerThreadCount, self.RetiredWorkerThreadCount, self.ThreadSuspended,
       self.AverageCPUUsage, self.CurrentCPUUsage]);
  end;

  procedure Load;
  begin
    TParallel.For(0, 99999999, procedure(i: Integer)
    var
      T:Single;
    begin
      T:=Sin(i/PI);
    end);
  end;

begin
  try
    Writeln('PPL Test ---------------');
    Writeln('Before: '+ TThreadPoolStats.Current.Formatted);
    SW:=TStopWatch.StartNew;
    Load;
    Writeln('Finished in '+SW.Elapsed.ToString);
    Sleep(1000);
    Writeln('After: '+TThreadPoolStats.Current.Formatted);
  except
    on E: Exception do
      Writeln(E.ClassName, ': ', E.Message);
  end;
  ReadLn;
end.

This is the output

 

32-bits

 

PPL Test ---------------
Before: Worker:  0, Min:  8, Max: 200, Idle:  0, Retired:  0, Suspended:  0, CPU(Avg):   0, CPU:   0
Finished in 00:00:00.7620933
After: Worker:  8, Min:  8, Max: 200, Idle:  7, Retired:  0, Suspended:  0, CPU(Avg):   8, CPU:  15

 

64-bits

PPL Test ---------------
Before: Worker:  0, Min:  8, Max: 200, Idle:  0, Retired:  0, Suspended:  0, CPU(Avg):   0, CPU:   0
Finished in 00:00:14.0655228
After: Worker:  8, Min:  8, Max: 200, Idle:  7, Retired:  0, Suspended:  0, CPU(Avg):  85, CPU:   1

 

Can anyone explain the huge difference in times? (it was consistent over many runs).

 

[pyscripter 579]

Oh I get it.  sin is highly optimized in 32-bits but apparently not in 64-bits.

[Vandrovnik 202]

May be https://github.com/neslib/FastMath and/or http://docwiki.embarcadero.com/RADStudio/Rio/en/Floating_point_precision_control_(Delphi_for_x64) can help.

[David Heffernan 2300]

The issue is that x64 trig functions are very slow for very large values. Nobody actually wants to know sin for 99999999/pi radians.  Put in sensible values for the argument to sin and it looks more reasonable. For instance try using      

 

T:=Sin(i/99999999);

[David Heffernan 2300]

7 hours ago, pyscripter said:

Oh I get it.  sin is highly optimized in 32-bits but apparently not in 64-bits.

 

No. That's wrong. In fact sin is quicker under x64 than under x86. Even though sin (and other trig) is implemented in hardware in the x87 unit, and in Pascal in x64 (because the SSE2 unit does not have built in trig).

[pyscripter 579]

8 minutes ago, David Heffernan said:

The issue is that x64 trig functions are very slow for very large values. Nobody actually wants to know sin for 99999999/pi radians.  Put in sensible values for the argument to sin and it looks more reasonable. For instance try using      

 

T:=Sin(i/99999999);

Yes you are right...