Delphi: fast Pos with 64-bit

问题:

Is there any code for a Pos() version that's as fast in 64-bit than the current 32-bit?

To my understanding, the 32-bit version in Delphi (tested up to XE5) adopted the FastCode assembler versions many years ago, but for 64-bit it uses a PurePascal version, which is around 5 to 10 times slower.

Some tests, same procedure in a long loop:

32-bit: 65..90ms

64-bit: 280..300ms

回答1:

Using Fastcoders purepascal PosEx_Sha_Pas_2 algorithm (modified to fit x64):

function PosEx_Sha_Pas_2(const SubStr, S: string; Offset: Integer = 1): Integer;
Type
  PInteger =^Integer;
var
  len, lenSub: Integer;
  ch: char;
  p, pSub, pStart, pStop: pchar;
label
  Loop0, Loop4,
  TestT, Test0, Test1, Test2, Test3, Test4,
  AfterTestT, AfterTest0,
  Ret, Exit;
begin;
  pSub := pointer(SubStr);
  p := pointer(S);

  if (p = nil) or (pSub = nil) or (Offset < 1) then
  begin;
    Result := 0;
    goto Exit;
  end;

  lenSub := PLongInt(PByte(pSub) - 4)^ - 1; // <- Modified
  len := PLongInt(PByte(p) - 4)^; // <- Modified
  if (len < lenSub + Offset) or (lenSub < 0) then
  begin;
    Result := 0;
    goto Exit;
  end;

  pStop := p + len;
  p := p + lenSub;
  pSub := pSub + lenSub;
  pStart := p;
  p := p + Offset + 3;

  ch := pSub[0];
  lenSub := -lenSub;
  if p < pStop then
    goto Loop4;
  p := p - 4;
  goto Loop0;

Loop4:
  if ch = p[-4] then
    goto Test4;
  if ch = p[-3] then
    goto Test3;
  if ch = p[-2] then
    goto Test2;
  if ch = p[-1] then
    goto Test1;
Loop0:
  if ch = p[0] then
    goto Test0;
AfterTest0:
  if ch = p[1] then
    goto TestT;
AfterTestT:
  p := p + 6;
  if p < pStop then
    goto Loop4;
  p := p - 4;
  if p < pStop then
    goto Loop0;
  Result := 0;
  goto Exit;

Test3:
  p := p - 2;
Test1:
  p := p - 2;
TestT:
  len := lenSub;
  if lenSub <> 0 then
    repeat
      ;
      if (pSub[len] <> p[len + 1]) or (pSub[len + 1] <> p[len + 2]) then
        goto AfterTestT;
      len := len + 2;
    until len >= 0;
  p := p + 2;
  if p <= pStop then
    goto Ret;
  Result := 0;
  goto Exit;

Test4:
  p := p - 2;
Test2:
  p := p - 2;
Test0:
  len := lenSub;
  if lenSub <> 0 then
    repeat
      ;
      if (pSub[len] <> p[len]) or (pSub[len + 1] <> p[len + 1]) then
        goto AfterTest0;
      len := len + 2;
    until len >= 0;
  Inc(p);
Ret:
  Result := p - pStart;
Exit:
end;

And the result using David's test case (x64 release):

System.Pos       18427
wcsstr            8122
PosEx_Sha_Pas_2   2282

For the x32 release the results are:

System.Pos        2171
wcsstr            9634
PosEx_Sha_Pas_2   1868

---

Conclusion:

PosEx_Sha_Pas_2 is almost as fast in x64 bit mode as Pos in x32 bit mode. Additionally it seems as PosEx_Sha_Pas_2 is faster than Pos in x32 bit mode.

All tests here is with the XE4 version.

---

Improve purepascal System.Pos() still open in Tokyo 10.2.

回答2:

The C runtime function wcsstr as implemented in the system provided msvcrt.dll is better than Delphi RTL Pos for 64 bit code.

{$APPTYPE CONSOLE}

uses
  SysUtils, Diagnostics;

function wcsstr(const str, strsearch: PWideChar): PWideChar; cdecl; external 'msvcrt.dll';

var
  i, j: Integer;
  Stopwatch: TStopwatch;
  str: string;
  P: PWideChar;

const
  N = 50000000;

begin
  str := 'Lorem ipsum dolor sit amet, consectetur adipisicing elit, sed do '
    + 'eiusmod tempor incididunt ut labore et dolore magna aliqua. Ut enim '
    + 'ad minim veniam, quis nostrud exercitation ullamco laboris nisi ut '
    + 'aliquip ex ea commodo consequat. Duis aute irure dolor in reprehenderit '
    + 'in voluptate velit esse cillum dolore eu fugiat nulla pariatur. Excepteur '
    + 'sint occaecat cupidatat non proident, sunt in culpa qui officia deserunt '
    + 'mollit anim id est laborum.';

  Stopwatch := TStopwatch.StartNew;
  for i := 1 to N do
  begin
    j := Pos('tempor', str);
    if j=0 then
      Beep;
  end;
  Writeln('Pos: ' + IntToStr(Stopwatch.ElapsedMilliseconds));

  Stopwatch := TStopwatch.StartNew;
  for i := 1 to N do
  begin
    P := wcsstr(PChar(str), 'tempor');
    if P=nil then
      Beep;
  end;
  Writeln('wcsstr: ' + IntToStr(Stopwatch.ElapsedMilliseconds));

  Readln;
end.

32 bit release build

Pos: 1930
wcsstr: 6951

64 bit release build

Pos: 18384
wcsstr: 6701

Interestingly, the 32 bit Pos is clearly very well implemented.

My system is running Win7 x64 on i5-2300.