fstrrep.pas

UDP聊天室

{其实还可以在FastReplace中先对FindStr进行判断，如果
FindStr完全是中文的话，直接可以在FastReplace中用
FastPos，这样可以提高速度。}

unit fStrRep;

interface

Type
  TFastPosProc = function(
    const aSourceString, aFindString : String;
    const aSourceLen, aFindLen, StartPos : integer
    ) : integer;

function FastReplace(
  var aSourceString : String;
  const aFindString, aReplaceString : String;
  CaseSensitive : Boolean = False) : String;

function FastPos(
  const aSourceString, aFindString : String;
  const aSourceLen, aFindLen, StartPos : integer
  ) : integer;

function FastPosNoCase(
  const aSourceString, aFindString : String;
  const aSourceLen, aFindLen, StartPos : integer
  ) : integer;

implementation

// This TYPE declaration will become apparent later.
//The first thing to note here is that I’m passing the SourceLength and FindL
//ength. As neither Source nor Find will alter at any point during FastReplace
//, there’s no need to call the LENGTH subroutine each time!
function FastPos(
  const aSourceString, aFindString : String;
  const aSourceLen, aFindLen, StartPos : integer
  ) : integer;
var
  SourceLen : integer;
begin
  // Next, we determine how many bytes we need to
  // scan to find the "start" of aFindString.
  SourceLen := aSourceLen;
  SourceLen := SourceLen - aFindLen;
  if (StartPos-1) > SourceLen then begin
    Result := 0;
    Exit;
  end;
  SourceLen := SourceLen - StartPos;
  SourceLen := SourceLen +2;
  // The ASM starts here.
  asm
    // Delphi uses ESI, EDI, and EBX a lot,
    // so we must preserve them.
    push ESI
    push EDI
    push EBX
    // Get the address of sourceString[1]
    // and Add (StartPos-1).
    // We do this for the purpose of finding
    // the NEXT occurrence, rather than
    // always the first!
    mov EDI, aSourceString
    add EDI, StartPos
    Dec EDI
    // Get the address of aFindString.
    mov ESI, aFindString
    // Note how many bytes we need to
    // look through in aSourceString
    // to find aFindString.
    mov ECX, SourceLen
    // Get the first char of aFindString;
    // note how it is done outside of the
    // main loop, as it never changes!
    Mov  Al, [ESI]
    // Now the FindFirstCharacter loop!
    @ScaSB:
    // Get the value of the current
    // character in aSourceString.
    // This is equal to ah := EDI^, that
    // is what the [] are around [EDI].
    Mov  Ah, [EDI]
    // Compare this character with aDestString[1].
    cmp  Ah,Al
    // If they're not equal we don't
    // compare the strings.
    jne  @NextChar
    // If they're equal, obviously we do!
    @CompareStrings:
    // Put the length of aFindLen in EBX.
    mov  EBX, aFindLen
    // We DEC EBX to point to the end of
    // the string; that is, we don't want to
    // add 1 if aFindString is 1 in length!
    dec  EBX

    // add by ShengQuanhu
    // If EBX is zero, then we've successfully
    // compared each character; i.e. it's A MATCH!
    // It will be happened when aFindLen=1
    Jz @EndOfMatch
    //add end

//Here’s another optimization tip. People at this point usually PUSH ESI and
//so on and then POP ESI and so forth at the end–instead, I opted not to chan
//ge ESI and so on at all. This saves lots of pushing and popping!
    @CompareNext:
    // Get aFindString character +
    // aFindStringLength (the last char).
    mov  Al, [ESI+EBX]
    // Get aSourceString character (current
    // position + aFindStringLength).
    mov  Ah, [EDI+EBX]
    // Compare them.
    cmp  Al, Ah
    Jz   @Matches
    // If they don't match, we put the first char
    // of aFindString into Al again to continue
    // looking for the first character.
    Mov  Al, [ESI]
    Jmp  @NextChar
    @Matches:
    // If they match, we DEC EBX (point to
    // previous character to compare).
    Dec  EBX
    // If EBX <> 0 ("J"ump "N"ot "Z"ero), we
    // continue comparing strings.
    Jnz  @CompareNext

    //add by Shengquanhu
    @EndOfMatch:
    //add end

    // If EBX is zero, then we've successfully
    // compared each character; i.e. it's A MATCH!
    // Move the address of the *current*
    // character in EDI.
    // Note, we haven't altered EDI since
    // the first char was found.
    mov  EAX, EDI
    // This is an address, so subtract the
    // address of aSourceString[1] to get
    // an actual character position.
    sub  EAX, aSourceString
    // Inc EAX to make it 1-based,
    // rather than 0-based.
    inc  EAX
    // Put it into result.
    mov  Result, EAX
    // Finish this routine!
    jmp  @TheEnd
    @NextChar:
//This is where I jump to when I want to continue searching for the first char
//acter of aFindString in aSearchString:
    // Point EDI (aFindString[X]) to
    // the next character.
    Inc  EDI
    // Dec ECX tells us that we've checked
    // another character, and that we're
    // fast running out of string to check!
    dec  ECX
    // If EBX <> 0, then continue scanning
    // for the first character.

    //add by shengquanhu
    //if ah is chinese char,jump again
    jz   @Result0
    cmp  ah, $80
    jb   @ScaSB
    Inc  EDI
    Dec  ECX
    //add by shengquanhu end

    jnz  @ScaSB

    //add by shengquanhu
    @Result0:
    //add by shengquanhu end

    // If EBX = 0, then move 0 into RESULT.
    mov  Result,0
    // Restore EBX, EDI, ESI for Delphi
    // to work correctly.
    // Note that they're POPped in the
    // opposite order they were PUSHed.
    @TheEnd:
    pop  EBX
    pop  EDI
    pop  ESI
  end;
end;

//This routine is an identical copy of FastPOS except where commented! The ide
//a is that when grabbing bytes, it ANDs them with $df, effectively making the
//m lowercase before comparing. Maybe this would be quicker if aFindString was
// made lowercase in one fell swoop at the beginning of the function, saving a
//n AND instruction each time.
function FastPosNoCase(
  const aSourceString, aFindString : String;
  const aSourceLen, aFindLen, StartPos : integer
  ) : integer;
var
  SourceLen : integer;
begin
  SourceLen := aSourceLen;
  SourceLen := SourceLen - aFindLen;
  if (StartPos-1) > SourceLen then begin
    Result := 0;
    Exit;
  end;
  SourceLen := SourceLen - StartPos;
  SourceLen := SourceLen +2;
  asm
    push ESI
    push EDI
    push EBX

    mov EDI, aSourceString
    add EDI, StartPos
    Dec EDI
    mov ESI, aFindString
    mov ECX, SourceLen
    Mov  Al, [ESI]

    //add by shengquanhu:just modified the lowercase 'a'..'z'
    cmp Al, $7A
    ja @ScaSB

    cmp Al, $61
    jb @ScaSB
    //end------------------------------------------

    // Make Al uppercase.
    and  Al, $df

    @ScaSB:
    Mov  Ah, [EDI]

    //add by shengquanhu:just modified the lowercase 'a'..'z'
    cmp Ah, $7A
    ja @CompareChar

    cmp Ah, $61
    jb @CompareChar
    //end------------------------------------------

    // Make Ah uppercase.
    and  Ah, $df

    @CompareChar:
    cmp  Ah,Al
    jne  @NextChar
    @CompareStrings:
    mov  EBX, aFindLen
    dec  EBX

    //add by ShengQuanhu
    Jz   @EndOfMatch
    //add end

    @CompareNext:
    mov  Al, [ESI+EBX]
    mov  Ah, [EDI+EBX]

    //add by shengquanhu:just modified the lowercase 'a'..'z'
    cmp Ah, $7A
    ja @LowerAh

    cmp Al, $61
    jb @LowerAh