Base64Decode With 16Bits-CoderTable

 整个“表”的构造过程，可能看起来会显得有些复杂

感谢 DelphiGuy 于 2010-10-08 17:27:37 给出的提醒

 

2010-10-08 增加 Base64Decode 用于编码数据中存在非BASE64有效字符时的解码

unit uDecoder4To3_14BitsCoderTable; { 2010-10-01 Create by Unsigned(僵哥) 2010-10-08 增加 Base64Decode 用于编码数据中存在非BASE64有效字符时的解码 by Unsigned(僵哥) } interface function Base64DecodeEx( const Input: Pointer ; const Output: Pointer ; InputCount: Integer ): Integer; assembler; function Base64Decode( const Input: Pointer ; const Output: Pointer ; InputCount: Integer ): Integer; assembler; function GetSizeDecoder4To3(InputCount: Integer): Integer; inline; implementation //根据DelphiGuy的提醒，将码表增大到足够安全 { const B_0 = -11051; const B_1 = 9253; const B_2 = 29557; const B_3 = 60912; var CoderTable: array[0..60911 + 256] of Byte; } const B_0 = 0; const B_1 = 65536; const B_2 = 131072; const B_3 = 196608; var CoderTable: array[0..196607 + 256] of Byte; function GetSizeDecoder4To3(InputCount: Integer): Integer; inline; begin Result := (InputCount) div 4 * 3; end; function Base64DecodeEx( const Input: Pointer ; const Output: Pointer ; InputCount: Integer ): Integer; assembler; asm //根据DelphiGuy的提醒，防止 "EAX and EDX = 0" 的情况 { TEST EAX, EDX JZ @RET } TEST EAX, EAX JZ @RET TEST EDX, EDX JZ @RET SHR ECX, 2 TEST ECX, ECX JZ @RET PUSH ESI PUSH EDI PUSH EBP PUSH EBX PUSH EDX LEA EBP, [CoderTable] MOV ESI, EAX MOV EDI, EDX XOR EDX, EDX XOR EAX, EAX @LOOP: MOV AX, [ESI] //Bits: 00222211 00111111 MOV DX, [ESI + 2] //Bits: 00333333 00332222 MOV BL, [EBP + EAX + B_0] //CoderTable[Bits: 00222211 00111111] MOV AL, DL //EAX := Bits: 0000000 0000000 00222211 00332222 MOV BH, [EBP + EAX + B_1] //CoderTable[Bits: 00222211 00332222] MOV [EDI], BX MOV BL, [EBP + EDX + B_2] //CoderTable[Bits: 00333333 00332222] MOV [EDI + 2], BL ADD ESI, 4 ADD EDI, 3 DEC ECX JNZ @LOOP POP EAX NEG EAX ADD EAX, EDI CMP DH, $3D JNZ @END DEC EAX CMP DL, $3D JNZ @END DEC EAX @END: POP EBX POP EBP POP EDI POP ESI @RET: end; function Base64Decode( const Input: Pointer ; const Output: Pointer ; InputCount: Integer ): Integer; assembler; asm //根据DelphiGuy的提醒，防止 "EAX and EDX" 的情况 { TEST EAX, EDX JZ @RET } TEST EAX, EAX JZ @RET TEST EDX, EDX JZ @RET SHR ECX, 2 TEST ECX, ECX JZ @RET PUSH ESI PUSH EDI PUSH EBP PUSH EBX PUSH EDX LEA EBP, [CoderTable] MOV ESI, EAX MOV EDI, EDX XOR EDX, EDX XOR EAX, EAX XOR EBX, EBX @@0: MOV BL, [ESI + 0] //Bits: 00111111 CMP Byte ptr [EBP + EBX + B_3], 1 JE @AL INC ESI DEC ECX JZ @ENDLOOP JMP @@0 @AL: MOV AL, BL @@1: MOV BL, [ESI + 1] //Bits: 00222211 CMP Byte ptr [EBP + EBX + B_3], 1 JE @AH INC ESI DEC ECX JNZ @@1 JMP @ENDLOOP @AH: MOV AH, BL @@2: MOV BL, [ESI + 2] //Bits: 00332222 CMP Byte ptr [EBP + EBX + B_3], 1 JE @DL INC ESI DEC ECX JNZ @@2 JMP @ENDLOOP @DL: MOV DL, BL @@3: MOV BL, [ESI + 3] //Bits: 00333333 CMP Byte ptr [EBP + EBX + B_3], 1 JE @DH INC ESI DEC ECX JNZ @@3 JMP @ENDLOOP @DH: MOV DH, BL MOV BL, [EBP + EAX + B_0] //CoderTable[Bits: 00222211 00111111] MOV [EDI + 0], BL MOV AL, DL //EAX := Bits: 0000000 0000000 00222211 00332222 MOV BL, [EBP + EAX + B_1] //CoderTable[Bits: 00222211 00332222] MOV [EDI + 1], BL MOV BL, [EBP + EDX + B_2] //CoderTable[Bits: 00333333 00332222] MOV [EDI + 2], BL ADD ESI, 4 ADD EDI, 3 SUB ECX, 4 JNZ @LOOP @ENDLOOP: POP EAX NEG EAX ADD EAX, EDI CMP DH, $3D JNZ @END DEC EAX CMP DL, $3D JNZ @END DEC EAX @END: POP EBX POP EBP POP EDI POP ESI @RET: end; {$REGION 'InitTable'} procedure InitTable; inline; var I,J: Integer; Index_I, Bits_I_0, Bits_I_1, Bits_I_2: Integer; begin FillChar(CoderTable, SizeOf(CoderTable), 0); for I := 0 to 64 do begin Bits_I_0 := (I shr 4) and $3; Bits_I_1 := (I shl 4) and $F0; Bits_I_2 := I; case I of 0..25: begin Index_I := (I + 65) shl 8; CoderTable[B_3 + I + 65] := 1; end; 26..51: begin Index_I := (I - 26 + 97) shl 8; CoderTable[B_3 + I - 26 + 97] := 1; end; 52..61: begin Index_I := (I - 52 + 48) shl 8; CoderTable[B_3 + I - 52 + 48] := 1; end; 62: begin Index_I := 43 shl 8; CoderTable[B_3 + 43] := 1; end; 63: begin Index_I := 47 shl 8; CoderTable[B_3 + 47] := 1; end else begin Index_I := 61 shl 8; CoderTable[B_3 + 61] := 1; Bits_I_0 := 0; Bits_I_1 := 0; Bits_I_2 := 0; end; end; for J := 0 to 9 do begin CoderTable[(Index_I or (J + 48)) + B_0] := Bits_I_0 or ((J + 52) shl 2); CoderTable[(Index_I or (J + 65)) + B_0] := Bits_I_0 or ((J + 0) shl 2); CoderTable[(Index_I or (J + 97)) + B_0] := Bits_I_0 or ((J + 26) shl 2 ); CoderTable[(Index_I or (J + 48)) + B_1] := Bits_I_1 or ((J + 52) shr 2); CoderTable[(Index_I or (J + 65)) + B_1] := Bits_I_1 or ((J + 0) shr 2); CoderTable[(Index_I or (J + 97)) + B_1] := Bits_I_1 or ((J + 26) shr 2); CoderTable[(Index_I or (J + 48)) + B_2] := Bits_I_2 or (((J + 52) shl 6) and $C0); CoderTable[(Index_I or (J + 65)) + B_2] := Bits_I_2 or (((J + 0) shl 6) and $C0); CoderTable[(Index_I or (J + 97)) + B_2] := Bits_I_2 or (((J + 26) shl 6) and $C0); end; for J := 10 to 25 do begin CoderTable[(Index_I or (J + 65)) + B_0] := Bits_I_0 or ((J + 0) shl 2); CoderTable[(Index_I or (J + 97)) + B_0] := Bits_I_0 or ((J + 26) shl 2); CoderTable[(Index_I or (J + 65)) + B_1] := Bits_I_1 or ((J + 0) shr 2); CoderTable[(Index_I or (J + 97)) + B_1] := Bits_I_1 or ((J + 26) shr 2); CoderTable[(Index_I or (J + 65)) + B_2] := Bits_I_2 or (((J + 0) shl 6) and $C0); CoderTable[(Index_I or (J + 97)) + B_2] := Bits_I_2 or (((J + 26) shl 6) and $C0); end; CoderTable[(Index_I or 43) + B_0] := Bits_I_0 or 248; CoderTable[(Index_I or 47) + B_0] := Bits_I_0 or 252; CoderTable[(Index_I or 61) + B_0] := Bits_I_0 or 0; CoderTable[(Index_I or 43) + B_1] := Bits_I_1 or 15; CoderTable[(Index_I or 47) + B_1] := Bits_I_1 or 15; CoderTable[(Index_I or 61) + B_1] := Bits_I_1 or 0; CoderTable[(Index_I or 43) + B_2] := Bits_I_2 or 128; CoderTable[(Index_I or 47) + B_2] := Bits_I_2 or 192; CoderTable[(Index_I or 61) + B_2] := Bits_I_2 or 0; end; end; {$ENDREGION} initialization InitTable; end.

对于某些流水线较长的高性能CPU，也可以试着把末尾的两条比较指令转换成运算，当然这个是不推荐的，对于绝大部分CPU都是相对低效的（虽然低不了多少，一亿次里面也就一秒左右）

function Base64DecodeEx( const Input: Pointer ; const Output: Pointer ; InputCount: Integer ): Integer; assembler; asm //根据DelphiGuy的提醒，防止 "EAX and EDX = 0" 的情况 { TEST EAX, EDX JZ @RET } TEST EAX, EAX JZ @RET TEST EDX, EDX JZ @RET SHR ECX, 2 TEST ECX, ECX JZ @RET PUSH ESI PUSH EDI PUSH EBP PUSH EBX PUSH EDX LEA EBP, [CoderTable] MOV ESI, EAX MOV EDI, EDX XOR EDX, EDX XOR EAX, EAX @LOOP: MOV DX, [ESI] //Bits: 00222211 00111111 MOV AX, [ESI + 2] //Bits: 00333333 00332222 MOV BL, [EBP + EDX + B_0] //CoderTable[Bits: 00222211 00111111] MOV DL, AL //EDX := Bits: 0000000 0000000 00222211 00332222 MOV BH, [EBP + EDX + B_1] //CoderTable[Bits: 00222211 00332222] MOV [EDI], BX MOV BL, [EBP + EAX + B_2] //CoderTable[Bits: 00333333 00332222] MOV [EDI + 2], BL ADD ESI, 4 ADD EDI, 3 DEC ECX JNZ @LOOP XOR EAX, $3D3D LEA EBX, [EAX + $101] NOT EAX AND EAX, EBX AND EAX, $101 MOV CL, AH ADD CL, AL POP EAX ADD EAX, ECX NEG EAX ADD EAX, EDI @END: POP EBX POP EBP POP EDI POP ESI @RET: end;

BTW: Base64Decode 相对 Base64DecodeEx 可以尽可能对有效的Base64编码串进行解码，根据测试的结果显示，从一次性解码数据的由少到多，性能损耗达到甚至超过了 30% ~ 80% (意思是说可能base64DecodeEx只需要不到一分钟时间做完的事情，换用Base64Decode需要三分钟)

测试代码

procedure TForm1.Button1Click(Sender: TObject); var strBase64, strRawData: AnsiString; begin strBase64 := 'SGVsbG8gV29yZCE='; SetLength(strRawData, GetSizeDecoder4To3(Length(strBase64))); SetLength(strRawData, Base64DecodeEx(Pointer(strBase64), Pointer(strRawData), Length(strBase64))); ShowMessage(strRawData); end;