The MASM Forum Archive 2004 to 2012

Here's the version I've got:
align 4

szRtrim proc src:DWORD,dst:DWORD

    push esi
    push edi

    mov esi, src
    mov edi, dst
    sub esi, 1
  @@:
    add esi, 1
    cmp BYTE PTR [esi], 32
    je @B
    cmp BYTE PTR [esi], 9
    je @B
    cmp BYTE PTR [esi], 0
    jne @F
    xor eax, eax            ; return zero on empty string
    mov BYTE PTR [edi], 0   ; set string length to zero
    jmp szLout

  @@:
    mov esi, src
    xor ecx, edx
    xor ecx, ecx        ; ECX as index and location counter

  @@:
    mov al, [esi+ecx]   ; copy bytes from src to dst
    mov [edi+ecx], al
    add ecx, 1
    test al, al
    je @F               ; exit on zero
    cmp al, 33
    jb @B
    mov edx, ecx        ; store count if asc 33 or greater
    jmp @B

  @@:
    mov BYTE PTR [edi+edx], 0

    mov eax, edx        ; return length of trimmed string
    mov ecx, dst

  szLout:
    pop edi
    pop esi

    ret

szRtrim endp

The instruction "xor ecx, edx" should be omitted.
"mov BYTE PTR [edi], 0" may be replaced by "mov [edi],al", and
"mov BYTE PTR [edi+edx], 0" by "mov [edi+edx],al"
because AL is zero in both cases and the AL variations are shorter.
A different style, equally fast according to some testing here:
szRtrim:  ;{lpsrc,lpdest}
    push ebx
    mov ebx,[esp+12]    ;dest
    mov ecx,[esp+8]   ;src
    mov edx,ebx    ;edx maintains 1+address of the last nonspace in the destination
    sub ecx,ebx
;align 8     ;for speed, this can make an appreciable difference
@@: mov al,[ecx+ebx]
    mov [ebx],al
    add ebx,1    ;on the P4 this is massively faster than INC EBX or LEA EBX,[EBX+1]
    test al,al      ;the two-byte INC EBX (db 0FFh,0C3h) is not great either
    jz short @F
    cmp al," "
    je short @B
    cmp al,9          ;tab
    cmovne edx,ebx    ;slower but equivalent is "JE @B / MOV EDX,EBX"
    jmp short @B
@@: mov [edx],al     ;zero
    mov ecx,[esp+12]       ;if desired
    mov eax,edx
    sub eax,ecx      ;return new string length
    pop ebx
    ret 8


Larry,

On late model Intel PIV I would do thes mods to try and get the speed up a little.


  @@:
    movzx eax, BYTE PTR [esi+ecx]   ; copy bytes from src to dst
    mov [edi+ecx], al
    add ecx, 1
    test eax, eax
    je @F                           ; exit on zero
    cmp al, 33
    jb @B
    mov edx, ecx                    ; store count if asc 33 or greater
    jmp @B