Answering msg from Paul Edwards to Matthew Parker,
on Wednesday December 27 1995 at 12:56

 PE>  002c  8a 01             L2              mov     al,[ecx]
 PE>  002e  88 02                             mov     [edx],al
 PE>  0030  41                                inc     ecx
 PE>  0031  42                                inc     edx
 PE>  0032  3c 0a                             cmp     al,0aH
 PE>  0034  75 f6                             jne     L2

; Entry requirements:
;  ESI = pointer to source  (ECX above)
;  EDI = pointer dest       (EDX above)

    push    es      ; save ES

    push    ds
    pop     es      ; ES=DS

L2:
    lodsb           ; load AL from DS:[SI] then increment SI
    stosb           ; save AL to ES:[DI] then increment DI
    cmp     al,0ah
    jne     L2

    pop     es      ; restore ES

 PE> This code is what will be the bottleneck in any text processing
 PE> applications using fgets() with PDPCLIB, and I was thinking it
 PE> might be possible/better to use some assembler here, to use a
 PE> MOVSB or something.  If the average length of a text line that is
 PE> processed by fgets is say 30 bytes, then this loop will be
 PE> executed 30 times.

I don't think you'll see much of a speed improvement. On the old 8088s the
string commands such as LODSB, STOSB, MOVSB were much faster, but now
"hand rolling" them discretely as above can be just as fast,
sometimes faster.

 PE> I was wondering if you had any feel for what would be best
 PE> time-wise.  This code is a C runtime library for OS/2 32-bit
 PE> 80386 and above.  The C runtime library, being of dire
 PE> importance, can be as dirty as you can make it, so long as it's
 PE> FAST FAST FAST!  BFN.  Paul.

Rethinking it with SCASB (scan for byte) then MOVSW/MOVSB might be a better idea.

---