Re: Any more graceful way to saturate adds?

Esra Sdrawkcab wrote:
> Might it be "better" (at least smaller code) to place the stosb after 
> the "next" label? but then you'd need a setup JMP to start,
> [and a "cx=0" test at some point!]
> 
>  xor dx,dx
>  jmp short start
> 
>  next:                   ; bytes
>    stosb
> start:
>    mov dl,[bx+si]                ; 2+1
>    lodsb                 ; 1+1
>    cbw                   ; 1     -128 to 127
>    add ax,dx             ; 2     -128 to 383
>    test ah,0ffh
>    jz next

Yes indeed, this is a very common/useful optimization, I didn't do this 
simply because I didn't know what the surrounding code looked like, I 
just wrote a minimal wrapper around the core clamping code.

BTW, there is a very sneaky way to avoid that initial jump:

Instead of

  jmp short start
  next:
    stosb
start:

you can instead write:

   mov al,0AAh
org $-1		; Back up one byte into the immediate constant above
next:
   stosb

I.e. on the first iteration we do a dummy two-byte instruction with a 
one-byte immediate, which we then make sure contains the STOSB opcode
byte.

The gain is that we save one opcode byte (MOV instead of JMP SHORT) and 
a taken branch, so the actual cost of rotating the loop like this is 8 
cycles on the first iteration, instead of 17, and each iteration saves 
two opcode bytes or 8 cycles.

Terje

-- 
- <Terje.Mathisen@[EMAIL PROTECTED]
>
"almost all programming can be viewed as an exercise in caching"