Date   : Wed, 14 Feb 1996 11:58:11 +0000 (GMT)
From   : Mark Cooke <mpc@...>
Subject: Re : Optimization

On Wed, 14 Feb 1996, Tom Seddon wrote:

> On Tue, 13 Feb 1996, Mark Cooke wrote:
> 
> > Depending on your compiler. A potentially better trick is to use
an array 
> > of function pointers. Mileage will vary with compiler.
> gcc does this:
> 
> mov eax,opcode
> mov eax,[opcodesjumptable+eax*4]
> jmp [eax]
> 
> which is about as fast as you can get I think!

Heh - that's not too bad :-). Almost on a par with having the opcodes 
aligned on page boundaries, and doing:

mov ah,opcode        1
xor al,al            1
jmp [ax]             5

Which is what my DOS assembly version does. It means your opcodes all 
have to be < 256 bytes, but that isn't a problem :-). Checking CPU clocks
it doesn't appear to make a difference on a 486, but I would guess 
intuitively that you stand more chance of a cache miss with the extra 
read from the table for the gcc code. Can't expect any better from gcc 
though. It's a nice solution - and using C is probably a lot easier than 
mesing around in assembler...especially 'gas' :-)

Mark

Mark Cooke                             The views expressed above are mine
Systems Programmer                     and do not reflect in any way the
University Of Birmingham               current policy of my employers.