Re: Switch Case Loop Slower For Emulator

["Redskull" <redskulldc@DESPAM.yahoo.com.au>]

"Bryan Parkoff" <BParkoff@satx.rr.com> wrote in message
SPidnUW3adArBjWjXTWQkQ@giganews.com">news:SPidnUW3adArBjWjXTWQkQ@giganews.com...
> > Go one step further and create a table which is word indexed according
> > to opcode and the following byte.
> > You can then do a direct jump to an opcode handling routine and pass it
> > the operand (in the case of single byte operands).
> > This saves an extra memory fetch for the operand as well.
>     I agree with you, but some opcodes are not used frequently.  For
> example, Intel has over 50,000 opcodes that it has to emulate on PC Power
or
> 68xxx.  It will be very slow.

Only 256 (or less, depending on which 6502 model you are emulating)

>     There is another way.  Add 8 bytes (Not Bit) in this table that uses
> DWORD indexed.  First DWORD will be opcode.  Second DWORD will be pointer
> address to each opcode's Jump Address Location.  It will use only one IF
> THEN statement instead of switch case statement.  It will run much faster
> with frequent opcodes better than infrequent opcodes.
>     Table with DWORD indexed should use dymanic memory instead of static
> memory.  New opcode will be added every time Table with DWORD indexed will
> be removed from dymanic memory before it will be created back to dymanic
> memory.  It might increase the performance with frequency opcodes.
>     What do you think?

I think the best speed increases would be gained from doing a 6502 JIT
emulator version.

Seems a lot of hard work for not a great deal in return.
My 2GHz P4 will emulate the GS (using Kegs) at well over 125MHz equivalent
speed.
I have to throttle it to 80MHz, otherwise there appear to be problems with
the pointer being correctly updated in GSOS.

Red