Re: generated decoder code question

[Dave Brolley <brolley at redhat dot com>]

Frank Ch. Eigler wrote:

> Hi -
>
>  
>
> > In the utils-sim.scm file, the "-gen-decode-insn-entry" function has
> > the following lines:
> > 		     ; Generate code to check that all of the opcode bits 
> > 		     for this insn match
> > [...]
> > It generates an IF-statement to check the opcode again inside the case
> > statement before doing the real extraction. This seems really
> > redundant 
> >    
>
>
> I believe that this check is done because the switch statement by
> itself is not necessarily sufficient.  The switch may simply
> disambiguate the instructions (so no two candidate instructions fall
> into the same switch case), but that could leave some additional
> decodeable bits.  Those bits would need to be tested to tell apart the
> remaining candidate instruction from illegal opcodes.
>  
This is correct. The switch only disambiguates the valid insns. 
Sometimes when a case has been reduced to only one insn, there are still 
some opcodes bits which have not been tested. In this case only, the 
additional test is generated.

>  
>
> > and make the decoder inefficient.
> >    
>
> Do you have a sense of how frequently such a test is completely
> redundant, and how much additional time this test takes?
>  
Please point out any cases where the test is not needed and I will see 
if I can eliminate it in those cases.

Thanks,
Dave