Re: [Sbcl-devel] [Sbcl-commits] master: Optimize the compiler a bit.

[Stas B. <sta...@gm...>]

Christophe Rhodes <cs...@ca...> writes:

> "stassats" <sta...@us...> writes:
>
>> +  (do () (())
>> +    (cond ((eql x y) (return t))
>> +          ((and (consp x)
>> +                (consp y)
>> +                (eql (pop x) (pop y))))
>> +          (t
>> +           (return)))))
>
> Is (do () (()) ...) the same as (loop ...)?
It is, but seeing  other parts not using LOOP, I opted for DO.
>>                                           (if rest
>>                                               (type-hash-value rest)
>>                                               42)
>> -                                         (sxhash allowp))
>> +                                         (if allowp
>> +                                             #.(logand #xFF (sxhash t))
>> +                                             #.(logand #xFF (sxhash nil))))
>>                                          #xFF)))
>
> I haven't looked at the code, but this looks wrong: there's no guarantee
> that SXHASH on the host (evaluated at #. read-time) is the same as
> SXHASH on the target; the read-time evaluation bakes-in a host value,
> and so is going to lead to changed behaviour depending on which host is
> used to build.  Is this the cause of the speedup you refer to on IRC?
> If it is, maybe LOAD-TIME-VALUE works?
This part isn't particular slow, but the sxhash itself value isn't
important, perhaps a random value can be selected instead, just like the
42 one.

-- 
With best regards, Stas.