Re: [bug-mes] mescc and Load/Store architectures

[Jan Nieuwenhuizen]

Danny Milosavljevic writes:

Hi Danny,

> On Thu, 07 Feb 2019 16:08:38 +0100
> Jan Nieuwenhuizen <address@hidden> wrote:
>
>> Jan Nieuwenhuizen writes:
>> 
>> Oh, I just realise that whether or not we prototype/use these
>> 
>> > (define (armv4:label->arg info label i)
>> >   `(("ldr____$i32,%s" (#:address ,label))
>> >     ("push___%s)))  
>> 
>> > (define (armv4:r-mem-add info v)
>> >   (let ((r (get-r info)))
>> >     `(,(if (< (abs v) #x1000) `(,(string-append "ldr___(%" "r),%s")
>> >                                 ,(string-append "add___$i8,(%" s ")") 
>> > (#:immediate1 ,v)
>> >                                 ,(string-append "str___%s,(" r ")"))
>> >            `(,(string-append "ldr___(%" "r),%s")
>> >              ,(string-append "add____$i32,(%" r ")") (#:immediate ,v)
>> >              ,(string-append "str___%s,(" r ")"))))))
>> >  
>> 
>> for ARM, we can still use the same for x86 and thus keep using the same
>> M1 alphabet across processors.  That was probably part of your
>> suggestion, right?
>> 
>> That could be a feature, it could even reduce the alphabet specified in
>> x86.M1.  Not sure if a performance loss, if any, is relevant.
>
> Yes, that would be ideal.
>
> Nowadays we have to measure the performance loss, the CPUs are too
> complicated for us to tell otherwise.  A wild speculation would say
> that CISC CPUs do the same RISC steps internally anyway (in
> microcode), so it shouldn't make a difference, except:

Hah!  It's hard to defeat the physics.

> If there are more instructions per program, the instruction cache will
> miss more and be eventually be full.  In general probably more loads
> from main memory need to be done, so it *could* be that it's slower
> with more instructions per program compared to the fewer instructions
> per program.

Yes, I can imagine that using more registers and avoiding re-fetching
from memory really helps, but using more registers also means a richer
alphabet in ??.M1, which is currently not really what we want.

I'd gladly see some optimizing work someday in MesCC, could be fun
too ;-)

janneke

-- 
Jan Nieuwenhuizen <address@hidden> | GNU LilyPond http://lilypond.org
Freelance IT http://JoyofSource.com | Avatar® http://AvatarAcademy.com