Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > comp.lang.forth > #47571 > unrolled thread

32 bit VS 64 bit modes

Back to article view | Back to comp.lang.forth

Contents

  32 bit VS 64 bit modes Jason Damisch <jasondamisch@yahoo.com> - 2016-05-27 11:21 -0700
    Re: 32 bit VS 64 bit modes Paul Rubin <no.email@nospam.invalid> - 2016-05-27 13:06 -0700
      Re: 32 bit VS 64 bit modes Jason Damisch <jasondamisch@yahoo.com> - 2016-05-27 13:22 -0700
        Re: 32 bit VS 64 bit modes hughaguilar96@gmail.com - 2016-05-27 13:59 -0700
    Re: 32 bit VS 64 bit modes anton@mips.complang.tuwien.ac.at (Anton Ertl) - 2016-05-28 10:07 +0000
      Re: 32 bit VS 64 bit modes Paul Rubin <no.email@nospam.invalid> - 2016-05-28 19:09 -0700
        Re: 32 bit VS 64 bit modes Paul Rubin <no.email@nospam.invalid> - 2016-05-28 20:00 -0700
        Re: 32 bit VS 64 bit modes Albert van der Horst <albert@cherry.spenarnc.xs4all.nl> - 2016-05-29 17:46 +0000
          Re: 32 bit VS 64 bit modes Paul Rubin <no.email@nospam.invalid> - 2016-05-29 11:27 -0700
    Re: 32 bit VS 64 bit modes Rod Pemberton <NoHaveNotOne@bcczxcfre.cmm> - 2016-05-28 23:42 -0400

#47571 — 32 bit VS 64 bit modes

> In another thread Anton Ertl stated

> I have now access to an Odroid C2 running Ubuntu 16.04 in
> 64-bit mode often called AArch64, but Gforth calls it arm64)

I admit not being the most adroit computer programmer or user
in the world.  So, I thought that I'd ask, if people don't mind
too much, what the various trade-offs and advantages are between
32 bit and 64 bit modes.  For 16 bit VS 32 bit the advantages
are obvious.  But, it seems to me that 32 VS 64 has diminishing
returns.  Maybe 40 bit would be ideal?  What reasons could there
possibly be for having more than 4 GB of RAM?

Jason

[toc] | [next] | [standalone]

#47574

Jason Damisch <jasondamisch@yahoo.com> writes:
> I thought that I'd ask... what the various trade-offs and advantages
> are between 32 bit and 64 bit modes.... it seems to me that 32 VS 64
> has diminishing returns.

Besides address space, 64 bits effectively gives more parallelism.  Like
if you're doing a memory block move without special instructions, doing
it 8 bytes at a time is much faster than with 4.  Similarly if you're
doing wide arithmetic.  

On the other hand, 64-bit programs consume more memory because pointers
are twice as wide, and (depending on language etc.) integers can be
twice as wide too.

> Maybe 40 bit would be ideal?

People want general purpose computers to have convenient byte addressing
these days, since they're often processing character data rather than
numerics.  So the word size is almost always a power of 2.  You see
sizes like 40 bits in specialized processors like DSPs where byte
addressing isn't as important.

> What reasons could there possibly be for having more than 4 GB of RAM?

"640k ought to be enough for anyone..." ;-).  More seriously, one
obvious benefit on PC's is that you can run lots of programs
simultaneously and still allot plenty of memory to each one.  As for
individual programs using GB's of ram, some programs like web browsers
do it, but that's an artifact of the ram being available and therefore
applications bloat up to use it.  Other programs like databases make
actual good use of the memory, though typically not on consumer
desktops.  

It's certainly true that bigger computers let you solve bigger problems.
I half-remember hearing OKAD could use up to a GB or so of ram when it
had the whole GA144 chip design in memory at once.  That's the most
memory intensive Forth program I can think of other than artificial
benchmarks, and of course it would use more with a bigger or more
complicated chip.  But in the benchmark category, Anton's program to
find the 10^9th Hamming number used an 8GB table, iirc.

[toc] | [prev] | [next] | [standalone]

#47575

On Friday, May 27, 2016 at 1:06:26 PM UTC-7, Paul Rubin wrote:

> Besides address space, 64 bits effectively gives more parallelism.  Like
> if you're doing a memory block move without special instructions, doing
> it 8 bytes at a time is much faster than with 4.  Similarly if you're
> doing wide arithmetic.  

Thanks for the input.  Others?

[toc] | [prev] | [next] | [standalone]

#47577

On Friday, May 27, 2016 at 1:22:46 PM UTC-7, Jason Damisch wrote:
> On Friday, May 27, 2016 at 1:06:26 PM UTC-7, Paul Rubin wrote:
> 
> > Besides address space, 64 bits effectively gives more parallelism.  Like
> > if you're doing a memory block move without special instructions, doing
> > it 8 bytes at a time is much faster than with 4.  Similarly if you're
> > doing wide arithmetic.  
> 
> Thanks for the input.  Others?

My language is 64-bit --- the stack is 64-bit wide and numbers are 64-bit --- addresses however are 32-bit and the whole thing must fit in 4GB.

[toc] | [prev] | [next] | [standalone]

#47583

Jason Damisch <jasondamisch@yahoo.com> writes:
>I admit not being the most adroit computer programmer or user
>in the world.  So, I thought that I'd ask, if people don't mind
>too much, what the various trade-offs and advantages are between
>32 bit and 64 bit modes.  For 16 bit VS 32 bit the advantages
>are obvious.  But, it seems to me that 32 VS 64 has diminishing
>returns.  Maybe 40 bit would be ideal?  What reasons could there
>possibly be for having more than 4 GB of RAM?

For having more?  Because you can afford it.  I recently bought a
cheap CPU, motherboard, cooler, and RAM.  I had to decide whether to
buy 4GB for something like EUR20 or 8GB for EUR30 to go with the EUR30
CPU and the EUR 46 mainboard.  I decided to buy 8GB.  And sizes are
growing over time: Moore's law in action.

For using more?  For some applications, you don't need more, for
others, using more helps.  E.g. the game Runes of Magic is a 32-bit
program that crashes if the RAM usage goes above about 1.6GB, even on
a machine with 16GB.  If it was a 64-bit application, these crashes
probably would not happen.  In-Memory databases are a big topic these
days, and they often use more than 4GB.  Or, in the Forth world I used
more than 4GB for something quite a while ago (IIRC it was
<https://www.complang.tuwien.ac.at/forth/programs/sbpcp.fs>).

One can also benefit from having larger integers.  Also dependent on
the application.

Finally, in the case of AMD64 and Aarch64, the 64-bit architectures
provide additional benefits, in particular, more registers than the
32-bit architecture.  That was the main point of my original posting:
Gforth uses the additional registers on Aarch64 for better stack
caching, i.e., faster execution.

- anton
-- 
M. Anton Ertl  http://www.complang.tuwien.ac.at/anton/home.html
comp.lang.forth FAQs: http://www.complang.tuwien.ac.at/forth/faq/toc.html
     New standard: http://www.forth200x.org/forth200x.html
   EuroForth 2016: http://www.euroforth.org/ef16/

[toc] | [prev] | [next] | [standalone]

#47598

anton@mips.complang.tuwien.ac.at (Anton Ertl) writes:
> Or, in the Forth world I used
> more than 4GB for something quite a while ago (IIRC it was
> <https://www.complang.tuwien.ac.at/forth/programs/sbpcp.fs>).

If that took more than 1MB or so, then something was terribly wrong ;).

I think it was actually
  http://rosettacode.org/wiki/Hamming_numbers#Forth
since I remember the discussion here.

[toc] | [prev] | [next] | [standalone]

#47600

Paul Rubin <no.email@nospam.invalid> writes:
> If that took more than 1MB or so, then something was terribly wrong ;).

Oh nm, pi(100000) was just an example and it goes much higher, using
more memory.  Though getting up to GB's would likely take a long time.

It looks like it has been computed with fancy algorithms up to 10**26:
https://en.wikipedia.org/wiki/Prime-counting_function

[toc] | [prev] | [next] | [standalone]

#47610

On 2016-05-29, Paul Rubin <no.email@nospam.invalid> wrote:
> anton@mips.complang.tuwien.ac.at (Anton Ertl) writes:
>> Or, in the Forth world I used
>> more than 4GB for something quite a while ago (IIRC it was
>> <https://www.complang.tuwien.ac.at/forth/programs/sbpcp.fs>).
>
> If that took more than 1MB or so, then something was terribly wrong ;).
>
> I think it was actually
>   http://rosettacode.org/wiki/Hamming_numbers#Forth
> since I remember the discussion here.

No, that was Anton's "slightly  better prime counting program" as an
answer to my "yet another prime counting program".

On 64 bits that program can use a couple of Gbytes. The larger the
prime table to start with the faster it goes.
It is probably worth dusting off, make it parallel.

Groetjes Albert

--
Albert van der Horst, UTRECHT,THE NETHERLANDS
Economic growth -- being exponential -- ultimately falters.
albert@spe&ar&c.xs4all.nl &=n http://home.hccnet.nl/a.w.m.van.der.horst

[toc] | [prev] | [next] | [standalone]

#47611

Albert van der Horst <albert@cherry.spenarnc.xs4all.nl> writes:
> On 64 bits that program can use a couple of Gbytes. The larger the
> prime table to start with the faster it goes.

Ok, fair enough.

> It is probably worth dusting off, make it parallel.

I'd be more interested in seeing better algorithms.

[toc] | [prev] | [next] | [standalone]

#47602

On Fri, 27 May 2016 11:21:29 -0700 (PDT)
Jason Damisch <jasondamisch@yahoo.com> wrote:

> > In another thread Anton Ertl stated  

> > I have now access to an Odroid C2 running Ubuntu 16.04 in
> > 64-bit mode often called AArch64, but Gforth calls it arm64)  
> 
> I admit not being the most adroit computer programmer or user
> in the world.  So, I thought that I'd ask, if people don't mind
> too much, what the various trade-offs and advantages are between
> 32 bit and 64 bit modes.  For 16 bit VS 32 bit the advantages
> are obvious.  But, it seems to me that 32 VS 64 has diminishing
> returns.

64-bit x86 provided more registers for 64-bit mode.

They also made some changes to the way instructions address memory.

I'm not up to date on this.  Maybe a post to alt.lang.asm,
alt.os.development, comp.lang.asm.x86, or comp.arch etc will yield more
answers.

From posts to some other groups where I only lurk, I thought Anton had
done some 64-bit x86 work.

> Maybe 40 bit would be ideal?  
...

> What reasons could there
> possibly be for having more than 4 GB of RAM?

Software just keeps becoming more bloated.  E.g., on Linux 4GB is more
than enough, as long as you don't run a web browser, which will slam
your entire 4GB of memory in no time flat.


Rod Pemberton

[toc] | [prev] | [standalone]

Back to top | Article view | comp.lang.forth

csiph-web