Re: [gentoo-amd64] Re: Giving up 64 platform

[Brandon Edens]

Having 16 (well actually only 12 according to the SBCL hackers) general purpose
registers (AMD64) versus x86's 8 GPRs is reason enough for me to run AMD64. 

I believe the major performance bottle-neck in modern computers is memory access
latencies and the like. Not just RAM btw, I'm talking about the memory hierarchy:
register -> cache -> RAM -> hard-drive
(ordered by speed of access: left being fastest, right being slowest)

Processor calculation speeds have improved at a faster rate than memory access
speeds. Your processor is essentially just sitting there idling as it waits for
instructions to be fed to it (starved). Having more registers means you have a
larger scratch-pad to perform calculations on. If you had 2 registers and you
wanted to add (w + x) - (y + z). You'd do something like this:

; (w + x)
load w to register1
load x to register2
add register1 and register2 then store result to register1
save register 1 to memory cell ABCD

; (y + z)
load y to register1
load z to register2
add register1 and register2 then store result to register1

; (w + x) - (y + z)
load contents of memory cell ABCD to register2
subtract register1 from register2 then store result to register1

So if I've done everything correctly I now have the result in register1 Now
suppose you had 8 registers instead of just 2. Now you could do something like
the following.

load w to register1
load x to register2
load y to register3
load z to register4
add register1 and register2 store the result in register5
add register3 and register4 store the result in register6
subtract register6 from register5 and store the result in register7

and we still have one register that wasn't used.


The first example?
5 loads
1 save

The second example?
4 loads

Even though the additional save and load is going to cache memory (hopefully),
it still is a performance penalty and those penalties will add up.

This is a trivial example but it might give you an idea as to why registers are
important.


So to continue, I'd thought for awhile that AMD64 running in 32 bit mode might
be using virtual or register windowing. So that when in 32 mode all of the 16
GPRs were being used by processes competing for the CPU.  ie process 1 is using
registers 0-7 and process 2 is using registers 8-15.  Sadly, this is not the
case.

So knowing all this, you make the call. I'd use AMD64 because it is the future.
Well that and that they could always use additional testers.

When a 64 bit flash and proprietary codecs are released this will hopefully all
be irrelevant.

Brandon Edens


On Mon, Apr 24, 2006 at 12:12:08PM -0300, Allan Spagnol Comar wrote:

> Hi all, Thanks for all the advises !!!
>
> I will try run some binary 32 bits programs ( until now I was using
> just source programs )
>
> I got curious with chroot enviroments, is there any literature that I
> can find ( like manuals and samples, I am newbee speaking of cross or
> dual platforms ).
>
> Duncan, you real are a little radical but I like some of your
> thoughts, one thing that make me curious it will be about the
> advantages of using a gentoo x86 on a AMD64, once I would not be able
> to make optimizations on compilation flags ?
>
> Thanks Again, Allan
> --
> An application asked:
> "Requeires Windows 9x, NT4 or better",
> so I?ve installed Linux
>
> --
> gentoo-amd64@g.o mailing list
>