Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss

ARM now as powerful as Intel chips

14 views
Skip to first unread message

7

unread,
May 3, 2012, 3:01:35 PM5/3/12
to
ARM now as powerful as Intel chips
----------------------------------

http://www.theinquirer.net/inquirer/news/2172329/tsmc-28nm-31ghz-cortex-a9-chip

A 3.1GHz dual core ARM is out which probably means it will outperform
any Intel CPU where instructions per second bottleneck is a problem.

Big news for Linux.

These devices ought to bring higher than desktop PC performance
to mediocre netbooks and tablets on a low power budget
and probably leave you wanting nothing more than beyond that
for all your every day office and browsing needs.

Since Linux runs on parallel supercomputers, putting 2 or 3 of
these ARM chips and getting a parallel Linux distro up and running
on a tablet with probably take your tablet beyond where gaming is stuck
at with windopws. Any electronics engineer could get this done.

Get coding I say! :-)


OldGoat

unread,
May 3, 2012, 5:30:10 PM5/3/12
to
But will the arms be data alignment tolerant like the Intels are?

David Brown

unread,
May 4, 2012, 3:05:09 AM5/4/12
to
It will probably be as tolerant as x86/amd64 - i.e., misaligned data
works for simple stuff, but at much lower performance, and completely
screws any difficult things like precise ordering for inter-process and
inter-thread communication.

Grant Edwards

unread,
May 4, 2012, 9:20:54 AM5/4/12
to
On 2012-05-04, David Brown <da...@westcontrol.removethisbit.com> wrote:

>> But will the arms be data alignment tolerant like the Intels are?
>
> It will probably be as tolerant as x86/amd64 - i.e., misaligned data
> works for simple stuff, but at much lower performance, and completely
> screws any difficult things like precise ordering for inter-process
> and inter-thread communication.

Finally! I've used a number ARM cores (ARM7, ARM9, StrongARM) and
none of them ever support misaligned data. But, neither did the
SPARC, 68K, PDP-11, and many others -- so it's not like anybody with a
clue expected misaligned data access to work without checking the
processor specs).

It looks like the ARMv6 and ARMv7 do -- at least for simple load/store
operations. There appear to still be alignment requirements for the
fancy bits like multiple-register load/store operations.

--
Grant Edwards grant.b.edwards Yow! Youth of today!
at Join me in a mass rally
gmail.com for traditional mental
attitudes!

David Brown

unread,
May 4, 2012, 10:09:44 AM5/4/12
to
On 04/05/2012 15:20, Grant Edwards wrote:
> On 2012-05-04, David Brown<da...@westcontrol.removethisbit.com> wrote:
>
>>> But will the arms be data alignment tolerant like the Intels are?
>>
>> It will probably be as tolerant as x86/amd64 - i.e., misaligned data
>> works for simple stuff, but at much lower performance, and completely
>> screws any difficult things like precise ordering for inter-process
>> and inter-thread communication.
>
> Finally! I've used a number ARM cores (ARM7, ARM9, StrongARM) and
> none of them ever support misaligned data. But, neither did the
> SPARC, 68K, PDP-11, and many others -- so it's not like anybody with a
> clue expected misaligned data access to work without checking the
> processor specs).
>
> It looks like the ARMv6 and ARMv7 do -- at least for simple load/store
> operations. There appear to still be alignment requirements for the
> fancy bits like multiple-register load/store operations.
>

That's correct, as far as I can see - misaligned accesses work but not
for multiple register operations. I'd guess you also need to keep the
stack aligned. So Cortex M3 (and M4) and Cortex Ax can work with
misaligned data.

Some 68K devices (including the 68332 and the Coldfire) can work with
misaligned data too.

I don't see it as a big issue, however.

OldGoat

unread,
May 4, 2012, 1:58:00 PM5/4/12
to
Sounds like the ARM community is making inroads then.

OldGoat

unread,
May 4, 2012, 2:00:49 PM5/4/12
to
Right now in the Apple camp there was an issue of misaligned data on
their ARM chips. IIRC, the Xcode newsletters I receive, had an article
and one poster had a major problem with it. Consequently, the Apple dev
team pointed him to an IBM site on misaligned data and how to code for
it properly. Rather interesting topic.

Theo Markettos

unread,
May 4, 2012, 7:45:41 PM5/4/12
to
In uk.comp.os.linux David Brown <da...@westcontrol.removethisbit.com> wrote:
> It will probably be as tolerant as x86/amd64 - i.e., misaligned data
> works for simple stuff, but at much lower performance, and completely
> screws any difficult things like precise ordering for inter-process and
> inter-thread communication.

ARMv7 changed the alignment semantics for reading words from
non-word-aligned addresses. This caused a whole pile of pain for people who
were using code from the ARMv3 era where compilers used non-word-aligned
accesses to achieve some speedup tricks (eg on 16-bit values before LDRH
existed) - after all, such accesses were explicitly documented in the ARM
ARM. Alignment exceptions on the ARMv7 kind-of cover this, but with a big
performance penalty (every memory access being trapped is not good news).

Bottom line is, don't rely on non-word-aligned accesses, because the
semantics may well change underneath you and bite you in the future.

Theo
0 new messages