Michael S (already5chosen@yahoo.com) on 1/13/11 wrote:
>
>Frankly, if you can require 100% ARMv7-AR compliance as a prerequisite for your
>software it does solves majority of the problems.
>You, most likely, can't do it, but your freinds from Redmond can, at least for Nt-based stuff.

Yeah, I do agree. The problem is that ARMv7 may be the
"modern" stuff, but the "ancient" stuff is ancient only
in design - it's still manufactured in mass quantities
(and ARM even designs new "ancient" stuff in the form of
Cortex-M0 etc).

And we do end up supporting them. Maybe we'll never support
M0, but who knows.

And what annoys me about Wilco's wild fanboism is that he
is perfectly happy to ignore all the incredible crap in
ARM ("it's obsolete"), but then when he talks about how
much better ARM is than x86 he certainly isn't willing to
ignore the x86 legacy.

And the thing is, x86 today is better than ARM
today. On x86, you really absolutely can ignore all
the worst insanities - nobody cares about the 16-bit modes
any more. Because there really aren't any chips that matter
out there. So the historical real baggage doesn't really
hurt.

With ARM, you really cannot just say "ARMv7+VFP". Because
there's a ton of chips being designed and manufactured that
still carry that historical baggae around in a way that
actually hurts you.

And yes, anybody who does pure virtual caches is just crazy.
There is no excuse. ARM is not alone in that heap,
but dammit, don't come to me and talk about a "clean
architecture" when you have sh*t like that.

Linus