[geeks] Itanium 32 bit performace.... hahahaha

Francisco Javier Mesa-Martinez lefa at ucsc.edu
Sat Dec 20 03:25:30 CST 2003


On Thu, 11 Dec 2003, Charles Shannon Hendrix wrote:
> Did the DEC compiler around 1996 pack bytes to make it more efficient?

Hum, that is a good question. I don't know how efficient is to pack bytes
into a word, since you may still need to do a few shifts on that word to
get the bits that you are interested in.

> Anyone know how GNU C handles it?

GNU C used to be pretty bad at it, but I have not look at the internals
for a while, I need to check on it. I know the first quick hack did use up
a whole word even for a single byte, so it chew memory like no body's
business but that was in the 2.6/7 days.

> I remember it eating RAM quickly on some projects, and not others.
> Maybe it was inconsistent.
>
> There were problems with using a lot of barrier instructions on the
> 21064 as well.

The 21064 was geared towards making the common case fast, barrier
instructions may have been deemed to not be that common.

> The later Alphas are so good at preduction, it almost doesn't matter.

I do not follow here, they do have prediction and some serious rebuffering
capabilities, so I assume that you wanted to say that the 21164 and later
are less dependant on good code ordering, which is true (when compared
with the 21064).



More information about the geeks mailing list