[tremor] Tremor on x86 vs ARM, performance highlights: CPU use 120x different

[slothead]

OK, MULT32 and its fellow callers are using 30% of my CPU time.  I replaced MULT32 with a simple if
(x > 1) return 1; return 0; and let her fly.  She flew, only 30% faster.  Dilb, you still talking to
me?  If so, maybe you see if that has such a dramtic affect for you.  I bet it doesn't.  A good
candidate for me to start looking at, anyway.  If I can get that 50% faster, just that single macro,
I shave off 15%.  If I had x-ray eyes would be even better.

To Dilb:

Tweaking compiler generated assembly usually won't get you more than a few percent.  I really only
glanced at it because it was very, very difficult to follow, like all compiler-generated output is.
If a real person had done it, sure from scratch, it makes for a lot better chance of creaming the
compiler, and it's a lot easier on the eyes, if not to follow.  Just wanted to make sure you didn't
that all by your lonesome.

x86 is relative to this because it's running 120x faster, per MHz, than the strongarm version.  That
says a lot.  They may be "different' but they both run the exact same tremor source here.  120x is
absurd.  That's 12000% (twelve thousand percent) faster, per mhz.

--- >8 ----
List archives:  http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/
To unsubscribe from this list, send a message to 'tremor-request at xiph.org'
containing only the word 'unsubscribe' in the body.  No subject is needed.
Unsubscribe messages sent to the list will be ignored/filtered.