Re: Optimized program four times slower.

[Tim Prince <n8tm at aol dot com>]

nightstrike@gmail.com wrote:
> On 7/14/07, Tim Prince <tprince@myrealbox.com> wrote:
> > Brian Dessent wrote:
>
> > you should file a PR.  Note that dec, rather than add -1, has been a
> > standard way to write code which might be OK on AMD but dead slow on
> > Intel, since 6 years ago.
>
> Why is add -1 slower than dec?
Did you look up partial flags stalls, if you haven't done so the last 6 
years?  dec has the same problem as inc.
dec is slower than add -1 or sub 1, on many Intel CPUs, beginning with 
P4.  The time required to restore the flags which are to be preserved by 
dec is outrageous.  Often, there may be an additional advantage in 
moving the add -1 further up the chain, so as to allow more opportunity 
to prepare the branch.
I doubt Intel foresaw the deficit which would be incurred by default 
options in the gcc 3.x of RedHat/CentOS 4.x using dec by default,even 
though it has been documented by Intel 
http://www.intel.com/intelpress/toc-soc.pdf
and others, for years.
If you are using these oldish gcc versions, the Intel options don't 
generally hurt AMD, but the default AMD options do hurt Intel.