[tree-ssa] Speeding things up

[Daniel Berlin]

On Friday, February 21, 2003, at 11:02  AM, law@redhat.com wrote:

> In message <Pine.LNX.4.33.0301301206530.17724-100000@wotan.suse.de>, 
> Michael Ma
> tz writes:
>
>> First, we were talking about get_immediate_dominator().  And of 
>> course you
>> wouldn't directly inline those whole functions, but instead write an
>> inlinable accessor to that function, which additionally caches the
>> result.  What you've done right now is to convert:
> My point was that there is this knee-jerk reaction by some folks to
> inline functions which are components of performance issues rather
> than taking a deeper look at the problem.  And that was precisely
> what was suggested at one point -- inlining of get_immediate_dominator.
>
> [ In fact, the caching that is done in get_immediate_dominator was 
> introduced
>   because it was being called billions of times, without the author of
>   that cache ever looking at why it was called so often.  ]
>
That's just bullshit.
It was slow when it was being called a smaller number of times.
Do you really think I didn't *look* at why/when/how it was called 
before introducing the cache.

If it was only slow when called billions of times, i would have simply 
fixed the callers not to directly call it a billion times, but *not* by 
introducing another cache.