Compilation performance comparison of 3.5.0 and TreeSSA trees on MICO sources as requested in: [tree-ssa] Merge status 2004-05-03

[Daniel Berlin]

On May 5, 2004, at 4:12 PM, Karel Gardas wrote:

> On Wed, 5 May 2004, Daniel Berlin wrote:
>
>> Note that in almost all of his cases, the parser takes at least 15%,
>> and usually 30%+ of the time of compilation.
>
> Yes, that's true. Parser and name lookup are most expensive tasks. But 
> how
> is it possible that trunk is so much (~20%) faster than tree-ssa, when
> tree-ssa is based on trunk?
Good question.
>
>
>> I only mention this because i'm pretty sure that right now, even if we
>> made gimplification take no time on these testcases, we'd still have a
>> large percent regression just from the parser times.
>
> Agree, but maybe it just tell something about not so good 
> optimizations on
> tree-ssa branch in comparison with trunk, i.e. trunk optimizes C++ 
> parser
> much better...

In one case, you are talking about it getting 50% slower. I have a very 
hard time believing we are optimizing things that much worse in an 
entire file.
It is also consistent across platforms (IE more than just x86), is it 
not (I don't have a 3.5 build handy on this powerbook)?


> Is it possible to not bootstrap tree-ssa, but just compile
> it by GCC3.4.0/3.5.0 and see if parser is faster? If so, how?


Sure.
Don't do make bootstrap, just do make
It'll compile gcc with your system compiler