Re: [patch] Speed up phi node insertion

[Daniel Berlin <dberlin at dberlin dot org>]

Jan Hubicka wrote:
>> Jan Hubicka wrote:
>>>> Zdenek Dvorak wrote on 07/30/06 06:53:
>> The changes i've got in store to fix this are not really stage3 material:
>>
>> dberlin@dannyb-corp0:~/gccstuff/gcc-pre-speed/gcc> svn diff
>> tree-ssa-pre.c|diffstat
>> tree-ssa-pre.c | 2435
>> ++++++++++++++++++++++++++++++++++++--------------------- 1 file
>> changed, 1546 insertions(+), 889 deletions(-)
>>
>>
>> I've been waiting for stage1 to put them in.
> 
> I would say that we do resonable job on that PR now (I will do some
> summary and propose postnotting rest of 4.2).  Aren't there some very
> low
> hanging fruits in your changes, like inlining get_value_handle that
> ought to help the constant factor at -O2?

I never measured that change separately, but without even more inlining,
it will end being a microoptimization.  The real way to improvement was
calling it less through checking TREE_CODE's, etc.

In fact, on some testcases i have, inlining get_value_handle made things
slightly slower :(

Sadly, what PRE could really use is IPCP to clone and specialize most of
it's functions.  A lot of the places we know things about what kind of
tree we are talking about, and thus, could eliminate most of the
branching in the inlined get_value_handle and friends.

Maybe some day.