Features

[Wei Li]

Dear all:

If I am correct, MILEPOST features are optimization independent as
they are extracted at the SSA level. However, I am getting different
features for a given program under different optimization levels ( O0,
O1, O2 and O3). Am I missing something here?

--
Regards,

Wei Li

[Joern Rennecke]

Quoting Wei Li <weil...@gmail.com>:

> Dear all:
>
> If I am correct, MILEPOST features are optimization independent as
> they are extracted at the SSA level.

Huh? Compiler flags and parameters are features, and they definitely
depend on the optimization level.

Or are you referring only to the features of the "ml-feat" pass?
Just from looking at the compiler sources I can't really tell exactly where
this pass is inserted, but it is most likely after the lowering passes;
the lowering passes already depend on the optimization level.
Moreover, some functions are generated as the result of intraprocedural
or auto-parallelization optimizations.

[Wei Li]

---------- Forwarded message ----------
From: Wei Li <weil...@gmail.com>
Date: Wed, Dec 9, 2009 at 8:19 AM
Subject: Re: Features
To: Joern Rennecke <amy...@spamcop.net>


I am talking about "ml-feat". I get different features if I use O0,
O1, O2 or O3. However, the features do not change if I use different
flags with O3.

--
Regards,

Wei Li



--
Regards,

Wei Li