Silvius Rus - Re: Patch: New implementation of -Wstrict-aliasing (original) (raw)

This is the mail archive of the gcc-patches@gcc.gnu.orgmailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

On 01 Feb 2007 05:01:00 -0600, Gabriel Dos Reis gdr@cs.tamu.edu wrote:

Do you have some data about the relative cost (in terms of compiler-time) of level 2 vs level 3? This is for decidiing which should be the default.

Hi Gaby,

The compile time difference from level 2 vs level 3 appears to be insignificant at both -O0 and -O2 (see timing results below). The only potential serious source of overhead would be the points-to based analysis, which only runs at -O2 or higher. However, the overhead turns out to be very low because the points-to information is highy preprocessed by pass_may_alias. The warning pass looks only at pointers that are both aliased and known to be dereferenced in the current function. Also, level 2 may end up looking at more conversion expressions because it does not check whether pointers are referenced, so in some cases level 2 could actually be slightly more expensive than level 3.

Let me know if you agree with default = level 3.

Is there a standard way to measure GCC's compilation overhead? Here are the timing results from building GMP on i686 (all times in seconds):

-O0: 54.069 -O0 -fstrict-aliasing: 54.948 -O0 -fstrict-aliasing -Wstrict-aliasing=2: 55.699 -O0 -fstrict-aliasing -Wstrict-aliasing=3: 55.657

-O2: 67.253 -O2 -fstrict-aliasing: 68.071 -O2 -fstrict-aliasing -Wstrict-aliasing=2: 68.852 -O2 -fstrict-aliasing -Wstrict-aliasing=3: 68.886

Many thanks for doing this. -- Gaby

I am happy to help. Silvius

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]