This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: stack variable liveness analysis
On Sun, Aug 23, 2009 at 12:24 PM, Xinliang David Li<davidxl@google.com> wrote:
> On Sun, Aug 23, 2009 at 9:03 AM, Steven Bosscher<stevenb.gcc@gmail.com> wrote:
>> Hi,
>>
>> +perform_stack_var_liveness_analysis (void)
>>
>> Please don't add local dataflow solvers. Just use DF instead, please.
>
> Is DF designed to work on tree/gimple level analysis?
The solvers are.
It hands you edges/basic blocks/etc, you fill in the bitmaps for local
computation, etc.
See struct df_problem in df.h
> The effort of
> adapting into existing DF framework seems much larger than having the
> private little solver which has very little overhead, but I could be
> wrong.
Honestly, i think you are wrong ;)
The problem is a lot of people felt this way, and we ended up with 30
little solvers here and there in GCC to maintain.
Most of them got quite slow as they ended up getting bigger and bigger
problems, got hacked to be faster in different ways, became harder and
harder to maintain.
--Dan