This is the mail archive of the gcc-patches@gcc.gnu.org mailing 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]

Re: [patch RFC] SH: Use FRAME_GROWS_DOWNWARD


Kaz Kojima wrote:

Joern RENNECKE <joern.rennecke@st.com> wrote:


Come to think of it, I can't actually recall why we ended up with
an inverted allocation pattern for -fomit-frame-pointer, and hence
I can't verify we still do by merely looking at the source.
The partial spilling code might also have changed the correlation
between offset and usage frequency for -fno-omit-frame-pointer.
This probably calls for some experiments.



I've done a simple experimentation with compiling gcc itself by
the original and the patched compiler. Some object sizes with
-O2 are here:


have you compiled identical sources each time, or did you compile
each compiler with itself?

	
It seems that there is no clear law between the original and
the patched one, though the size of the object may be not so
good measure.

Regards,
kaz


Yes, gcc has a notion of the cost of individual accesses - the ones
that are more frequent ae considered more costly, and are thus expected
to be spilled later.

I tried to sum up your numbers, but still the differences
might be just noise:

bash-2.05b$ cat sizes | awk '{ sum += $2 } END {print sum }'
940888
bash-2.05b$ cat sizes | awk '{ sum += $4 } END {print sum }'
940976
bash-2.05b$ cat sizes | awk '{ sum += $3 } END {print sum }'
927892
bash-2.05b$ cat sizes | awk '{ sum += $5 } END {print sum }'
926884
bash-2.05b$ bc -l
bc 1.06
Copyright 1991-1994, 1997, 1998, 2000 Free Software Foundation, Inc.
This is free software with ABSOLUTELY NO WARRANTY.
For details type `warranty'.
940976/940888
1.00009352866653629337
926884/927892
.99891366667672530854



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