This is the mail archive of the
mailing list for the GCC project.
Re: Speedup CSE by 5%
> * cse.c (find_best_addr): Don't call copy_rtx before calling
> fold_rtx. Save cost recomputation if fold_rtx did nothing.
This is a great patch!
Here is the timing in seconds for five runs of ./cc1 -quiet -O2".
original patched diff
combine.i 18.186 18.043 -0.786%
fold-const.i 40.025 39.379 -1.613%
reload.i 14.524 14.374 -1.032%
cc1-i files 218.025 216.066 -0.898% <- wow!
The cc1-i files test was run only once for the unpatched and patched
versions of cc1.
Although you have already tested that your patch has no effect except
on the compile time, you may want to do a regression check just to be
sure and convince whoever is approving this patch.