[ira] better allocation for -O0
Vladimir Makarov
vmakarov@redhat.com
Mon May 5 21:31:00 GMT 2008
Hi, this patch is a fast allocation algorithm rewriting. Instead of
possible conflict info, it uses live ranges which is more accurate
info. The patch makes IRA in -O0 even a more faster (93.86 user and
1:56.98 wall for IRA vs 97.71 and 1:59.88 for the old allocator on
all_cp2k_fortran.f90) and better allocator. So now IRA with -O0 works
as fast as the old allocator but generates smaller (2.5% smaller on
all_cp2k_fortran.f90) and faster (up to 2 times on some benchmarks)
code in -O0 mode.
The patch was successfully bootstrapped on x86_64 and tortured on
x86/x86_64.
2008-05-05 Vladimir Makarov <vmakarov@redhat.com>
* ira-color.c (allocno_assign_compare_func): Remove.
(ira_fast_allocation): Rewrite.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: O0better.patch
Type: text/x-patch
Size: 5760 bytes
Desc: not available
URL: <http://gcc.gnu.org/pipermail/gcc-patches/attachments/20080505/b66c31a2/attachment.bin>
More information about the Gcc-patches
mailing list