On Thu, 2009-02-26 at 21:04 +0000, Richard Sandiford wrote: > When I did some performance measurements a few years back -- probably > on a VR413x -- n32 did give a noticeable improvement over o32.
For the fun of it I did some tests on gzip and bzip2 (sources patched debian lenny - except for bzip2 I forced static linking) on gcc51 (loongson 2F). system gzip and bzip2 are compiled (I assume) using debian GCC 4.3.2 for MIPS I with abi=32. My versions are compiled with trunk rev 144425 with abi=xx and with or without -march=native. Time reported is best of 4 sequential runs and where stable. For bzip2 trunk regress badly in performance against 4.3.2 but n32 is indeed relatively faster than 32 (still slower than 4.3.2). For gzip trunk matches 4.3.2 but n32 is slower than 32 and 64 even slower. Sincerely, Laurent gue...@gcc51:~$ time gzip -9c gmp-4.2.4.tar > /dev/null real 0m4.508s user 0m4.457s sys 0m0.050s gue...@gcc51:~$ time bin/gzip-trunk-144425-32 -9c gmp-4.2.4.tar > /dev/null real 0m4.515s user 0m4.455s sys 0m0.057s gue...@gcc51:~$ time bin/gzip-trunk-144425-n32 -9c gmp-4.2.4.tar > /dev/null real 0m4.742s user 0m4.697s sys 0m0.043s gue...@gcc51:~$ time bin/gzip-trunk-144425-64 -9c gmp-4.2.4.tar > /dev/null real 0m5.457s user 0m5.400s sys 0m0.052s gue...@gcc51:~$ time bzip2 -9c gmp-4.2.4.tar > /dev/null real 0m11.274s user 0m11.200s sys 0m0.069s gue...@gcc51:~$ time bin/bzip2-trunk-144425-32 -9c gmp-4.2.4.tar > /dev/null real 0m18.697s user 0m18.618s sys 0m0.072s gue...@gcc51:~$ time bin/bzip2-trunk-144425-32-native -9c gmp-4.2.4.tar > /dev/null real 0m19.800s user 0m19.704s sys 0m0.083s gue...@gcc51:~$ time bin/bzip2-trunk-144425-n32 -9c gmp-4.2.4.tar > /dev/null real 0m14.873s user 0m14.796s sys 0m0.069s gue...@gcc51:~$ time bin/bzip2-trunk-144425-n32-native -9c gmp-4.2.4.tar > /dev/null real 0m19.773s user 0m19.703s sys 0m0.062s gue...@gcc51:~$ time bin/bzip2-trunk-144425-64 -9c gmp-4.2.4.tar > /dev/null real 0m19.792s user 0m19.701s sys 0m0.082s gue...@gcc51:~$ time bin/bzip2-trunk-144425-64-native -9c gmp-4.2.4.tar > /dev/null real 0m16.388s user 0m16.305s sys 0m0.075s