new-ra code sizes on Hitachi H8

[tm_gccmail@mail.kloo.net]

I was curious about the effect of new-ra on the H8 codegen, so I built a
table. The baseline is GNUPro 2000r1, using options -mint32 -mh.
 
.text section sizes (in hexadecimal):

                                       -fnew-ra
                GNUPro   10/22   2/12    2/12
file            2000r1   2002    2003    2003
----------------------------------------------------
* layer3.i        7bbc   79d0    768c    6e32
* l3bitstream.i   28a4   29ec    287e    2750
* navion_aero.i    5d6    5c8     596     59a
@ advdomestic.i   21e4   2104    20fa    1f70
@ aiunit.i        5826   53aa    52d0    5050
@ chanserv.i      e6c0   df84    cc06    c8d0
@ melee2.i        65f4   608a    649c    5d14
@ wizard1.i       4268   3e6e    3ce0    3c5e
+ adler32.i        1b8    19c     188     19e
+ blowfish.i      24b6   21a2    18f0    19bc
+ imdecode.i     22f2e  220fc   215c4   1e9ea
+ map_fog.i       93c4   9b84    8fea    8adc
+ scanline.i      11d6   1188    113e    1020
+ tif_fax3.i      2106   213a    2044    1ffc
. quantize.i      2bce   2b52    2aac    29ac
. tif_packbits.i   a44    5a4     51e     57c
. s_serv.i        6b2c   6902    663c    6494

* = heavily floating-point testcase
@ = branchy testcase
+ = loopy testcase
. = mixed testcase

The Hitachi H8 is a fairly difficult register-allocation situation - only
8 general-purpose registers. So new-ra seems to be doing fairly well.
Only 4 testcases increased in size; the other 12 decreased in size.

Toshi