Re: 3.2 vs 3.1 speed

[Dan Nicolaescu <dann at godzilla dot ICS dot UCI dot EDU>]

One of the reasons that 3.2 is slower than 3.1 is that 3.2 just
generates a lot more RTL (also see the huge increase in time spent in
expand).

Here are the size of some dumps: 
for gcc -O2

-rw-------   1 dann     grad     26647634 Jun  5 11:46 3.2/orb.i.00.rtl
-rw-------   1 dann     grad     16073122 Jun  5 11:46 3.1/orb.i.00.rtl


and for gcc -O2 -fno-inline
-rw-------   1 dann     grad     5969090 Jun  5 12:16 3.2/orb.i.00.rtl
-rw-------   1 dann     grad     5890276 Jun  5 11:54 3.1/orb.i.00.rtl

When using -O2 -fno-inline the difference between the times spent in
the scheduler (and the absolute times) decreases a lot. 

3.1 scheduling            :   3.32 ( 3%) usr   0.04 ( 1%) sys   3.00 ( 3%) wall
3.2 scheduling            :   3.41 ( 3%) usr   0.00 ( 0%) sys   2.00 ( 2%) wall

Also note that the 3.2 I used in all tests predates the PCH merge. 

Is this increase in the size of the generated RTL OK?