How long should -O1 compiles take?

[Brad Lucier <lucier at math dot purdue dot edu>]

In August I started a thread about slowdowns in global-alloc going from
egcs-1.1.2 -> gcc-2.95.1, and Joern Rennecke came up with a design for
a fix, which I have not implemented (nor has anyone else); the last
message was

http://egcs.cygnus.com/ml/gcc/1999-08/msg00416.html

The last time I got the cvs mainline to compile, it seemed even slower
than 2.95.1 with -O1.

So I want to ask the more general question---how long should -O1 compiles
take?  In other words, what are the design goals for gcc -O1?  Is it OK
to add algorithms to -O1 that double or triple the compile times for some
programs?  Is -O1 supposed to be a balance between local optimizations
and fast compile times, or is it supposed to be something else?  A person
whose judgement I respect said that gcc was one of the few compilers that
works hard to keep resource requirements down. (I see this in the recent
GC work, for example.)  Is this now a design consideration for -O1,
and will it be in the future?

Brad Lucier