- wrote corresponding fast versions in fast.arith.s
- implemented in-place operations for some routines
- updated existing code to be compatible with in-place
  routines

These changes allow the pidigits benchmark to run
approx. 30% faster. Enabling the assembly routines
in fast.arith.s will give another approx. 3%.

R=r
DELTA=486  (252 added, 68 deleted, 166 changed)
OCL=32980
CL=33003