Changes by: Aleksey Nogin (nogin at cs.caltech.edu)
Date: 2006-01-13 14:42:11 -0800 (Fri, 13 Jan 2006)
Revision: 8469
Log message:
- Do pro-active normalization of the numerical expressions (so that they are
normalized before they are duplicated). This is done by changing the
annotation processor for all the hoas_lof/hoas_normalize resources.
- substl coalescing (substl_substl_lof2) is not a part of the normalizeLofC
(it has the exactly right style and it is wasteful to dedicate a separate
sweepUpC to it).
- Tried making the "constant reduction" more robust. In particular, when
normalizeLofC performs a "constant reduction", it will now follow up by
using reduceC to fully propagate the constant. I've removed Jason's "I
believe they are too fragile" comment because I think it no longer applies.
Also, moved the hd_lof rewrite into the "constant reduction" section (where
it belongs).
This reduces the THEORIES=all expansion time and #of primitive steps by >10%.