[Patch, fortran] Inline DOT_PRODUCT

[Dominique Dhumieres]

Some results with the latest inlined dot_product
(all timing on a 1.8Ghz G5):

fatigue   from  34s to 26s  
induct    from 292s to 68s  
kepler    from  92s to 48s  
test_fpu  from  42s to 39s (CROUT from 10.3s and 16.5s to 8.3s and 15.4s). 

Also the logical test now works with logical 1, 2, 4, and 8.

Note that nf went from 116s (snapshot 4.2-20060218) to 77s (snapshot 
4.2-20060225). Does anyone know why?

Possible drawbacks:

(1) when compiled without -On n>0, I get the following timing per element
(length=1024 elements)

f77 do loop   18.26ns
library        3.55ns
inline        25.09ns

It would probably be nice to use the library with -O0 (may be -O1);
this is another typical use of compilers for some users.

(2) Although I seldom use debuggers, the impact of inlining of their use 
may have been overlooked(?).

Dominique