Enhancements to expt

[jpellegrini]

• Don't use Scheme arithmentic where not needed in exponentiation by squaring (the exponent is known to be a fixnum).
• Fast path for exponent=2 (much much faster).
• Change name exact_exponent_expt -> fixnum_exponent_expt
• Move the bignum check to my_expt, so the exponent is always a fixnum when calling fixnum_exponent_expt.
• Since the exponent is always a fixnum in fixnum_exponent_expt, compare with MAKE_INT(0) instead of using zerop.

Results:

• for some integer exponents, expt is faster
• specifically, for exponent = 2, it is more than twice as fast

[egallesio]

Hi @jpellegrini,

Just realized that I forgot to merge thi PR before working on expt.
I'll try to do the adaptation.

[egallesio]

Merged by hand. It seems OK.

Furthermore, exact_exponent_expt take a long instead of a SCM now

The function signature was

static inline SCM fixnum_exponent_expt(SCM x, SCM y)                    

It is now

static inline SCM fixnum_exponent_expt(SCM x, long y)                   

We have fewer conversions from Scheme fixnums to longs now.
Not tested if it is really more efficient (it should).

[jpellegrini]

It seems generally faster!

This:

(let ((a 0)
      (b 10)
      (e 2.0))
  (time
   (repeat 10_000_000
     (set! a (expt b e))))
  a)

Runs i nsomething like half the time, compared to the STklos checkout from last week. 😀

[jpellegrini]

Runs i nsomething like half the time, compared to the STklos checkout from last week. 😀

Even with the float exponent!

[egallesio]

That's great!
It's crazy the number of places and things to optimize/correct on a single function.