mat64: omit final unused pow squaring

[kortschak]

@btracey PTAL

name       old mean              new mean              delta
Pow10_3    27.6µs × (0.92,1.07)  21.1µs × (0.98,1.03)   -23.54% (p=0.000 n=10+9)
Pow100_3   2.93ms × (1.00,1.01)  2.03ms × (0.98,1.02)   -30.67% (p=0.000 n=8+10)
Pow1000_3   2.67s × (1.00,1.00)   1.75s × (0.99,1.01)  -34.33% (p=0.000 n=10+10)
Pow10_4    33.6µs × (1.00,1.00)  27.5µs × (0.98,1.03)   -18.14% (p=0.000 n=8+10)
Pow100_4   3.83ms × (0.99,1.01)  2.96ms × (0.99,1.01)    -22.80% (p=0.000 n=9+9)
Pow1000_4   3.55s × (1.00,1.01)   2.67s × (1.00,1.00)  -24.62% (p=0.000 n=10+10)
Pow10_5    34.0µs × (0.99,1.02)  28.7µs × (0.98,1.03)  -15.46% (p=0.000 n=10+10)
Pow100_5   3.83ms × (1.00,1.00)  2.97ms × (0.99,1.01)   -22.31% (p=0.000 n=10+9)
Pow1000_5   3.55s × (1.00,1.00)   2.67s × (1.00,1.00)   -24.98% (p=0.000 n=8+10)
Pow10_6    40.5µs × (1.00,1.00)  35.0µs × (1.00,1.00)    -13.38% (p=0.000 n=9+8)
Pow100_6   4.76ms × (1.00,1.00)  3.92ms × (0.99,1.01)  -17.71% (p=0.000 n=10+10)
Pow1000_6   4.43s × (1.00,1.00)   3.55s × (1.00,1.00)  -19.83% (p=0.000 n=10+10)
Pow10_7    40.7µs × (0.99,1.01)  35.9µs × (0.98,1.03)   -11.75% (p=0.000 n=9+10)
Pow100_7   4.75ms × (1.00,1.00)  3.93ms × (1.00,1.01)   -17.14% (p=0.000 n=8+10)
Pow1000_7   4.43s × (1.00,1.00)   3.55s × (1.00,1.00)  -19.93% (p=0.000 n=10+10)
Pow10_8    48.1µs × (1.00,1.00)  42.6µs × (0.98,1.03)  -11.47% (p=0.000 n=10+10)
Pow100_8   5.72ms × (1.00,1.00)  4.87ms × (1.00,1.00)    -14.92% (p=0.000 n=9+9)
Pow1000_8   5.33s × (1.00,1.00)   4.44s × (1.00,1.00)   -16.63% (p=0.000 n=8+10)
Pow10_9    40.9µs × (1.00,1.00)  35.5µs × (1.00,1.01)   -13.36% (p=0.000 n=10+9)
Pow100_9   4.76ms × (1.00,1.00)  3.93ms × (0.99,1.00)  -17.55% (p=0.000 n=10+10)
Pow1000_9   4.43s × (1.00,1.00)   3.55s × (1.00,1.00)   -19.88% (p=0.000 n=9+10)

[btracey]

Isn't it just better to say n > 1 in the for loop?

[kortschak]

That's not correct; you don't get the last power multiplication either.

[btracey]

Oh, right, sorry.

[kortschak]

Having said that, there is another optimisation here that I will try later today.

[btracey]

LGTM after the one question

I like speedups