Fix pow performance

[hersle]

Addresses #105. Similar to #85. Now the example code in the issue on my machine gives:

T = Float32, sign = 1:
  10.300 ns (0 allocations: 0 bytes)
  10.300 ns (0 allocations: 0 bytes)
T = Float32, sign = -1:
  2.700 ns (0 allocations: 0 bytes)
  2.500 ns (0 allocations: 0 bytes)
T = Float64, sign = 1:
  16.200 ns (0 allocations: 0 bytes)
  16.100 ns (0 allocations: 0 bytes)
T = Float64, sign = -1:
  2.700 ns (0 allocations: 0 bytes)
  2.500 ns (0 allocations: 0 bytes)

Anything I am overlooking here?

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 94.63%. Comparing base (e62c849) to head (27fde84).

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #106      +/-   ##
==========================================
+ Coverage   94.15%   94.63%   +0.47%     
==========================================
  Files           1        1              
  Lines         154      149       -5     
==========================================
- Hits          145      141       -4     
+ Misses          9        8       -1     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[hersle]

Added @devmotion's suggestion and removed unreliable performance tests. Instead, here are results from my machine:

using NaNMath, BenchmarkTools
for T in (Float16, Float32, Float64)
    x, y = rand(T), rand(T) # in (0, 1)
    println("T=$T: ", (@belapsed NaNMath.pow($x, $y)) / (@belapsed Base.:^($x, $y)))
end

Before this PR:

T=Float16: 0.9996094659628969
T=Float32: 3.3202716041996956
T=Float64: 3.0019657223244507

After this PR:

T=Float16: 1.0047619047619047
T=Float32: 1.0098039215686276
T=Float64: 0.9878048780487806

[devmotion]

LGTM, can you update the version number?

[hersle]

Sure, like so? I can't tag the release, though.