Log2 edge cases

[Ka-zam]

Prior fix only handles neon kernels, the same error existed for x64 kernels. Also increased accuracy and added edge case testing. Refactored polynomial to intrinsics.

RUN_VOLK_TESTS: volk_32f_log2_32f(vlen=131071, iter=1000, tol=5e-06)
arch                       |         time |     throughput |    max_abs |
---------------------------+--------------+----------------+------------+
generic                    |  583.3820 ms |    1797.5 MB/s |          - |
u_sse4_1                   |   39.6278 ms |   26461.4 MB/s |    4.3e-06 |
a_sse4_1                   |   39.7973 ms |   26348.7 MB/s |    4.3e-06 |
u_avx2                     |   20.2697 ms |   51732.9 MB/s |    4.3e-06 |
a_avx2                     |   20.3882 ms |   51432.1 MB/s |    4.3e-06 |
u_avx2_fma                 |   20.3909 ms |   51425.4 MB/s |    4.3e-06 |
a_avx2_fma                 |   20.3383 ms |   51558.4 MB/s |    4.3e-06 |
u_avx512                   |   11.8731 ms |   88317.9 MB/s |    4.3e-06 |
a_avx512                   |   11.8860 ms |   88222.2 MB/s |    4.3e-06 |
u_avx512dq                 |   11.8220 ms |   88699.5 MB/s |    4.3e-06 | *
a_avx512dq                 |   11.8691 ms |   88347.6 MB/s |    4.3e-06 |
Best aligned arch          | u_avx512dq (49.35x)
Best unaligned arch        | u_avx512dq (49.35x)
--------------------------------------------------------------------------------

RUN_VOLK_TESTS: volk_32f_log2_32f(vlen=131071, iter=1000, tol=5e-06)
arch                       |         time |     throughput |    max_abs |
---------------------------+--------------+----------------+------------+
generic                    |  992.7362 ms |    1056.3 MB/s |          - |
u_sse4_1                   |  193.5862 ms |    5416.7 MB/s |    4.1e-06 |
a_sse4_1                   |  192.6163 ms |    5444.0 MB/s |    4.1e-06 |
u_avx2                     |  174.0292 ms |    6025.5 MB/s |    4.1e-06 | *
a_avx2                     |  173.9084 ms |    6029.7 MB/s |    4.1e-06 | *
u_avx2_fma                 |  175.2591 ms |    5983.2 MB/s |    4.1e-06 |
a_avx2_fma                 |  174.2142 ms |    6019.1 MB/s |    4.1e-06 |
Best aligned arch          | a_avx2 (5.71x)
Best unaligned arch        | u_avx2 (5.70x)
--------------------------------------------------------------------------------

RUN_VOLK_TESTS: volk_32f_log2_32f(vlen=131071, iter=1000, tol=5e-06)
arch                       |         time |     throughput |    max_abs |
---------------------------+--------------+----------------+------------+
generic                    | 1060.9555 ms |     988.4 MB/s |          - |
neon                       |  230.3564 ms |    4552.1 MB/s |    4.8e-06 |
neonv8                     |  194.7116 ms |    5385.4 MB/s |    3.9e-06 | *
Best aligned arch          | neonv8 (5.45x)
Best unaligned arch        | neonv8 (5.45x)
--------------------------------------------------------------------------------

[jdemel]

LGTM

I really appreciate that you add those intrinsics. Helps to understand the code better.