B14 Generating Too Many Tests

[fourth-bit]

It looks like B14 is generating 15 tests for each possible shift value (i.e. 64000 * 15 tests for quads). Aharoni seems to suggest that we only need 1 value outside of the range (-2p + 1, p + 1) for the high and low ends.

[coreyqh]

@sww0000 Please take a look at this when you get a chance.

[sww0000]

On it.

B14 was originally written in somewhat of a brute force/flawed way, as exponent values and mantissa values are both randomly generated. Because of this reason, in some cases, FMA operations may result in unwanted shifts causing some bins to not be hit, so I increased the iteration amount to cover for this which resulted in B14 generating too many tests. I believe I overcomplicated the implementation method back then when I first wrote this test.

For a while I've wanted to rewrite this model but didn't get time to. I think I'll rewrite the python implementation sometimes during this weekend so every shift value of exponent in the range (-2p+1, p+1) has exactly one set of test vectors, and the # of tests should be as according to the B14's documentation test count analysis.

[sww0000]

Fixed + updated B14.py and added B14.adoc