Unit test clean-up required.

[edanor]

Originally reported by: edanor (Bitbucket: edanor, GitHub: edanor)

---

There are multiple unit test fails even when running with full scalar emulation. This can be as well caused by incorrect test data sets.

---

• Bitbucket: https://bitbucket.org/edanor/umesimd/issue/22

[edanor]

Original comment by edanor (Bitbucket: edanor, GitHub: edanor):

---

Unit tests require all test to be done using randomly generated data (if possible). This would increase the overall coverage of tests.

[edanor]

Original comment by edanor (Bitbucket: edanor, GitHub: edanor):

---

Some operations, such as SINCOS will by design have some error against reference value (e.g. std::sin/std::cos). This error should be expressed in ULP instead of relative error, and it is not possible to tell what is the 'correct' result.

Such operations should, instead of testing relative error, calculate ULP error and give errors only when some user-defined error threshold is passed.