[SYCL][ESIMD] Add test on simd bitwise not operator

[vasilytric]

No description provided.

[yuriykoch]

Actually, I'm not quite sure it's safe to use bitwise operators over signed integral values in general.

The reasons behind are:

• general rule: INT13-C
• trap values and padding bits possibility: n2218 mentions all details
• possible existence of compiler optimizations relying on such restrictions (like the ones for arithmetical operations mentioned in cppcon 2018)

Tagging @kbobrovs for awareness

[yuriykoch]

Looks like while there is possibility of trap values and compiler optimizations in general C++17, the llvm itself states that signed integer types are guaranteed to be two's complement:

• https://llvm.org/docs/LangRef.html
• https://bugs.llvm.org/show_bug.cgi?id=950

Given that information I would suggest to do the following:

1. provide coverage for signed integral types, making it as generic as possible from the point of C++17 spec (using For_bitwise_not),
2. but split operator_bitwise_not.cpp into the operator_bitwise_not_uint.cpp and operator_bitwise_not_sint.cpp to have specific test on signed integral types;
3. mention all links from the discussion in operator_bitwise_not_sint.cpp description to make the future support easier

@vasilytric Feel free to state any concerns for such solution. The other possible option is to make a two's complement representation a test precondition (run assert in operator_bitwise_not_sint.cpp, so there would be no need in For_bitwise_not in such case)

[vasilytric]

Looks like while there is possibility of trap values and compiler optimizations in general C++17, the llvm itself states that signed integer types are guaranteed to be two's complement:

The test has been split in two in this commit: 66e0d60

[vasilytric]

@v-klochkov Would you mind to review this PR?