[SYCL] Add static cast for half builtins relational

[AidanBeltonS]

This PR adds a static cast to builtins_relational functions which apply a negative operation to a boolean type.
This is only applied to functions which are specialised for cl_half. The cast prevents windows error message C4804.

This resolves a windows build issue introduced by #6061

[AidanBeltonS]

@yanfeng3721, ping

[dkhaldi]

should these return a bool or T?
Where in the spec, this is stated?

[AidanBeltonS]

__vFOrdEqual is not stated in the spec. I am not sure as to if this should return a bool or T.
The goal of this change to resolve an issue with building DPCPP on windows and not to change any functionality.

[dkhaldi]

All these are part of 4.17.9. Relational functions. Some of them should return bool. However, you are right that the PR does not change any functionality. If there is a a problem, it was already there in the original code.

[aelovikov-intel]

I think functions modified here are all ok, because they're vector and vector versions don't return bool at all (even all/any).

As for the scalar version (not touched in this PR), this is slightly more complicated as SYCL1.2.1 and SYCL2020 differ here. Those, however, are all internal details and not the user-visible routines so technically we can have them defined whatever way we want to be able to implement user-visible wrappers in a way dependent on the presence of SYCL2020_CONFORMANT_APIS macro.

[v-klochkov]

BTW, my guess why it returns -1 is because it is often good for vectorizer to have element filled with ones or with zeroes for using in vector mask operations like VANDPS/VORPS,
but I also was puzzled seeing return of -1.0 for float comparison.
It would be logically more correct to returning int32_t{-1} for float32 rather than -1.0

[aelovikov-intel]

User-visible methods must do exactly that according to the spec. I guess/think/hope that only those internal functions behave differently for some reason.

[dkhaldi]

While there seem to be ambiguity on the return type for these SYCL relational functions, this specific change to solve the build bugs on Windows LGTM.
I would suggest opening a new PR to check and correct the return type if something is wrong.

[v-klochkov]

This PR is supposed to fix the build errors on Windows, but CI still fails: http://llvm-ci-test2.intel.com:8080/blue/organizations/jenkins/SYCL_CI%2Fintel%2FWin%2FTest_Suite/detail/Test_Suite/1869/pipeline

`[2022-05-10T16:33:14.773Z] W:\jenkins-dir\workspace\SYCL_CI\intel\Win\Test_Suite\llvm.src\sycl\source\detail\builtins_relational.cpp(230): error C2220: the following warning is treated as an error

[2022-05-10T16:33:14.773Z] W:\jenkins-dir\workspace\SYCL_CI\intel\Win\Test_Suite\llvm.src\sycl\source\detail\builtins_relational.cpp(230): warning C4804: '-': unsafe use of type 'bool' in operation

[2022-05-10T16:33:14.773Z] W:\jenkins-dir\workspace\SYCL_CI\intel\Win\Test_Suite\llvm.src\sycl\source\detail\builtins_relational.cpp(24): warning C4804: '-': unsafe use of type 'bool' in operation

[2022-05-10T16:33:14.773Z] W:\jenkins-dir\workspace\SYCL_CI\intel\Win\Test_Suite\llvm.src\sycl\source\detail\builtins_relational.cpp(155): note: see reference to function template instantiation 'T cl::__host_std::`anonymous-namespace'::__vFOrdEqualcl::sycl::cl_half(T,T)' being compiled

[2022-05-10T16:33:14.773Z] with

[2022-05-10T16:33:14.773Z] [

[2022-05-10T16:33:14.773Z] T=cl::sycl::cl_half

[2022-05-10T16:33:14.773Z] ]

[2022-05-10T16:33:14.773Z] W:\jenkins-dir\workspace\SYCL_CI\intel\Win\Test_Suite\llvm.src\sycl\source\detail\builtins_relational.cpp(28): warning C4804: '-': unsafe use of type 'bool' in operation`

[pvchupin]

I've checked locally that it helps before merging. I think it fails for reference compiler, that's the only possible explanation I have.

[bader]

I think it fails for reference compiler, that's the only possible explanation I have.

Whether it fails for reference compiler or not must be clear from the logs.

[pvchupin]

@bader, I can't understand that from logs. Tagging @tfzhu.
In any case post-commit passed, so I believe original issue is resolved.