[SYCL][NFCI] Move abs and div to cstdlib header

[bader]

The C++ specification defines that abs(<integer>) and div functions are provided by cstdlib.
Whether they're available or not after #include <cmath> is an implementation detail of the underlying C++ standard library implementation.

With this change we move the definition to the correct wrapper file
aligned with the C++ specification.

[bader]

run_prebuilt_e2e_tests (E2E tests on Intel Ponte Vecchio GPU, ["Linux", "pvc"], -u 1001 --device=... / E2E tests on Intel Ponte Vecchio GPU failure doesn't seem to be related to the change. The test doesn't use abs or div functions and there is known issue with this test on PVC. See #19662.

[jinge90]

Hi, @bader
Can we keep the functions in libdevice? They are used by OMP offloading as well.
If can't, we can only build them for OMP offloading side.
Thanks very much.

[bader]

Can we keep the functions in libdevice? They are used by OMP offloading as well. If can't, we can only build them for OMP offloading side. Thanks very much.

There is no sense in making any changes for OpenMP offloading in this repository because OpenMP offloading is not supported/tested here.

In the upstream OpenMP offloading uses clang's headers to implement abs on AMD GPU. I suggest using the same approach for SPIR-V targets in other repositories as well.
The same applies to div if it's needed for OpenMP offloading.

@sarnex, are you working on something similar in the upstream?

[Maetveis]

This will conflict with #19620. Can you apply the changes here I did to these functions, I don't want it to be lost during conflict resolution.

[sarnex]

@sarnex, are you working on something similar in the upstream?

I haven't begun to think about how we will deal with libdevice in OMP upstream, however I agree we should match whatever the community does as best as possible.

[Maetveis]

LGTM, thanks!

[bader]

@intel/llvm-reviewers-runtime, ping.

[aelovikov-intel]

These functions are declared in cmath and use data type defined in
cstdlib.

This is ambiguous as cmath/cstdlib might refer to either system implementation or our "wrappers". Also, if they do refer to our wrappers, then it should have been something like "were declared in our cmath wrapper before this PR".

It seems you're trying to say something like

These functions are only guaranteed to be provided by `cstdlib` 
by the specification and whether they're available or not after 
`#include <cmath>` is an implementation detail of the underlying 
libc implementation.

The fact that we've been providing them as part of our `cmath` wrapper
was a hack (see pre-existing comments in <insert_file>) prior to this PR.
With this change we move the definition to the correct wrapper file
aligned with the specification/host libc implementation/<insert proper term>.

[bader]

@aelovikov-intel, thanks. I've updated the pull request description. Please, take a look.