sycl::vec::convert has no scalar equivalent

[tdavidcl]

Specification Version

SYCL 2020 (Revision 11)

Section Number(s)

4.14.2.1. Vec interface

Issue Description

Currently the sycl::vec interface provide the .convert function with rounding modes.
As is the spec does not provide a scalar equivalent for the conversion using the same rounding modes.
As such one can not reliably ensure that the scalar conversion results match the one of sycl::vec.

Code Example (Optional)

No response

[TApplencourt]

• Does Cuda have some scalar conversion? ( math function may do it, but no free-function)
• How does OpenCL work? (global state + anotation as an extension?)
  • https://registry.khronos.org/OpenCL/specs/3.0-unified/html/OpenCL_C.html#explicit-conversions
• Are you using convert for vec @tdavidcl ?
• Example of possible implementation is necessary to move forward (in cuda, opencl seem trivial)

[tdavidcl]

• Does Cuda have some scalar conversion? ( math function may do it, but no free-function)

  • How does OpenCL work? (global state + anotation as an extension?)

    • https://registry.khronos.org/OpenCL/specs/3.0-unified/html/OpenCL_C.html#explicit-conversions

I don't know for OpenCL, but for CUDA there is no such function so the implementation would be probably static cast for rounding down, ... (standard c++ way)

* Are you using `convert` for vec  [@tdavidcl](https://github.com/tdavidcl) ?

Yes i do use it quite a lot, even though some recently discovered inconsistencies between Acpp and DPc++ are making me slowly fallback onto a custom version of it.