[SYCL] Introduce external arbitrary fixed point functions

[vmaksimo]

This change enables support of functions that operate on arbitrary precision
fixed point numbers. This datatype and its corresponding operations can be
useful on targets that can take advantage of narrower representation such as
FPGAs.

Arbitrary precision fixed point numbers are represented using arbitrary
precision integers (_ExtInt(N) wrapped in ap_int). Additional parameters are:

• W - total width of the datatype. It's encoded in the width of the OpTypeInt.
• I - determines the position of the decimal point.
• S - determines if this is a signed or an unsigned number.

Usage example:
ap_int a;
auto ap_fixed_Sqrt = __spirv_FixedSqrtINTEL<W,rW>(a, S, I, rI);
// auto to be deducted to ap_int

[keryell]

That sounds like a good generic extension useful for others. :-)
Perhaps something to add as a KHR extension in SPIR-V?

[AlexeySachkov]

That sounds like a good generic extension useful for others. :-)
Perhaps something to add as a KHR extension in SPIR-V?

@keryell, here you go: #1934. This is not a KHR one, but at least some starting point

[keryell]

Do you have a host (& by side effect a host device) implementation for this too?
Actually all the data-types should be available under sycl::... to be able to do something useful globally and keep the single-source approach.
There might be some alignment constraints to be compatible with C++ on the host and also have a kind of packed representation.

[MrSidims]

@keryell there will a wrapper headers around these SPIR-V builtins and this header will include host implementation.

[rbegam]

LGTM.