[Clang] Add __builtin_elementwise_fshl/fshr builtin intrinsics

[RKSimon]

Providing __builtin_elementwise_fshl / __builtin_elementwise_fshr helper builtins would allow us to represent both integer rotates and funnel/double shift patterns on scalar and vector types.

LLVM represents these as:

 fshl(X,Y,Z): (X << (Z % BW)) | (Y >> (BW - (Z % BW)))
 fshr(X,Y,Z): (X << (BW - (Z % BW))) | (Y >> (Z % BW))

X and Y have to be same type (and is the return type as well). technically Z (the shift amount) type doesn't have to match, but should be able to represent the full (modulo) shift amount range.

[llvmbot]

@llvm/issue-subscribers-clang-frontend

Author: Simon Pilgrim (RKSimon)

Providing `__builtin_elementwise_fshl` / `__builtin_elementwise_fshr` helper builtins would allow us to represent both integer rotates and funnel/double shift patterns on scalar and vector types.

LLVM represents these as:

 fshl(X,Y,Z): (X << (Z % BW)) | (Y >> (BW - (Z % BW)))
 fshr(X,Y,Z): (X << (BW - (Z % BW))) | (Y >> (Z % BW))

X and Y have to be same type (and is the return type as well). technically Z (the shift amount) type doesn't have to match, but should be able to represent the full (modulo) shift amount range.

[ckoparkar]

@RKSimon I'd like to take a stab at this. I realize this could be a bit more complicated than the patches I've worked on so far, but I can use clang: Add __builtin_elementwise_fma for reference?

[RKSimon]

A mixture of the fma and the addsubsat patches c00db97 might be helpful - although I think things have been refactored a bit since both of those