add support for llvm.experimental.vector.reduce intrinsics

[gnzlbg]

Bugzilla Link	36691
Version	trunk
OS	All
Depends On	#46849
CC	@hfinkel,@RKSimon,@zygoloid

Extended Description

Currently clang has no front-end support for the llvm.experimental.vector.reduce intrinsics. It would be great to be able to use these from clang directly.

[RKSimon]

[Issue #46850] adds x86-specific intrinsics (just for 512-bit vectors) but it shouldn't be much work to generalize these to work for any vector type.

[RKSimon]

mentioned in issue #46849

[RKSimon]

We currently have the following:

 __builtin_reduce_max(x)
 __builtin_reduce_min(x)
 __builtin_reduce_and(x)
 __builtin_reduce_or(x)
 __builtin_reduce_xor(x)

[fhahn]

The available vector reduction builtins support by Clang are documented here: https://clang.llvm.org/docs/LanguageExtensions.html#vector-builtins

[RKSimon]

__builtin_reduce_add(x) now works for integer types - a23291b

[RKSimon]

__builtin_reduce_mul(x) now works for integer types - 8a92c45