Add population count primop

[tibbe]

Modern CPUs have a POPCNT instruction for efficient population count. This instruction can be used to implement various data structures.

I propose we add the following set of primops

popCnt8# :: Word# -> Word#
popCnt16# :: Word# -> Word#
popCnt32# :: Word# -> Word#
popCnt64# :: Word64# -> Word#
popCnt# :: Word# -> Word#

(We use Word# for all functions except the 64 bit version as there are no Word8, Word16# and Word32# types).

Each primop compiles into either a single POPCNT instruction or a call to some fallback function, implemented in C.

[tibbe]

I've implemented the primops. Optionally we might want to create a small static library in ghc-prim containing the C fallbacks, to avoid the overhead of dynamic linking for these "fat machine instructions".

[simonmar]

Shouldn't -msse4.2 imply -msse2?

[tibbe]

Replying to simonmar:

Shouldn't -msse4.2 imply -msse2?

Yes and I think it does. Check the helpers in the native code gen.

[tibbe]

Fixed in 2d0438f329ac153f9e59155f405d27fac0c43d65