[llvm] [X86] X86FixupVectorConstants - shrink vector load to movsd/movsd/movd/movq 'zero upper' instructions (PR #79000)

[Simon Pilgrim via llvm-commits]

RKSimon wrote:

@phoebewang I've added missing general VectorConstant decodes (where before we only needed support for splats/broadcasts) and now prefer vzload over broadcasts (even if we would have been able to use a smaller broadcast constant), which seems to give consistent better codegen.

https://github.com/llvm/llvm-project/pull/79000