llvm-project

96d2cb41 - [LLVM][CodeGen][DirectX] Fix scalarisation when vector ConstantFP is used. (#172684)

[LLVM][CodeGen][DirectX] Fix scalarisation when vector ConstantFP is used. (#172684) When using -use-constant-fp-for-fixed-length-splat `splat (float C)` becomes ConstantFP(C) rather than ConstantVector(C, C, C...).