llvm-project

b9984f56 - [X86] combineConcatVectorOps - add handling for FMA/FMSUB/FNMSUB/FNMADD vector ops (#172989)

[X86] combineConcatVectorOps - add handling for FMA/FMSUB/FNMSUB/FNMADD vector ops (#172989) This involves some analysis of the operands to ensure we don't concat more subvectors than before the fold