[llvm] [X86] combineConcatVectorOps - concat per-lane v2f64/v4f64 shuffles into vXf64 vshufpd (PR #143017)

[via llvm-commits]

bgra8 wrote:

@RKSimon we're seeing many significant increases in memory usage during compilation of certain translation units inside google. In several cases the memory consumption goes above 12GB (might go above this, we just stopped the compilation  at this limit). The compilation times also increase significantly for such cases.

https://github.com/llvm/llvm-project/commit/a6ace2801e8900a6fe8c3b8295938f3b3c1e4466 does not fix the issue.

We're working on a reproducer.

https://github.com/llvm/llvm-project/pull/143017