[llvm] [X86][CostModel] Add per-shape gather/scatter cost tables for AMD znver4+ (PR #199488)

[Phoebe Wang via llvm-commits]

phoebewang wrote:

That's a good example why I prefer to a per-target tuning leverage schedule model. The current granularity of cost tables is too coarse. I think it's a good start to improve instructions like gather/scatter that have large variations from target to target.

> This is a function of Vector Factor, Data Type, and the instruction. So, deducing a number from the latency of scheduler is not applicable directly.

It's a fixed map under given feature (as well as target). That's also how we calculate cost for each type, see comments accompanied with the table.

https://github.com/llvm/llvm-project/pull/199488