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

[via llvm-commits]

Andarwinux wrote:

> I'd much prefer we just use these new costs as the default for AVX512 targets, AFAICT a lot of Intel numbers aren't that dissimilar for many of these, once all the ucode changes have been applied for speculative exploits etc.

but GDS didn't affect scatter, and Intel scatter is significantly faster than AMD.

https://uops.info/table.html?search=scatter&cb_lat=on&cb_tp=on&cb_uops=on&cb_ports=on&cb_TGL=on&cb_ZEN5=on&cb_measurements=on&cb_avx512=on

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