[all-commits] [llvm/llvm-project] c963bd: [CostModel] remove cost-kind predicate for scatter...

[RotateRight via All-commits]

  Branch: refs/heads/master
  Home:   https://github.com/llvm/llvm-project
  Commit: c963bde0152a741d58b9e1e9ac485d8f2e6c6d58
      https://github.com/llvm/llvm-project/commit/c963bde0152a741d58b9e1e9ac485d8f2e6c6d58
  Author: Sanjay Patel <spatel at rotateright.com>
  Date:   2020-10-21 (Wed, 21 Oct 2020)

  Changed paths:
    M llvm/include/llvm/CodeGen/BasicTTIImpl.h
    M llvm/test/Analysis/CostModel/ARM/intrinsic-cost-kinds.ll

  Log Message:
  -----------
  [CostModel] remove cost-kind predicate for scatter/gather cost

This is similar in spirit to 01ea93d85d6e (memcpy) except that
here the underlying caller assumptions were created for vectorizer
use (throughput) rather than other passes.

That meant ARM could have an enormous throughput cost with no
corresponding size, latency, or blended cost increase. X86 has
the same throughput restriction as the basic implementation, so
it is still unchanged.

Paraphrasing from the previous commit:
This may not make sense for some callers, but at least now the
costs will be consistently wrong instead of mysteriously wrong.

Targets should provide better overrides if the current modeling
is not accurate.