[PATCH] D114951: [Analysis][AArch64] Add on the address computational cost for gathers/scatters

[David Sherwood via Phabricator via llvm-commits]

david-arm added a comment.

In X86TTIImpl::getAddressComputationCost there is code like this to analyse strided accesses:



================
Comment at: llvm/lib/Target/AArch64/AArch64TargetTransformInfo.cpp:1791
+  MemOpCost += 1;
   return LT.first * MemOpCost * getMaxNumElements(LegalVF);
 }
----------------
peterwaller-arm wrote:
> This appears to be increasing the cost by a factor of the legalization cost, in addition to the 1 x element count as stated in the PR body, is that intended? I might instead have expected this to read `return (LT.first * MemOpCost + 1) * getMaxNumElements(LegalVF)`.
Yeah that's intended, because we'll still have the address computational cost for each legal part too. For each legal gather we'll have to construct a set of indices. Maybe I can clarify the commit message - I meant the total element count for the whole vector, even it's not legal.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D114951/new/

https://reviews.llvm.org/D114951