[PATCH] D98512: [LoopVectorize] Simplify scalar cost calculation in getInstructionCost

[David Sherwood via Phabricator via llvm-commits]

david-arm added a comment.

Hi @fhahn, I tested the example you pasted above before and after applying my patch and got this debug output:

  Before:
  LV: Found an estimated cost of 0 for VF 4 For instruction:   %i = phi i32 [ 0, %entry ], [ %i.next, %for.body ]
  LV: Found an estimated cost of 0 for VF 4 For instruction:   %tmp0 = getelementptr half, half* %b, i32 %i
  LV: Found an estimated cost of 3 for VF 4 For instruction:   %lv = load half, half* %tmp0, align 2
  LV: Found an estimated cost of 2 for VF 4 For instruction:   %add = fadd half %lv, 0xH5100
  LV: Found an estimated cost of 23 for VF 4 For instruction:   store half %add, half* %tmp0, align 1
  LV: Found an estimated cost of 4 for VF 4 For instruction:   %i.next = add nuw nsw i32 %i, 3
  LV: Found an estimated cost of 1 for VF 4 For instruction:   %cond = icmp eq i32 %i.next, %n
  LV: Found an estimated cost of 0 for VF 4 For instruction:   br i1 %cond, label %for.end, label %for.body
  
  After:
  LV: Found an estimated cost of 0 for VF 4 For instruction:   %i = phi i32 [ 0, %entry ], [ %i.next, %for.body ]
  LV: Found an estimated cost of 0 for VF 4 For instruction:   %tmp0 = getelementptr half, half* %b, i32 %i
  LV: Found an estimated cost of 3 for VF 4 For instruction:   %lv = load half, half* %tmp0, align 2
  LV: Found an estimated cost of 2 for VF 4 For instruction:   %add = fadd half %lv, 0xH5100
  LV: Found an estimated cost of 23 for VF 4 For instruction:   store half %add, half* %tmp0, align 1
  LV: Found an estimated cost of 1 for VF 4 For instruction:   %i.next = add nuw nsw i32 %i, 3
  LV: Found an estimated cost of 1 for VF 4 For instruction:   %cond = icmp eq i32 %i.next, %n
  LV: Found an estimated cost of 0 for VF 4 For instruction:   br i1 %cond, label %for.end, label %for.body

It looks like the only change is the IV update which goes from 4 -> 1. I think the adds you mentioned are included as part of the load/store cost.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D98512/new/

https://reviews.llvm.org/D98512