[llvm] [LV] Change loops' interleave count computation (PR #73766)

David Sherwood via llvm-commits llvm-commits at lists.llvm.org
Wed Nov 29 06:49:29 PST 2023


================
@@ -5737,10 +5741,15 @@ LoopVectorizationCostModel::selectInterleaveCount(ElementCount VF,
   // the InterleaveCount as if vscale is '1', although if some information about
   // the vector is known (e.g. min vector size), we can make a better decision.
   if (BestKnownTC) {
-    MaxInterleaveCount =
-        std::min(*BestKnownTC / VF.getKnownMinValue(), MaxInterleaveCount);
-    // Make sure MaxInterleaveCount is greater than 0.
-    MaxInterleaveCount = std::max(1u, MaxInterleaveCount);
+    if (InterleaveSmallLoopScalarReduction ||
+        (*BestKnownTC % VF.getKnownMinValue() == 0))
----------------
david-arm wrote:

I think it makes sense to make a better guess of the runtime VF here by doing

```
  unsigned EstimatedVF = VF.getKnownMinValue();
  if (std::optional<unsigned> VScale = getVScaleForTuning())
    EstimatedVF *= *VScale;
```

so that you're algorithm is doing what you expect for scalable vectors. It's possible that vscale may be quite large for the CPU we're targetting.

https://github.com/llvm/llvm-project/pull/73766


More information about the llvm-commits mailing list