[PATCH] D98362: [AMDGPU] Fix -amdgpu-inline-arg-alloca-cost

[Stanislav Mekhanoshin via Phabricator via llvm-commits]

rampitec added a comment.

In D98362#2619375 <https://reviews.llvm.org/D98362#2619375>, @dfukalov wrote:

> Do you have any test for the fix?

These tests tend to be either unreliable or huge. We cannot measure performance with lit tests.



================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUTargetTransformInfo.cpp:1192
   if (AllocaSize)
-    return ArgAllocaCost;
+    return ArgAllocaCost * getInliningThresholdMultiplier();
   return 0;
----------------
dfukalov wrote:
> rampitec wrote:
> > arsenm wrote:
> > > Should the just adjust for the scale then?
> > I thought about this, but whenever we will adjust the scale the next time we will have to visit it again.
> Nit: It seems instead of this modification you can just swap two lines
> ```
> 1582:  Threshold *= TTI.getInliningThresholdMultiplier();
> 1583:  Threshold += TTI.adjustInliningThreshold(&Call);
> 
> ```
> [[ https://reviews.llvm.org/source/llvm-github/browse/main/llvm/lib/Analysis/InlineCost.cpp$1582-1583 | in InlineCost.cpp ]] so we'll stay with just one place of `* getInliningThresholdMultiplier()`.
That would change behavior for all targets.


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D98362/new/

https://reviews.llvm.org/D98362