[llvm-branch-commits] [llvm] [AMDGPU] Introduce subtarget features `FeatureLDSAllocGranularity<size>` and expose them to the frontend (PR #220394)

[Chinmay Deshpande via llvm-branch-commits]

================
@@ -476,14 +476,44 @@ collectVisibleFeatures(const Record *GPU,
   return Visible;
 }
 
+// Return the LDS allocation granularity in \p GPU's feature closure. Two
+// features setting different granularities is an error: SubtargetFeature
+// silently takes the larger.
+static int64_t getLDSAllocGranularity(const Record *GPU) {
----------------
chinmaydd wrote:

I think implementing a general solution for this makes sense. Moving over to stack at https://github.com/llvm/llvm-project/pull/223177

https://github.com/llvm/llvm-project/pull/220394