[Mlir-commits] [mlir] [mlir][sparse] support sparsifying batch levels (PR #83898)

[Peiming Liu]

================
@@ -374,7 +374,7 @@ Level SparseTensorEncodingAttr::getLvlRank() const {
 
 LevelType SparseTensorEncodingAttr::getLvlType(Level l) const {
   if (!getImpl())
-    return LevelFormat::Dense;
+    return LevelFormat::Batch;
----------------
PeimingLiu wrote:

I think `Batch` make more sense than `Dense` for non-annotated tensor here, because it would require keeping all non-linearized coordinates.

https://github.com/llvm/llvm-project/pull/83898