[Mlir-commits] [mlir] [mlir][sparse] implement direct IR alloc/empty/new for non-permutations (PR #72585)

[Aart Bik]

================
@@ -766,17 +768,22 @@ class SparseTensorAllocConverter
       return success();
     }
 
+    // Construct the dim/lvl sizes and the (unused) dim2lvl/lvl2dim buffers.
----------------
aartbik wrote:

As discussed offline, we probably can even change the alloc/empty cases with assert isIdentity, and use dimsizes directly. I was playing it safe during the migration, since the code is now always going through the same steps. But indeed, cleanup TBD when we are comfortable.

https://github.com/llvm/llvm-project/pull/72585