[Mlir-commits] [mlir] [mlir][sparse] support type conversion from batched sparse tensors to… (PR #83163)

[Aart Bik]

================
@@ -182,6 +186,16 @@ func.func @sparse_csr(%arg0: tensor<?x?xf64, #CSR>) {
   return
 }
 
+// CHECK-LABEL: func @sparse_bcsr(
+//  CHECK-SAME: %[[A1:.*0]]: memref<?x2x?xindex>,
+//  CHECK-SAME: %[[A2:.*1]]: memref<?x2x?xindex>,
+//  CHECK-SAME: %[[A3:.*]]: memref<?x2x?xf64>,
+//  CHECK-SAME: %[[A4:.*]]: !sparse_tensor.storage_specifier
+//       CHECK: return
+func.func @sparse_bcsr(%arg0: tensor<?x2x?x?xf64, #BCSR>) {
----------------
aartbik wrote:

can we also add a ?x?x?x? and a 16x8x4x2 case just to make sure we are not confused at the batch/dense boundary?

https://github.com/llvm/llvm-project/pull/83163