[Mlir-commits] [mlir] [mlir][sparse] replace specialized buffer setup with util code (PR #68461)

[Aart Bik]

================
@@ -136,18 +136,16 @@ func.func @sparse_new3d(%arg0: !llvm.ptr<i8>) -> tensor<?x?x?xf32, #SparseTensor
 //   CHECK-DAG: %[[Empty:.*]] = arith.constant 0 : i32
 //   CHECK-DAG: %[[C0:.*]] = arith.constant 0 : index
 //   CHECK-DAG: %[[C1:.*]] = arith.constant 1 : index
-//   CHECK-DAG: %[[DimSizes0:.*]] = memref.alloca() : memref<2xindex>
-//   CHECK-DAG: %[[LvlSizes0:.*]] = memref.alloca() : memref<2xindex>
 //   CHECK-DAG: %[[LvlTypes0:.*]] = memref.alloca() : memref<2xi8>
+//   CHECK-DAG: %[[Sizes0:.*]] = memref.alloca() : memref<2xindex>
 //   CHECK-DAG: %[[Iota0:.*]] = memref.alloca() : memref<2xindex>
-//   CHECK-DAG: %[[DimSizes:.*]] = memref.cast %[[DimSizes0]] : memref<2xindex> to memref<?xindex>
-//   CHECK-DAG: %[[LvlSizes:.*]] = memref.cast %[[LvlSizes0]] : memref<2xindex> to memref<?xindex>
 //   CHECK-DAG: %[[LvlTypes:.*]] = memref.cast %[[LvlTypes0]] : memref<2xi8> to memref<?xi8>
+//   CHECK-DAG: %[[Sizes:.*]] = memref.cast %[[Sizes0]] : memref<2xindex> to memref<?xindex>
 //   CHECK-DAG: %[[Iota:.*]] = memref.cast %[[Iota0]] : memref<2xindex> to memref<?xindex>
-//   CHECK-DAG: memref.store %[[I]], %[[DimSizes0]][%[[C0]]] : memref<2xindex>
-//   CHECK-DAG: memref.store %[[J]], %[[DimSizes0]][%[[C1]]] : memref<2xindex>
+//   CHECK-DAG: memref.store %[[I]], %[[Sizes0]][%[[C0]]] : memref<2xindex>
+//   CHECK-DAG: memref.store %[[J]], %[[Sizes0]][%[[C1]]] : memref<2xindex>
----------------
aartbik wrote:

also note how the sequence in general has become shorter, since the shared util knows when to share dim/lvl sizes for id permutation

https://github.com/llvm/llvm-project/pull/68461