[mlir][sparse] fixed pack op documentation and purity

[aartbik]

Summary

Example did not have correct shapes. Also the operation
has no side effects (since it slaps a clean SSA tensor
around the data).

[wrengr]

mlir/include/mlir/Dialect/SparseTensor/IR/SparseTensorOps.td

90–91

I'd leave the values in the format 1.0, 2.0, 3.0 to help keep it clear how the result relates to the inputs. Even better would be to change the values to something else entirely (e.g., [4,5,6]) so that there's no possible confusion (similar to how the example uses 3x2 rather than 2x2 which would be ambiguous).

[Peiming]

mlir/include/mlir/Dialect/SparseTensor/IR/SparseTensorOps.td

85–92

Oh, I found the typo and but changed it to tensor<2xf64> in unpack revision ;-) I will accept your change since Wren made a good point that 2x2 can be ambiguous.

[aartbik]

let fp values jump out more

mlir/include/mlir/Dialect/SparseTensor/IR/SparseTensorOps.td

string description = [{
     sparse_tensor.new %source : !Source to tensor<1024x1024xf64, #CSR>
     ```
   }];
   let assemblyFormat = "(`expand_symmetry` $expandSymmetry^)? $source attr-dict"
                        "`:` type($source) `to` type($result)";
   let hasVerifier = 1;
 }
 
-def SparseTensor_PackOp : SparseTensor_Op<"pack">,
+def SparseTensor_PackOp : SparseTensor_Op<"pack", [Pure]>,
     Arguments<(ins AnyRankedTensor:$data,
                    AnyRankedTensor:$indices)>,
     Results<(outs AnySparseTensor: $result)> {
   let summary = "Returns a sparse tensor from the given (data, indices) pair";
 
   let description = [{
     Packs the data/indices into a COO sparse tensor. The coordinates in `indices`
-    shall not exceed the dimension sizes of the returned sparse tensor.
-    Note that the returned tensor must be statically
-    shaped because it is impossible to infer the shape from sparse coordinates.
+    shall not exceed the dimension sizes of the returned sparse tensor. Note
+    that the returned tensor must be statically shaped because it is impossible
+    to infer the shape from sparse coordinates.
 
     `$indices`: stored via a 2-D tensor of integer elements with shape [N, ndims],
     which specifies the indices of the elements in the sparse tensor that contains
     non-zero values.
 
     `$data`: stored via a 1-D tensor with shape [N], that supplies the corresponding
     values for the indices.
 
-    The operation can be used to materialize a sparse tensor from external sources. E.g.,
-    when passing from Python as two numpy arrays for data and indices.
+    The operation can be used to materialize a sparse tensor from external sources.
+    E.g., when passing from Python as two numpy arrays for data and indices.
 
     Example:
     ```mlir
-    %data    = arith.constant dense<[ 1     , 5     ]> : tensor<3xf64>
-    %indices = arith.constant dense<[[0,  0],[1,  2]]> : tensor<3x2xindex>
+    %data    = arith.constant dense<[ 1.1,   2.2,   3.3 ]> : tensor<3xf64>
+    %indices = arith.constant dense<[[0,0], [1,2], [1,3]]> : tensor<3x2xindex>
 
-    %st = sparse_tensor.pack %data, %indices : tensor<6xf64>, tensor<6x2xi32
-                                            to tensor<100x100xf64, #COO>
-    // %st = [[1, 0, 0, 0],
-    //        [0, 0, 5, 0],
-    //        [0, 0, 0, 0]]
+    %st = sparse_tensor.pack %data, %indices : tensor<3xf64>, tensor<3x2xindex
+                                            to tensor<3x4xf64, #COO>
+    // yields COO format |1.1, 0.0, 0.0, 0.0|
+    //     of 3x4 matrix |0.0, 0.0, 2.2, 3.3|
+    //                   |0.0, 0.0, 0.0, 0.0|
     ```
   }];
 
   let assemblyFormat = "$data `,` $indices attr-dict `:` type($data) `,` type($indices)"
                                                    "`to` type($result)";
 
   let hasVerifier = 1;
 }