Avoid creating null mask in groupby aggregation M2 (#20726)

This stops creating a null mask during the computation of groupby aggregation `M2`, similar to #20716. Such null mask is unnecessary as `M2` values are used together with valid count  which already contains enough information for further computation (for `std`/`variance`).

Authors:
  - Nghia Truong (https://github.com/ttnghia)

Approvers:
  - David Wendt (https://github.com/davidwendt)
  - Bradley Dice (https://github.com/bdice)
  - Gera Shegalov (https://github.com/gerashegalov)
  - Alfred Xu (https://github.com/sperlingxx)

URL: #20726

Author: ttnghia

cpp/src/groupby/common/m2_var_std.cu

@@ -91,8 +91,7 @@ std::unique_ptr<column> compute_m2(data_type source_type,
 {
   auto output = make_numeric_column(cudf::detail::target_type(source_type, aggregation::M2),
                                     sum.size(),
-                                    cudf::detail::copy_bitmask(sum, stream, mr),
-                                    sum.null_count(),
+                                    mask_state::UNALLOCATED,
                                     stream,
                                     mr);
   type_dispatcher(source_type, m2_functor{}, output->mutable_view(), sum_sqr, sum, count, stream);

cpp/src/groupby/sort/group_m2.cu

@@ -101,12 +101,6 @@ struct m2_functor {
       compute_m2_fn(d_values, values_iter, group_labels, d_means, d_result, stream);
     }
 
-    // M2 column values should have the same bitmask as means's.
-    if (group_means.nullable()) {
-      result->set_null_mask(cudf::detail::copy_bitmask(group_means, stream, mr),
-                            group_means.null_count());
-    }
-
     return result;
   }
 

cpp/tests/groupby/m2_tests.cpp

@@ -114,7 +114,7 @@ TYPED_TEST(GroupbyM2TypedTest, AllNullValuesInput)
   auto const vals = vals_col<T>{{3, 4, 5}, all_nulls()};
 
   auto const [out_keys, out_M2s] = compute_M2(keys, vals);
-  auto const expected_M2s        = M2s_col<R>{{null, null, null}, all_nulls()};
+  auto const expected_M2s        = M2s_col<R>{0, 0, 0};
 
   CUDF_TEST_EXPECT_COLUMNS_EQUIVALENT(keys, *out_keys, verbosity);
   CUDF_TEST_EXPECT_COLUMNS_EQUIVALENT(expected_M2s, *out_M2s, verbosity);
@@ -168,7 +168,7 @@ TYPED_TEST(GroupbyM2TypedTest, ValuesHaveNulls)
 
   auto const [out_keys, out_M2s] = compute_M2(keys, vals);
   auto const expected_keys       = keys_col<T>{1, 2, 3, 4, 5};
-  auto const expected_M2s        = M2s_col<R>{{0.0, 2.0, 8.0, 0.0, 0.0 /*NULL*/}, null_at(4)};
+  auto const expected_M2s        = M2s_col<R>{0.0, 2.0, 8.0, 0.0, 0.0};
 
   CUDF_TEST_EXPECT_COLUMNS_EQUIVALENT(expected_keys, *out_keys, verbosity);
   CUDF_TEST_EXPECT_COLUMNS_EQUIVALENT(expected_M2s, *out_M2s, verbosity);
@@ -188,7 +188,7 @@ TYPED_TEST(GroupbyM2TypedTest, KeysAndValuesHaveNulls)
 
   auto const [out_keys, out_M2s] = compute_M2(keys, vals);
   auto const expected_keys       = keys_col<T>{1, 2, 3, 4};
-  auto const expected_M2s = M2s_col<R>{{4.5, 32.0 + 2.0 / 3.0, 18.0, 0.0 /*NULL*/}, null_at(3)};
+  auto const expected_M2s        = M2s_col<R>{4.5, 32.0 + 2.0 / 3.0, 18.0, 0.0};
 
   CUDF_TEST_EXPECT_COLUMNS_EQUIVALENT(expected_keys, *out_keys, verbosity);
   CUDF_TEST_EXPECT_COLUMNS_EQUIVALENT(expected_M2s, *out_M2s, verbosity);

java/src/test/java/ai/rapids/cudf/TableTest.java

@@ -7804,7 +7804,7 @@ void testGroupByM2() {
              .onColumn(1));
          Table resultsSorted = results.orderBy(OrderByArg.asc(0));
          Table expected = new Table.TestBuilder().column(1, 2, 3, 4, 5)
-             .column(0.0, 2.0, 8.0, 0.0, null)
+             .column(0.0, 2.0, 8.0, 0.0, 0.0)
              .build()) {
       assertTablesAreEqual(expected, resultsSorted);
     }