Allow runtime has_nulls parameter for row operators (#9623)

Closes #6952 

This PR allows the `has_nulls` template parameter for row operators to be used a runtime parameter in places where the null-handling logic has little to no affect on runtime performance. 
This can improve compile time as described in #6952.

This will also close #9152 and #9580

Authors:
  - David Wendt (https://github.com/davidwendt)

Approvers:
  - Jake Hemstad (https://github.com/jrhemstad)
  - Nghia Truong (https://github.com/ttnghia)
  - Conor Hoekstra (https://github.com/codereport)

URL: #9623

cpp/benchmarks/groupby/group_sum_benchmark.cu

@@ -44,7 +44,6 @@ void BM_basic_sum(benchmark::State& state)
 {
   using wrapper = cudf::test::fixed_width_column_wrapper<int64_t>;
 
-  // const cudf::size_type num_columns{(cudf::size_type)state.range(0)};
   const cudf::size_type column_size{(cudf::size_type)state.range(0)};
 
   auto data_it = cudf::detail::make_counting_transform_iterator(
@@ -53,7 +52,7 @@ void BM_basic_sum(benchmark::State& state)
   wrapper keys(data_it, data_it + column_size);
   wrapper vals(data_it, data_it + column_size);
 
-  cudf::groupby::groupby gb_obj(cudf::table_view({keys}));
+  cudf::groupby::groupby gb_obj(cudf::table_view({keys, keys, keys}));
 
   std::vector<cudf::groupby::aggregation_request> requests;
   requests.emplace_back(cudf::groupby::aggregation_request());
@@ -73,7 +72,9 @@ BENCHMARK_REGISTER_F(Groupby, Basic)
   ->UseManualTime()
   ->Unit(benchmark::kMillisecond)
   ->Arg(10000)
-  ->Arg(10000000);
+  ->Arg(1000000)
+  ->Arg(10000000)
+  ->Arg(100000000);
 
 void BM_pre_sorted_sum(benchmark::State& state)
 {

cpp/benchmarks/hashing/hash_benchmark.cpp

@@ -25,27 +25,40 @@
 class HashBenchmark : public cudf::benchmark {
 };
 
-static void BM_hash(benchmark::State& state, cudf::hash_id hid)
+enum contains_nulls { no_nulls, nulls };
+
+static void BM_hash(benchmark::State& state, cudf::hash_id hid, contains_nulls has_nulls)
 {
   cudf::size_type const n_rows{(cudf::size_type)state.range(0)};
   auto const data = create_random_table({cudf::type_id::INT64}, 1, row_count{n_rows});
+  if (has_nulls == contains_nulls::no_nulls)
+    data->get_column(0).set_null_mask(rmm::device_buffer{}, 0);
 
   for (auto _ : state) {
     cuda_event_timer raii(state, true, rmm::cuda_stream_default);
     cudf::hash(data->view(), hid);
   }
 }
 
-#define HASH_BENCHMARK_DEFINE(name)                               \
-  BENCHMARK_DEFINE_F(HashBenchmark, name)                         \
-  (::benchmark::State & st) { BM_hash(st, cudf::hash_id::name); } \
-  BENCHMARK_REGISTER_F(HashBenchmark, name)                       \
-    ->RangeMultiplier(4)                                          \
-    ->Ranges({{1 << 14, 1 << 24}})                                \
-    ->UseManualTime()                                             \
+#define concat(a, b, c) a##b##c
+
+#define H_BENCHMARK_DEFINE(name, hid, n)                                            \
+  BENCHMARK_DEFINE_F(HashBenchmark, name)                                           \
+  (::benchmark::State & st) { BM_hash(st, cudf::hash_id::hid, contains_nulls::n); } \
+  BENCHMARK_REGISTER_F(HashBenchmark, name)                                         \
+    ->RangeMultiplier(4)                                                            \
+    ->Ranges({{1 << 14, 1 << 24}})                                                  \
+    ->UseManualTime()                                                               \
     ->Unit(benchmark::kMillisecond);
 
-HASH_BENCHMARK_DEFINE(HASH_MURMUR3)
-HASH_BENCHMARK_DEFINE(HASH_MD5)
-HASH_BENCHMARK_DEFINE(HASH_SERIAL_MURMUR3)
-HASH_BENCHMARK_DEFINE(HASH_SPARK_MURMUR3)
+#define HASH_BENCHMARK_DEFINE(hid, n) H_BENCHMARK_DEFINE(concat(hid, _, n), hid, n)
+
+HASH_BENCHMARK_DEFINE(HASH_MURMUR3, nulls)
+HASH_BENCHMARK_DEFINE(HASH_MD5, nulls)
+HASH_BENCHMARK_DEFINE(HASH_SERIAL_MURMUR3, nulls)
+HASH_BENCHMARK_DEFINE(HASH_SPARK_MURMUR3, nulls)
+
+HASH_BENCHMARK_DEFINE(HASH_MURMUR3, no_nulls)
+HASH_BENCHMARK_DEFINE(HASH_MD5, no_nulls)
+HASH_BENCHMARK_DEFINE(HASH_SERIAL_MURMUR3, no_nulls)
+HASH_BENCHMARK_DEFINE(HASH_SPARK_MURMUR3, no_nulls)