From a7f5261208f67e414ea0fdab924792bdb1283b50 Mon Sep 17 00:00:00 2001
From: Martin Marenz <martin.marenz@webfleet.com>
Date: Fri, 25 Aug 2023 20:22:00 +0200
Subject: [PATCH] Enable fractional null probability for hashing benchmark

In the past, the HASING_NVBENCH benchmark treated the nulls parameter as
a boolean. Any value other than 0.0 resulted in a null probability of
1.0.

Now, the nulls parameter directly determines the null probability. For
instance, a value of 0.1 will generate 10% of the data as null.
Moreover, setting nulls to 0.0 produces data without a null bitmask.

Additionally, `bytes_per_second` are added to the benchmark.

This patch relates to #13735.
---
 cpp/benchmarks/hashing/hash.cpp | 33 +++++++++++++++++++++++++++++----
 1 file changed, 29 insertions(+), 4 deletions(-)

diff --git a/cpp/benchmarks/hashing/hash.cpp b/cpp/benchmarks/hashing/hash.cpp
index f0e9202612e..86ea17316b8 100644
--- a/cpp/benchmarks/hashing/hash.cpp
+++ b/cpp/benchmarks/hashing/hash.cpp
@@ -17,6 +17,7 @@
 #include <benchmarks/common/generate_input.hpp>
 
 #include <cudf/hashing.hpp>
+#include <cudf/strings/strings_column_view.hpp>
 #include <cudf/table/table.hpp>
 #include <cudf/utilities/default_stream.hpp>
 
@@ -24,25 +25,49 @@
 
 static void bench_hash(nvbench::state& state)
 {
-  auto const num_rows  = static_cast<cudf::size_type>(state.get_int64("num_rows"));
-  auto const nulls     = static_cast<bool>(state.get_float64("nulls"));
+  auto const num_rows = static_cast<cudf::size_type>(state.get_int64("num_rows"));
+  auto const nulls    = state.get_float64("nulls");
+  // disable null bitmask if probability is exactly 0.0
+  bool const no_nulls  = nulls == 0.0;
   auto const hash_name = state.get_string("hash_name");
 
-  data_profile const profile = data_profile_builder().null_probability(nulls);
-  auto const data            = create_random_table(
+  data_profile const profile =
+    data_profile_builder().null_probability(no_nulls ? std::nullopt : std::optional<double>{nulls});
+  auto const data = create_random_table(
     {cudf::type_id::INT64, cudf::type_id::STRING}, row_count{num_rows}, profile);
 
   auto stream = cudf::get_default_stream();
   state.set_cuda_stream(nvbench::make_cuda_stream_view(stream.value()));
 
+  // collect statistics
+  cudf::strings_column_view input(data->get_column(1).view());
+  auto const chars_size = input.chars_size();
+  // add memory read from string column
+  state.add_global_memory_reads<nvbench::int8_t>(chars_size);
+  // add memory read from int64_t column
+  state.add_global_memory_reads<nvbench::int64_t>(num_rows);
+  // add memory read from bitmaks
+  if (!no_nulls) {
+    state.add_global_memory_reads<nvbench::int8_t>(2 *
+                                                   cudf::bitmask_allocation_size_bytes(num_rows));
+  }
+  // memory written depends on used hash
+
   if (hash_name == "murmurhash3_x86_32") {
+    state.add_global_memory_reads<nvbench::uint32_t>(num_rows);
+
     state.exec(nvbench::exec_tag::sync, [&](nvbench::launch& launch) {
       auto result = cudf::hashing::murmurhash3_x86_32(data->view());
     });
   } else if (hash_name == "md5") {
+    // md5 writes out string with 32bytes
+    state.add_global_memory_reads<nvbench::int8_t>(32 * num_rows);
+
     state.exec(nvbench::exec_tag::sync,
                [&](nvbench::launch& launch) { auto result = cudf::hashing::md5(data->view()); });
   } else if (hash_name == "spark_murmurhash3_x86_32") {
+    state.add_global_memory_reads<nvbench::int32_t>(num_rows);
+
     state.exec(nvbench::exec_tag::sync, [&](nvbench::launch& launch) {
       auto result = cudf::hashing::spark_murmurhash3_x86_32(data->view());
     });