From 8a5a1f46641f15dabd1b17b8eb63cdb4cb3a325c Mon Sep 17 00:00:00 2001
From: Yunsong Wang <yunsongw@nvidia.com>
Date: Thu, 11 May 2023 10:10:19 -0700
Subject: [PATCH] Clean up `distinct_count` benchmark (#13321)

This PR moves the `distinct_count` benchmark to the `stream_compaction` folder. It also adds peak memory usage information and targets the public API for performance evaluation.

Authors:
  - Yunsong Wang (https://github.com/PointKernel)

Approvers:
  - Nghia Truong (https://github.com/ttnghia)
  - Karthikeyan (https://github.com/karthikeyann)

URL: https://github.com/rapidsai/cudf/pull/13321
---
 cpp/benchmarks/CMakeLists.txt                   |  7 +++----
 .../distinct_count.cpp                          | 17 ++++++++++-------
 2 files changed, 13 insertions(+), 11 deletions(-)
 rename cpp/benchmarks/{reduction => stream_compaction}/distinct_count.cpp (75%)

diff --git a/cpp/benchmarks/CMakeLists.txt b/cpp/benchmarks/CMakeLists.txt
index 90225b8537a..dcc70a4b6d9 100644
--- a/cpp/benchmarks/CMakeLists.txt
+++ b/cpp/benchmarks/CMakeLists.txt
@@ -153,8 +153,8 @@ ConfigureBench(APPLY_BOOLEAN_MASK_BENCH stream_compaction/apply_boolean_mask.cpp
 # ##################################################################################################
 # * stream_compaction benchmark -------------------------------------------------------------------
 ConfigureNVBench(
-  STREAM_COMPACTION_NVBENCH stream_compaction/distinct.cpp stream_compaction/unique.cpp
-  stream_compaction/unique_count.cpp
+  STREAM_COMPACTION_NVBENCH stream_compaction/distinct.cpp stream_compaction/distinct_count.cpp
+  stream_compaction/unique.cpp stream_compaction/unique_count.cpp
 )
 
 # ##################################################################################################
@@ -195,8 +195,7 @@ ConfigureBench(
   reduction/reduce.cpp reduction/scan.cpp
 )
 ConfigureNVBench(
-  REDUCTION_NVBENCH reduction/distinct_count.cpp reduction/rank.cpp reduction/scan_structs.cpp
-  reduction/segmented_reduce.cpp
+  REDUCTION_NVBENCH reduction/rank.cpp reduction/scan_structs.cpp reduction/segmented_reduce.cpp
 )
 
 # ##################################################################################################
diff --git a/cpp/benchmarks/reduction/distinct_count.cpp b/cpp/benchmarks/stream_compaction/distinct_count.cpp
similarity index 75%
rename from cpp/benchmarks/reduction/distinct_count.cpp
rename to cpp/benchmarks/stream_compaction/distinct_count.cpp
index d2218c270a8..2b2c901b90f 100644
--- a/cpp/benchmarks/reduction/distinct_count.cpp
+++ b/cpp/benchmarks/stream_compaction/distinct_count.cpp
@@ -15,14 +15,14 @@
  */
 
 #include <benchmarks/common/generate_input.hpp>
-#include <benchmarks/fixture/rmm_pool_raii.hpp>
+#include <benchmarks/fixture/benchmark_fixture.hpp>
 
-#include <cudf/detail/stream_compaction.hpp>
+#include <cudf/stream_compaction.hpp>
 
 #include <nvbench/nvbench.cuh>
 
 template <typename Type>
-static void bench_reduction_distinct_count(nvbench::state& state, nvbench::type_list<Type>)
+static void bench_distinct_count(nvbench::state& state, nvbench::type_list<Type>)
 {
   auto const dtype            = cudf::type_to_id<Type>();
   auto const size             = static_cast<cudf::size_type>(state.get_int64("num_rows"));
@@ -40,16 +40,19 @@ static void bench_reduction_distinct_count(nvbench::state& state, nvbench::type_
   auto const& data_column = data_table->get_column(0);
   auto const input_table  = cudf::table_view{{data_column, data_column, data_column}};
 
+  auto mem_stats_logger = cudf::memory_stats_logger();  // init stats logger
+  state.set_cuda_stream(nvbench::make_cuda_stream_view(cudf::get_default_stream().value()));
   state.exec(nvbench::exec_tag::sync, [&](nvbench::launch& launch) {
-    rmm::cuda_stream_view stream{launch.get_stream()};
-    cudf::detail::distinct_count(input_table, cudf::null_equality::EQUAL, stream);
+    cudf::distinct_count(input_table, cudf::null_equality::EQUAL);
   });
+  state.add_buffer_size(
+    mem_stats_logger.peak_memory_usage(), "peak_memory_usage", "peak_memory_usage");
 }
 
 using data_type = nvbench::type_list<int32_t, int64_t, float, double>;
 
-NVBENCH_BENCH_TYPES(bench_reduction_distinct_count, NVBENCH_TYPE_AXES(data_type))
-  .set_name("reduction_distinct_count")
+NVBENCH_BENCH_TYPES(bench_distinct_count, NVBENCH_TYPE_AXES(data_type))
+  .set_name("distinct_count")
   .add_int64_axis("num_rows",
                   {
                     10000,      // 10k