Move nvtext ngrams benchmarks to nvbench (#17173)

Moves the `nvtext::generate_ngrams` and `nvtext::generate_character_ngrams` benchmarks from google-bench to nvbench.
Target parameters are exposed to help with profiling.

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

Approvers:
  - Bradley Dice (https://github.com/bdice)
  - Yunsong Wang (https://github.com/PointKernel)

URL: #17173

cpp/benchmarks/CMakeLists.txt

@@ -345,11 +345,11 @@ ConfigureNVBench(BINARYOP_NVBENCH binaryop/binaryop.cpp binaryop/compiled_binary
 
 # ##################################################################################################
 # * nvtext benchmark -------------------------------------------------------------------
-ConfigureBench(TEXT_BENCH text/ngrams.cpp text/subword.cpp)
+ConfigureBench(TEXT_BENCH text/subword.cpp)
 
 ConfigureNVBench(
-  TEXT_NVBENCH text/edit_distance.cpp text/hash_ngrams.cpp text/jaccard.cpp text/normalize.cpp
-  text/replace.cpp text/tokenize.cpp text/vocab.cpp
+  TEXT_NVBENCH text/edit_distance.cpp text/hash_ngrams.cpp text/jaccard.cpp text/ngrams.cpp
+  text/normalize.cpp text/replace.cpp text/tokenize.cpp text/vocab.cpp
 )
 
 # ##################################################################################################

cpp/benchmarks/text/ngrams.cpp

@@ -15,58 +15,45 @@
  */
 
 #include <benchmarks/common/generate_input.hpp>
-#include <benchmarks/fixture/benchmark_fixture.hpp>
-#include <benchmarks/string/string_bench_args.hpp>
-#include <benchmarks/synchronization/synchronization.hpp>
 
 #include <cudf/scalar/scalar.hpp>
 #include <cudf/strings/strings_column_view.hpp>
 
 #include <nvtext/generate_ngrams.hpp>
 
-class TextNGrams : public cudf::benchmark {};
+#include <nvbench/nvbench.cuh>
 
-enum class ngrams_type { tokens, characters };
-
-static void BM_ngrams(benchmark::State& state, ngrams_type nt)
+static void bench_ngrams(nvbench::state& state)
 {
-  auto const n_rows          = static_cast<cudf::size_type>(state.range(0));
-  auto const max_str_length  = static_cast<cudf::size_type>(state.range(1));
+  auto const num_rows   = static_cast<cudf::size_type>(state.get_int64("num_rows"));
+  auto const row_width  = static_cast<cudf::size_type>(state.get_int64("row_width"));
+  auto const ngram_type = state.get_string("type");
+
   data_profile const profile = data_profile_builder().distribution(
-    cudf::type_id::STRING, distribution_id::NORMAL, 0, max_str_length);
-  auto const column = create_random_column(cudf::type_id::STRING, row_count{n_rows}, profile);
+    cudf::type_id::STRING, distribution_id::NORMAL, 0, row_width);
+  auto const column = create_random_column(cudf::type_id::STRING, row_count{num_rows}, profile);
   cudf::strings_column_view input(column->view());
   auto const separator = cudf::string_scalar("_");
 
-  for (auto _ : state) {
-    cuda_event_timer raii(state, true);
-    switch (nt) {
-      case ngrams_type::tokens: nvtext::generate_ngrams(input, 2, separator); break;
-      case ngrams_type::characters: nvtext::generate_character_ngrams(input); break;
-    }
-  }
+  state.set_cuda_stream(nvbench::make_cuda_stream_view(cudf::get_default_stream().value()));
 
-  state.SetBytesProcessed(state.iterations() * input.chars_size(cudf::get_default_stream()));
-}
+  auto chars_size = input.chars_size(cudf::get_default_stream());
+  state.add_global_memory_reads<nvbench::int8_t>(chars_size);
+  state.add_global_memory_writes<nvbench::int8_t>(chars_size * 2);
 
-static void generate_bench_args(benchmark::internal::Benchmark* b)
-{
-  int const min_rows   = 1 << 12;
-  int const max_rows   = 1 << 24;
-  int const row_mult   = 8;
-  int const min_rowlen = 5;
-  int const max_rowlen = 40;
-  int const len_mult   = 2;
-  generate_string_bench_args(b, min_rows, max_rows, row_mult, min_rowlen, max_rowlen, len_mult);
+  if (ngram_type == "chars") {
+    state.exec(nvbench::exec_tag::sync, [&](nvbench::launch& launch) {
+      auto result = nvtext::generate_character_ngrams(input);
+    });
+  } else {
+    state.exec(nvbench::exec_tag::sync, [&](nvbench::launch& launch) {
+      auto result = nvtext::generate_ngrams(input, 2, separator);
+    });
+  }
 }
 
-#define NVTEXT_BENCHMARK_DEFINE(name)                             \
-  BENCHMARK_DEFINE_F(TextNGrams, name)                            \
-  (::benchmark::State & st) { BM_ngrams(st, ngrams_type::name); } \
-  BENCHMARK_REGISTER_F(TextNGrams, name)                          \
-    ->Apply(generate_bench_args)                                  \
-    ->UseManualTime()                                             \
-    ->Unit(benchmark::kMillisecond);
-
-NVTEXT_BENCHMARK_DEFINE(tokens)
-NVTEXT_BENCHMARK_DEFINE(characters)
+NVBENCH_BENCH(bench_ngrams)
+  .set_name("ngrams")
+  .add_int64_axis("num_rows", {131072, 262144, 524288, 1048578})
+  .add_int64_axis("row_width", {10, 20, 40, 100})
+  .add_string_axis("type", {"chars", "tokens"});