[GitHub] [arrow] wjones127 commented on a diff in pull request #34323: GH-34322: [C++][Parquet] Encoding Microbench for ByteArray

["wjones127 (via GitHub)" <gi...@apache.org>]

wjones127 commented on code in PR #34323:
URL: https://github.com/apache/arrow/pull/34323#discussion_r1128329500


##########
cpp/src/parquet/encoding_benchmark.cc:
##########
@@ -569,6 +569,130 @@ BENCHMARK(BM_DeltaBitPackingDecode_Int64_Narrow)->Range(MIN_RANGE, MAX_RANGE);
 BENCHMARK(BM_DeltaBitPackingDecode_Int32_Wide)->Range(MIN_RANGE, MAX_RANGE);
 BENCHMARK(BM_DeltaBitPackingDecode_Int64_Wide)->Range(MIN_RANGE, MAX_RANGE);
 
+static void ByteArrayCustomArguments(benchmark::internal::Benchmark* b) {
+  b->ArgsProduct({{8, 64, 1024}, {512, 2048}})
+      ->ArgNames({"max-string-length", "batch-size"});
+}
+
+void EncodingByteArrayBenchmark(benchmark::State& state, Encoding::type encoding) {
+  ::arrow::random::RandomArrayGenerator rag(0);
+  // Using arrow generator to generate random data.
+  int32_t max_length = static_cast<int32_t>(state.range(0));
+  int32_t array_size = static_cast<int32_t>(state.range(1));
+  auto array =
+      rag.String(/* size */ array_size, /* min_length */ 0, /* max_length */ max_length,
+                 /* null_probability */ 0);
+  const auto array_actual =
+      ::arrow::internal::checked_pointer_cast<::arrow::StringArray>(array);
+  auto encoder = MakeTypedEncoder<ByteArrayType>(encoding);
+  std::vector<ByteArray> values;
+  for (int i = 0; i < array_actual->length(); ++i) {
+    values.emplace_back(array_actual->GetView(i));
+  }
+
+  for (auto _ : state) {
+    encoder->Put(values.data(), static_cast<int>(values.size()));
+    encoder->FlushValues();
+  }
+  state.SetItemsProcessed(state.iterations() * array_actual->length());
+  state.SetBytesProcessed(state.iterations() * array_actual->total_values_length());
+}
+
+static void BM_DeltaLengthEncodingByteArray(benchmark::State& state) {
+  EncodingByteArrayBenchmark(state, Encoding::DELTA_LENGTH_BYTE_ARRAY);
+}
+
+static void BM_PlainEncodingByteArray(benchmark::State& state) {
+  EncodingByteArrayBenchmark(state, Encoding::PLAIN);
+}
+
+void DecodingByteArrayBenchmark(benchmark::State& state, Encoding::type encoding) {
+  ::arrow::random::RandomArrayGenerator rag(0);
+  int32_t max_length = static_cast<int32_t>(state.range(0));
+  int32_t array_size = static_cast<int32_t>(state.range(1));
+  // Using arrow to write, because we just benchmark decoding here.
+  auto array =
+      rag.String(/* size */ array_size, /* min_length */ 0, /* max_length */ max_length,
+                 /* null_probability */ 0);
+  const auto array_actual =
+      ::arrow::internal::checked_pointer_cast<::arrow::StringArray>(array);
+  auto encoder = MakeTypedEncoder<ByteArrayType>(encoding);
+  encoder->Put(*array);
+  std::shared_ptr<Buffer> buf = encoder->FlushValues();
+
+  std::vector<ByteArray> values;
+  values.resize(array->length());
+  for (auto _ : state) {
+    auto decoder = MakeTypedDecoder<ByteArrayType>(encoding);
+    decoder->SetData(static_cast<int>(array->length()), buf->data(),
+                     static_cast<int>(buf->size()));
+    decoder->Decode(values.data(), static_cast<int>(values.size()));
+    ::benchmark::DoNotOptimize(values);
+  }
+  state.SetItemsProcessed(state.iterations() * array->length());
+  state.SetBytesProcessed(state.iterations() * array_actual->total_values_length());
+}
+
+static void BM_PlainDecodingByteArray(benchmark::State& state) {
+  DecodingByteArrayBenchmark(state, Encoding::PLAIN);
+}
+
+static void BM_DeltaLengthDecodingByteArray(benchmark::State& state) {
+  DecodingByteArrayBenchmark(state, Encoding::DELTA_LENGTH_BYTE_ARRAY);
+}
+
+BENCHMARK(BM_PlainEncodingByteArray)->Apply(ByteArrayCustomArguments);
+BENCHMARK(BM_DeltaLengthEncodingByteArray)->Apply(ByteArrayCustomArguments);
+BENCHMARK(BM_PlainDecodingByteArray)->Apply(ByteArrayCustomArguments);
+BENCHMARK(BM_DeltaLengthDecodingByteArray)->Apply(ByteArrayCustomArguments);
+
+static void BM_DecodingByteArraySpaced(benchmark::State& state, Encoding::type encoding) {
+  const double null_percent = 0.02;
+
+  auto rand = ::arrow::random::RandomArrayGenerator(0);
+  int32_t max_length = static_cast<int32_t>(state.range(0));
+  int32_t num_values = static_cast<int32_t>(state.range(1));
+  const auto array = rand.String(num_values, /* min_length */ 0,
+                                 /* max_length */ max_length, null_percent);
+  const auto valid_bits = array->null_bitmap_data();
+  const int null_count = static_cast<int>(array->null_count());
+  const auto array_actual =
+      ::arrow::internal::checked_pointer_cast<::arrow::StringArray>(array);
+
+  std::vector<ByteArray> byte_arrays;
+  byte_arrays.reserve(array_actual->length());
+  for (int i = 0; i < array_actual->length(); ++i) {
+    byte_arrays.emplace_back(array_actual->GetView(i));
+  }
+
+  auto encoder = MakeTypedEncoder<ByteArrayType>(encoding);
+  encoder->PutSpaced(byte_arrays.data(), num_values, valid_bits, 0);
+  std::shared_ptr<Buffer> buf = encoder->FlushValues();
+
+  auto decoder = MakeTypedDecoder<ByteArrayType>(encoding);
+  std::vector<uint8_t> decode_values(num_values * sizeof(ByteArray));
+  auto decode_buf = reinterpret_cast<ByteArray*>(decode_values.data());
+  for (auto _ : state) {
+    decoder->SetData(num_values - null_count, buf->data(), static_cast<int>(buf->size()));
+    decoder->DecodeSpaced(decode_buf, num_values, null_count, valid_bits, 0);
+    ::benchmark::DoNotOptimize(decode_buf);
+  }
+  state.counters["null_percent"] = null_percent * 100;
+  state.SetItemsProcessed(state.iterations() * array_actual->length());
+  state.SetBytesProcessed(state.iterations() * array_actual->total_values_length());

Review Comment:
   Elsewhere I know we consider both the size of values *and* offsets in the bytes processed. But not sure if it's necessarily worth doing in these benchmarks.
   
   https://github.com/apache/arrow/blob/864ba8221e0e852a9013d0692310c0f91b9b3b80/cpp/src/parquet/arrow/reader_writer_benchmark.cc#L230-L234



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: github-unsubscribe@arrow.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org