From be9ba6c4d999fca1588f845b825d26e79cece621 Mon Sep 17 00:00:00 2001 From: Basit Ayantunde Date: Wed, 20 Nov 2024 15:38:54 +0000 Subject: [PATCH] Added Arrow Interop Benchmarks (#17194) This merge request adds benchmarks for the Arrow Interop APIs: - `from_arrow_host` - `to_arrow_host` - `from_arrow_device` - `to_arrow_device` Closes https://github.com/rapidsai/cudf/issues/17104 Authors: - Basit Ayantunde (https://github.com/lamarrr) Approvers: - David Wendt (https://github.com/davidwendt) URL: https://github.com/rapidsai/cudf/pull/17194 --- cpp/benchmarks/CMakeLists.txt | 6 + cpp/benchmarks/interop/interop.cpp | 244 +++++++++++++++++++++++++++++ 2 files changed, 250 insertions(+) create mode 100644 cpp/benchmarks/interop/interop.cpp diff --git a/cpp/benchmarks/CMakeLists.txt b/cpp/benchmarks/CMakeLists.txt index 7fdaff35525..ca2bdc24b25 100644 --- a/cpp/benchmarks/CMakeLists.txt +++ b/cpp/benchmarks/CMakeLists.txt @@ -286,6 +286,12 @@ ConfigureNVBench( ConfigureBench(HASHING_BENCH hashing/partition.cpp) ConfigureNVBench(HASHING_NVBENCH hashing/hash.cpp) +# ################################################################################################## +# * interop benchmark ------------------------------------------------------------------------------ +ConfigureNVBench(INTEROP_NVBENCH interop/interop.cpp) +target_link_libraries(INTEROP_NVBENCH PRIVATE nanoarrow) +target_include_directories(INTEROP_NVBENCH PRIVATE ${CMAKE_SOURCE_DIR}/tests/interop) + # ################################################################################################## # * merge benchmark ------------------------------------------------------------------------------- ConfigureBench(MERGE_BENCH merge/merge.cpp) diff --git a/cpp/benchmarks/interop/interop.cpp b/cpp/benchmarks/interop/interop.cpp new file mode 100644 index 00000000000..dad7e6f429e --- /dev/null +++ b/cpp/benchmarks/interop/interop.cpp @@ -0,0 +1,244 @@ +/* + * Copyright (c) 2024, NVIDIA CORPORATION. + * + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ + +#include +#include + +#include + +#include + +#include +#include +#include +#include + +#include +#include +#include + +template +void BM_to_arrow_device(nvbench::state& state, nvbench::type_list>) +{ + auto const num_rows = static_cast(state.get_int64("num_rows")); + auto const num_columns = static_cast(state.get_int64("num_columns")); + auto const num_elements = static_cast(num_rows) * num_columns; + + std::vector types(num_columns, data_type); + + auto const table = create_random_table(types, row_count{num_rows}); + int64_t const size_bytes = estimate_size(table->view()); + + state.add_element_count(num_elements, "num_elements"); + state.add_global_memory_reads(size_bytes); + state.add_global_memory_writes(size_bytes); + + state.exec(nvbench::exec_tag::sync, [&](nvbench::launch& launch) { + cudf::to_arrow_device(table->view(), rmm::cuda_stream_view{launch.get_stream()}); + }); +} + +template +void BM_to_arrow_host(nvbench::state& state, nvbench::type_list>) +{ + auto const num_rows = static_cast(state.get_int64("num_rows")); + auto const num_columns = static_cast(state.get_int64("num_columns")); + auto const num_elements = static_cast(num_rows) * num_columns; + + std::vector types(num_columns, data_type); + + auto const table = create_random_table(types, row_count{num_rows}); + int64_t const size_bytes = estimate_size(table->view()); + + state.add_element_count(num_elements, "num_elements"); + state.add_global_memory_reads(size_bytes); + state.add_global_memory_writes(size_bytes); + + state.exec(nvbench::exec_tag::sync, [&](nvbench::launch& launch) { + cudf::to_arrow_host(table->view(), rmm::cuda_stream_view{launch.get_stream()}); + }); +} + +template +void BM_from_arrow_device(nvbench::state& state, nvbench::type_list>) +{ + auto const num_rows = static_cast(state.get_int64("num_rows")); + auto const num_columns = static_cast(state.get_int64("num_columns")); + auto const num_elements = static_cast(num_rows) * num_columns; + + std::vector types(num_columns, data_type); + + data_profile profile; + profile.set_struct_depth(1); + profile.set_list_depth(1); + + auto const table = create_random_table(types, row_count{num_rows}, profile); + cudf::table_view table_view = table->view(); + int64_t const size_bytes = estimate_size(table_view); + + std::vector table_metadata; + + std::transform(thrust::make_counting_iterator(0), + thrust::make_counting_iterator(num_columns), + std::back_inserter(table_metadata), + [&](auto const column) { + cudf::column_metadata column_metadata{""}; + column_metadata.children_meta = std::vector( + table->get_column(column).num_children(), cudf::column_metadata{""}); + return column_metadata; + }); + + cudf::unique_schema_t schema = cudf::to_arrow_schema(table_view, table_metadata); + cudf::unique_device_array_t input = cudf::to_arrow_device(table_view); + + state.add_element_count(num_elements, "num_elements"); + state.add_global_memory_reads(size_bytes); + state.add_global_memory_writes(size_bytes); + + state.exec(nvbench::exec_tag::sync, [&](nvbench::launch& launch) { + cudf::from_arrow_device_column( + schema.get(), input.get(), rmm::cuda_stream_view{launch.get_stream()}); + }); +} + +template +void BM_from_arrow_host(nvbench::state& state, nvbench::type_list>) +{ + auto const num_rows = static_cast(state.get_int64("num_rows")); + auto const num_columns = static_cast(state.get_int64("num_columns")); + auto const num_elements = static_cast(num_rows) * num_columns; + + std::vector types(num_columns, data_type); + + data_profile profile; + profile.set_struct_depth(1); + profile.set_list_depth(1); + + auto const table = create_random_table(types, row_count{num_rows}, profile); + cudf::table_view table_view = table->view(); + int64_t const size_bytes = estimate_size(table_view); + + std::vector table_metadata; + + std::transform(thrust::make_counting_iterator(0), + thrust::make_counting_iterator(num_columns), + std::back_inserter(table_metadata), + [&](auto const column) { + cudf::column_metadata column_metadata{""}; + column_metadata.children_meta = std::vector( + table->get_column(column).num_children(), cudf::column_metadata{""}); + return column_metadata; + }); + + cudf::unique_schema_t schema = cudf::to_arrow_schema(table_view, table_metadata); + cudf::unique_device_array_t input = cudf::to_arrow_host(table_view); + + state.add_element_count(num_elements, "num_elements"); + state.add_global_memory_reads(size_bytes); + state.add_global_memory_writes(size_bytes); + + state.exec(nvbench::exec_tag::sync, [&](nvbench::launch& launch) { + cudf::from_arrow_host_column( + schema.get(), input.get(), rmm::cuda_stream_view{launch.get_stream()}); + }); +} + +using data_types = nvbench::enum_type_list; + +static char const* stringify_type(cudf::type_id value) +{ + switch (value) { + case cudf::type_id::INT8: return "INT8"; + case cudf::type_id::INT16: return "INT16"; + case cudf::type_id::INT32: return "INT32"; + case cudf::type_id::INT64: return "INT64"; + case cudf::type_id::UINT8: return "UINT8"; + case cudf::type_id::UINT16: return "UINT16"; + case cudf::type_id::UINT32: return "UINT32"; + case cudf::type_id::UINT64: return "UINT64"; + case cudf::type_id::FLOAT32: return "FLOAT32"; + case cudf::type_id::FLOAT64: return "FLOAT64"; + case cudf::type_id::BOOL8: return "BOOL8"; + case cudf::type_id::TIMESTAMP_DAYS: return "TIMESTAMP_DAYS"; + case cudf::type_id::TIMESTAMP_SECONDS: return "TIMESTAMP_SECONDS"; + case cudf::type_id::TIMESTAMP_MILLISECONDS: return "TIMESTAMP_MILLISECONDS"; + case cudf::type_id::TIMESTAMP_MICROSECONDS: return "TIMESTAMP_MICROSECONDS"; + case cudf::type_id::TIMESTAMP_NANOSECONDS: return "TIMESTAMP_NANOSECONDS"; + case cudf::type_id::DURATION_DAYS: return "DURATION_DAYS"; + case cudf::type_id::DURATION_SECONDS: return "DURATION_SECONDS"; + case cudf::type_id::DURATION_MILLISECONDS: return "DURATION_MILLISECONDS"; + case cudf::type_id::DURATION_MICROSECONDS: return "DURATION_MICROSECONDS"; + case cudf::type_id::DURATION_NANOSECONDS: return "DURATION_NANOSECONDS"; + case cudf::type_id::DICTIONARY32: return "DICTIONARY32"; + case cudf::type_id::STRING: return "STRING"; + case cudf::type_id::LIST: return "LIST"; + case cudf::type_id::DECIMAL32: return "DECIMAL32"; + case cudf::type_id::DECIMAL64: return "DECIMAL64"; + case cudf::type_id::DECIMAL128: return "DECIMAL128"; + case cudf::type_id::STRUCT: return "STRUCT"; + default: return "unknown"; + } +} + +NVBENCH_DECLARE_ENUM_TYPE_STRINGS(cudf::type_id, stringify_type, stringify_type) + +NVBENCH_BENCH_TYPES(BM_to_arrow_host, NVBENCH_TYPE_AXES(data_types)) + .set_type_axes_names({"data_type"}) + .set_name("to_arrow_host") + .add_int64_axis("num_rows", {10'000, 100'000, 1'000'000, 10'000'000}) + .add_int64_axis("num_columns", {1}); + +NVBENCH_BENCH_TYPES(BM_to_arrow_device, NVBENCH_TYPE_AXES(data_types)) + .set_type_axes_names({"data_type"}) + .set_name("to_arrow_device") + .add_int64_axis("num_rows", {10'000, 100'000, 1'000'000, 10'000'000}) + .add_int64_axis("num_columns", {1}); + +NVBENCH_BENCH_TYPES(BM_from_arrow_host, NVBENCH_TYPE_AXES(data_types)) + .set_type_axes_names({"data_type"}) + .set_name("from_arrow_host") + .add_int64_axis("num_rows", {10'000, 100'000, 1'000'000, 10'000'000}) + .add_int64_axis("num_columns", {1}); + +NVBENCH_BENCH_TYPES(BM_from_arrow_device, NVBENCH_TYPE_AXES(data_types)) + .set_type_axes_names({"data_type"}) + .set_name("from_arrow_device") + .add_int64_axis("num_rows", {10'000, 100'000, 1'000'000, 10'000'000}) + .add_int64_axis("num_columns", {1});