Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

[test] compression benchmarks #17145

Draft
wants to merge 15 commits into
base: branch-24.12
Choose a base branch
from
Draft

[test] compression benchmarks #17145

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
15 commits
Select commit Hold shift + click to select a range
ac65ccf
don't set compression
vuule Oct 22, 2024
79252f4
GZIP stable
vuule Oct 22, 2024
0b6ba52
all pinned & device source
vuule Oct 22, 2024
6e70606
8 iterations
vuule Oct 22, 2024
4953625
Merge branch 'branch-24.12' of https://github.com/rapidsai/cudf into …
vuule Nov 4, 2024
e1a4165
remove pinned_buffer
vuule Nov 4, 2024
1feca3b
logging :P
vuule Nov 4, 2024
714011a
back to more cases
vuule Nov 5, 2024
6356b72
remove prints
vuule Nov 5, 2024
3cdce23
Merge branch 'branch-24.12' of https://github.com/rapidsai/cudf into …
vuule Nov 7, 2024
f225d6f
double up
vuule Nov 7, 2024
6ec8ed4
limit bm
vuule Nov 7, 2024
b832eee
Merge branch 'branch-24.12' of https://github.com/rapidsai/cudf into …
vuule Nov 7, 2024
e9b8c81
Revert "limit bm"
vuule Nov 8, 2024
a1beb19
log second calls
vuule Nov 8, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
72 changes: 36 additions & 36 deletions cpp/benchmarks/io/parquet/parquet_reader_input.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -43,11 +43,14 @@ void parquet_read_common(cudf::size_type num_rows_to_read,
state.exec(
nvbench::exec_tag::sync | nvbench::exec_tag::timer, [&](nvbench::launch& launch, auto& timer) {
try_drop_l3_cache();

timer.start();
auto const result = cudf::io::read_parquet(read_opts);
timer.stop();

setenv("LOG_COMP", "1", 1);
cudf::io::read_parquet(read_opts);
setenv("LOG_COMP", "0", 1);

CUDF_EXPECTS(result.tbl->num_columns() == num_cols_to_read, "Unexpected number of columns");
CUDF_EXPECTS(result.tbl->num_rows() == num_rows_to_read, "Unexpected number of rows");
});
Expand All @@ -66,7 +69,6 @@ void BM_parquet_read_data_common(nvbench::state& state,
{
auto const d_type = get_type_or_group(static_cast<int32_t>(DataType));
auto const source_type = retrieve_io_type_enum(state.get_string("io_type"));
auto const compression = cudf::io::compression_type::SNAPPY;
cuio_source_sink_pair source_sink(source_type);

auto const num_rows_written = [&]() {
Expand All @@ -75,8 +77,7 @@ void BM_parquet_read_data_common(nvbench::state& state,
auto const view = tbl->view();

cudf::io::parquet_writer_options write_opts =
cudf::io::parquet_writer_options::builder(source_sink.make_sink_info(), view)
.compression(compression);
cudf::io::parquet_writer_options::builder(source_sink.make_sink_info(), view);
cudf::io::write_parquet(write_opts);
return view.num_rows();
}();
Expand Down Expand Up @@ -125,7 +126,6 @@ void BM_parquet_read_io_compression(nvbench::state& state)
auto const cardinality = static_cast<cudf::size_type>(state.get_int64("cardinality"));
auto const run_length = static_cast<cudf::size_type>(state.get_int64("run_length"));
auto const source_type = retrieve_io_type_enum(state.get_string("io_type"));
auto const compression = retrieve_compression_type_enum(state.get_string("compression_type"));
cuio_source_sink_pair source_sink(source_type);

auto const num_rows_written = [&]() {
Expand All @@ -136,8 +136,7 @@ void BM_parquet_read_io_compression(nvbench::state& state)
auto const view = tbl->view();

cudf::io::parquet_writer_options write_opts =
cudf::io::parquet_writer_options::builder(source_sink.make_sink_info(), view)
.compression(compression);
cudf::io::parquet_writer_options::builder(source_sink.make_sink_info(), view);
cudf::io::write_parquet(write_opts);
return view.num_rows();
}();
Expand Down Expand Up @@ -170,8 +169,7 @@ void BM_parquet_read_io_small_mixed(nvbench::state& state)

cudf::io::parquet_writer_options write_opts =
cudf::io::parquet_writer_options::builder(source_sink.make_sink_info(), view)
.max_page_size_rows(10'000)
.compression(cudf::io::compression_type::NONE);
.max_page_size_rows(10'000);
cudf::io::write_parquet(write_opts);
}

Expand All @@ -186,7 +184,6 @@ void BM_parquet_read_chunks(nvbench::state& state, nvbench::type_list<nvbench::e
auto const run_length = static_cast<cudf::size_type>(state.get_int64("run_length"));
auto const byte_limit = static_cast<cudf::size_type>(state.get_int64("byte_limit"));
auto const source_type = retrieve_io_type_enum(state.get_string("io_type"));
auto const compression = cudf::io::compression_type::SNAPPY;
cuio_source_sink_pair source_sink(source_type);

auto const num_rows_written = [&]() {
Expand All @@ -197,8 +194,7 @@ void BM_parquet_read_chunks(nvbench::state& state, nvbench::type_list<nvbench::e
auto const view = tbl->view();

cudf::io::parquet_writer_options write_opts =
cudf::io::parquet_writer_options::builder(source_sink.make_sink_info(), view)
.compression(compression);
cudf::io::parquet_writer_options::builder(source_sink.make_sink_info(), view);

cudf::io::write_parquet(write_opts);
return view.num_rows();
Expand All @@ -223,6 +219,12 @@ void BM_parquet_read_chunks(nvbench::state& state, nvbench::type_list<nvbench::e
} while (reader.has_next());
timer.stop();

setenv("LOG_COMP", "1", 1);
do {
auto const _ = reader.read_chunk();
} while (reader.has_next());
setenv("LOG_COMP", "0", 1);

CUDF_EXPECTS(num_rows_read == num_rows_written, "Benchmark did not read the entire table");
});

Expand All @@ -243,7 +245,7 @@ void BM_parquet_read_wide_tables(nvbench::state& state,
auto const data_size_bytes = static_cast<size_t>(state.get_int64("data_size_mb") << 20);
auto const cardinality = static_cast<cudf::size_type>(state.get_int64("cardinality"));
auto const run_length = static_cast<cudf::size_type>(state.get_int64("run_length"));
auto const source_type = io_type::DEVICE_BUFFER;
auto const source_type = retrieve_io_type_enum(state.get_string("io_type"));
cuio_source_sink_pair source_sink(source_type);

auto const num_rows_written = [&]() {
Expand All @@ -254,8 +256,7 @@ void BM_parquet_read_wide_tables(nvbench::state& state,
auto const view = tbl->view();

cudf::io::parquet_writer_options write_opts =
cudf::io::parquet_writer_options::builder(source_sink.make_sink_info(), view)
.compression(cudf::io::compression_type::NONE);
cudf::io::parquet_writer_options::builder(source_sink.make_sink_info(), view);
cudf::io::write_parquet(write_opts);
return view.num_rows();
}();
Expand All @@ -277,7 +278,7 @@ void BM_parquet_read_wide_tables_mixed(nvbench::state& state)
auto const data_size_bytes = static_cast<size_t>(state.get_int64("data_size_mb") << 20);
auto const cardinality = static_cast<cudf::size_type>(state.get_int64("cardinality"));
auto const run_length = static_cast<cudf::size_type>(state.get_int64("run_length"));
auto const source_type = io_type::DEVICE_BUFFER;
auto const source_type = retrieve_io_type_enum(state.get_string("io_type"));
cuio_source_sink_pair source_sink(source_type);

auto const num_rows_written = [&]() {
Expand All @@ -288,8 +289,7 @@ void BM_parquet_read_wide_tables_mixed(nvbench::state& state)
auto const view = tbl->view();

cudf::io::parquet_writer_options write_opts =
cudf::io::parquet_writer_options::builder(source_sink.make_sink_info(), view)
.compression(cudf::io::compression_type::NONE);
cudf::io::parquet_writer_options::builder(source_sink.make_sink_info(), view);
cudf::io::write_parquet(write_opts);
return view.num_rows();
}();
Expand All @@ -299,7 +299,6 @@ void BM_parquet_read_wide_tables_mixed(nvbench::state& state)

using d_type_list = nvbench::enum_type_list<data_type::INTEGRAL,
data_type::FLOAT,
data_type::BOOL8,
data_type::DECIMAL,
data_type::TIMESTAMP,
data_type::DURATION,
Expand All @@ -312,50 +311,51 @@ NVBENCH_BENCH_TYPES(BM_parquet_read_data, NVBENCH_TYPE_AXES(d_type_list))
.set_type_axes_names({"data_type"})
.add_string_axis("io_type", {"DEVICE_BUFFER"})
.set_min_samples(4)
.add_int64_axis("cardinality", {0, 1000})
.add_int64_axis("run_length", {1, 32});
.add_int64_axis("cardinality", {0, 2000})
.add_int64_axis("run_length", {1, 16});

NVBENCH_BENCH(BM_parquet_read_io_compression)
.set_name("parquet_read_io_compression")
.add_string_axis("io_type", {"FILEPATH", "HOST_BUFFER", "DEVICE_BUFFER"})
.add_string_axis("compression_type", {"SNAPPY", "NONE"})
.add_string_axis("io_type", {"DEVICE_BUFFER"})
.set_min_samples(4)
.add_int64_axis("cardinality", {0, 1000})
.add_int64_axis("run_length", {1, 32});
.add_int64_axis("cardinality", {0, 2000})
.add_int64_axis("run_length", {1, 16});

NVBENCH_BENCH_TYPES(BM_parquet_read_chunks, NVBENCH_TYPE_AXES(d_type_list))
.set_name("parquet_read_chunks")
.add_string_axis("io_type", {"DEVICE_BUFFER"})
.set_min_samples(4)
.add_int64_axis("cardinality", {0, 1000})
.add_int64_axis("run_length", {1, 32})
.add_int64_axis("cardinality", {0, 2000})
.add_int64_axis("run_length", {1, 16})
.add_int64_axis("byte_limit", {0, 500'000});

NVBENCH_BENCH(BM_parquet_read_io_small_mixed)
.set_name("parquet_read_io_small_mixed")
.add_string_axis("io_type", {"FILEPATH"})
.add_string_axis("io_type", {"DEVICE_BUFFER"})
.set_min_samples(4)
.add_int64_axis("cardinality", {0, 1000})
.add_int64_axis("run_length", {1, 32})
.add_int64_axis("cardinality", {0, 2000})
.add_int64_axis("run_length", {1, 16})
.add_int64_axis("num_string_cols", {1, 2, 3});

using d_type_list_wide_table = nvbench::enum_type_list<data_type::DECIMAL, data_type::STRING>;
NVBENCH_BENCH_TYPES(BM_parquet_read_wide_tables, NVBENCH_TYPE_AXES(d_type_list_wide_table))
.set_name("parquet_read_wide_tables")
.set_min_samples(4)
.set_type_axes_names({"data_type"})
.add_string_axis("io_type", {"DEVICE_BUFFER"})
.add_int64_axis("data_size_mb", {1024, 2048, 4096})
.add_int64_axis("num_cols", {256, 512, 1024})
.add_int64_axis("cardinality", {0, 1000})
.add_int64_axis("run_length", {1, 32});
.add_int64_axis("cardinality", {0, 2000})
.add_int64_axis("run_length", {1, 16});

NVBENCH_BENCH(BM_parquet_read_wide_tables_mixed)
.set_name("parquet_read_wide_tables_mixed")
.add_string_axis("io_type", {"DEVICE_BUFFER"})
.set_min_samples(4)
.add_int64_axis("data_size_mb", {1024, 2048, 4096})
.add_int64_axis("num_cols", {256, 512, 1024})
.add_int64_axis("cardinality", {0, 1000})
.add_int64_axis("run_length", {1, 32});
.add_int64_axis("cardinality", {0, 2000})
.add_int64_axis("run_length", {1, 16});

// a benchmark for structs that only contain fixed-width types
using d_type_list_struct_only = nvbench::enum_type_list<data_type::STRUCT>;
Expand All @@ -364,5 +364,5 @@ NVBENCH_BENCH_TYPES(BM_parquet_read_fixed_width_struct, NVBENCH_TYPE_AXES(d_type
.set_type_axes_names({"data_type"})
.add_string_axis("io_type", {"DEVICE_BUFFER"})
.set_min_samples(4)
.add_int64_axis("cardinality", {0, 1000})
.add_int64_axis("run_length", {1, 32});
.add_int64_axis("cardinality", {0, 2000})
.add_int64_axis("run_length", {1, 16});
17 changes: 8 additions & 9 deletions cpp/benchmarks/io/parquet/parquet_reader_multithread.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -74,7 +74,6 @@ std::tuple<std::vector<cuio_source_sink_pair>, size_t, size_t> write_file_data(

cudf::io::parquet_writer_options write_opts =
cudf::io::parquet_writer_options::builder(source_sink.make_sink_info(), view)
.compression(cudf::io::compression_type::SNAPPY)
.max_page_size_rows(50000)
.max_page_size_bytes(1024 * 1024);

Expand Down Expand Up @@ -281,7 +280,7 @@ NVBENCH_BENCH(BM_parquet_multithreaded_read_mixed)
.add_int64_axis("num_iterations", {1})
.add_int64_axis("num_cols", {4})
.add_int64_axis("run_length", {8})
.add_string_axis("io_type", {"PINNED_BUFFER"});
.add_string_axis("io_type", {"DEVICE_BUFFER"});

NVBENCH_BENCH(BM_parquet_multithreaded_read_fixed_width)
.set_name("parquet_multithreaded_read_decode_fixed_width")
Expand All @@ -292,7 +291,7 @@ NVBENCH_BENCH(BM_parquet_multithreaded_read_fixed_width)
.add_int64_axis("num_iterations", {1})
.add_int64_axis("num_cols", {4})
.add_int64_axis("run_length", {8})
.add_string_axis("io_type", {"PINNED_BUFFER"});
.add_string_axis("io_type", {"DEVICE_BUFFER"});

NVBENCH_BENCH(BM_parquet_multithreaded_read_string)
.set_name("parquet_multithreaded_read_decode_string")
Expand All @@ -303,7 +302,7 @@ NVBENCH_BENCH(BM_parquet_multithreaded_read_string)
.add_int64_axis("num_iterations", {1})
.add_int64_axis("num_cols", {4})
.add_int64_axis("run_length", {8})
.add_string_axis("io_type", {"PINNED_BUFFER"});
.add_string_axis("io_type", {"DEVICE_BUFFER"});

NVBENCH_BENCH(BM_parquet_multithreaded_read_list)
.set_name("parquet_multithreaded_read_decode_list")
Expand All @@ -314,7 +313,7 @@ NVBENCH_BENCH(BM_parquet_multithreaded_read_list)
.add_int64_axis("num_iterations", {1})
.add_int64_axis("num_cols", {4})
.add_int64_axis("run_length", {8})
.add_string_axis("io_type", {"PINNED_BUFFER"});
.add_string_axis("io_type", {"DEVICE_BUFFER"});

// mixed data types: fixed width, strings
NVBENCH_BENCH(BM_parquet_multithreaded_read_chunked_mixed)
Expand All @@ -328,7 +327,7 @@ NVBENCH_BENCH(BM_parquet_multithreaded_read_chunked_mixed)
.add_int64_axis("run_length", {8})
.add_int64_axis("input_limit", {640 * 1024 * 1024})
.add_int64_axis("output_limit", {640 * 1024 * 1024})
.add_string_axis("io_type", {"PINNED_BUFFER"});
.add_string_axis("io_type", {"DEVICE_BUFFER"});

NVBENCH_BENCH(BM_parquet_multithreaded_read_chunked_fixed_width)
.set_name("parquet_multithreaded_read_decode_chunked_fixed_width")
Expand All @@ -341,7 +340,7 @@ NVBENCH_BENCH(BM_parquet_multithreaded_read_chunked_fixed_width)
.add_int64_axis("run_length", {8})
.add_int64_axis("input_limit", {640 * 1024 * 1024})
.add_int64_axis("output_limit", {640 * 1024 * 1024})
.add_string_axis("io_type", {"PINNED_BUFFER"});
.add_string_axis("io_type", {"DEVICE_BUFFER"});

NVBENCH_BENCH(BM_parquet_multithreaded_read_chunked_string)
.set_name("parquet_multithreaded_read_decode_chunked_string")
Expand All @@ -354,7 +353,7 @@ NVBENCH_BENCH(BM_parquet_multithreaded_read_chunked_string)
.add_int64_axis("run_length", {8})
.add_int64_axis("input_limit", {640 * 1024 * 1024})
.add_int64_axis("output_limit", {640 * 1024 * 1024})
.add_string_axis("io_type", {"PINNED_BUFFER"});
.add_string_axis("io_type", {"DEVICE_BUFFER"});

NVBENCH_BENCH(BM_parquet_multithreaded_read_chunked_list)
.set_name("parquet_multithreaded_read_decode_chunked_list")
Expand All @@ -367,4 +366,4 @@ NVBENCH_BENCH(BM_parquet_multithreaded_read_chunked_list)
.add_int64_axis("run_length", {8})
.add_int64_axis("input_limit", {640 * 1024 * 1024})
.add_int64_axis("output_limit", {640 * 1024 * 1024})
.add_string_axis("io_type", {"PINNED_BUFFER"});
.add_string_axis("io_type", {"DEVICE_BUFFER"});
52 changes: 2 additions & 50 deletions cpp/benchmarks/io/parquet/parquet_reader_options.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -77,7 +77,8 @@ void BM_parquet_read_options(nvbench::state& state,
auto const tbl = create_random_table(data_types, table_size_bytes{data_size});
auto const view = tbl->view();

cuio_source_sink_pair source_sink(io_type::HOST_BUFFER);
auto const source_type = retrieve_io_type_enum(state.get_string("io_type"));
cuio_source_sink_pair source_sink(source_type);
cudf::io::parquet_writer_options options =
cudf::io::parquet_writer_options::builder(source_sink.make_sink_info(), view);
cudf::io::write_parquet(options);
Expand Down Expand Up @@ -133,52 +134,3 @@ void BM_parquet_read_options(nvbench::state& state,
mem_stats_logger.peak_memory_usage(), "peak_memory_usage", "peak_memory_usage");
state.add_buffer_size(source_sink.size(), "encoded_file_size", "encoded_file_size");
}

using row_selections =
nvbench::enum_type_list<row_selection::ALL, row_selection::NROWS, row_selection::ROW_GROUPS>;
NVBENCH_BENCH_TYPES(BM_parquet_read_options,
NVBENCH_TYPE_AXES(nvbench::enum_type_list<column_selection::ALL>,
row_selections,
nvbench::enum_type_list<converts_strings::YES>,
nvbench::enum_type_list<uses_pandas_metadata::YES>,
nvbench::enum_type_list<cudf::type_id::EMPTY>))
.set_name("parquet_read_row_selection")
.set_type_axes_names({"column_selection",
"row_selection",
"str_to_categories",
"uses_pandas_metadata",
"timestamp_type"})
.set_min_samples(4);

using col_selections = nvbench::enum_type_list<column_selection::ALL,
column_selection::ALTERNATE,
column_selection::FIRST_HALF,
column_selection::SECOND_HALF>;
NVBENCH_BENCH_TYPES(BM_parquet_read_options,
NVBENCH_TYPE_AXES(col_selections,
nvbench::enum_type_list<row_selection::ALL>,
nvbench::enum_type_list<converts_strings::YES>,
nvbench::enum_type_list<uses_pandas_metadata::YES>,
nvbench::enum_type_list<cudf::type_id::EMPTY>))
.set_name("parquet_read_column_selection")
.set_type_axes_names({"column_selection",
"row_selection",
"str_to_categories",
"uses_pandas_metadata",
"timestamp_type"})
.set_min_samples(4);

NVBENCH_BENCH_TYPES(
BM_parquet_read_options,
NVBENCH_TYPE_AXES(nvbench::enum_type_list<column_selection::ALL>,
nvbench::enum_type_list<row_selection::ALL>,
nvbench::enum_type_list<converts_strings::YES, converts_strings::NO>,
nvbench::enum_type_list<uses_pandas_metadata::YES, uses_pandas_metadata::NO>,
nvbench::enum_type_list<cudf::type_id::EMPTY>))
.set_name("parquet_read_misc_options")
.set_type_axes_names({"column_selection",
"row_selection",
"str_to_categories",
"uses_pandas_metadata",
"timestamp_type"})
.set_min_samples(4);
Loading
Loading