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

Fix decimal benchmark input data generation #11863

Merged
merged 4 commits into from
Oct 6, 2022
Merged

Fix decimal benchmark input data generation #11863

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
07e6cc1
fix decimal benchmark input data generation (scale also was in data)
karthikeyann Oct 5, 2022
0f36e7e
remove T::rep usage completely
karthikeyann Oct 6, 2022
91bbdfe
IILE constexpr if for datatype scale arg for only decimal types.
karthikeyann Oct 6, 2022
99ff519
remove unnecessary transform clamp
karthikeyann Oct 6, 2022
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
52 changes: 27 additions & 25 deletions cpp/benchmarks/common/generate_input.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -247,40 +247,33 @@ struct random_value_fn<T, std::enable_if_t<cudf::is_chrono<T>()>> {
*/
template <typename T>
struct random_value_fn<T, std::enable_if_t<cudf::is_fixed_point<T>()>> {
using rep = typename T::rep;
rep const lower_bound;
rep const upper_bound;
distribution_fn<rep> dist;
using DeviceType = cudf::device_storage_type_t<T>;
DeviceType const lower_bound;
DeviceType const upper_bound;
distribution_fn<DeviceType> dist;
std::optional<numeric::scale_type> scale;

random_value_fn(distribution_params<rep> const& desc)
random_value_fn(distribution_params<DeviceType> const& desc)
: lower_bound{desc.lower_bound},
upper_bound{desc.upper_bound},
dist{make_distribution<rep>(desc.id, desc.lower_bound, desc.upper_bound)}
dist{make_distribution<DeviceType>(desc.id, desc.lower_bound, desc.upper_bound)}
{
}

rmm::device_uvector<T> operator()(thrust::minstd_rand& engine, unsigned size)
[[nodiscard]] numeric::scale_type get_scale(thrust::minstd_rand& engine)
{
if (not scale.has_value()) {
int const max_scale = std::numeric_limits<rep>::digits10;
constexpr int max_scale = std::numeric_limits<DeviceType>::digits10;
std::uniform_int_distribution<int> scale_dist{-max_scale, max_scale};
std::mt19937 engine_scale(engine());
scale = numeric::scale_type{scale_dist(engine_scale)};
}
auto const ints = dist(engine, size);
rmm::device_uvector<T> result(size, cudf::default_stream_value);
// Clamp the generated random value to the specified range
thrust::transform(thrust::device,
ints.begin(),
ints.end(),
result.begin(),
[scale = *(this->scale),
upper_bound = this->upper_bound,
lower_bound = this->lower_bound] __device__(auto int_value) {
return T{std::clamp(int_value, lower_bound, upper_bound), scale};
});
return result;
return scale.value_or(numeric::scale_type{0});
}

rmm::device_uvector<DeviceType> operator()(thrust::minstd_rand& engine, unsigned size)
{
return dist(engine, size);
}
};

Expand Down Expand Up @@ -398,9 +391,17 @@ std::unique_ptr<cudf::column> create_random_column(data_profile const& profile,
distribution_params<bool>{1. - profile.get_null_probability().value_or(0)});
auto value_dist = random_value_fn<T>{profile.get_distribution_params<T>()};

using DeviceType = cudf::device_storage_type_t<T>;
cudf::data_type const dtype = [&]() {
if constexpr (cudf::is_fixed_point<T>())
return cudf::data_type{cudf::type_to_id<T>(), value_dist.get_scale(engine)};
else
return cudf::data_type{cudf::type_to_id<T>()};
}();

// Distribution for picking elements from the array of samples
auto const avg_run_len = profile.get_avg_run_length();
rmm::device_uvector<T> data(0, cudf::default_stream_value);
rmm::device_uvector<DeviceType> data(0, cudf::default_stream_value);
rmm::device_uvector<bool> null_mask(0, cudf::default_stream_value);

if (profile.get_cardinality() == 0 and avg_run_len == 1) {
Expand All @@ -412,11 +413,12 @@ std::unique_ptr<cudf::column> create_random_column(data_profile const& profile,
: profile_cardinality;
}();
rmm::device_uvector<bool> samples_null_mask = valid_dist(engine, cardinality);
rmm::device_uvector<T> samples = value_dist(engine, cardinality);
rmm::device_uvector<DeviceType> samples = value_dist(engine, cardinality);

// generate n samples and gather.
auto const sample_indices =
sample_indices_with_run_length(avg_run_len, cardinality, num_rows, engine);
data = rmm::device_uvector<T>(num_rows, cudf::default_stream_value);
data = rmm::device_uvector<DeviceType>(num_rows, cudf::default_stream_value);
null_mask = rmm::device_uvector<bool>(num_rows, cudf::default_stream_value);
thrust::gather(
thrust::device, sample_indices.begin(), sample_indices.end(), samples.begin(), data.begin());
Expand All @@ -431,7 +433,7 @@ std::unique_ptr<cudf::column> create_random_column(data_profile const& profile,
cudf::detail::valid_if(null_mask.begin(), null_mask.end(), thrust::identity<bool>{});

return std::make_unique<cudf::column>(
cudf::data_type{cudf::type_to_id<T>()},
dtype,
num_rows,
data.release(),
profile.get_null_probability().has_value() ? std::move(result_bitmask) : rmm::device_buffer{});
Expand Down