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.

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

Simplify type dispatch with device_storage_dispatch #7419

Merged
merged 9 commits into from
Feb 23, 2021
Merged
Show file tree
Hide file tree
Changes from 8 commits
Commits
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
24 changes: 11 additions & 13 deletions cpp/include/cudf/detail/gather.cuh
Original file line number Diff line number Diff line change
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2019-2020, NVIDIA CORPORATION.
* Copyright (c) 2019-2021, NVIDIA CORPORATION.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
Expand Down Expand Up @@ -176,11 +176,9 @@ struct column_gatherer_impl {
auto destination_column =
cudf::detail::allocate_like(source_column, num_rows, policy, stream, mr);

using Type = device_storage_type_t<Element>;

gather_helper(source_column.data<Type>(),
gather_helper(source_column.data<Element>(),
source_column.size(),
destination_column->mutable_view().template begin<Type>(),
destination_column->mutable_view().template begin<Element>(),
gather_map_begin,
gather_map_end,
nullify_out_of_bounds,
Expand Down Expand Up @@ -633,14 +631,14 @@ std::unique_ptr<table> gather(
for (auto const& source_column : source_table) {
// The data gather for n columns will be put on the first n streams
destination_columns.push_back(
cudf::type_dispatcher(source_column.type(),
column_gatherer{},
source_column,
gather_map_begin,
gather_map_end,
bounds_policy == out_of_bounds_policy::NULLIFY,
stream,
mr));
cudf::type_dispatcher<dispatch_storage_type>(source_column.type(),
column_gatherer{},
source_column,
gather_map_begin,
gather_map_end,
bounds_policy == out_of_bounds_policy::NULLIFY,
stream,
mr));
}

gather_bitmask_op const op = bounds_policy == out_of_bounds_policy::NULLIFY
Expand Down
24 changes: 11 additions & 13 deletions cpp/include/cudf/detail/scatter.cuh
Original file line number Diff line number Diff line change
Expand Up @@ -91,15 +91,13 @@ struct column_scatterer_impl {
auto result = std::make_unique<column>(target, stream, mr);
auto result_view = result->mutable_view();

using Type = device_storage_type_t<Element>;

// NOTE use source.begin + scatter rows rather than source.end in case the
// scatter map is smaller than the number of source rows
thrust::scatter(rmm::exec_policy(stream),
source.begin<Type>(),
source.begin<Type>() + cudf::distance(scatter_map_begin, scatter_map_end),
source.begin<Element>(),
source.begin<Element>() + cudf::distance(scatter_map_begin, scatter_map_end),
scatter_map_begin,
result_view.begin<Type>());
result_view.begin<Element>());

return result;
}
Expand Down Expand Up @@ -286,14 +284,14 @@ std::unique_ptr<table> scatter(
target.begin(),
result.begin(),
[=](auto const& source_col, auto const& target_col) {
return type_dispatcher(source_col.type(),
scatter_functor,
source_col,
updated_scatter_map_begin,
updated_scatter_map_end,
target_col,
stream,
mr);
return type_dispatcher<dispatch_storage_type>(source_col.type(),
scatter_functor,
source_col,
updated_scatter_map_begin,
updated_scatter_map_end,
target_col,
stream,
mr);
});

auto gather_map = scatter_to_gather(
Expand Down
12 changes: 12 additions & 0 deletions cpp/include/cudf/utilities/type_dispatcher.hpp
Original file line number Diff line number Diff line change
Expand Up @@ -190,6 +190,18 @@ CUDF_TYPE_MAPPING(numeric::decimal32, type_id::DECIMAL32);
CUDF_TYPE_MAPPING(numeric::decimal64, type_id::DECIMAL64);
CUDF_TYPE_MAPPING(cudf::struct_view, type_id::STRUCT);

/**
* @brief Use this specialization on `type_dispatcher` whenever you only need to operate on the
* underlying stored type.
*
* For example, `cudf::sort` in sort.cu uses `cudf::type_dispatcher<dispatch_storage_type>(...)`
codereport marked this conversation as resolved.
Show resolved Hide resolved
* However, reductions needs both `data_type` and underlying type, so can not use this.
*/
template <cudf::type_id Id>
codereport marked this conversation as resolved.
Show resolved Hide resolved
struct dispatch_storage_type {
using type = device_storage_type_t<typename id_to_type_impl<Id>::type>;
};

template <typename T>
struct type_to_scalar_type_impl {
using ScalarType = cudf::scalar;
Expand Down
12 changes: 5 additions & 7 deletions cpp/src/copying/concatenate.cu
Original file line number Diff line number Diff line change
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2020, NVIDIA CORPORATION.
* Copyright (c) 2020-2021, NVIDIA CORPORATION.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
Expand Down Expand Up @@ -300,13 +300,11 @@ struct concatenate_dispatch {
bool const has_nulls =
std::any_of(views.cbegin(), views.cend(), [](auto const& col) { return col.has_nulls(); });

using Type = device_storage_type_t<T>;

// Use a heuristic to guess when the fused kernel will be faster
if (use_fused_kernel_heuristic(has_nulls, views.size())) {
return fused_concatenate<Type>(views, has_nulls, stream, mr);
return fused_concatenate<T>(views, has_nulls, stream, mr);
} else {
return for_each_concatenate<Type>(views, has_nulls, stream, mr);
return for_each_concatenate<T>(views, has_nulls, stream, mr);
}
}
};
Expand Down Expand Up @@ -409,8 +407,8 @@ std::unique_ptr<column> concatenate(std::vector<column_view> const& columns_to_c
return empty_like(columns_to_concat.front());
}

return type_dispatcher(columns_to_concat.front().type(),
concatenate_dispatch{columns_to_concat, stream, mr});
return type_dispatcher<dispatch_storage_type>(
columns_to_concat.front().type(), concatenate_dispatch{columns_to_concat, stream, mr});
}

std::unique_ptr<table> concatenate(std::vector<table_view> const& tables_to_concat,
Expand Down
60 changes: 29 additions & 31 deletions cpp/src/copying/copy.cu
Original file line number Diff line number Diff line change
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2019-2020, NVIDIA CORPORATION.
* Copyright (c) 2019-2021, NVIDIA CORPORATION.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
Expand Down Expand Up @@ -41,28 +41,26 @@ struct copy_if_else_functor_impl {
rmm::cuda_stream_view stream,
rmm::mr::device_memory_resource* mr)
{
using Type = device_storage_type_t<T>;

if (left_nullable) {
if (right_nullable) {
auto lhs_iter = cudf::detail::make_pair_iterator<Type, true>(lhs);
auto rhs_iter = cudf::detail::make_pair_iterator<Type, true>(rhs);
auto lhs_iter = cudf::detail::make_pair_iterator<T, true>(lhs);
auto rhs_iter = cudf::detail::make_pair_iterator<T, true>(rhs);
return detail::copy_if_else(
true, lhs_iter, lhs_iter + size, rhs_iter, filter, lhs.type(), stream, mr);
}
auto lhs_iter = cudf::detail::make_pair_iterator<Type, true>(lhs);
auto rhs_iter = cudf::detail::make_pair_iterator<Type, false>(rhs);
auto lhs_iter = cudf::detail::make_pair_iterator<T, true>(lhs);
auto rhs_iter = cudf::detail::make_pair_iterator<T, false>(rhs);
return detail::copy_if_else(
true, lhs_iter, lhs_iter + size, rhs_iter, filter, lhs.type(), stream, mr);
}
if (right_nullable) {
auto lhs_iter = cudf::detail::make_pair_iterator<Type, false>(lhs);
auto rhs_iter = cudf::detail::make_pair_iterator<Type, true>(rhs);
auto lhs_iter = cudf::detail::make_pair_iterator<T, false>(lhs);
auto rhs_iter = cudf::detail::make_pair_iterator<T, true>(rhs);
return detail::copy_if_else(
true, lhs_iter, lhs_iter + size, rhs_iter, filter, lhs.type(), stream, mr);
}
auto lhs_iter = cudf::detail::make_pair_iterator<Type, false>(lhs);
auto rhs_iter = cudf::detail::make_pair_iterator<Type, false>(rhs);
auto lhs_iter = cudf::detail::make_pair_iterator<T, false>(lhs);
auto rhs_iter = cudf::detail::make_pair_iterator<T, false>(rhs);
return detail::copy_if_else(
false, lhs_iter, lhs_iter + size, rhs_iter, filter, lhs.type(), stream, mr);
}
Expand Down Expand Up @@ -182,30 +180,30 @@ std::unique_ptr<column> copy_if_else(Left const& lhs,
auto filter = [bool_mask_device] __device__(cudf::size_type i) {
return bool_mask_device.is_valid_nocheck(i) and bool_mask_device.element<bool>(i);
};
return cudf::type_dispatcher(lhs.type(),
copy_if_else_functor{},
lhs,
rhs,
boolean_mask.size(),
left_nullable,
right_nullable,
filter,
stream,
mr);
return cudf::type_dispatcher<dispatch_storage_type>(lhs.type(),
copy_if_else_functor{},
lhs,
rhs,
boolean_mask.size(),
left_nullable,
right_nullable,
filter,
stream,
mr);
} else {
auto filter = [bool_mask_device] __device__(cudf::size_type i) {
return bool_mask_device.element<bool>(i);
};
return cudf::type_dispatcher(lhs.type(),
copy_if_else_functor{},
lhs,
rhs,
boolean_mask.size(),
left_nullable,
right_nullable,
filter,
stream,
mr);
return cudf::type_dispatcher<dispatch_storage_type>(lhs.type(),
copy_if_else_functor{},
lhs,
rhs,
boolean_mask.size(),
left_nullable,
right_nullable,
filter,
stream,
mr);
}
}

Expand Down
20 changes: 10 additions & 10 deletions cpp/src/copying/copy_range.cu
Original file line number Diff line number Diff line change
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2019-2020, NVIDIA CORPORATION.
* Copyright (c) 2019-2021, NVIDIA CORPORATION.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
Expand Down Expand Up @@ -108,9 +108,8 @@ struct out_of_place_copy_range_dispatch {
}

if (source_end != source_begin) { // otherwise no-op
using Type = cudf::device_storage_type_t<T>;
auto ret_view = p_ret->mutable_view();
in_place_copy_range<Type>(source, ret_view, source_begin, source_end, target_begin, stream);
in_place_copy_range<T>(source, ret_view, source_begin, source_end, target_begin, stream);
}

return p_ret;
Expand Down Expand Up @@ -261,13 +260,14 @@ std::unique_ptr<column> copy_range(column_view const& source,
"Range is out of bounds.");
CUDF_EXPECTS(target.type() == source.type(), "Data type mismatch.");

return cudf::type_dispatcher(target.type(),
out_of_place_copy_range_dispatch{source, target},
source_begin,
source_end,
target_begin,
stream,
mr);
return cudf::type_dispatcher<dispatch_storage_type>(
target.type(),
out_of_place_copy_range_dispatch{source, target},
source_begin,
source_end,
target_begin,
stream,
mr);
}

} // namespace detail
Expand Down
24 changes: 11 additions & 13 deletions cpp/src/copying/scatter.cu
Original file line number Diff line number Diff line change
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2019-2020, NVIDIA CORPORATION.
* Copyright (c) 2019-2021, NVIDIA CORPORATION.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
Expand Down Expand Up @@ -106,18 +106,16 @@ struct column_scalar_scatterer_impl {
auto result = std::make_unique<column>(target, stream, mr);
auto result_view = result->mutable_view();

using Type = device_storage_type_t<Element>;

// Use permutation iterator with constant index to dereference scalar data
auto scalar_impl = static_cast<const scalar_type_t<Type>*>(&source.get());
auto scalar_impl = static_cast<const scalar_type_t<Element>*>(&source.get());
auto scalar_iter =
thrust::make_permutation_iterator(scalar_impl->data(), thrust::make_constant_iterator(0));

thrust::scatter(rmm::exec_policy(stream),
scalar_iter,
scalar_iter + scatter_rows,
scatter_iter,
result_view.begin<Type>());
result_view.begin<Element>());

return result;
}
Expand Down Expand Up @@ -300,14 +298,14 @@ std::unique_ptr<table> scatter(std::vector<std::reference_wrapper<const scalar>>
target.begin(),
result.begin(),
[=](auto const& source_scalar, auto const& target_col) {
return type_dispatcher(target_col.type(),
scatter_functor,
source_scalar,
scatter_iter,
scatter_rows,
target_col,
stream,
mr);
return type_dispatcher<dispatch_storage_type>(target_col.type(),
scatter_functor,
source_scalar,
scatter_iter,
scatter_rows,
target_col,
stream,
mr);
});

scatter_scalar_bitmask(source, scatter_iter, scatter_rows, result, stream, mr);
Expand Down
12 changes: 6 additions & 6 deletions cpp/src/copying/shift.cu
Original file line number Diff line number Diff line change
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2019-2020, NVIDIA CORPORATION.
* Copyright (c) 2019-2021, NVIDIA CORPORATION.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
Expand Down Expand Up @@ -60,8 +60,7 @@ struct shift_functor {
rmm::cuda_stream_view stream,
rmm::mr::device_memory_resource* mr)
{
using Type = device_storage_type_t<T>;
using ScalarType = cudf::scalar_type_t<Type>;
using ScalarType = cudf::scalar_type_t<T>;
auto& scalar = static_cast<ScalarType const&>(fill_value);

auto device_input = column_device_view::create(input);
Expand All @@ -88,7 +87,7 @@ struct shift_functor {
output->set_null_count(std::get<1>(mask_pair));
}

auto data = device_output->data<Type>();
auto data = device_output->data<T>();

// avoid assigning elements we know to be invalid.
if (not scalar.is_valid()) {
Expand All @@ -103,7 +102,7 @@ struct shift_functor {
auto func_value =
[size, offset, fill = scalar.data(), input = *device_input] __device__(size_type idx) {
auto src_idx = idx - offset;
return out_of_bounds(size, src_idx) ? *fill : input.element<Type>(src_idx);
return out_of_bounds(size, src_idx) ? *fill : input.element<T>(src_idx);
};

thrust::transform(rmm::exec_policy(stream), index_begin, index_end, data, func_value);
Expand All @@ -128,7 +127,8 @@ std::unique_ptr<column> shift(column_view const& input,

if (input.is_empty()) { return empty_like(input); }

return type_dispatcher(input.type(), shift_functor{}, input, offset, fill_value, stream, mr);
return type_dispatcher<dispatch_storage_type>(
input.type(), shift_functor{}, input, offset, fill_value, stream, mr);
}

} // namespace detail
Expand Down
Loading