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 cudf::test::to_host to handle both offset types for strings columns #15073

Merged
merged 2 commits into from
Feb 29, 2024
Merged

Fix cudf::test::to_host to handle both offset types for strings columns #15073

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
eb20e37
Fix cudf::test::to_host to handle offset types
davidwendt Feb 16, 2024
e0d6f91
fix doxygen error
davidwendt Feb 16, 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
43 changes: 3 additions & 40 deletions cpp/include/cudf_test/column_utilities.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -194,23 +194,7 @@ std::pair<thrust::host_vector<T>, std::vector<bitmask_type>> to_host(column_view
* `column_view`'s data, and second is the column's bitmask.
*/
template <typename T, std::enable_if_t<cudf::is_fixed_point<T>()>* = nullptr>
std::pair<thrust::host_vector<T>, std::vector<bitmask_type>> to_host(column_view c)
{
using namespace numeric;
using Rep = typename T::rep;

auto host_rep_types = thrust::host_vector<Rep>(c.size());

CUDF_CUDA_TRY(
cudaMemcpy(host_rep_types.data(), c.begin<Rep>(), c.size() * sizeof(Rep), cudaMemcpyDefault));

auto to_fp = [&](Rep val) { return T{scaled_integer<Rep>{val, scale_type{c.type().scale()}}}; };
auto begin = thrust::make_transform_iterator(std::cbegin(host_rep_types), to_fp);
auto const host_fixed_points = thrust::host_vector<T>(begin, begin + c.size());

return {host_fixed_points, bitmask_to_host(c)};
}
//! @endcond
std::pair<thrust::host_vector<T>, std::vector<bitmask_type>> to_host(column_view c);

/**
* @brief Copies the data and bitmask of a `column_view` of strings
Expand All @@ -223,29 +207,8 @@ std::pair<thrust::host_vector<T>, std::vector<bitmask_type>> to_host(column_view
* and second is the column's bitmask.
*/
template <>
inline std::pair<thrust::host_vector<std::string>, std::vector<bitmask_type>> to_host(column_view c)
{
thrust::host_vector<std::string> host_data(c.size());
auto stream = cudf::get_default_stream();
if (c.size() > c.null_count()) {
auto const scv = strings_column_view(c);
auto const h_chars = cudf::detail::make_std_vector_sync<char>(
cudf::device_span<char const>(scv.chars_begin(stream), scv.chars_size(stream)), stream);
auto const h_offsets = cudf::detail::make_std_vector_sync(
cudf::device_span<cudf::size_type const>(scv.offsets().data<cudf::size_type>() + scv.offset(),
scv.size() + 1),
stream);

// build std::string vector from chars and offsets
std::transform(
std::begin(h_offsets),
std::end(h_offsets) - 1,
std::begin(h_offsets) + 1,
host_data.begin(),
[&](auto start, auto end) { return std::string(h_chars.data() + start, end - start); });
}
return {std::move(host_data), bitmask_to_host(c)};
}
std::pair<thrust::host_vector<std::string>, std::vector<bitmask_type>> to_host(column_view c);
//! @endcond

} // namespace cudf::test

Expand Down
77 changes: 76 additions & 1 deletion cpp/tests/utilities/column_utilities.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,5 +1,5 @@
/*
* Copyright (c) 2019-2023, NVIDIA CORPORATION.
* Copyright (c) 2019-2024, NVIDIA CORPORATION.
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
Expand Down Expand Up @@ -941,5 +941,80 @@ bool validate_host_masks(std::vector<bitmask_type> const& expected_mask,
});
}

template <typename T, std::enable_if_t<cudf::is_fixed_point<T>()>*>
std::pair<thrust::host_vector<T>, std::vector<bitmask_type>> to_host(column_view c)
{
using namespace numeric;
using Rep = typename T::rep;

auto host_rep_types = thrust::host_vector<Rep>(c.size());

CUDF_CUDA_TRY(
cudaMemcpy(host_rep_types.data(), c.begin<Rep>(), c.size() * sizeof(Rep), cudaMemcpyDefault));
Comment on lines +952 to +953
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can we use cudf::detail::make_std_vector_sync or some vector factory like that, rather than using raw cudaMemcpy calls? We're using that function below, so it seems reasonable to do the same here.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This cudaMemcpy was not introduced here but merely moved from .hpp file.
All of the to_host overloads return thrust::host_vector objects and the vector factories return std::vector.
I think the right thing would be change all of these since there are more than this one cudMemcpy due to the to_host signature. I believe this would be out of scope for this PR and I can create follow on one to correct all the to_host functions and callers.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Great. Let's do a follow-up PR.


auto to_fp = [&](Rep val) { return T{scaled_integer<Rep>{val, scale_type{c.type().scale()}}}; };
auto begin = thrust::make_transform_iterator(std::cbegin(host_rep_types), to_fp);
auto const host_fixed_points = thrust::host_vector<T>(begin, begin + c.size());

return {host_fixed_points, bitmask_to_host(c)};
Comment on lines +957 to +959
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

nit: copy elision does not happen here. (for both const or non-const)
Tried a few codes in compiler explorer; https://godbolt.org/z/jMGn8qK31
Found that copy elision (NRVO) doesn't work while calling constructor of std::pair.
It's probably better to usestd::move() here.

It could be a follow up PR to optimize all specializations in this file.

}

template std::pair<thrust::host_vector<numeric::decimal32>, std::vector<bitmask_type>> to_host(
column_view c);
template std::pair<thrust::host_vector<numeric::decimal64>, std::vector<bitmask_type>> to_host(
column_view c);
template std::pair<thrust::host_vector<numeric::decimal128>, std::vector<bitmask_type>> to_host(
column_view c);

namespace {
struct strings_to_host_fn {
template <typename OffsetType,
std::enable_if_t<std::is_same_v<OffsetType, int32_t> ||
std::is_same_v<OffsetType, int64_t>>* = nullptr>
void operator()(thrust::host_vector<std::string>& host_data,
char const* chars,
cudf::column_view const& offsets,
rmm::cuda_stream_view stream)
{
auto const h_offsets = cudf::detail::make_std_vector_sync(
cudf::device_span<OffsetType const>(offsets.data<OffsetType>(), offsets.size()), stream);
// build std::string vector from chars and offsets
std::transform(std::begin(h_offsets),
std::end(h_offsets) - 1,
std::begin(h_offsets) + 1,
host_data.begin(),
[&](auto start, auto end) { return std::string(chars + start, end - start); });
}

template <typename OffsetType,
std::enable_if_t<!std::is_same_v<OffsetType, int32_t> &&
!std::is_same_v<OffsetType, int64_t>>* = nullptr>
void operator()(thrust::host_vector<std::string>&,
char const*,
cudf::column_view const&,
rmm::cuda_stream_view)
{
CUDF_FAIL("invalid offsets type");
}
};
} // namespace

template <>
std::pair<thrust::host_vector<std::string>, std::vector<bitmask_type>> to_host(column_view c)
{
thrust::host_vector<std::string> host_data(c.size());
auto stream = cudf::get_default_stream();
if (c.size() > c.null_count()) {
auto const scv = strings_column_view(c);
auto const h_chars = cudf::detail::make_std_vector_sync<char>(
cudf::device_span<char const>(scv.chars_begin(stream), scv.chars_size(stream)), stream);
auto offsets =
cudf::slice(scv.offsets(), {scv.offset(), scv.offset() + scv.size() + 1}).front();
cudf::type_dispatcher(
offsets.type(), strings_to_host_fn{}, host_data, h_chars.data(), offsets, stream);
}
return {std::move(host_data), bitmask_to_host(c)};
}

} // namespace test
} // namespace cudf
Loading