Skip to content

Fix large-message initialization in communication benchmark #1016

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

Open
pentschev wants to merge 2 commits into
base: main
Choose a base branch
from
Open

Fix large-message initialization in communication benchmark #1016

Show file tree
Hide file tree
Changes from 1 commit
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
7f6168b
Fix large-message initialization in communication benchmark
pentschev May 6, 2026
34e04ad
Use safe_cast, fix bugs
pentschev May 7, 2026
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
31 changes: 23 additions & 8 deletions cpp/benchmarks/utils/random_data.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,6 +3,9 @@
* SPDX-License-Identifier: Apache-2.0
*/

#include <cstdint>
#include <limits>

#include <thrust/random.h>
#include <thrust/transform.h>

Expand All @@ -17,21 +20,30 @@
#include "random_data.hpp"

rmm::device_uvector<std::int32_t> random_device_vector(
cudf::size_type nelem,
std::size_t nelem,
std::int32_t min_val,
std::int32_t max_val,
rmm::cuda_stream_view stream,
rmm::device_async_resource_ref mr
) {
// Fill vector with random data.
rmm::device_uvector<std::int32_t> vec(static_cast<std::size_t>(nelem), stream, mr);
using index_t = std::int64_t;
RAPIDSMPF_EXPECTS(
nelem <= static_cast<std::size_t>(std::numeric_limits<index_t>::max()),
"random_device_vector size exceeds signed iterator range"
);
Copy link
Copy Markdown
Contributor

@nirandaperera nirandaperera May 6, 2026

Choose a reason for hiding this comment

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

Nit. Do we consider 32-bit systems? For 64 bit nelem > INT64_MAX seems unlikely isnt it?

Copy link
Copy Markdown
Member

@madsbk madsbk May 7, 2026

Choose a reason for hiding this comment

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

Always use safe_cast()

Copy link
Copy Markdown
Member Author

@pentschev pentschev May 7, 2026

Choose a reason for hiding this comment

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

Nit. Do we consider 32-bit systems?

No: https://forums.developer.nvidia.com/t/whats-the-last-version-of-the-cuda-toolkit-to-support-32-bit-applications/323106/4

For 64 bit nelem > INT64_MAX seems unlikely isnt it?

Removed that.

Always use safe_cast()

Done.

rmm::device_uvector<std::int32_t> vec(nelem, stream, mr);
thrust::counting_iterator<index_t> const begin(0);
thrust::counting_iterator<index_t> const end(static_cast<index_t>(nelem));
thrust::transform(
rmm::exec_policy(stream),
thrust::make_counting_iterator(0),
thrust::make_counting_iterator(nelem),
begin,
end,
vec.begin(),
[min_val, max_val] __device__(cudf::size_type index) {
thrust::default_random_engine engine(index); // HACK: use the seed as index
[min_val, max_val] __device__(index_t index) {
thrust::default_random_engine engine(
static_cast<thrust::default_random_engine::result_type>(index)
);
thrust::uniform_int_distribution<std::int32_t> dist(min_val, max_val);
return dist(engine);
}
Expand All @@ -46,7 +58,10 @@ std::unique_ptr<cudf::column> random_column(
rmm::cuda_stream_view stream,
rmm::device_async_resource_ref mr
) {
auto vec = random_device_vector(nrows, min_val, max_val, stream, mr);
RAPIDSMPF_EXPECTS(nrows >= 0, "nrows must be non-negative");
auto vec = random_device_vector(
static_cast<std::size_t>(nrows), min_val, max_val, stream, mr
);
return std::make_unique<cudf::column>(
std::move(vec), rmm::device_buffer{0, stream, mr}, 0
);
Expand All @@ -72,7 +87,7 @@ void random_fill(rapidsmpf::Buffer& buffer, rmm::device_async_resource_ref mr) {
case rapidsmpf::MemoryType::DEVICE:
{
auto vec = random_device_vector(
buffer.size / sizeof(std::int32_t) + sizeof(std::int32_t),
(buffer.size + sizeof(std::int32_t) - 1) / sizeof(std::int32_t),
Copy link
Copy Markdown
Contributor

@nirandaperera nirandaperera May 6, 2026

Choose a reason for hiding this comment

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

if buffer is empty now, we will be passing 0 to random_device_vector. I think random_fill was trying to prevent that before? I'm not sure.

Copy link
Copy Markdown
Member Author

@pentschev pentschev May 7, 2026

Choose a reason for hiding this comment

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

Should be fixed now.

std::numeric_limits<std::int32_t>::min(),
std::numeric_limits<std::int32_t>::max(),
buffer.stream(),
Expand Down
5 changes: 4 additions & 1 deletion cpp/benchmarks/utils/random_data.hpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4,6 +4,9 @@
*/
#pragma once

#include <cstddef>
#include <cstdint>

#include <cudf/column/column.hpp>
#include <cudf/table/table.hpp>
#include <cudf/types.hpp>
Expand Down Expand Up @@ -44,7 +47,7 @@ std::size_t constexpr random_table_size_lower_bound(
* @note The function uses the specified CUDA stream for asynchronous operations.
*/
rmm::device_uvector<std::int32_t> random_device_vector(
cudf::size_type nelem,
std::size_t nelem,
std::int32_t min_val,
std::int32_t max_val,
rmm::cuda_stream_view stream,
Expand Down
Loading