Require cudaMemcpyDefault policy

[bdice]

Description

This PR requires the use of cudaMemcpyDefault instead of explicit host/device policies. This ensures that copies do not fail: we only care about memory accessibility, not memory residency. We must support cases where the source is device-accessible but resident on the host, such as host pinned memory.

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: 138c61bc-4e90-451a-aaaf-ea8d43fbd945

📥 Commits

Reviewing files that changed from the base of the PR and between 0ab2f73 and 4c989bc.

📒 Files selected for processing (2)

• cpp/doxygen/developer_guide/DEVELOPER_GUIDE.md
• cpp/tests/io/io_test_utils.hpp

🚧 Files skipped from review as they are similar to previous changes (1)

• cpp/tests/io/io_test_utils.hpp

---

📝 Walkthrough

Summary by CodeRabbit

• Bug Fixes

  • Improved CUDA memory-copy handling across data transfers, including asynchronous reads and writes.
  • Standardized copy-direction detection to reduce transfer configuration errors across C++, CUDA, Cython, Java, Python, and tests.

• Documentation

  • Added guidance for consistent CUDA memory-copy usage.

• Chores

  • Added automated checks to help prevent incorrect CUDA memory-copy configurations.
  • Updated applicable copyright notices through 2026.

Walkthrough

The change replaces explicit CUDA memcpy direction kinds with cudaMemcpyDefault across runtime code, Java and Python bindings, tests, and evaluation code. A pre-commit hook and developer guide enforce the updated usage.

Changes

CUDA memcpy direction normalization

Layer / File(s)	Summary
Memcpy direction enforcement .pre-commit-config.yaml, cpp/doxygen/developer_guide/DEVELOPER_GUIDE.md	Adds a pre-commit hook and documentation for detecting and avoiding explicit CUDA memcpy direction kinds.
Runtime memcpy updates cpp/include/cudf_test/tdigest_utilities.hpp, cpp/src/io/utilities/data_sink.cpp, cpp/src/strings/utilities.cu, java/src/main/native/src/*, python/pylibcudf/pylibcudf/contiguous_split.pyx	Uses cudaMemcpyDefault for tdigest validation, IO transfers, string table initialization, Java native buffer operations, and staged Python transfers. Updates affected copyright notices.
Validation and test updates cpp/tests/io/cudftable_test.cpp, cpp/tests/io/io_test_utils.hpp, cpp/tests/io/json/json_quote_normalization_test.cpp, skills/accelerated-computing-cudf/evals/files/cudf-native-stream-handoff-boundary/code/threaded_handoff.cu	Updates CUDA copies in tests and evaluation code. Adds CUDA error handling and stream-view synchronization in IO test utilities. Updates affected copyright notices.

Estimated code review effort: 2 (Simple) | ~10 minutes

Possibly related PRs

• rapidsai/cudf#23228: Both changes modify .pre-commit-config.yaml hooks.

Suggested reviewers: jameslamb, brandon-b-miller, qbacpey

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly summarizes the main change: enforcing cudaMemcpyDefault usage.
Description check	✅ Passed	The description explains the cudaMemcpyDefault policy, its purpose, and related test and documentation updates.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@cpp/tests/io/io_test_utils.hpp`:
- Around line 62-64: Update device_read to replace the raw cudaMemcpyAsync call
with cudf::detail::memcpy_async and wrap the result in CUDF_CUDA_TRY; also wrap
cudaStreamSynchronize(stream.value()) with CUDF_CUDA_TRY so both CUDA operations
propagate errors.

🪄 Autofix

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Enterprise

Run ID: afd14363-619d-4ca3-9d6d-1262d1bcc40b

📥 Commits

Reviewing files that changed from the base of the PR and between 6a22d1d and 0ab2f73.

📒 Files selected for processing (11)

• .pre-commit-config.yaml
• cpp/include/cudf_test/tdigest_utilities.hpp
• cpp/src/io/utilities/data_sink.cpp
• cpp/src/strings/utilities.cu
• cpp/tests/io/cudftable_test.cpp
• cpp/tests/io/io_test_utils.hpp
• cpp/tests/io/json/json_quote_normalization_test.cpp
• java/src/main/native/src/jni_writer_data_sink.hpp
• java/src/main/native/src/multi_host_buffer_source.cpp
• python/pylibcudf/pylibcudf/contiguous_split.pyx
• skills/accelerated-computing-cudf/evals/files/cudf-native-stream-handoff-boundary/code/threaded_handoff.cu

[coderabbitai]

🩺 Stability & Availability | 🟡 Minor | ⚡ Quick win

🧩 Analysis chain

🏁 Script executed:

#!/bin/bash
set -eu

printf '%s\n' '--- target file context ---'
sed -n '1,110p' cpp/tests/io/io_test_utils.hpp

printf '%s\n' '--- CUDA error macros and related usage ---'
rg -n --glob '*.{h,hpp,cpp,cu,cuh}' 'CUDF_CUDA_TRY_NO_THROW|CUDF_CUDA_TRY|cudaMemcpyAsync|cudaStreamSynchronize' cpp | head -200

Repository: rapidsai/cudf

Length of output: 21884

---

🏁 Script executed:

#!/bin/bash
set -eu

printf '%s\n' '--- CUDA error macro definition ---'
sed -n '260,315p' cpp/include/cudf/utilities/error.hpp

printf '%s\n' '--- memcpy helper declaration and implementation ---'
sed -n '1,90p' cpp/include/cudf/detail/utilities/cuda_memcpy.hpp
sed -n '1,75p' cpp/src/utilities/cuda_memcpy.cu

printf '%s\n' '--- includes used by comparable test helpers ---'
sed -n '35,70p' cpp/tests/copying/concatenate_tests.cpp
sed -n '75,105p' cpp/tests/io/comp/comp_test.cpp
sed -n '95,125p' cpp/benchmarks/io/cuio_common.cpp

printf '%s\n' '--- direct includes in target or nearby test utilities ---'
rg -n '^`#include` .*error.hpp|^`#include` .*cuda_memcpy.hpp|^`#include` .*cuda_runtime' cpp/tests/io cpp/tests/copying/concatenate_tests.cpp

Repository: rapidsai/cudf

Length of output: 13712

---

🏁 Script executed:

#!/bin/bash
set -eu

python3 - <<'PY'
from pathlib import Path
import re

target = Path("cpp/tests/io/io_test_utils.hpp").read_text()
error_h = Path("cpp/include/cudf/utilities/error.hpp").read_text()
memcpy_h = Path("cpp/include/cudf/detail/utilities/cuda_memcpy.hpp").read_text()

device_read = re.search(
    r'std::unique_ptr<cudf::io::datasource::buffer> device_read\(.*?\n  \}',
    target,
    re.S,
)
assert device_read, "device_read definition not found"
body = device_read.group(0)

calls = {
    "cudaMemcpyAsync": len(re.findall(r'\bcudaMemcpyAsync\s*\(', body)),
    "cudaStreamSynchronize": len(re.findall(r'\bcudaStreamSynchronize\s*\(', body)),
}
wrapped = {
    name: bool(re.search(rf'CUDF_CUDA_TRY\s*\(\s*{name}\s*\(', body))
    for name in calls
}

macro = re.search(
    r'`#define` CUDF_CUDA_TRY\(call\).*?cudaSuccess != status',
    error_h,
    re.S,
)
helper = re.search(
    r'\[\[nodiscard\]\]\s+cudaError_t\s+memcpy_async\(.*?\);',
    memcpy_h,
    re.S,
)

print("device_read calls:", calls)
print("device_read calls wrapped:", wrapped)
print("CUDF_CUDA_TRY checks cudaSuccess:", bool(macro))
print("detail::memcpy_async returns cudaError_t:", bool(helper))
PY

Repository: rapidsai/cudf

Length of output: 393

---

Handle CUDA errors in device_read.

Use cudf::detail::memcpy_async for the copy and wrap its result with CUDF_CUDA_TRY. Wrap cudaStreamSynchronize with CUDF_CUDA_TRY as well.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@cpp/tests/io/io_test_utils.hpp` around lines 62 - 64, Update device_read to
replace the raw cudaMemcpyAsync call with cudf::detail::memcpy_async and wrap
the result in CUDF_CUDA_TRY; also wrap cudaStreamSynchronize(stream.value())
with CUDF_CUDA_TRY so both CUDA operations propagate errors.

Source: Coding guidelines