[PR #661] [MERGED] fix(bench): use correct GPU ID in memory registration #1725

New issue

Closed

opened 2026-10-29 21:29:50 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-29 21:29:50 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/kvcache-ai/Mooncake/pull/661
Author: @staryxchen
Created: 7/23/2025
Status: ✅ Merged
Merged: 7/24/2025
Merged by: @alogfans

Base: main ← Head: bugfix/memory_location

📝 Commits (1)

64a9289 fix: use correct GPU ID in memory registration

📊 Changes

4 files changed (+18 additions, -9 deletions)

View changed files

📝 mooncake-transfer-engine/example/transfer_engine_bench.cpp (+8 -4)
📝 mooncake-transfer-engine/example/transfer_engine_bench_with_retry.cpp (+4 -2)
📝 mooncake-transfer-engine/tests/nvlink_transport_test.cpp (+2 -1)
📝 mooncake-transfer-engine/tests/rdma_transport_test.cpp (+4 -2)

📄 Description

Summary

This PR fixes inconsistent GPU ID usage in memory registration across test and benchmark files. Previously, some files were using loop index i instead of the configured FLAGS_gpu_id for VRAM buffer registration.

Changes

transfer_engine_bench.cpp: Use FLAGS_gpu_id for VRAM buffer registration in both initiator and target functions
transfer_engine_bench_with_retry.cpp: Use FLAGS_gpu_id for VRAM buffer registration in initiator function
nvlink_transport_test.cpp: Use gpu_id variable instead of hardcoded "cuda:0"
rdma_transport_test.cpp: Use FLAGS_gpu_id for VRAM buffer registration

Impact

Ensures consistent GPU device selection across all test files
Fixes potential issues where tests might register memory on wrong GPU device

_{🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.}

## 📋 Pull Request Information **Original PR:** https://github.com/kvcache-ai/Mooncake/pull/661 **Author:** [@staryxchen](https://github.com/staryxchen) **Created:** 7/23/2025 **Status:** ✅ Merged **Merged:** 7/24/2025 **Merged by:** [@alogfans](https://github.com/alogfans) **Base:** `main` ← **Head:** `bugfix/memory_location` --- ### 📝 Commits (1) - [`64a9289`](https://github.com/kvcache-ai/Mooncake/commit/64a9289f2a7676c4ae9c3f24a9a90cae6d0b5480) fix: use correct GPU ID in memory registration ### 📊 Changes **4 files changed** (+18 additions, -9 deletions) <details> <summary>View changed files</summary> 📝 `mooncake-transfer-engine/example/transfer_engine_bench.cpp` (+8 -4) 📝 `mooncake-transfer-engine/example/transfer_engine_bench_with_retry.cpp` (+4 -2) 📝 `mooncake-transfer-engine/tests/nvlink_transport_test.cpp` (+2 -1) 📝 `mooncake-transfer-engine/tests/rdma_transport_test.cpp` (+4 -2) </details> ### 📄 Description # Summary This PR fixes inconsistent GPU ID usage in memory registration across test and benchmark files. Previously, some files were using loop index i instead of the configured `FLAGS_gpu_id` for VRAM buffer registration. # Changes - transfer_engine_bench.cpp: Use FLAGS_gpu_id for VRAM buffer registration in both initiator and target functions - transfer_engine_bench_with_retry.cpp: Use FLAGS_gpu_id for VRAM buffer registration in initiator function - nvlink_transport_test.cpp: Use gpu_id variable instead of hardcoded "cuda:0" - rdma_transport_test.cpp: Use FLAGS_gpu_id for VRAM buffer registration # Impact - Ensures consistent GPU device selection across all test files - Fixes potential issues where tests might register memory on wrong GPU device --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi

2026-10-29 21:29:50 +00:00