Add GpuIpcMemHandle

[chhwang]

Add GpuIpcMemHandle that is a generic GPU memory handle that covers all existing methods for GPU memory mapping. This PR fixes issues that fail to properly fallback to a feasible type of memory handle on the importing environment. It also consolidates code for creating or destroying various memory handles into a single RAII wrapper.

[Copilot]

Pull request overview

This pull request introduces a new GpuIpcMem class to encapsulate GPU IPC (Inter-Process Communication) memory management, replacing the previous inline implementation in RegisteredMemory. The refactoring simplifies memory handle management by consolidating three different handle types (RuntimeIpc, PosixFd, and Fabric) into a unified abstraction layer.

Key changes include:

• Introduction of GpuIpcMemHandle struct and GpuIpcMem class for managing GPU IPC memory handles
• Significant simplification of RegisteredMemory implementation by delegating IPC handle management to the new classes
• Migration from old debug.h logging system to new logger.hpp system with GPU subsystem support
• Test infrastructure improvements including MPI group caching and better resource cleanup

Reviewed changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
src/gpu_ipc_mem.cc	New file implementing GpuIpcMem class with support for RuntimeIpc, PosixFd, and Fabric handle types
src/include/gpu_ipc_mem.hpp	New header defining GpuIpcMemHandle struct and GpuIpcMem class interfaces
src/registered_memory.cc	Refactored to use GpuIpcMem abstraction, removing ~100 lines of inline IPC handle management code
src/include/registered_memory.hpp	Updated to use GpuIpcMemHandle in TransportInfo union and removed obsolete fields
src/include/logger.hpp	Added GPU subsystem to LogSubsys enum for GPU-related logging
test/mp_unit/executor_tests.cc	Moved test prerequisites check from test body to SetUp() method
python/test/mscclpp_mpi.py	Added MPI group caching and improved fixture cleanup with try-finally
python/test/conftest.py	New conftest for MPI initialization before test collection
python/mscclpp/utils.py	Added stdin=subprocess.DEVNULL to prevent subprocess hanging
include/mscclpp/gpu.hpp	Added CUDA_ERROR_NOT_SUPPORTED macro mapping for HIP compatibility
src/gpu_utils.cc	Removed unused nvlsCompatibleMemHandleType variable

Comments suppressed due to low confidence (1)

src/registered_memory.cc:27

• The WARN macro call should use mscclpp::LogSubsys::P2P enum value instead of mscclpp::P2P. The logger.hpp defines LogSubsys enum with GPU, NET, CONN, EXEC, NCCL values. P2P is not defined in LogSubsys. Consider using GPU subsystem here since this is GPU IPC related code.

      WARN("Call to " #cmd " failed, error is %s", errStr); \

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 6 out of 6 changed files in this pull request and generated 5 comments.

[chhwang]

/azp run

[azure-pipelines]

Azure Pipelines successfully started running 3 pipeline(s).

[Binyang2014]

Are you going to update switch_channel.cc?

[chhwang]

/azp run mscclpp-ut

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).