Month: 2025-10 — Nanhu-V5 stability and data-path improvements focused on MissQueue and StoreQueue. No new user-facing features delivered this month; two critical bug fixes improved correctness and reliability of the core data pipeline. Fixes address race conditions in the MissQueue CMO validity flag handling and ensure proper allocation-before-merge sequencing for diff tests, with traceability to NHV5-1454. Business value: higher data integrity, reduced risk of incorrect cache states, and more reliable data flow for downstream components and verification tests.

September 2025: Delivered a focused set of correctness and stability improvements across Nanhu-V5 and LinkNan, centering on memory subsystem integrity, error handling, and replay logic. The work reduces data corruption risk, improves cache/memory reliability, and provides clearer commit traceability for maintenance and audits.

August 2025 monthly summary for OpenXiangShan-Nanhu/Nanhu-V5. Focused on memory subsystem reliability, timing improvements, and correct exception handling. Delivered pipeline-based buffering to memory paths that improved LSQ–DCache timing, and refined signaling flows for memory operations. Also addressed correctness in fault reporting and memory ordering to enhance reliability for MMIO, AMO, and related memory paths.

Month: 2025-07 — Nanhu-V5 memory subsystem robustness and performance improvements. Focused on memory pipeline correctness, stability, and efficiency with targeted fixes, data-path optimizations, LR/SC consistency, and validation coverage enhancements. Result is increased reliability and throughput for production workloads, with clearer failure signaling and reduced stall scenarios across the memory subsystem.

June 2025 focus for OpenXiangShan-Nanhu/Nanhu-V5 was strengthening memory subsystem correctness, cache coherence, and build reliability, translating design changes into measurable stability and developer productivity improvements. Delivered CMO support across the data path with new debug signals, fixed PMA/MMIO accuracy, and synchronized submodules to resolve downstream issues; and improved prefetch, load/store correctness and TLB handling.

May 2025 monthly highlights for OpenXiangShan-Nanhu/Nanhu-V5 focused on memory subsystem correctness and uncache state management. Key improvements include a comprehensive set of memory subsystem fixes addressing CBO, MMIO, and uncache behaviors, and a refactor of the Uncache module (LsqWrapper) to simplify state handling and prevent incorrect pending-state transitions. These changes improve correctness, exception reporting, MMIO handling, and maintainability, delivering business value through more reliable memory operations and safer code paths.

April 2025 performance summary for OpenXiangShan-Nanhu/Nanhu-V5. The month focused on strengthening IO robustness, improving memory subsystem efficiency, and enabling advanced cache operations to support higher-throughput workloads. Key work spanned MMIO path enhancements in the Load Unit, replay-path optimizations to reduce interference, and the introduction of Cache Block Operations (CBO) support. Critical correctness fixes in MMIO handling, atomic response validity, and PMA gating were completed to improve reliability and safety in production workloads. The combined efforts deliver measurable business value through lower latency for IO-heavy tasks, higher cache operation density, and improved fault visibility and permissions checks across the IO and memory subsystems.

March 2025 monthly summary focusing on reliability and correctness of MMIO handling in the Nanhu-V5 replay path. Implemented a targeted fix for MMIO address handling in the replay queue, introducing virtual-address storage for MMIO and performing physical-address lookups within the virtual queue. Changes span LoadQueueReplay and VirtualLoadQueue modules, reducing risk of incorrect memory translations during replay and improving simulator fidelity for hardware validation.

In December 2024, delivered stabilization and correctness fixes for the Nanhu-V5 memory subsystem focusing on UncacheBuffer and LoadQueueReplay. The work improved replay interactions, prevented repeated enqueues, corrected mask generation, and simplified the replay flow with more robust MMIO processing, enhancing overall reliability and maintainability of the hardware model.

Month 2024-11 — Nanhu-V5: Delivered memory-subsystem enhancements and reliability improvements across DCache and queue subsystems, with a focus on maintainability, correctness, and business value. Key features delivered include DCache wrapper refactor with MainPipe simplification, LoadQueue integration of rarqueue, and unified StoreQueue address handling. Major defects resolved in the DCache refill path and multiple rar-related issues to improve correctness under edge cases and higher load. Overall impact includes a cleaner, more maintainable design, reduced risk for future changes, and a stronger foundation for performance optimization.