跳转至

Homepage

[XiangShan Biweekly 114] 20261028

Welcome to the XiangShan biweekly column! Through this column, we will regularly share the latest development progress of XiangShan. This is the 114th issue of the biweekly report.

RISC-V Summit China 2026 was successfully held from October 18 to 20 in Shenzhen. Researcher Bao Yungang, Deputy Director of the Institute of Computing Technology, Chinese Academy of Sciences, and Chief Scientist of the Beijing Institute of Open Source Chip, delivered a talk at the summit titled "RISC-V: A 12-Year Journey of Understanding and a 10-Year Outlook on Trends (2015-2035)" (translated from the Chinese title). He reviewed the development of RISC-V, discussed industry trends through 2035, and shared his perspectives on the evolution of open-source chip technology and the industry.

Bao Yungang presenting at the XiangShan Ruyi Ecosystem Conference

During the summit, the XiangShan team hosted the XiangShan Processor Session, featuring presentations on the evolution of XiangShan's high-performance architecture and industrial-grade validation practices. All presentation slides from the session are now publicly available in Talks-and-Publications.

Team photo

Recent XiangShan development includes frontend fixes for address sign extension, ICache ECC checks, and exception propagation; backend support for mnret and dret in the new decoder and an increase in the default ROB capacity to 160 entries; banked CoupledL2 SRAMs and a custom L1–L2 interconnect, alongside fixes for StoreQueue forwarding and fault-address reporting; and an independent asynchronous CUTE clock domain, improved multi-channel memory access scheduling and matrix register writes, and enhanced matrix verification and debugging tools in XSAI. Infrastructure updates strengthen FPGA DiffTest parallel checking and trace recording/replay, improve Kunminghu V2 reference-model integration, fix vector-store exception handling, and optimize store checking.

【香山双周报 114】20261028 期

欢迎来到香山双周报专栏,我们将通过这一专栏定期介绍香山的开发进展。本次是第 114 期双周报。

2026 RISC-V 中国峰会已于 10 月 18 日至 20 日在深圳成功举办。中国科学院计算技术研究所副所长、北京开源芯片研究院首席科学家包云岗研究员在大会上作了题为《RISC-V:十二年认知之路与十年趋势展望(2015-2035)》的报告,回顾 RISC-V 的发展历程,展望至 2035 年的产业趋势,并分享了对开源芯片技术与产业演进的见解。

包云岗研究员在香山如意生态大会上作报告

峰会期间,香山团队举办了香山处理器专场,介绍了香山处理器的高性能架构演进与工业级验证实践。该专场的演讲 PPT 现已全部公开,欢迎访问 Talks-and-Publications。

团队合影

关于香山近期开发进展,前端修复地址符号扩展、ICache ECC 检查及异常传递问题;后端补齐新译码单元对 mnret、dret 指令的支持,并将默认 ROB 容量扩大至 160 项;访存与缓存推进 CoupledL2 分 bank 设计和专用 L1–L2 互连,修复 StoreQueue 转发与异常地址报告问题;XSAI 支持 CUTE 独立异步时钟域,优化多通道访存调度与矩阵寄存器写入,并完善矩阵验证与调试工具;基础设施增强 FPGA DiffTest 并行检查和 trace 录制回放,完善昆明湖 V2 参考模型集成,修复向量存储异常处理问题并优化存储校验。

[XiangShan Biweekly 113] 20261014

Welcome to the XiangShan biweekly column! Through this column, we will regularly share the latest development progress of XiangShan. This is the 113th issue of the biweekly report.

Regarding the recent development progress of XiangShan, the frontend added VictimBTB to reuse evicted MainBTB entries for branch prediction; the backend implemented a hybrid compressed ROB (CROB), optimized early wakeup and bypass for FMA, and fixed issues in floating-point conversions and instruction legality checks; the memory and cache subsystem added softTLB/softPTW simulation paths supporting two-stage address translation and fixed issues with TLB global attributes and rollback on memory access violations; XSAI integrated a matrix-guided L2 prefetcher and a standalone transpose load engine, while improving matrix debugging and hash comparison; infrastructure updates enhanced FPGA DiffTest, NEMU, Palladium, and AI-assisted debugging tools and fixed GSIM code generation issues; XS-GEM5 advanced model alignment for L2 prefetching and SBuffer behavior, fixed out-of-bounds register access during vector loads, and improved MMIO support, performance regression testing, and build checks.

During RISC-V Summit China 2026, the XiangShan Processor Session will be held on October 18. The session will feature presentations on the evolution of XiangShan's high-performance architecture and industrial-grade validation practices, the application of the UCAgent intelligent agent to XiangShan processors, and a hybrid RISC-V security verification platform for XiangShan processors.

Following the processor session, the XiangShan Developer Conference and XiangShan Ecosystem Conference will both be held on October 20. The XiangShan Developer Conference (09:00–12:30) will focus on technical co-development and sharing, featuring a review of the community’s annual development achievements, recognition of outstanding contributors, an SMT enterprise technology seminar, and the first presentation of the full-stack architectural evolution of the third-generation Kunminghu processor core IP. The XiangShan Ecosystem Conference (13:30–17:30) will focus on collaborative ecosystem building, with presentations on XiangShan and Ruyi technologies, in-depth discussions of foundational RISC-V + AI technologies, AI-themed talks, the launch of an open-source chip community, and a ceremony marking the launch of the Ruyi community co-construction initiative. Ecosystem partners and developers are welcome to follow the event updates and join us in person.

【香山双周报 113】20261014 期

欢迎来到香山双周报专栏,我们将通过这一专栏定期介绍香山的开发进展。本次是第 113 期双周报。

关于香山近期开发进展,前端新增 VictimBTB,复用 MainBTB 淘汰表项参与分支预测;后端实现混合压缩 ROB(CROB),优化 FMA 提前唤醒与旁路,并修复浮点转换和指令合法性检查等问题;访存与缓存新增支持两阶段地址翻译的 softTLB/softPTW 仿真路径,并修复 TLB 全局属性与访存违规回滚相关问题;XSAI 集成矩阵引导的 L2 预取器和独立转置加载引擎,并完善矩阵调试与 hash 对比;基础设施增强 FPGA DiffTest、NEMU、Palladium 和 AI 辅助调试工具,修复 GSIM 代码生成问题;XS-GEM5 推进 L2 预取与 SBuffer 行为对齐,修复向量 load 寄存器访问越界问题,并完善 MMIO 支持、性能回归和构建检查。

2026 RISC-V 中国峰会期间,香山处理器专场将于 10 月 18 日举行,会议将围绕香山处理器的高性能架构演进与工业级验证实践、UCAgent 智能体在香山处理器中的实践,以及面向香山处理器的混合式 RISC-V 安全验证平台展开报告。

继处理器专场之后,香山开发者大会和香山生态大会将于 10 月 20 日同日举行。香山开发者大会(09:00–12:30)将围绕技术共建与分享,开展社区年度建设成果复盘、杰出贡献者表彰和 SMT 企业技术专题研讨,并首发第三代昆明湖处理器核 IP 全栈架构演进成果;香山生态大会(13:30–17:30)将聚焦生态协同共建,带来香山与如意技术成果分享、RISC-V+AI 根技术深度研讨和 AI 专题报告,并同步举行开源芯片社区成立发布及如意社区共建点亮仪式。欢迎生态伙伴和开发者关注活动安排并参加现场交流。

[XiangShan Biweekly 112] 20260930

Welcome to XiangShan biweekly column! Through this column, we will regularly share the latest development progress of XiangShan. This is the 112th issue of the biweekly report.

Regarding the recent development progress of XiangShan, the frontend fixed issues in CommonHR history and instruction-fetch address checks and optimized uBTB timing; the backend added a unified vector floating-point functional unit and scalar floating-point source operands, while fixing issues related to vector state, CBO, and DiffTest; the memory and cache subsystem integrated ZhuJiang NoC, fixed multiple vector-memory, StoreQueue, exception-handling, and coherence issues, and optimized memory-dependence tracking and timing; XSAI improved matrix debugging support across CUTE, NEMU, and DiffTest; infrastructure updates enhanced FPGA DiffTest, the NEMU reference model, test workloads, sampling and checkpointing, and GSIM; XS-GEM5 continued model alignment and exploration for vector memory access, prefetching, branch prediction, and execution queues.

RISC-V Summit China 2026 will take place from October 18 to 20, 2026, at the Shenzhen Convention and Exhibition Center (Futian). The XiangShan team will present the latest microarchitecture developments and host a XiangShan tutorial at the summit. We look forward to seeing you in Shenzhen!

【香山双周报 112】20260930 期

欢迎来到香山双周报专栏,我们将通过这一专栏定期介绍香山的开发进展。本次是第 112 期双周报。

关于香山近期开发进展,前端修复了 CommonHR 历史、取指地址检查等问题并优化了 uBTB 时序;后端新增统一的向量浮点功能单元和标量浮点源操作数支持,并修复了多项向量状态、CBO 和 DiffTest 相关问题;访存与缓存集成了 ZhuJiang NoC,修复了多项向量访存、StoreQueue、异常处理和一致性相关问题,并优化了访存依赖跟踪及时序;XSAI 完善了 CUTE、NEMU 和 DiffTest 的矩阵调试能力;基础设施增强了 FPGA DiffTest、NEMU 参考模型、测试负载、采样切片和 GSIM;XS-GEM5 持续推进向量访存、预取、分支预测和执行队列等功能的模型对齐与探索。

同时向大家预告一则消息:2026 RISC-V 中国峰会将于 2026 年 10 月 18 日至 20 日在深圳会展中心(福田)举行。香山团队将在峰会上介绍最新的微架构进展,并举办香山 tutorial,期待与大家在深圳相见!

[XiangShan Biweekly 111] 20260916

Welcome to the XiangShan biweekly column! Through this column, we will regularly share the latest development progress of XiangShan. This is the 111th issue of the biweekly report.

Regarding the recent development progress of XiangShan, the frontend added BPU S2 override, increased its runahead distance, and fixed an ICache replacement-policy bug; the backend expanded support for vector widening, narrowing, multiply-add, and division, and fixed decoding, bypass, and floating-point bugs; in memory access and caching, F-POP and teacher-student BOP prefetchers were integrated, while LoadQueueReplay gained early wakeup and pipeline optimizations; XSAI added matrix prefetch descriptors to CUTE memory requests, fixed a prefetch livelock, and improved AME DiffTest checks; infrastructure updates improved FPGA DiffTest workflows, optimized the NEMU reference model, expanded test workloads, and fixed GSIM simulation bugs; XS-GEM5 continued aligning Sbuffer, vector writeback, and BPU behavior, and optimized SMT pipeline policies and LSQ contention logic.

【香山双周报 111】20260916 期

欢迎来到香山双周报专栏,我们将通过这一专栏定期介绍香山的开发进展。本次是第 111 期双周报。

关于香山近期开发进展,前端新增 BPU S2 override 并扩大超前距离,修复了 ICache 替换策略相关 bug;后端完善了向量宽化、窄化、乘加及除法支持,并修复了多项解码、旁路和浮点处理相关 bug;访存与缓存方面,集成了 F-POP 和师生结构 BOP 预取器,新增 LoadQueueReplay 提前唤醒并优化其流水线;XSAI 支持在 CUTE 访存请求中传递矩阵预取描述信息,修复了预取活锁,并完善了 AME DiffTest 检查;基础设施完善了 FPGA DiffTest 流程,优化 NEMU 参考模型,扩展测试负载并修复 GSIM 仿真问题;XS-GEM5 持续对齐 Sbuffer、向量写回和 BPU 等行为,并优化 SMT 流水线与 LSQ 竞争逻辑。

[XiangShan Biweekly 110] 20260902

Welcome to XiangShan biweekly column! Through this column, we will regularly share the latest development progress of XiangShan. This is the 110th issue of the biweekly report.

Regarding the recent development progress of XiangShan, the frontend has fixed several branch-prediction and instruction-fetch issues and optimized uBTB training and IFU/ICache timing; the backend has advanced V3 register-file refactoring; the memory and cache subsystem has fixed several issues while continuing V3 development; XSAI has fixed matrix decoding, memory-access, and operand-wakeup issues and improved DiffTest and CI; in terms of infrastructure, FPGA DiffTest, NEMU/QEMU checkpointing, and GSIM build and simulation capabilities have been enhanced, while XS-GEM5 model calibration and performance exploration have continued.

Special thanks to community contributor muke101 for providing an open-source gem5 implementation of the PHAST memory-dependence prediction algorithm through a GitHub issue. XS-GEM5 further refactored and optimized this implementation.

We are also pleased to announce that the XiangShan team and Professor Trevor E. Carlson of the National University of Singapore will jointly host a workshop entitled Agile Development and Verification for Chips during MICRO 2026 in Greece. Scheduled for October 31, the workshop is now accepting submissions from the community. Please visit the submission website for details and to submit your work. We look forward to meeting you in Greece for an engaging exchange of ideas!

【香山双周报 110】20260902 期

欢迎来到香山双周报专栏,我们将通过这一专栏定期介绍香山的开发进展。本次是第 110 期双周报。

关于香山近期开发进展,前端修复了多个分支预测与取指问题,并优化了 uBTB 训练及 IFU/ICache 时序;后端推进了 V3 寄存器堆相关重构;访存与缓存修复了多个问题,并继续推进 V3 开发;XSAI 修复了矩阵译码、访存和操作数唤醒问题,并完善了 DiffTest 与 CI;基础设施方面,增强了 FPGA DiffTest、NEMU/QEMU checkpoint 以及 GSIM 构建和仿真能力,并持续推进 XS-GEM5 模型对齐和性能探索。

社区贡献者 muke101 通过 GitHub Issue 向我们提供了 PHAST 内存依赖预测算法的gem5 开源实现,在此特别致谢。XS-GEM5 在该实现基础上进行了进一步的重构与优化。

同时向大家预告一则消息:香山团队将联合新加坡国立大学的 Trevor E. Carlson 教授,在希腊举行的 MICRO 2026 会议期间举办题为 Agile Development and Verification for Chips 的专题研讨会(workshop)。研讨会定于 10 月 31 日举行,现面向社区公开征稿。欢迎访问投稿网站(https://workshop.xiangshan.cc)了解详情并投稿,期待届时与大家相聚希腊,共同交流与探讨!