OpenXiangShan/GEM5
Oct 2025 – Jul 2026
8 Months active
Languages Used
C++Python
Technical Skills
Branch PredictionCPU ArchitecturePerformance OptimizationSystem SimulationC++ programmingdebugging
Cao Jiaming
PROFILE
Cao Jiaming
Worked on the OpenXiangShan/GEM5 repository to enhance CPU branch prediction accuracy and simulation performance through targeted improvements to MicroTAGE, TAGE, and BTBTAGE predictors. Leveraged C++ and Python to refactor prediction logic, streamline BTB entry handling, and introduce new statistics for better observability and diagnostics. Addressed stability by reverting configuration changes when necessary and fixing trace handling to ensure reliable simulation results. Focused on maintainability by cleaning up legacy code and consolidating prediction paths, enabling faster iteration and easier integration of new features. Emphasized performance optimization and system simulation, supporting architectural exploration and more accurate hardware modeling.
Overall Statistics
Feature vs Bugs
82%Features
Repository Contributions
21Total
Bugs
2
Commits
21
Features
9
Lines of code
4,329
Activity Months8
Your Network
48 peopleWork History
July 2026
1 Commits • 1 Features
Jul 1, 2026July 2026 OpenXiangShan/GEM5 monthly summary focusing on key architectural and performance enhancements. Delivered MicroTAGE branch predictor improvements with a teacher-based S3 update, enhanced observability through uBTB override attribution stats, and performance tuning by disabling MicroTAGE bank conflicts. Changes contribute to higher branch-prediction accuracy, lower misprediction-related stalls, and improved simulation throughput for hardware research.
1 Commits • 1 Features
Jul 1, 2026July 2026 OpenXiangShan/GEM5 monthly summary focusing on key architectural and performance enhancements. Delivered MicroTAGE branch predictor improvements with a teacher-based S3 update, enhanced observability through uBTB override attribution stats, and performance tuning by disabling MicroTAGE bank conflicts. Changes contribute to higher branch-prediction accuracy, lower misprediction-related stalls, and improved simulation throughput for hardware research.
July 2026
May 2026
1 Commits • 1 Features
May 1, 2026May 2026 monthly summary for OpenXiangShan/GEM5. Focused on stabilizing and enhancing CPU branch prediction performance in the O3 path by reverting TAGE table sizes and numWays to a proven configuration. Implemented via commit 0b41e02dba724e4757881b939d4864ce5a3a9abf (cpu-o3: tage back to 16k), with Change-Id I92133c1f0223041e4f7ae63f14d4c3cf1ba0faf3. Co-authored-by: Cao Jiaming <caojiaming@bosc.ac.cn>. No major bugs fixed this month. The change provides more predictable performance in CPU simulations and benchmarks, with minimal risk and clear traceability.
May 2026
1 Commits • 1 Features
May 1, 2026May 2026 monthly summary for OpenXiangShan/GEM5. Focused on stabilizing and enhancing CPU branch prediction performance in the O3 path by reverting TAGE table sizes and numWays to a proven configuration. Implemented via commit 0b41e02dba724e4757881b939d4864ce5a3a9abf (cpu-o3: tage back to 16k), with Change-Id I92133c1f0223041e4f7ae63f14d4c3cf1ba0faf3. Co-authored-by: Cao Jiaming <caojiaming@bosc.ac.cn>. No major bugs fixed this month. The change provides more predictable performance in CPU simulations and benchmarks, with minimal risk and clear traceability.
April 2026
1 Commits
Apr 1, 2026OpenXiangShan/GEM5 (April 2026) - Focused on stability and correctness of the BTBTAGE/MicroTAGE branch-predictor path. Implemented a set of fixes to ensure reliable trace handling and consistent latency assumptions. Specifically, disabled update-on-read in BTBTAGE, refactored trace handling for robustness, and reverted the MicroTAGE latency change to preserve 0-cycle latency. The patch improves simulator stability, reduces misprediction risk, and maintains consistent performance metrics across runs. Co-authored by Cao Jiaming.
1 Commits
Apr 1, 2026OpenXiangShan/GEM5 (April 2026) - Focused on stability and correctness of the BTBTAGE/MicroTAGE branch-predictor path. Implemented a set of fixes to ensure reliable trace handling and consistent latency assumptions. Specifically, disabled update-on-read in BTBTAGE, refactored trace handling for robustness, and reverted the MicroTAGE latency change to preserve 0-cycle latency. The patch improves simulator stability, reduces misprediction risk, and maintains consistent performance metrics across runs. Co-authored by Cao Jiaming.
April 2026
March 2026
4 Commits • 2 Features
Mar 1, 2026Month: 2026-03 - GEM5 contributions focused on branch predictor enhancements and stability improvements. Key work delivered includes refactoring the MicroTAGE branch predictor for performance and clarity, updating history management and index calculations to improve prediction accuracy, and introducing a per-table configuration pathway for BTBTAGE (with a subsequent rollback to a stable fixed-per-set configuration to preserve reliability). These changes were complemented by targeted bug fixes to ensure consistent behavior across predictor configurations.
March 2026
4 Commits • 2 Features
Mar 1, 2026Month: 2026-03 - GEM5 contributions focused on branch predictor enhancements and stability improvements. Key work delivered includes refactoring the MicroTAGE branch predictor for performance and clarity, updating history management and index calculations to improve prediction accuracy, and introducing a per-table configuration pathway for BTBTAGE (with a subsequent rollback to a stable fixed-per-set configuration to preserve reliability). These changes were complemented by targeted bug fixes to ensure consistent behavior across predictor configurations.
January 2026
2 Commits • 1 Features
Jan 1, 2026January 2026 (OpenXiangShan/GEM5) focused on streamlining the branch prediction and BTB (branch target buffer) pipeline to improve simulation accuracy and maintainability. Core efforts were centered around consolidating predwrongSource handling for wrong-source predictions, aligning BTB entry processing with stage-based predictions, and removing legacy basetable logic to simplify the code path. Key results include a refactored BTB/prediction path that consolidates source handling, the introduction of wrong-prediction statistics for improved diagnostic visibility, and a marked reduction in dead/unused code by removing basetable logic from BTBTAGE and MBTB and related components.
2 Commits • 1 Features
Jan 1, 2026January 2026 (OpenXiangShan/GEM5) focused on streamlining the branch prediction and BTB (branch target buffer) pipeline to improve simulation accuracy and maintainability. Core efforts were centered around consolidating predwrongSource handling for wrong-source predictions, aligning BTB entry processing with stage-based predictions, and removing legacy basetable logic to simplify the code path. Key results include a refactored BTB/prediction path that consolidates source handling, the introduction of wrong-prediction statistics for improved diagnostic visibility, and a marked reduction in dead/unused code by removing basetable logic from BTBTAGE and MBTB and related components.
January 2026
December 2025
9 Commits • 2 Features
Dec 1, 2025December 2025: Delivered substantial branch predictor enhancements in GEM5 (OpenXiangShan) with focus on accuracy, configurability, and observability across ITTAGE/TAGE and UBTB/BTB. Key outcomes include ABTB configurability, S3 prediction handling, misprediction counters, and expanded commit statistics across ITTAGE/BTB components, supported by cleanup efforts. MBTB basetable alignment and integration progressed with fixed-target statistics and support for a choice MBTB basetable, boosting accuracy and analytics. Backend refactors and bug fixes across ABTB/MBTB/UBTB improved coherence and reduced debt. Overall business impact: more reliable performance modeling, faster tuning, and better decision support for architecture exploration. Technologies demonstrated: C++, hardware simulation internals, performance analytics, and collaborative git workflows across multiple contributors.
December 2025
9 Commits • 2 Features
Dec 1, 2025December 2025: Delivered substantial branch predictor enhancements in GEM5 (OpenXiangShan) with focus on accuracy, configurability, and observability across ITTAGE/TAGE and UBTB/BTB. Key outcomes include ABTB configurability, S3 prediction handling, misprediction counters, and expanded commit statistics across ITTAGE/BTB components, supported by cleanup efforts. MBTB basetable alignment and integration progressed with fixed-target statistics and support for a choice MBTB basetable, boosting accuracy and analytics. Backend refactors and bug fixes across ABTB/MBTB/UBTB improved coherence and reduced debt. Overall business impact: more reliable performance modeling, faster tuning, and better decision support for architecture exploration. Technologies demonstrated: C++, hardware simulation internals, performance analytics, and collaborative git workflows across multiple contributors.
November 2025
2 Commits • 1 Features
Nov 1, 2025November 2025 — OpenXiangShan/GEM5: Delivered internal branch prediction (ABTB) debugging enhancements and targeted UBTB/TAGE improvements to strengthen debugging capabilities, reliability, and microarchitectural performance. Implemented a dedicated ABTB debug flag, cleaned up legacy code for maintainability, and improved misprediction handling within the TAGE path. Addressed UBTB-related bugs to improve accuracy and measurement fidelity, enabling faster diagnosis and iteration for future optimizations.
2 Commits • 1 Features
Nov 1, 2025November 2025 — OpenXiangShan/GEM5: Delivered internal branch prediction (ABTB) debugging enhancements and targeted UBTB/TAGE improvements to strengthen debugging capabilities, reliability, and microarchitectural performance. Implemented a dedicated ABTB debug flag, cleaned up legacy code for maintainability, and improved misprediction handling within the TAGE path. Addressed UBTB-related bugs to improve accuracy and measurement fidelity, enabling faster diagnosis and iteration for future optimizations.
November 2025
October 2025
1 Commits • 1 Features
Oct 1, 2025OpenXiangShan/GEM5 (2025-10) focused on a targeted performance enhancement by integrating MicroTAGE into the S1 stage of the CPU pipeline. This work adds a MicroTAGE branch predictor to S1, updates predictor configuration, and refactors code for maintainability and easier integration with existing branch-prediction logic. The effort delivers a concrete architectural advancement for dynamic branch prediction, ready for broader evaluation and future optimization. Delivered commit: 340af13e8310dbbcdf9797b1d77ead2159321d2b ("Microtage on s1 perf (#567)").
October 2025
1 Commits • 1 Features
Oct 1, 2025OpenXiangShan/GEM5 (2025-10) focused on a targeted performance enhancement by integrating MicroTAGE into the S1 stage of the CPU pipeline. This work adds a MicroTAGE branch predictor to S1, updates predictor configuration, and refactors code for maintainability and easier integration with existing branch-prediction logic. The effort delivers a concrete architectural advancement for dynamic branch prediction, ready for broader evaluation and future optimization. Delivered commit: 340af13e8310dbbcdf9797b1d77ead2159321d2b ("Microtage on s1 perf (#567)").
Activity
Loading activity data...
Quality Metrics
Correctness86.8%
Maintainability81.8%
Architecture82.8%
Performance81.8%
AI Usage32.4%
Skills & Technologies
Programming Languages
C++Python
Technical Skills
Branch PredictionC++C++ DevelopmentC++ developmentC++ programmingCPU ArchitectureComputer ArchitecturePerformance AnalysisPerformance OptimizationPythonSystem Simulationalgorithm designbranch prediction algorithmsdebuggingmicroarchitecture
Repositories Contributed To
Overview of all repositories you've contributed to across your timeline