tenstorrent/tt-llk
Feb 2026 – Feb 2026
1 Month active
Languages Used
C++Python
Technical Skills
C++ developmentPython scriptingdata visualizationhardware integrationperformance analysis
Marko Vlahovic
PROFILE
Marko Vlahovic
Developed and delivered end-to-end hardware performance observability systems for the tenstorrent/tt-llk and tenstorrent/tt-metal repositories, focusing on C++ and Python. Built a C++ performance counter infrastructure with per-thread and later shared L1 buffer architectures, enabling detailed tracking and analysis of hardware metrics across UNPACK, MATH, and PACK threads. Enhanced data collection and reporting through Python scripting, providing automated summaries and derived metrics for performance analysis. Refactored the subsystem to reduce memory usage by 67% and improved synchronization using multithreading techniques. Simplified metrics output for clearer insights, supporting reproducible analysis, targeted optimization, and maintainable system architecture in CI environments.
Overall Statistics
Feature vs Bugs
100%Features
Repository Contributions
2Total
Bugs
0
Commits
2
Features
2
Lines of code
3,329
Activity Months2
Your Network
846 peopleWork History
March 2026
1 Commits • 1 Features
Mar 1, 2026March 2026 monthly work summary focusing on delivering a high-impact optimization in the performance counter subsystem of tt-metal, with an emphasis on memory efficiency, data integrity, and maintainability.
1 Commits • 1 Features
Mar 1, 2026March 2026 monthly work summary focusing on delivering a high-impact optimization in the performance counter subsystem of tt-metal, with an emphasis on memory efficiency, data integrity, and maintainability.
March 2026
February 2026
1 Commits • 1 Features
Feb 1, 2026February 2026: Delivered the Tensix Performance Counter System for tt-llk, establishing end-to-end hardware performance observability across UNPACK/MATH/PACK threads and enabling data-driven optimization. Core features include a C++ PerfCounters plumbing with per-thread L1 buffers, Python tooling for configuration and readout, and derived metrics with automated summaries. Added matmul kernel instrumentation and kernel-level integration to provide side-by-side REQUESTS vs GRANTS analysis, improving visibility into arbitration, stalls, and bottlenecks. This work lays the foundation for reproducible performance analysis, targeted tuning, and better capacity planning.
February 2026
1 Commits • 1 Features
Feb 1, 2026February 2026: Delivered the Tensix Performance Counter System for tt-llk, establishing end-to-end hardware performance observability across UNPACK/MATH/PACK threads and enabling data-driven optimization. Core features include a C++ PerfCounters plumbing with per-thread L1 buffers, Python tooling for configuration and readout, and derived metrics with automated summaries. Added matmul kernel instrumentation and kernel-level integration to provide side-by-side REQUESTS vs GRANTS analysis, improving visibility into arbitration, stalls, and bottlenecks. This work lays the foundation for reproducible performance analysis, targeted tuning, and better capacity planning.
Activity
Loading activity data...
Quality Metrics
Correctness100.0%
Maintainability80.0%
Architecture100.0%
Performance80.0%
AI Usage50.0%
Skills & Technologies
Programming Languages
C++Python
Technical Skills
C++ developmentPython scriptingdata analysisdata visualizationhardware integrationmultithreadingperformance analysisperformance optimizationsystem architecture
Repositories Contributed To
Overview of all repositories you've contributed to across your timeline
tenstorrent/tt-metal
Mar 2026 – Mar 2026
1 Month active
Languages Used
C++Python
Technical Skills
data analysismultithreadingperformance optimizationsystem architecture