tenstorrent/tt-mlir
Jul 2025 – Dec 2025
4 Months active
Languages Used
C++MLIRPythonTableGenMarkdown
Technical Skills
Compiler DevelopmentDomain-Specific Languages (DSLs)IR DesignTestingMLIROptimization
Mo Bagherbeik
PROFILE
Mo Bagherbeik
Worked on the tenstorrent/tt-mlir repository to deliver a series of compiler features focused on elementwise fusion, operation scheduling, and custom print formatting for MLIR-based tensor operations. Leveraged C++, MLIR, and Python to implement optimization passes such as elementwise fusion and loop fission, improving compute efficiency and resource utilization. Developed an OpScheduler to manage register pressure and enable aggressive fusion, while updating documentation and tests to ensure maintainability and correctness. Enhanced the readability and debugging of TTKernel operations by introducing a custom assembly format, and consistently aligned code, tests, and documentation to support scalable, high-performance compiler workflows.
Overall Statistics
Feature vs Bugs
100%Features
Repository Contributions
7Total
Bugs
0
Commits
7
Features
4
Lines of code
5,114
Activity Months4
Your Network
452 peopleWork History
December 2025
4 Commits • 1 Features
Dec 1, 2025December 2025 monthly summary for tenstorrent/tt-mlir: Delivered a major TTNN Elementwise/Op Fusion feature with scheduling improvements, expanded JIT fusion documentation, and expanded test coverage. Implemented an OpScheduler to optimize scheduling for D2M elementwise fusion, enabling more aggressive fusion while managing DST register usage; introduced scheduling and loop-nesting techniques to minimize register pressure and added tagging for fused/nested loops. Strengthened code quality and visibility via updated docs and tests, setting the foundation for higher-performance fusion paths and scalable support for fused eltwise operations.
4 Commits • 1 Features
Dec 1, 2025December 2025 monthly summary for tenstorrent/tt-mlir: Delivered a major TTNN Elementwise/Op Fusion feature with scheduling improvements, expanded JIT fusion documentation, and expanded test coverage. Implemented an OpScheduler to optimize scheduling for D2M elementwise fusion, enabling more aggressive fusion while managing DST register usage; introduced scheduling and loop-nesting techniques to minimize register pressure and added tagging for fused/nested loops. Strengthened code quality and visibility via updated docs and tests, setting the foundation for higher-performance fusion paths and scalable support for fused eltwise operations.
December 2025
November 2025
1 Commits • 1 Features
Nov 1, 2025Month: 2025-11 — Summary: Delivered Elementwise Fusion Efficiency Enhancement in tenstorrent/tt-mlir that rewired intermediate values to the consumer's output control buffer and reworked reserve semantics to avoid duplicate reserves, improving efficiency and correctness of the elementwise fusion path in MLIR for tensor operations. Implemented corrected CB routing and resource reuse to align with the computation graph; leveraged d2m.wait/d2m.reserve, linalg.generic, and bf16 tiled operations (tile_mul, tile_add). Note: TTMLIR custom_sharding_rule (custom_op_sdpa.mlir) test remains failing as of this commit, guiding next steps.
November 2025
1 Commits • 1 Features
Nov 1, 2025Month: 2025-11 — Summary: Delivered Elementwise Fusion Efficiency Enhancement in tenstorrent/tt-mlir that rewired intermediate values to the consumer's output control buffer and reworked reserve semantics to avoid duplicate reserves, improving efficiency and correctness of the elementwise fusion path in MLIR for tensor operations. Implemented corrected CB routing and resource reuse to align with the computation graph; leveraged d2m.wait/d2m.reserve, linalg.generic, and bf16 tiled operations (tile_mul, tile_add). Note: TTMLIR custom_sharding_rule (custom_op_sdpa.mlir) test remains failing as of this commit, guiding next steps.
October 2025
1 Commits • 1 Features
Oct 1, 2025Month: 2025-10 — Focused on performance optimization in tenstorrent/tt-mlir, delivering two MLIR optimization passes that materially improve compute-bound D2M.generic paths: elementwise fusion and loop fission. These changes reduce intermediate tensors, improve locality, and enhance load-compute-store patterns, contributing to faster inference/training workloads and better resource utilization. No major bug fixes were reported this month; ongoing stability improvements continue via optimization passes.
1 Commits • 1 Features
Oct 1, 2025Month: 2025-10 — Focused on performance optimization in tenstorrent/tt-mlir, delivering two MLIR optimization passes that materially improve compute-bound D2M.generic paths: elementwise fusion and loop fission. These changes reduce intermediate tensors, improve locality, and enhance load-compute-store patterns, contributing to faster inference/training workloads and better resource utilization. No major bug fixes were reported this month; ongoing stability improvements continue via optimization passes.
October 2025
July 2025
1 Commits • 1 Features
Jul 1, 2025Concise monthly summary for 2025-07 focused on delivering business value and technical excellence for the tt-mlir repository.
July 2025
1 Commits • 1 Features
Jul 1, 2025Concise monthly summary for 2025-07 focused on delivering business value and technical excellence for the tt-mlir repository.
Activity
Loading activity data...
Quality Metrics
Correctness88.6%
Maintainability88.6%
Architecture87.2%
Performance84.2%
AI Usage31.4%
Skills & Technologies
Programming Languages
C++MLIRMarkdownPythonTableGen
Technical Skills
C++ ProgrammingCompiler DesignCompiler DevelopmentDomain-Specific Languages (DSLs)IR DesignMLIRMLIR DevelopmentOptimizationOptimization TechniquesPass DevelopmentPythonTensor OperationsTestingcompiler optimizationdocumentation
Repositories Contributed To
Overview of all repositories you've contributed to across your timeline