Abhishek Varma
Member Of Technical Staff at AMD
Kolkata, West Bengal, India
Summary
🤩
Rockstar🎓
Top SchoolAbhishek Varma is a compiler engineer with nine years of experience building ML and AI-oriented compiler infrastructure, currently a Member of Technical Staff at AMD. He specializes in MLIR/LLVM toolchains, contributing notable patches to llvm-project, torch-mlir, TensorFlow, IREE and Hugging Face Diffusers to improve lowering, canonicalization, and codegen for real-world ML workloads. At nod.ai (acquired by AMD) and PolyMage Labs he drove Linalg lowering, tiling/decomposition and bufferization improvements that moved models closer to efficient GPU execution. His work blends deep compiler internals with practical ML deployment needs—evident in contributions to high-profile repos like llvm/llvm-project and iree-org/iree. He holds an MTech in Computer Science from IIT Bombay and brings a persistent interest in game development that informs his performance-oriented, systems-level thinking. Colleagues describe him as the kind of engineer who finds elegant transformations in IR that measurably reduce runtime costs.
9 years of coding experience
4 years of employment as a software developer
Indian Institute of Technology BombayBachelor of Technology (B.Tech.) Computer Science & Engineering, Bachelor of Technology (B.Tech.) Computer Science & Engineering at Heritage Institute of Technology, Kolkata
Stackoverflow
1reputation
221reached
0answers
2questions
Github Skills (18)
c-language10
llvm10
machine-learning10
mlr10
compiler-design10
compiler-compiler10
tensorflow10
compiler10
cprogramming-language10
lg9
cuda9
vulkan8
tensor8
pytorch8
mlops7
Programming languages (4)
Github contributions (5)
llvm/llvm-project
Aug 2023 - Mar 2025
The LLVM Project is a collection of modular and reusable compiler and toolchain technologies.
Role in this project:
Back-end Developer
Back-end DeveloperContributions:49 reviews, 15 PRs, 5 pushes in 1 year 7 months
Contributions summary:Abhishek's contributions primarily focus on enhancing the MLIR (Multi-Level Intermediate Representation) compiler infrastructure within the LLVM project. They implemented and refined canonicalization patterns, specifically for the SCF (Structured Control Flow) dialect, including folding iterator arguments in `scf.forall` operations and adding APIs to fuse consumers and producers within SCF loops. Furthermore, the user addressed data layout propagation issues related to `tensor.unpack` operations in the Linalg dialect. They also worked on improvements to the Affine dialect's folding mechanism, ensuring that constant attributes are used where applicable.
compilerstechnologiesclangsubmittoolchain
iree-org/iree
Mar 2023 - Feb 2025
A retargetable MLIR-based machine learning compiler and runtime toolkit.
Role in this project:ML Engineer
ML EngineerContributions:15 reviews, 11 PRs, 1 push in 1 year 11 months
Contributions summary:Abhishek primarily contributed to the IREE compiler project, focusing on machine learning related tasks. Their work involved implementing and adapting operations within the IREE compiler, specifically targeting the `TM_TENSOR` dialect by adding and lowering a `sort` operation and making improvements to the `GPUReduceBankConflictsPass` to avoid padding of zero rank memref.alloc ops, and cleaning up bufferization pass for LinalgExt ops. They also removed the `LinalgExt::Softmax` and used upstream `linalg::softmax`. Furthermore, the user adapted the attention and winograd ops for tiling, implementing tiling and decomposition.
mlirspirvvulkantensorflowcompiler
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.