Principal Software Development Engineer – Compiler & ML Acceleration

AMD AMD · Semiconductors · San Jose, CA · Engineering

Principal Software Development Engineer focused on compiler technology, MLIR-based infrastructure, and model-to-hardware optimization for accelerating LLMs and ML workloads on emerging accelerator architectures, specifically targeting high-performance inference.

What you'd actually do

  1. Lead architecture design and development of compiler components and optimization pipelines for machine learning
  2. Design and implement MLIR-based compiler flows to lower high-level ML representations into highly optimized hardware-specific code
  3. Drive model compilation and data movement optimization for ML inference workloads
  4. Define and implement compiler strategies for operator fusion, memory planning, scheduling, and performance optimization
  5. Work cross-functionally with hardware, runtime, frontend, and systems teams to align compiler capabilities with evolving accelerator architectures

Skills

Required

  • compiler development (front-end, middle-end, and/or back-end)
  • code optimization
  • MLIR and/or LLVM-based compiler infrastructure
  • neural network workloads
  • graph-level and compiler-level optimizations for ML models
  • C++ development
  • large, complex codebases
  • Principal-level technical scope
  • influencing architecture and direction
  • targeting or optimizing for NPUs or specialized AI accelerators
  • model compilation stacks
  • custom lowering pipelines
  • compiler or ML infrastructure in production environments

Nice to have

  • AMD is building the next generation of compiler and software infrastructure to accelerate Large Language Models (LLMs) and ML workloads on emerging accelerator architectures.
  • This role sits at the intersection of compiler development, ML frameworks, and AI model execution, supporting advanced acceleration pipelines targeting NPUs and other specialized compute engines.
  • Provide technical leadership and mentorship, influencing compiler direction and best practices across the organization
  • Contribute to long-term roadmap decisions for compiler and ML acceleration software

What the JD emphasized

  • compiler technology
  • MLIR-based infrastructure
  • model-to-hardware optimization
  • high-performance execution for AI workloads
  • ML inference workloads
  • compiler development
  • ML frameworks
  • AI model execution
  • accelerator architectures
  • compiler components
  • optimization pipelines
  • MLIR-based compiler flows
  • hardware-specific code
  • model compilation
  • data movement optimization
  • operator fusion
  • memory planning
  • scheduling
  • performance optimization
  • hardware
  • runtime
  • frontend
  • systems teams
  • compiler capabilities
  • accelerator architectures
  • compiler development
  • MLIR and/or LLVM-based compiler infrastructure
  • neural network workloads
  • graph-level and compiler-level optimizations for ML models
  • ML models
  • model compilation stacks
  • custom lowering pipelines
  • compiler or ML infrastructure in production environments

Other signals

  • accelerate Large Language Models (LLMs) and ML workloads on emerging accelerator architectures
  • enabling high-performance execution for AI workloads
  • compiler development, ML frameworks, and AI model execution
  • model compilation and data movement optimization for ML inference workloads