AI/ML Compiler & Runtime Software Engineer
GlobalFoundries2 hours ago
Pune, IndiaStaff+
Responsibilities
- Architect, design, and develop AI/ML compiler and runtime software for RISC-V IP, NPUs, and SoC platforms.
- Develop IREE-based compiler flows covering MLIR lowering, code generation, runtime integration, and edge-AI deployment.
- Create custom MLIR dialects, compiler passes, lowering pipelines, pattern rewrites, and accelerator backend integrations.
- Enable AI framework import and lowering paths for PyTorch, ONNX, TensorFlow Lite, torch-mlir, TOSA, and Linalg.
- Optimize neural-network workloads using operator fusion, tiling, memory planning, quantization, layout transformation, vectorization, and accelerator-aware scheduling.
- Enable efficient execution across CPU, vector, matrix, NPU, DSP, and custom accelerator paths while balancing latency, throughput, memory, and power.
- Collaborate with architecture, hardware, firmware, FPGA, validation, product, and customer-facing teams on workload bring-up and deployment.
- Analyze compiler, runtime, generated-code, and hardware/software bottlenecks and drive graph-, operator-, and kernel-level optimizations.
- Define technical direction for AI SDK compiler pipelines, runtime interfaces, model deployment flows, and accelerator integration.
- Build correctness, performance, regression, benchmarking, validation, and CI infrastructure.
- Provide technical leadership, mentorship, debugging support, performance tuning, and software enablement across teams.
Requirements
- Three to twelve years of hands-on software engineering experience in compiler, runtime, embedded software, or AI/ML systems.
- Strong hands-on experience with IREE, LLVM, and MLIR compiler infrastructure.
- Experience developing MLIR dialects, compiler passes, lowering pipelines, pattern rewrites, code-generation flows, or custom-hardware backend integrations.
- Understanding of IREE code-generation flows, dispatch formation, executable generation, HAL/runtime concepts, and target-specific lowering.
- Experience with AI compiler and runtime stacks for edge AI or accelerator-backed inference.
- Experience with PyTorch, ONNX, TensorFlow Lite/TFLite, and related model conversion or import flows.
- Working knowledge of torch-mlir, TOSA, Linalg, tensor dialects, bufferization, quantization dialects, and MLIR model lowering.
- Strong understanding of neural-network execution and optimization, including quantization, fusion, tensor layouts, memory planning, tiling, vectorization, and kernel selection.
- Experience optimizing workloads for AI accelerators, NPUs, DSPs, vector processors, matrix engines, or custom SoC IP.
- Strong C/C++ programming skills and Python scripting ability for compiler tooling, testing, automation, and model workflows.
- Experience with Linux development, cross-compilation, debugging, profiling, build systems, and runtime bring-up.
- Ability to translate accelerator capabilities into compiler and runtime enablement with architecture and hardware teams.
- Proven ability to technically lead complex software modules, mentor engineers, and drive cross-functional execution.
- Preferred experience includes RISC-V, ARM, x86, DSP, GPU, RISC-V Vector, matrix acceleration, custom instructions, FPGA prototyping, Linux bring-up, board debugging, emulation, simulation, early silicon, or edge-AI deployment.
- Preferred familiarity with llama.cpp, GGML/GGUF, ONNX Runtime, TensorFlow Lite, TVM, XNNPACK, AI benchmarking, runtime systems, kernel libraries, microkernels, DMA, scratchpad memory, cache behavior, and accelerator data movement.
- Preferred familiarity with Jenkins, Git, CMake, Bazel, Jira, CI/CD, customer-facing enablement, silicon bring-up, platform software, and SDK delivery.
Benefits
- Benefits information is provided through GlobalFoundries' careers site.
- The role is located in Pune or Bangalore, India.
- Employment offers are subject to successful background checks, applicable medical screenings, and local laws.
- GlobalFoundries states that it is an equal opportunity and affirmative employer.