GPU Compute & MLIR Compiler Engineer – AI Workloads

Company Overview

Qualcomm India Private Limited seeks an accomplished engineer specializing in GPU compute, MLIR compiler development, and kernel optimization to enhance AI workload performance. This role centers on designing high-performance GPU compute kernels, advancing MLIR compiler backends, and collaborating cross-functionally to maximize GPU architecture capabilities.

Key Responsibilities

• Architect and fine-tune GPU compute kernels for OpenCL and Vulkan frameworks aimed at executing intensive AI and ML tasks.
• Design, develop, and enhance MLIR dialects across various abstraction layers including frontend, graph-level IR, tensor IR (like Linalg, Tensor, Tosa), and runtime/low-level dialects to support efficient comprehensive model compilation.
• Implement and sustain MLIR compiler passes such as tiling, fusion, bufferization, vectorization, and develop lowering pipelines targeting OpenCL and Vulkan GPU backends.
• Analyze compiled kernels through profiling tools and GPU counters to identify bottlenecks and drive compiler-based performance enhancements.
• Build and uphold GPU runtime components—including memory management, pipeline configuration, command buffer orchestration, and resource scheduling—for OpenCL and Vulkan platforms.
• Develop code generation pipelines enabling automatic transition from tensor IR via MLIR to optimized OpenCL and Vulkan GPU kernel code.
• Create performance-focused execution schedules involving tiling, loop fusion, parallelism, and caching within MLIR backend targeting OpenCL and Vulkan environments.
• Work closely with framework teams to optimize end-to-end model lowering for computer vision and large language model workloads utilizing MLIR compilation stacks.
• Design and implement reliable compiler and runtime elements using modern C/C++ with sophisticated programming approaches tailored for high-performance system development.

Required Qualifications

• Proven practical experience with the MLIR compiler framework, including custom dialect creation, compiler pass authoring, and constructing comprehensive lowering pipelines.
• Expertise across MLIR layers: frontend dialects for ML model ingestion (e.g., TOSA, StableHLO, ONNX-MLIR); graph-level transformations like operation fusion and shape inference; tensor IR manipulations with Linalg, Tensor, Vector dialects; runtime and low-level dialects including bufferization, MemRef, SCF, GPU, and LLVM for code generation.
• Hands-on skills in OpenCL programming encompassing kernel development, memory architecture, work-group and work-item tuning, and OpenCL runtime management.
• Solid grasp of Vulkan compute APIs covering descriptor management, pipeline creation, synchronization, and runtime internals.
• In-depth knowledge of GPU hardware architecture, memory hierarchies, and asynchronous compute mechanisms.
• Proficiency in C and C++ programming for system-level software development.
• Experience performing kernel profiling and bottleneck diagnostics on GPUs.
• Strong foundation in machine learning fundamentals relevant to computer vision and large language model tasks.

Preferred Skills

• Experience with IREE or related MLIR-based deployment frameworks such as TVM, XLA, or LLVM.
• Familiarity with IREE runtime and compiler architecture including hardware abstraction layers, executable compilation, and dispatch, especially regarding OpenCL/Vulkan backends.
• Contributions to MLIR or IREE open-source projects.
• Knowledge of quantization, mixed-precision inference, and model optimizations targeting edge and server GPUs.
• Exposure to compiling across multiple hardware targets (CPU, GPU, NPU) using MLIR toolchains.
• Experience with GPU profiling tools like ARM Streamline, Qualcomm Snapdragon Profiler, or Intel VTune for OpenCL/Vulkan environments.

Educational and Experience Requirements

• Bachelor's degree in Engineering, Computer Science, Information Systems, or a related field plus at least 4 years of relevant experience.
• Alternatively, a Master's degree with a minimum of 3 years of applicable work experience.
• Or a PhD with at least 2 years’ experience in systems engineering or a related discipline.

Diversity, Equality, and Inclusion Commitment

Qualcomm is committed to equal employment opportunities and offers accommodations for applicants with disabilities during the hiring process. Reasonable adjustments can be requested via the specified email or toll-free contact to support participation.

Additional Information for Applicants

Applicants must comply with all Qualcomm policies including confidentiality and proprietary data protection. Recruiting agencies are advised not to submit unsolicited applications or resumes as Qualcomm does not accept them.