Deep Learning Performance Architect, CUTLASS DSL

Description

Are you passionate about programming languages, compiler technology, and GPU performance? Do you want to help shape the future of high-performance kernel development for AI? We are looking for outstanding engineers to build   CUTLASS   DSL, a Python-native language for GPU kernel development, along with the MLIR dialects and lowering passes behind it. In this role, you will   also   help accelerate kernel compilation while delivering performance comparable to CUTLASS C++, enabling efficient hardware-software co-design for NVIDIA's next generation of AI platforms. What   you'll   be doing: - Design, develop, and   optimize   C UTLASS   DSL, a Python-native language for high-performance GPU kernel development - Build and advance the MLIR dialects, lowering passes, and code generation flows that power the   C UTLASS   DSL stack - Drive innovations that improve kernel compilation speed while   maintaining   performance on par with CUTLASS C++ - Collaborate closely with architecture, research, software product teams, and the open-source community to bring   cutting-edge   optimizations into real products What we need to see: - MS, PhD, or equivalent experience in Computer Science, Software Engineering, or a related field - 2+ years of   relevant work experience - Excellent programming skills in Python and strong   proficiency   in C++ - Hands-on experience with DSLs, compilers, or code generation systems - Strong command of the MLIR/LLVM stack, including IR design and pass optimization - Strong communication   skills and the ability to thrive in a highly collaborative environment Ways to stand out from the crowd: - Deep understanding of the CUDA GPU programming model, GPU microarchitecture, and performance analysis and optimization techniques - Familiarity with key high-performance computing abstractions such as Layout, Tile, MMA, and TMA in the   CuTe   ecosystem