发现心仪岗位

Role Overview

Develop and optimize compute kernels that form the foundation of deep-learning operations. Ensure performance consistency and predictable integration with compilers and frameworks.

Key Responsibilities

1. Implement high-performance kernels for convolution, GEMM, and attention.

2. Optimize data layout, tiling, and precision modes.

3. Provide reusable and well-documented primitives.

4. Partner with compiler engineers for seamless integration.

5. Build tools for profiling and benchmarking operators.

Qualifications

1. 3–7 years in performance or numerical computing.

2. Strong C/C++ and profiling experience.

3. Familiar with SIMD, GEMM, or deep-learning operators.

Preferred Attributes

1. Understanding of low-level instruction tuning (RVV, RVM, NEON, CUDA, etc.).

2. Experience with quantized or mixed-precision kernels.

3. Familiarity with MLPerf-style performance measurement.

4. Strong grasp of memory bandwidth, latency, and tiling optimization strategies.