AMD ROCm is an ecosystem of libraries and tools for developing applications that harness the processing power of AMD graphics cards. It is a series of libraries with primitives rocmCUB, rocBLAS, rocSparse, rocThrust. These are also special purpose libraries such as MIOpen which provides basic algorithms from the area of machine learning running on GPU. In addition to libraries, ROCm provides development tools, a C++ and Fortran language compiler.
The main programming tool is HIP. The HIP is a C++ programming interface and runtime layer for writing software to execute on AMD and NVIDIA GPUs. The HIP framework includes many components based on the LLVM infrastructure.
In Gdańsk, we are building a team of 7 people who will work on different aspects of the HIP platform. These include GPU C++ compiler, profiler, OpenMP, FORTRAN compiler. The team will be part of a larger international AMD team working on ROCm components.
Application profiling is one of the essential parts of creating high-performance applications running on the GPU. The more straightforward and more reliable the tools are, the more enjoyable the profiling process is. If you are a C++ engineer with experience in application profiling, knowledge about performance metrics, counters, CPU, and GPU architectures, then you are the person we are looking for. The right candidate will be working on a GPU profiler which is one of the toolchain components of the HIP platform.