Masters Thesis: Measurement-driven DSP Superoptimization
Core
Develop a measurement-driven superoptimizer for the Ericsson Many Core Architecture (EMCA) DSP to improve compiler-generated code performance and memory footprint.
Role type
Master's thesis researcher (compiler optimization)
Builds
A superoptimizer tool that integrates with the EMCA DSP compiler toolchain
Domain
Embedded systems, compiler development, digital signal processing (DSP)
Required skills
C/C++, low-level programming, compiler development, assembly-level optimization, LLVM development, formal methods (SMT solvers)
Preferred skills
Experience with VLIW instruction sets, knowledge of Souper or similar superoptimization tools
Technologies
LLVM, Z3, C/C++, VLIW, EMCA DSP
Responsibilities
Survey state-of-the-art in superoptimization; implement a compiler back-end as a dynamic cost model; expand support for native DSP instructions; evaluate optimizations on production DSP code
Seniority
Master's student, hands-on research
