Masters Thesis: Measurement-driven DSP Superoptimization

Teknik, data och digitalt · Mjukvaru- och webbutveckling · Mjukvaruutveckling · Inbyggda system

I korthet

This Master's thesis project focuses on measurement-driven DSP superoptimization for Ericsson's Many Core Architecture (EMCA) compiler. You will survey state-of-the-art techniques, develop a compiler back-end cost model for enhanced optimization, expand DSP instruction support, and evaluate performance improvements on production code.

Ansvarsområden

  • Survey the state of the art in superoptimization.
  • Use the compiler back-end as the cost model, compiling candidates for the target DSP and measuring generated code properties.
  • Expand the set of DSP instructions the superoptimizer can reason about, enabling discovery of efficient DSP-specific instruction uses.
  • Evaluate offline on production DSP code, considering performance improvements and memory footprint reductions.

Krav

  • Master’s student in Computer Science or Computer Engineering.
  • Experience in C/C++.
  • Experience working with low-level programming, compilers, and assembly-level optimization.
  • Willingness to develop skills in LLVM development and formal methods.

Önskade kvalifikationer

  • Experience in LLVM development.
  • Skills in formal methods.

Förmåner

  • Outstanding opportunity to use your skills and imagination to push boundaries.
  • Build solutions never seen before.
  • Be challenged but not alone.
  • Join a team of diverse innovators.
  • Work directly on a production compiler and DSP program performance.
#Compiler#DSP#Superoptimization#VLIW#LLVM#SMT Solver#Real-time#Performance Optimization#Memory Footprint#ASIC Architecture#Master Thesis
Ericsson Logo

Företag

Ericsson

Publicerade jobb

för 3 dagar sedan

Anställningstyp

Praktik

Arbetsform

På plats

Erfarenhetsnivå

Student

Platser

Linköping, Sweden

Kvalifikation

Masterexamen

Sökande

Ansök tidigt