# Research: High-level DaCe

###### tags: `cycle 1`

*Note: This is a research project. The goal is not production-level code, but exploration of the research topic. Eventually, this project (or follow-up projects) will result in shapable work.*

## Appetite

full cycle (Linus?/Felix? + Hannes?/Rico?)

## Problem

So far, work on DaCe focused on low-level optimizations (on DaCe internal nodes) where a lot of the domain knowledge is already lost. In this project we target a high-level representation of gtc.oir (eve) nodes and explore data flow optimizations.

## Goal

Understand how DaCe and Eve can interact on a high-level (domain specific) representation. Especially, transformation to DaCe **and back** should be investigated. The GTC IRs can be adapted as needed, e.g. to better represent data flow already in the GTC IR.

## Ideas

Concrete results could be
- gtc.oir -> DaCe -> fuse some HorizontalExecutions -> gtc.oir