English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

The efficiency of a spatial DNN accelerator depends heavily on the compilerand its cost model ability to generate optimized mappings for various operatorsof DNN models on to the accelerator's compute and memory resources. But,existing cost models lack a formal boundary over the operators for precise andtractable analysis, which poses adaptability challenges for new DNN operators.To address this challenge, we leverage the recently introduced MaestroData-Centric (MDC) notation. We develop a formal understanding of DNN operatorswhose mappings can be described in the MDC notation, because any mappingadhering to the notation is always analyzable by the MDC's cost model.Furthermore, we introduce a transformation for translating mappings into theMDC notation for exploring the mapping space.  Searching for the optimal mappings is challenging because of the large spaceof mappings, and this challenge gets exacerbated with new operators and diverseaccelerator configurations.To address this challenge, we propose a decoupledoff-chip/on-chip approach that decomposes the mapping space into off-chip andon-chip subspaces, and first optimizes the off-chip subspace followed by theon-chip subspace. The motivation for this decomposition is to reduce the sizeof the search space dramatically and also to prioritize the optimization ofoff-chip data movement, which is 2-3 orders of magnitude more compared to theon-chip data movement. We implemented our approach in a tool called {\emMarvel}, and another major benefit of our approach is that it is applicable toany DNN operator conformable with the MDC notation.

Marvel: A Data-centric Compiler for DNN Operators on Spatial Accelerators