Monomorphism-Based CGRA Mapping Via Space and Time Decoupling

2025-03-31

Design, Automation and Test in Europe

AI Summary

To address poor compile-time scalability in mapping applications onto large coarse-grained reconfigurable arrays (CGRAs), this paper proposes a spatiotemporal decoupled mapping methodology. First, it employs an SMT solver for formal, constraint-based scheduling along the temporal dimension; second, it leverages monomorphism analysis to achieve efficient spatial mapping. This is the first approach to fully decouple time and space mapping into independent, formally verifiable subproblems—thereby simultaneously ensuring correctness guarantees and high search efficiency. Experimental evaluation on a 20×20 CGRA demonstrates an average 105× speedup in compilation time over state-of-the-art methods, while preserving mapping quality—achieving comparable performance in terms of latency, resource utilization, and throughput. The proposed framework significantly enhances both the scalability and practical deployability of compiler toolchains for large-scale CGRAs.

Technology Category

• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems
• Constraint Satisfaction and Optimization: Distributed CSP/Optimization
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning

Application Category

• Search and Retrieval-Augmented AI: Efficiency and scalability of Web search engines
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Security and Privacy: Large-scale security measurements

Abstract

Coarse-Grain Reconfigurable Arrays (CGRAs) provide flexibility and energy efficiency in accelerating compute-intensive loops. Existing compilation techniques often struggle with scalability, unable to map code onto large CGRAs. To address this, we propose a novel approach to the mapping problem where the time and space dimensions are decoupled and explored separately. We leverage an SMT formulation to traverse the time dimension first, and then perform a monomorphism-based search to find a valid spatial solution. Experimental results show that our approach achieves the same mapping quality of state-of-the-art techniques while significantly reducing compilation time, with this reduction being particularly tangible when compiling for large CGRAs. We achieve approximately 105 × average compilation speedup for the benchmarks evaluated on a 20 × 20 CGRA.

Problem

• Decouples time and space dimensions for CGRA mapping
• Addresses scalability issues in compiling for large CGRAs
• Reduces compilation time while maintaining mapping quality

Innovation

• Decouples time and space mapping dimensions
• Uses SMT formulation for time dimension exploration
• Employs monomorphism-based search for spatial solution