Modular SAIL: dream or reality?

Abstract

This paper addresses the critical challenge of compositionality in the RISC-V development flow, which is necessary to fully leverage the modular nature of the ISA specification. The authors introduce "modular SAIL," an experiment injecting compositionality directly into the SAIL-RISCV golden model to support modularity across emulation, simulation, and verification. They demonstrate the feasibility of adapting the SAIL-RISCV flow to create pluggable emulators using both static and dynamic binding techniques. Comparative study confirms that the resulting modular emulator maintains exact functional equivalence with the original monolithic RISC-V Instruction Set Simulator (ISS).

Report

Key Highlights

• The core goal is to inject compositionality into the RISC-V development ecosystem to maximize the benefits of the ISA's inherent modularity.
• The innovation, "modular SAIL," targets the compositionality issue by extending module support beyond mere specification to practical development stages (emulation, simulation, verification).
• The work focuses specifically on adapting the existing SAIL-RISCV golden model flow.
• The resulting architecture successfully yielded a pluggable emulator (aka RISC-V ISS) that uses modular components.
• Testing confirmed that the modular emulator, regardless of whether static or dynamic binding was used, exhibits identical functional behavior compared to the original, monolithic emulator.

Technical Details

• Target System: SAIL-RISCV golden model (the reference ISA specification tool).
• Technical Goal: Adapt the SAIL-RISCV flow, and potentially the SAIL compiler itself, to natively support modular components at the emulator level.
• Methodology: The implementation explores injecting compositionality mechanisms directly into the golden model derivation process.
• Binding Techniques: The pluggable emulator's performance and functional parity were studied using both static binding and dynamic binding mechanisms.
• Flow Coverage: The focus is on implementing compositionality across the crucial stages of the development flow, specifically mentioning emulation, simulation, and verification.

Implications

• Enabling RISC-V Modularity: This work directly tackles a significant tooling challenge preventing the RISC-V community from easily combining and verifying various standard and custom ISA extensions, thus accelerating innovation and customization.
• Improved Golden Model Efficiency: By transforming the monolithic golden model into a modular, pluggable system, development, maintenance, and verification of new ISA extensions become significantly easier and less error-prone.
• Tool Chain Advancement: Proving that modularity can be successfully injected at the emulator level using standard binding techniques (static/dynamic) provides critical feedback and a potential blueprint for future advancements in the SAIL compiler and related RISC-V tools.
• Verification Confidence: Establishing functional equivalence between the modular and monolithic ISS implementations ensures that the shift to compositionality does not introduce verification gaps or functional inconsistencies.