The current ev constraint model and fixtures are toy examples. The tool must validate itself against actual production RISC-V cores to become a complete verification platform.
Goal
Establish ev as a self-sufficient verification tool by modeling, testing, and exhaustively verifying the custom extension interfaces of at least two top-tier RISC-V cores.
Candidates
- CVA6 (OpenHW Group) — XIF interface, documented, production-grade. Already name-checked in
sample.xif.yaml target field but never actually modeled.
- Ibex (lowRISC) — simpler ALU extension, well-documented, good for demonstrating the value of exhaustive over random.
Deliverables per core
- XIF / extension interface spec analyzed and translated into ev constraint YAML
- Realistic fixture files (e.g.,
rv32i_vector_mac.xif.yaml, simple_alu_ext.xif.yaml)
- ev check passes with deterministic, auditable results
- Any coverage gaps vs. existing random-simulation approaches documented
- Results shared with the corresponding open-source core community
This is self-contained
SSCCS and Nexus do not need to change for this. ev produces Facts as JSON output. When the verification results are valid and complete, Nexus consumes them naturally through the existing ingestion pipeline (Facts in object storage → sync worker → knowledge graph). No cross-project coordination required.
Stacking order
- Evaluate current constraint/projector gaps against real XIF specs
- Implement missing constraint/projector types
- Create CVA6 and/or Ibex fixture files
- Run, validate, document
The current ev constraint model and fixtures are toy examples. The tool must validate itself against actual production RISC-V cores to become a complete verification platform.
Goal
Establish ev as a self-sufficient verification tool by modeling, testing, and exhaustively verifying the custom extension interfaces of at least two top-tier RISC-V cores.
Candidates
sample.xif.yamltarget field but never actually modeled.Deliverables per core
rv32i_vector_mac.xif.yaml,simple_alu_ext.xif.yaml)This is self-contained
SSCCS and Nexus do not need to change for this. ev produces Facts as JSON output. When the verification results are valid and complete, Nexus consumes them naturally through the existing ingestion pipeline (Facts in object storage → sync worker → knowledge graph). No cross-project coordination required.
Stacking order