Roadmap: Bytecode Equivalence Testing & New IR Design

[gakonst]

Summary

This issue outlines a roadmap for Solar to achieve bytecode generation parity with solc and eventually produce optimal EVM bytecode through a new IR design. This builds on and references the existing Roadmap #1.

---

Implementation Status

MIR Foundation (✅ Complete)

The MIR (Mid-level IR) structure has been implemented with:

• SSA-based design with phi nodes
• EVM-aware type system (UInt, Address, MemPtr, StoragePtr, etc.)
• Basic block structure with terminators
• FunctionBuilder API for construction
• HIR → MIR lowering (basic expressions, some builtins)
• MIR → EVM codegen skeleton (dispatcher, basic instructions)

Active Implementation Issues

Issue	Title	Priority	Status
#694	Liveness Analysis	High	🔴 Not started
#695	Phi Elimination Pass	High	🔴 Not started
#696	Stack Model & Basic Scheduling (≤16 values)	Critical	🔴 Not started
#697	Full Stack Scheduling with Spilling	Critical	🔴 Blocked on #696
#698	Assembler Label & Jump Resolution	High	🔴 Not started
#699	Complete HIR→MIR for Complex Types	High	🔴 Not started
#700	Dead Code Elimination (DCE)	Medium	🔴 Not started
#701	Constant Folding & Propagation	Medium	🔴 Not started
#702	Sparse Conditional Constant Propagation (SCCP)	Medium	🔴 Not started
#703	Common Subexpression Elimination (CSE)	Medium	🔴 Not started
#704	Bytecode Equivalence Testing Infrastructure	High	🔴 Not started

Dependency Graph

                    ┌─────────────┐
                    │ MIR (done)  │
                    └──────┬──────┘
                           │
              ┌────────────┼────────────┐
              ▼            ▼            ▼
         ┌────────┐   ┌────────┐   ┌────────┐
         │ #694   │   │ #698   │   │ #699   │
         │Liveness│   │Assembler│  │Complex │
         └───┬────┘   └────────┘   │ Types  │
             │                      └────────┘
             ▼
         ┌────────┐
         │ #695   │
         │Phi Elim│
         └───┬────┘
             │
             ▼
         ┌────────┐
         │ #696   │◄──── CRITICAL PATH
         │Basic   │
         │Stack   │
         └───┬────┘
             │
             ▼
         ┌────────┐
         │ #697   │
         │Full    │
         │Stack + │
         │Spilling│
         └───┬────┘
             │
             ▼
    ┌────────────────────┐
    │ Working Bytecode!  │
    └────────────────────┘
             │
    ┌────────┴────────┐
    ▼                 ▼
┌────────┐       ┌────────┐
│ #700   │       │ #704   │
│ DCE    │       │Equiv   │
└────────┘       │Testing │
                 └────────┘

---

Phase 0: Complete Frontend (Current Focus)

Before codegen, finish the semantic analysis layer. Many of these are in progress or have PRs open.

Type Checking (#615)

• Implicit conversions (Tracking: Implicit conversions #617) - PR feat(sema): complete implicit type conversions #678
• Explicit conversions (Tracking: Explicit conversions #609) - PR feat(sema): complete explicit type conversions #677
• Call type checking (Tracking: Call type checking #661) - PR feat(sema): complete call type checking #684
• Variable declaration rules (Variable declaration rules #657) - PR feat(sema): implement variable declaration rules #681
• OverrideChecker (Implement OverrideChecker #301) - PR feat(sema): implement override checker #685
• ContractLevelChecker (Implement ContractLevelChecker #303) - PR feat(sema): complete contract level checks #679
• View/pure checker - PR feat(sema): implement view/pure checker #680
• Static analysis + lint infrastructure - PR feat(sema): implement static analyzer warnings #683
• ICE bug fixes ([ICE]: identifier this already declared #216, [ICE]: Type error in error statement. #219, Incorrect Handling of pure Function #221, Incorrect Handling of ** #222) - PR fix(sema): fix typechecker ICE bugs #216 #219 #221 #222 #682
• Enable -Ztypeck for solc tests (Enable -Ztypeck for some solc tests #663) - PR feat(sema): enable -Ztypeck for solc test suite #686
• Parameter/argument validation helpers (Helpers to validate parameters and arguments #662)
• Functions implicit conversions (Functions implicit conversions #620)

HIR Improvements (#91)

• Lower Yul to HIR (Lower Yul to HIR #415) - P-high, E-hard
• HIR pretty-printer (Implement a simple HIR pretty-printer #86) - ✅ Done
• HIR builder infrastructure

LSP Support (#394)

Blocked on typeck completion:

• Basic LSP lifecycle (Basic LSP lifecycle #417)
• Symbol tables (LSP symbol tables #418)
• Go to definition (LSP go to definition #420)
• Autocomplete (LSP autocomplete #419)
• Inlay hints (LSP inlay hints #421)
• Flychecks (Add LSP flychecks #416)

Other Frontend

• AST pretty-printer (Implement a simple AST pretty-printer #27)
• Document divergence from solc (Explicitly document divergence from solc #547) - ✅ Done
• ABI: include referenced events (ABI output should include referenced events, not just defined #305)
• WebAssembly/WASI support (WebAssembly/WASI support #211)

---

Phase 1: Closed-Loop Bytecode Equivalence Testing

Tracking issue: #704

Build a test harness that verifies Solar-generated bytecode is behaviorally equivalent to solc-generated bytecode.

Approach

• Use Foundry property tests (fuzz testing) to compare runtime behavior
• For each Solidity contract:
  1. Compile with solc → bytecode A
  2. Compile with Solar → bytecode B
  3. Deploy both to Anvil
  4. Run identical transactions/calls against both
  5. Assert: same return values, same state changes, same reverts, same gas (within tolerance)

Deliverables

• Test harness infrastructure (Testing: Bytecode Equivalence Infrastructure #704)
• Integration with Foundry
• CI pipeline for equivalence testing
• Coverage across Solidity features (structs, mappings, inheritance, etc.)

---

Phase 2: New IR Design ✅ COMPLETE

Decision: Build custom IR inspired by Venom (Vyper) and Sonatina (Fe).

What was built

The MIR is located in crates/codegen/src/mir/ with:

Component	Description
Module	Top-level container with functions, data segments, storage layout
Function	SSA function with basic blocks, values, instructions
BasicBlock	Linear instruction sequence with terminator
Inst / InstKind	All EVM operations + SSA-specific (Phi, Select)
Value	SSA values: Inst result, Arg, Immediate, Phi, Undef
Type	EVM-aware: UInt(bits), Address, MemPtr, StoragePtr, etc.
FunctionBuilder	Convenient construction API

Design influences adopted

From Venom (Vyper):

• SSA form with phi nodes
• Basic block structure with terminators
• Stack scheduling approach (to be implemented in MIR Codegen: Stack Model & Basic Scheduling (≤16 values) #696, MIR Codegen: Full Stack Scheduling with Spilling #697)
• Multi-pass optimization architecture

From Sonatina (Fe):

• Rust-native types and builder patterns
• Dense entity IDs for values/instructions/blocks
• Clean separation of concerns

---

Phase 3: HIR → IR → Bytecode Pipeline

Core Pipeline (Critical Path)

Step	Issue	Description
1	#694	Liveness analysis - which values are live at each point
2	#695	Phi elimination - convert SSA to move-based form
3	#696	Basic stack scheduling - map SSA to EVM stack (≤16 values)
4	#697	Full stack scheduling - handle >16 values with spilling
5	#698	Assembler - resolve labels and jumps to bytecode

Supporting Work

Issue	Description
#699	Complete lowering for structs, arrays, mappings

Pipeline Visualization

Source (.sol)
    ↓
[Parser] → AST
    ↓
[Sema] → Typed HIR     ← Current Solar
    ↓
[Lowering] → MIR        ← Done (basic), #699 (complex types)
    ↓
[Phi Elim] → MIR'       ← #695
    ↓
[Stack Sched] → Scheduled MIR  ← #696, #697
    ↓
[Codegen] → EVM Assembly  ← Partially done
    ↓
[Assembler] → Bytecode    ← #698

---

Phase 4: Optimizations

Issue	Optimization	Description
#700	DCE	Remove unused instructions
#701	Constant Folding	Evaluate constant expressions at compile time
#702	SCCP	Propagate constants through CFG, eliminate dead branches
#703	CSE	Eliminate redundant computations

Optimization Pipeline Order

const_fold → sccp → dce → cse → dce

Future Optimizations (not yet tracked)

• Function inlining
• Loop optimizations
• Storage access coalescing (batch SLOAD/SSTORE)
• Memory layout optimization
• Jump threading
• Peephole optimizations

Success Criteria

• Solar bytecode is smaller than solc (fewer bytes)
• Solar bytecode uses less gas for common operations
• Solar bytecode passes all equivalence tests from Phase 1

---

Performance Enhancements (Parallel Track)

These can proceed independently:

• Add project benches (Add project benches #508)
• SIMD-Optimized Lexer (Add SIMD-Optimized Lexer for Performance Improvements #475)
• Enable modern SIMD features (Enable modern SIMD features for per-platform builds #494)

---

References

• Main Roadmap: Roadmap #1
• Venom Integration: Integrate with Vyper's Venom CLI #174
• LSP Roadmap: LSP roadmap #394
• Typeck Tracking: Tracking: Typeck #615
• Venom IR: https://github.com/vyperlang/vyper/tree/master/vyper/venom
• Sonatina: https://github.com/fe-lang/sonatina
• solx: https://github.com/matter-labs/solx
• Fe Language: https://github.com/ethereum/fe
• EVM specification: https://ethereum.github.io/yellowpaper/

cc @gakonst @DaniPopes @onbjerg