# Souper: A Synthesizing Superoptimizer

arXiv: https://arxiv.org/abs/1711.04422 · HTML: https://mlsanity.com/paper/1711.04422/

## Key facts

- Cited by 12 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 13 Nov 2017.

## Built on by (most cited first, 12 of 12)

- [Learning Performance-Improving Code Edits](https://mlsanity.com/paper/2302.07867/) (2023)
- [Equality Saturation for Tensor Graph Superoptimization](https://mlsanity.com/paper/2101.01332/) (2021)
- [FormulaCode: Evaluating Agentic Optimization on Large Codebases](https://mlsanity.com/paper/2603.16011/) (2026)
- [Learning to Superoptimize Real-world Programs](https://mlsanity.com/paper/2109.13498/) (2021)
- [Bootstrapping Fuzzers for Compilers of Low-Resource Language Dialects Using Language Models](https://mlsanity.com/paper/2512.05887/) (2025)
- [Securing Optimized Code Against Power Side Channels](https://mlsanity.com/paper/2207.02614/) (2022)
- [Compiling by Proving: Language-Agnostic Automatic Optimization from Formal Semantics](https://mlsanity.com/paper/2509.21793/) (2025)
- [GPU-Accelerated Synthesis of Mixed-Boolean Arithmetic: Beyond Caching](https://mlsanity.com/paper/2605.08243/) (2026)
- [Understanding Agent-Based Patching of Compiler Missed Optimizations](https://mlsanity.com/paper/2607.02370/) (2026)
- [Can Coding Agents Implement Missed Compiler Optimizations? Evaluating LLM Agents on LLVM Peephole Optimizations](https://mlsanity.com/paper/2607.02684/) (2026)
- [Can Large Language Models Recover Semantic Optimization Opportunities That Compilers Miss?](https://mlsanity.com/paper/2608.03983/) (2026)
- [AsmEvo: Agentic Assembly-Level Optimization of AMD GPU Kernels with Functional Equivalence Verification](https://mlsanity.com/paper/2608.20711/) (2026)

## Abstract (arXiv)

If we can automatically derive compiler optimizations, we might be able to sidestep some of the substantial engineering challenges involved in creating and maintaining a high-quality compiler. We developed Souper, a synthesizing superoptimizer, to see how far these ideas might be pushed in the context of LLVM. Along the way, we discovered that Souper's intermediate representation was sufficiently similar to the one in Microsoft Visual C++ that we applied Souper to that compiler as well. Shipping, or about-to-ship, versions of both compilers contain optimizations suggested by Souper but implemented by hand. Alternately, when Souper is used as a fully automated optimization pass it compiles a Clang compiler binary that is about 3 MB (4.4%) smaller than the one compiled by LLVM.