Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

From CISC to RISC: language-model guided assembly transpilation

Ahmed Heakl, Chaimaa Abi, Rania Hossam, Abdulrahman Mahmoud

TL;DR

CRT is introduced, a lightweight LLM-based transpiler that automatically converts x86 assembly to ARM assembly and successfully navigates the CISC/RISC divide and generates correctly executable RISC code despite machine ``language'' barriers.

Abstract

The transition from x86 to ARM architecture is becoming increasingly common across various domains, primarily driven by ARM's energy efficiency and improved performance across traditional sectors. However, this ISA shift poses significant challenges, mainly due to the extensive legacy ecosystem of x86 software and lack of portability across proprietary ecosystems and software stacks. This paper introduces CRT, a lightweight LLM-based transpiler that automatically converts x86 assembly to ARM assembly. Our approach bridges the fundamental architectural gap between x86's CISC-based and ARM's RISC-based computing paradigms while preserving program semantics and optimizing performance. We evaluate CRT on diverse real-world applications, achieving 79.25% translation accuracy from x86 to ARMv5 on our comprehensive test suite, and an 88.68% accuracy from x86 to RISC-V. In practical deployments on Apple M2 hardware (ARMv8), our transpiled code achieves 1.73$\times$ speedup compared to Apple's Rosetta 2 virtualization engine, while delivering 2.41$\times$ memory efficiency and 1.47$\times$ better energy consumption. Through testing and analysis, we show that CRT successfully navigates the CISC/RISC divide and generates correctly executable RISC code despite machine ``language'' barriers. We release our code, models, training datasets, and benchmarks at: \url{https://ahmedheakl.github.io/asm2asm/}.

From CISC to RISC: language-model guided assembly transpilation

TL;DR

CRT is introduced, a lightweight LLM-based transpiler that automatically converts x86 assembly to ARM assembly and successfully navigates the CISC/RISC divide and generates correctly executable RISC code despite machine ``language'' barriers.

Abstract

The transition from x86 to ARM architecture is becoming increasingly common across various domains, primarily driven by ARM's energy efficiency and improved performance across traditional sectors. However, this ISA shift poses significant challenges, mainly due to the extensive legacy ecosystem of x86 software and lack of portability across proprietary ecosystems and software stacks. This paper introduces CRT, a lightweight LLM-based transpiler that automatically converts x86 assembly to ARM assembly. Our approach bridges the fundamental architectural gap between x86's CISC-based and ARM's RISC-based computing paradigms while preserving program semantics and optimizing performance. We evaluate CRT on diverse real-world applications, achieving 79.25% translation accuracy from x86 to ARMv5 on our comprehensive test suite, and an 88.68% accuracy from x86 to RISC-V. In practical deployments on Apple M2 hardware (ARMv8), our transpiled code achieves 1.73 speedup compared to Apple's Rosetta 2 virtualization engine, while delivering 2.41 memory efficiency and 1.47 better energy consumption. Through testing and analysis, we show that CRT successfully navigates the CISC/RISC divide and generates correctly executable RISC code despite machine ``language'' barriers. We release our code, models, training datasets, and benchmarks at: \url{https://ahmedheakl.github.io/asm2asm/}.

Paper Structure

This paper contains 28 sections, 6 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Background & Related Work
  3. Approach
  4. Problem Definition
  5. Training Stages and Model Selection
  6. Tokenizer Extension
  7. Experiments and Evaluation
  8. Training Data
  9. Experimental Setup
  10. Evaluation Benchmark
  11. Evaluation Metrics
  12. Results
  13. Transpiler Validation
  14. Impact of Tokenizer and Quantization
  15. Training Performance
  16. ...and 13 more sections

Figures (6)

  • Figure 1: Conceptual representation of an asm-to-asm transpiler, which would enable direct "translation" from one machine language to another without needing the source code and by-passing the software stack.
  • Figure 2: CRT pipeline stages: Data (AnghaBench data curation), Experimentation (model tuning and accuracy), and Optimization & Deployment (final training and Rosetta evaluation).
  • Figure 3: DeepSeek-1.3B performance: (a) Accuracy across beam sizes (1, 2, 4, 8) for different training data sizes. (b) Accuracy progression over training steps with a logarithmic trend. (c) Quantization impact (float32, bfloat16, int8, int4) on ARM and RISC-V64.
  • Figure 4: Measured execution time, energy utilization, and RAM usage across different settings on Apple M2 Macbook.
  • Figure 5: Confusion matrix of the proposed approach executed on QEMU (ARMv5) and M2 (ARMv8) for HumanEval programs.
  • ...and 1 more figures