The growing demand for high-performance tensor programs on GPUs, especially for large language models (LLMs), necessitates advanced compilation and optimization techniques. However, the critical task of analyzing optimized, low-level PTX code for performance tuning or understanding poses significant challenges. While LLMs hold promise for PTX-to-CUDA de-compilation to improve code intelligibility, their effectiveness is severely limited by the scarcity of aligned training data and the inherent complexity of highly optimized, unrolled PTX code.In this work, we explore methodologies to significantly enhance LLM capabilities for accurate and readable PTX-to-CUDA decompilation and present PtxDec, a decompilation prototype implementing our approach. To overcome the critical barrier of data scarcity, we develop a compiler-based data augmentation framework coupled with rigorous post-processing, enabling the creation of a large-scale, high-quality dataset of 400K aligned CUDA-PTX kernel pairs for effective LLM training. Furthermore, to empower LLMs to handle the complexity of optimized PTX, we introduce Rolled-PTX—an intermediate representation generated through heuristic loop rerolling during preprocessing. Rolled-PTX condenses unrolled patterns, drastically simplifying the input structure presented to the LLM and aligning it better with higher-level loop constructs.Comprehensive evaluation demonstrates that PtxDec achieves substantial performance gains: our approach yields a 2.3×–3.1× improvement in functional accuracy over baseline methods, alongside significant enhancements in generated code readability and scheduling consistency with the original optimized kernels. Ablation studies further validate the contribution of each proposed component to the overall performance.To the best of our knowledge, this is the first work tackling PTX-to-CUDA decompilation, specifically focusing on and demonstrating effective strategies for augmenting LLMs to overcome the key challenges in this domain.

Escape analysis plays a crucial role in garbage-collected languages as it enables the allocation of non-escaping variables on the stack by identifying the dynamic lifetimes of objects and pointers. This helps in reducing heap allocations and alleviating garbage collection pressure. However, Go, as a garbage-collected language, employs a fast yet conservative escape analysis, which is field-insensitive and omits point-to-set calculation to expedite compilation. This results in more variables being allocated on the heap. Empirical statistics reveal that field access and indirect memory access are prevalent in real-world Go programs, suggesting potential opportunities for escape analysis to enhance program performance. In this paper, we propose MEA 2 , an escape analysis framework atop GoLLVM (an LLVM-based Go compiler), which combines field sensitivity and points-to analysis. Moreover, a novel generic function summary representation is designed to facilitate fast inter-procedural analysis. We evaluated it by using MEA 2 to perform stack allocation in 12 wildly-use open-source projects. The results show that, compared to Go’s escape analysis, MEA 2 can reduce heap allocation sites by 7.9 % on average (up to 25.7 % ) while reducing the dynamic memory allocation size by 11.6 % on average (up to 35.5 % ). All this is achieved while keeping the time overhead of escape analysis within 1 % of the compilation process.