This paper introduces two extensions to the popular PyTorch machine learning framework, TorchDynamo and TorchInductor, which implement the torch.compile feature released in PyTorch 2. TorchDynamo is a Python-level just-in-time (JIT) compiler that enables graph compilation in PyTorch programs without sacrificing the flexibility of Python. It achieves this by dynamically modifying Python bytecode before execution and extracting sequences of PyTorch operations into an FX graph, which is then JIT compiled using one of many extensible backends. TorchInductor is the default compiler backend for TorchDynamo, which translates PyTorch programs into OpenAI's Triton for GPUs and C++ for CPUs. Results show that TorchDynamo is able to capture graphs more robustly than prior approaches while adding minimal overhead, and TorchInductor is able to provide a 2.27× inference and 1.41× training geometric mean speedup on an NVIDIA A100 GPU across 180+ real-world models, which outperforms six other compilers. These extensions provide a new way to apply optimizations through compilers in eager mode frameworks like PyTorch.

This paper presents an efficient algorithm for post-synthesis logic simulation of digital circuits with oscillatory combinational loops. Oscillatory combinational loops can significantly degrade the performance of cycle accurate logic simulators. We provide an algorithm that first, dynamically detects oscillatory loops. Then, we introduce a novel approach to compute a multiple of their oscillation period which is used to optimize the efficiency of the simulation by reducing the number of time points that need to be evaluated. Finally, we provide the experimental results of our optimized algorithm measured on a cycle accurate simulator used in conjunction with a hardware emulator.