🐛 [Bug] "Cannot convert symbols to int" when compiling graph that outputs tensor with dynamic shape

[dgcnz]

Bug Description

When doing parse_graph_io on a graph that outputs a tensor with dynamic shapes, extract_var_range_info(node.meta["val"]) fails because tensor.shape[0] in opt_val = int(tensor.shape[0].node.shape_env.get(expr) is an unbacked SymInt without an actual value attached.

This sort of makes sense, so I'm not sure if it's a bug or a non-feature.

Anyway, (I think) another possible source to get a value for that unbacked symint is to use tensor.shape[0].node.shape_env.dim_constraints._static_results which returns {"L['x'].size()[0] == 10"}.

To Reproduce

MRE and logs in this gist.

Expected behavior

Not sure.

Environment

Build information about Torch-TensorRT can be found by turning on debug messages

• Torch-TensorRT Version (e.g. 1.0.0): 2.6.0.dev20241030+cu124
• PyTorch Version (e.g. 1.0): 2.6.0.dev20241028+cu124
• CPU Architecture: x86_64
• OS (e.g., Linux): Linux
• How you installed PyTorch (conda, pip, libtorch, source): pip
• Build command you used (if compiling from source): -
• Are you using local sources or building from archives: -
• Python version: 3.10.0
• CUDA version: 12.4
• GPU models and configuration: 4060 TI
• Any other relevant information: -