🐛 Describe the bug

It turns out some ops, like torch.Tensor.expand accept -1 as size: "Passing -1 as the size for a dimension means not changing the size of that dimension."

The present implementation of SymInt disallows negative numbers and reserves the most significant bit of the data_ member variable for handling symbolic shapes.

This issue blocks expand.SymInt shape inference to handle -1 correctly.

Proposed solution:

• Treat data_ values "smaller" than -1 to represent symbolic dimensions - assuming we know of no other use case to go below -1.

CC @Krovatkin @ezyang @zou3519 @Gamrix @wconstab @JackCaoG @shauheen

Versions

Collecting environment information...
PyTorch version: N/A
Is debug build: N/A
CUDA used to build PyTorch: N/A
ROCM used to build PyTorch: N/A

OS: Debian GNU/Linux rodete (x86_64)
GCC version: (Debian 11.2.0-19) 11.2.0
Clang version: 13.0.1-3+build2
CMake version: version 3.22.1
Libc version: glibc-2.33

Python version: 3.8.13 (default, Mar 28 2022, 11:38:47) [GCC 7.5.0] (64-bit runtime)
Python platform: Linux-5.16.18-1rodete2-amd64-x86_64-with-glibc2.17
Is CUDA available: N/A
CUDA runtime version: Could not collect
GPU models and configuration: Could not collect
Nvidia driver version: Could not collect
cuDNN version: Could not collect
HIP runtime version: N/A
MIOpen runtime version: N/A
Is XNNPACK available: N/A

Versions of relevant libraries:
[pip3] numpy==1.22.3
[pip3] torch==1.12.0a0+gitb32758f
[conda] magma-cuda110 2.5.2 1 pytorch
[conda] mkl 2022.0.1 h06a4308_117
[conda] mkl-include 2022.0.1 h06a4308_117
[conda] numpy 1.22.3 py38h7a5d4dd_0
[conda] numpy-base 1.22.3 py38hb8be1f0_0
[conda] torch 1.12.0a0+gitb32758f pypi_0 pypi