Add Triton installation for Windows CUDA, Linux ROCm/XPU

[godnight10061]

Summary

• Installs Triton automatically where PyTorch doesn’t provide it by default:
  • Windows + CUDA (cu* / nightly/cu*): installs triton-windows
  • Linux + ROCm 6.x+: installs pytorch-triton-rocm (from https://download.pytorch.org/whl)
  • Linux + XPU: installs pytorch-triton-xpu (from https://download.pytorch.org/whl)
• Adds missing runtime dependency packaging (required by torchruntime.platform_detection).
• Updates docs and adds installer unit tests.

Refs: #5

Why

torch.compile (and many third-party kernels) require Triton. On some platforms Torch bundles it (e.g. Linux CUDA), but on others users end up without Triton even after installing a GPU build of Torch.

Implementation

• torchruntime/installer.py appends an extra pip install command for the platform-specific Triton package.
• Kept logic minimal and platform-gated (no changes to platform detection).

Testing

• Unit tests: python -m pytest -q.
• Linux Mint 22.1 (Cinnamon): Verified with Python 3.9.23
• Real GPU smoke test (Windows, RTX 3060 Ti, Python 3.10):
  1. Create clean venv
  2. Install torch/torchvision/torchaudio from https://download.pytorch.org/whl/cu128
  3. Verify import triton fails
  4. Run python -m torchruntime install -> installs triton-windows
  5. Run a torch.compile CUDA smoke test -> OK

Request For Testing (hardware help wanted)

If you have one of these setups, please try:

• Linux ROCm (6.x): python -m torchruntime install then verify import triton and run a small torch.compile test.
• Linux Intel XPU: same as above.
  Also welcome: Windows CUDA users on different GPUs/Python versions.

[godnight10061]

Hi @iwr-redmond, since you mentioned ROCm support in Issue #5, could you help test if this installation logic works on your AMD setup?

[cmdr2]

Thanks @godnight10061 ! You can also ask on the #development channel in Easy Diffusion's discord server, since a few helpful users (with Linux) also hang out there - https://discord.com/invite/u9yhsFmEkB

You may also want to provide a simple script file that they can run to test torch.compile.

[cmdr2]

I'll give it a try with Windows 11 + WSL2 (Ubuntu) soon

[iwr-redmond]

Hi @iwr-redmond, since you mentioned ROCm support in Issue #5, could you help test if this installation logic works on your AMD setup?

My RTX 4070 is as useful for testing ROCm as the second buggy in a one-horse town.

[godnight10061]

Hi @cmdr2, this PR is ready for your review. I have implemented the automatic Triton installation logic for Windows (CUDA), Linux (ROCm), and Linux (XPU).
​

To make verification easier for the community, I also added a built-in self-test command in the latest commit: python -m torchruntime test compile. This will automatically verify if Triton is correctly installed and functional with torch.compile.
​

I have successfully smoke-tested this on Windows with an RTX 3060 Ti. since I lack AMD and Intel hardware, I’ve also reached out on Discord to find testers for the ROCm and XPU paths.

[godnight10061]

Quick update: A community member just verified this on Linux Mint 22.1 (Python 3.9). The test compile command passed successfully on their system, confirming that the ROCm-based Triton installation logic works as expected on Debian-based Linux distributions.

[iwr-redmond]

FYI, the XPU build of Triton also works on Windows. IIRC this wasn't the case when I wrote my original notes in #11.