Repository navigation
Home
ZhaoqiongZ edited this page Sep 28, 2026
·
3 revisions
Using PyTorch on Intel GPUs? See Getting Started on Intel GPU. This wiki is for developing torch-xpu-ops itself.
Developer documentation for intel/torch-xpu-ops — the SYCL/XPU ATen operator implementations consumed by PyTorch as a submodule.
New here? Start with the repo README and CONTRIBUTING, then use the pages below.
| Page | What it covers |
|---|---|
| Windows build cmd | Windows set flags for a source build; oneAPI toolkit download; nightly XPU wheel install |
| Page | What it covers |
|---|---|
| Guide for XPU native functions registration using Codegen | Source/generated file layout, native_functions.yaml reference decls, adding and modifying kernels, yaml/xpu_functions.yaml registration |
| Best Practice in development |
rebase over merge; mixed Python/C++ debugging in VS Code over remote SSH (DEBUG=1, BUILD_SEPARATE_OPS=ON, ptrace_scope fix) |
| Page | What it covers |
|---|---|
| Guide for operator level performance debugging | Setting up torchvision/torchaudio, Triton for XPU, and pytorch/benchmark from source |
| Page | What it covers |
|---|---|
| ATen Operator Coverage Report | Running tools/check_ops.py to report XPU ATen op coverage and gaps (basic / sparse / sparse_csr / nested_tensor) |
| Page | What it covers |
|---|---|
| torchxpubot User Guide | Bot commands on PRs and issues (fix, review, item=, run=), runner limits, rules before triggering a fix, end-to-end worked examples |
| Agentic Skills User Guide | OpenCode setup, repo/gh permissions, and the skills for nightly CI triage, single-issue UT fixes, roofline perf analysis, and operator alignment |
- Issues · Pull requests · Discussions
- AGENTS.md — in-repo instructions for coding agents
When adding a page, put it under one of the sections above and keep the one-line description in the table.