메뉴
Use when setting up a new torch-xpu-ops release branch corresponding to a PyTorch release. Covers branch creation, release tracker issue, workflow default changes PR, and tracker comment.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Analyze UT (unit test) results for a torch-xpu-ops PR. Use when asked to check test results, analyze CI failures, or evaluate test coverage for a PR. Produces a structured report of new failures, failure relevance, and new test coverage.
Rules for automatically determining which disable_* labels to apply to a PR based on the file paths changed. Used by the auto-label workflow.
Extract GPU ISA from oneDNN ngen-JIT kernels. This is the ONLY codegen path that bypasses the standard SYCL/SPIR-V/IGC stack. oneDNN uses its own native code generator (ngen) that directly emits GPU ISA bytes — no SPIR-V, no IGC, no zebin ELF, no .debug_line. Use when extracting ASM from oneDNN kernels (gemm_kernel, gen_conv_kernel), matmul, linear, conv, or SDPA-graph ops dispatched via mkldnn.
Extract GPU ISA from AOT-compiled SYCL binaries using NEO driver runtime dump. Use when the kernel is a SYCL symbol (_ZTS...) from an AOT binary such as libtorch_xpu.so, SYCL-TLA FMHA, or any DPC++ binary compiled with -fsycl-targets=spir64_gen -Xs "-device <gpu>".
Extract GPU ISA from JIT-compiled SYCL kernels. The binary only contains SPIR-V; IGC compiles it to native ISA at first launch. The zebin only exists in memory at runtime, so IGC_ShaderDumpEnable=1 is required to capture it. Use when extracting ASM from SYCL JIT kernels, torch-xpu-ops built with TORCH_XPU_ARCH_LIST=none, or standalone DPC++ without AOT flags.
Extract GPU ISA from Triton kernels on XPU. Triton compiles through the same IGC backend as SYCL (Triton IR → SPIR-V → IGC → zebin). Extraction is identical to sycl-jit: IGC_ShaderDumpEnable=1 captures the zebin at runtime. Use when extracting ASM from torch.compile fusions (triton_per_fused, triton_poi, triton_red), standalone @triton.jit kernels, or Inductor-generated XPU kernels.
| name | release-branching |
| description | Use when setting up a new torch-xpu-ops release branch corresponding to a PyTorch release. Covers branch creation, release tracker issue, workflow default changes PR, and tracker comment. |
Set up torch-xpu-ops for a new PyTorch release cycle: create the release branch, open the tracker issue, submit the workflow-defaults PR, and post the initial tracker comment.
Invocation: /xpu-release-branching release/X.Y
Given release/2.12:
BRANCH = release/2.12MAJOR_MINOR = 2.12VERSION = v.2.12.0Confirm the pytorch release branch exists and locate the release tracker issue:
# Branch must exist
gh api repos/pytorch/pytorch/branches/release/MAJOR_MINOR --jq '.name'
# Find the pytorch release tracker issue
gh issue list -R pytorch/pytorch -s all \
--search "[v.MAJOR_MINOR.0] Release Tracker in:title" \
--json number,title,url -L 1
Read the pytorch release tracker issue body to extract:
These dates are needed for the torch-xpu-ops tracker issue.
Get the pinned commit from pytorch's release branch and create the torch-xpu-ops branch from it. If the branch already exists (e.g. retrying after a partial failure), skip this step and verify the branch directly.
# Get the pinned torch-xpu-ops commit on the pytorch release branch
PINNED=$(gh api repos/pytorch/pytorch/contents/third_party/xpu.txt \
--jq '.content' -H 'Accept: application/vnd.github.v3+json' \
-f ref=release/MAJOR_MINOR | base64 -d | tr -d '[:space:]')
# Create the release branch at that commit
gh api repos/intel/torch-xpu-ops/git/refs -f ref=refs/heads/release/MAJOR_MINOR -f sha=$PINNED
Verify the branch was created:
gh api repos/intel/torch-xpu-ops/branches/release/MAJOR_MINOR --jq '.name'
Create a milestone PTMAJOR_MINOR in torch-xpu-ops if it does not exist yet:
# Check if milestone exists; create if not
gh api repos/intel/torch-xpu-ops/milestones --jq '.[] | select(.title == "PTMAJOR_MINOR") | .title' \
|| gh api repos/intel/torch-xpu-ops/milestones -f title="PTMAJOR_MINOR" -f state=open
Open the tracker issue with this exact body template:
Title: [VERSION] Release Tracker
Milestone: PTMAJOR_MINOR
Issue body (replace placeholders):
We cut a [release branch](https://github.com/intel/torch-xpu-ops/tree/BRANCH) for the MAJOR_MINOR.0 release.
Our plan from this point is roughly:
- Phase 1 (until <PHASE1_DEADLINE>): work on finalizing the release branch
- Phase 2 (after <PHASE2_START>): perform extended integration/stability/performance testing based on Release Candidate builds.
This issue is for tracking cherry-picks to the release branch.
Refer <PYTORCH_TRACKER_URL>
Where:
<PHASE1_DEADLINE> / <PHASE2_START> come from the pytorch tracker (Step 1)<PYTORCH_TRACKER_URL> is https://github.com/pytorch/pytorch/issues/<NUMBER>gh issue create -R intel/torch-xpu-ops \
--title "[VERSION] Release Tracker" \
--milestone "PTMAJOR_MINOR" \
--body "$(cat <<'EOF'
<body from template above>
EOF
)"
Record the created issue number for Step 5.
This PR changes the default branch parameters in workflow files from main
to BRANCH.
git fetch origin BRANCH
git checkout -b release_MAJOR_MINOR_change origin/BRANCH
(Branch naming convention: release_X.Y_change, e.g. release_2.12_change)
Two files need updating — .github/workflows/_linux_build.yml and
.github/workflows/_windows_ut.yml:
For each file, change the pytorch and torch_xpu_ops input defaults:
# Before
pytorch:
...
default: 'main'
description: Pytorch main by default, ...
torch_xpu_ops:
...
default: 'main'
description: Torch-xpu-ops main by default, ...
# After
pytorch:
...
default: 'BRANCH'
description: Pytorch BRANCH by default, ...
torch_xpu_ops:
...
default: 'BRANCH'
description: Torch-xpu-ops BRANCH by default, ...
Only change the default value and update main to BRANCH in the
description string. Do not touch any other fields.
git add .github/workflows/_linux_build.yml .github/workflows/_windows_ut.yml
git commit -m "[RELEASE MAJOR_MINOR] Release only changes"
git push origin release_MAJOR_MINOR_change
gh pr create -R intel/torch-xpu-ops \
--base BRANCH \
--head release_MAJOR_MINOR_change \
--title "[RELEASE MAJOR_MINOR] Release only changes"
No description body is needed.
Post the cherry-pick tracking comment on the torch-xpu-ops tracker issue created in Step 3:
gh issue comment <ISSUE_NUMBER> -R intel/torch-xpu-ops --body "$(cat <<'EOF'
Link to landed trunk PR (if applicable):
* NA
Link to release branch PR:
* <PR_URL>
Criteria Category:
* Release only changes
EOF
)"
Where <PR_URL> is the URL from Step 4d.