commit 9e9797a9a4a2bcbdb724337385a17a69f3d899ba
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Sep 10 03:00:39 2026 +0200

dev-python/triton-bin: trim 3.6.0 comment prose

commit f8bb65953de3a6241637d725c579064dc3bd5857
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Sep 10 02:59:57 2026 +0200

dev-python/triton-bin: trim 3.7.0 comment prose

commit b644e75373f80a4ec0a64b35f125f93d159c76f6
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Sep 10 01:37:11 2026 +0200

dev-python/triton-bin: trim 3.7.1 comment prose

commit 074d0f89b0b4405073831e77421045bd4de623d7
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Wed Sep 9 17:17:28 2026 +0200

dev-python/triton-bin: refresh the version-pairing note

The note was written to stop this rationale rotting and had itself gone
stale on three counts.

It claimed the consumers still pin ~virtual/triton-3.6.0, which stopped
being true when vllm 0.27.1 and 0.28.0 moved to the pairing for the torch
they actually depend on. It also said comfyui is on torch 2.11 and
therefore on 3.6.0, wrong on both halves: comfyui depends on caffe2 and
torchvision without a version, so it pins no torch at all, and it now takes
an unversioned virtual to match. And it dismissed triton 3.8.0 as pairing
with a torch that is not in the tree, but pytorch 2.14.0 landed on
2026-09-03 and its tarball names 3.8.0.

So the reason not to bump on drift alone is no longer "unreachable" but
"belongs to a torch line nothing needs Triton on yet", and the note now
says that instead. The mapping is re-read from the release tarballs for
every entry except 2.12.0, whose distfile is not local.

A line claiming this pairing was verified against pytorch v2.12.0 is
retired rather than restated: it cannot describe this ebuild, since 2.12.0
pairs with 3.7.0 per the table directly above it. What remains is the
accurate part -- no end-to-end run exists on this wheel, and that gap now
matters because vllm selects it rather than 3.6.0.

commit b28fef600884f818ac0cc9675cc798ec899cefbd
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sat Aug 29 12:05:16 2026 +0200

dev-python/triton-bin: record why 3.8.0 is not bumped

No version change. triton 3.8.0 is released and will show as drift on every
nvchecker run; this records why the answer is "no" so it is not re-litigated
each time.

triton tracks pytorch's .ci/docker/triton_version.txt, not the newest triton
release. Re-read from the pytorch tags: torch 2.11.0 -> 3.6.0,
2.12.0 -> 3.7.0, 2.13.0 -> 3.7.1. 2.13.0 is the newest torch in this
overlay, so 3.7.1 is the correct head of the line and 3.8.0 pairs with a
torch we do not ship.

The existing comment was also stale in a way that hid this: it stopped at
"torch-2.12.0 pairs with triton 3.7.0" while the tree had already moved to
torch 2.13.0 / triton 3.7.1, so it no longer described the version it was
attached to.

Structurally, a 3.8.0 would be unreachable regardless -- every consumer goes
through virtual/triton, so it would need a virtual/triton-3.8.0, and the
actual consumers (vllm 0.26.0/0.27.1/0.28.0, comfyui 0.33.4/0.34.x) still
pin ~virtual/triton-3.6.0. Bumping triton is a step inside a torch bump,
never a standalone drift fix.

commit 0a3cfe43b214c6d365585a291e5ef5340975bf66
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun Aug 23 12:42:45 2026 +0200

dev-python/triton-bin: block triton

commit 4a6c268f4e6175a21f7f31e859b675d14499aa4a
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Thu Jun 18 08:00:57 2026 +0200

dev-python/triton-bin: add 3.7.1

commit 7a6588a2044164910e43e4d206cf8e79d9d7b330
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Mon Jun 15 14:27:25 2026 +0200

dev-python/triton-bin: add 3.7.0

torch-2.12.0 pairs with triton 3.7.0 (pytorch
.ci/docker/triton_version.txt); the existing 3.6.0 is torch-2.11.0's
triton. humming-kernels (unversioned triton-bin dep) picks 3.7.0 up to
match the torch-2.12 stack. 3.6.0 stays for vllm and comfyui, which
still pin ~3.6.0 -- those pins predate the torch-2.12 bump and want
re-evaluating against 3.7.0, but that needs a build-verify pass.

commit 994121024687b6fd737cb5548664f4f8e03e81ab
Author: Ivan S. Titov <iohann.s.titov@gmail.com>
Date: Sun Jun 14 09:26:23 2026 +0200

dev-python/triton-bin: add 3.6.0

GPU programming language and compiler, packaged from the upstream PyPI
wheels (cp312/cp313/cp314, x86_64). Triton builds from an LLVM/MLIR
source tree that is impractical to compile in-tree, so the binary wheel
is the supported form, mirroring dev-python/cuda-tile-bin. The wheel is
self-contained (no python RDEPEND) and bundles its own LLVM plus the
NVIDIA ptxas toolchain.

3.6.0 is the Triton release that pytorch-2.11.0 pairs with
(pytorch .ci/docker/triton_version.txt), which the source-built
sci-ml/pytorch needs for its torch.compile/inductor backend. Pulled by
dev-python/vllm on CUDA, whose @triton.jit kernels otherwise fail at
first GPU inference.

Drop Triton's proton / proton-viewer profiler console scripts from
/usr/bin: the names are generic enough to collide (e.g. with Valve
Proton) and torch/vllm do not use them; the triton.profiler.proton
module stays importable.