llvm-toolchain-22 (1:22.1.8-1) [PTS] [DDPO]

NEW: VCS has unreleased changes: 1:22.1.8-3 > 1:22.1.8-1

Git: https://salsa.debian.org/pkg-llvm-team/llvm-toolchain.git -b 22
JSON
Enter a full Vcs-* header, supported schemes are: Vcs-Bzr, Vcs-Cvs, Vcs-Darcs, Vcs-Git, Vcs-Hg, Vcs-Mtn, Vcs-Svn.
For Vcs-Git append -b branch to select the branch name; append [subdir] if the debian/ directory is not in the repository root.
The next upload will reset the information from the source package headers.
Branch: 22
Path: debian/changelog
Repo size: 2301952
Browser: https://salsa.debian.org/pkg-llvm-team/llvm-toolchain/tree/22
Last scan: 2026-07-15 03:01:13+00
Next scan: 2026-07-20 17:32:00+00
Merge requests: 2

Debian changelog in Git:

llvm-toolchain-22 (1:22.1.8-3) UNRELEASED; urgency=medium

  * d/p/amdgpu-fix-simulated-trap-continuation-block.diff: cherry-pick
    llvm/llvm-project@9429a1e809fd (PR #174774,
    "[AMDGPU] Fix insertSimulatedTrap to return correct continuation block")
    Fixes a codegen crash on gfx1100/gfx1101/gfx1102 (RDNA3 desktop:
    RX 7900/7800/7700/7600) where clang aborts with
    "fatal error: error in backend: Unsupported instruction : <MCInst NNNN ...>"
    from the AMDGPU AsmPrinter whenever a kernel contains a block that ends in
    llvm.trap() with no successors — an unlowered pseudo reaches the MC layer.
    Notably affects pytorch-rocm 2.12's Loss.hip (FTBFS). Reported upstream as
    llvm/llvm-project#208912. Fix landed on main branch on 2026-01-21 (LLVM 23);
    release/22.x was cut 8 days before and never received it. This cherry-pick
    lets 22.1.x carry the fix without waiting for a 22.1.9 or 22.2.0 point
    release upstream. (Closes: #1141868)

 -- Talha Can Havadar <havadartalha@gmail.com>  Sat, 11 Jul 2026 23:01:25 +0200

This branch is 4 commits ahead of tag debian/1%22.1.8-1

Git log:

commit 4b66b8b493796a31e1ce4a5101bdfbf000225f71
Merge: fa997e3 4d8cd1b
Author: Sylvestre Ledru <sylvestre@debian.org>
Date:   Sat Jul 11 21:41:39 2026 +0000

    Merge branch '22' into '22'
    
    #1141868 d/p/amdgpu-fix-simulated-trap-continuation-block.diff: cherry-pick...
    
    See merge request pkg-llvm-team/llvm-toolchain!223

commit 4d8cd1b8b6389f80a91bd15bbed11c2f311eb546
Author: Talha Can Havadar <havadartalha@gmail.com>
Date:   Sat Jul 11 23:02:11 2026 +0200

    d/changelog: update changelog with AMDGPU RDNA3 fix

commit 1707a05759a754f895b9db89a570b39b892d41a8
Author: Talha Can Havadar <havadartalha@gmail.com>
Date:   Sat Jul 11 18:16:55 2026 +0200

    d/p/amdgpu-fix-simulated-trap-continuation-block.diff: cherry-pick llvm/llvm-project@9429a1e809fd (PR #174774, "[AMDGPU] Fix insertSimulatedTrap to return correct continuation block")
    
    Fixes a codegen crash on gfx1100/gfx1101/gfx1102 (RDNA3 desktop:
    RX 7900/7800/7700/7600) where clang aborts with
    "fatal error: error in backend: Unsupported instruction : <MCInst NNNN ...>"
    from the AMDGPU AsmPrinter whenever a kernel contains a block that ends in
    llvm.trap() with no successors — an unlowered pseudo reaches the MC layer.
    Notably affects pytorch-rocm 2.12's Loss.hip (FTBFS). Reported upstream as
    llvm/llvm-project#208912. Fix landed on main branch on 2026-01-21 (LLVM 23);
    release/22.x was cut 8 days before and never received it. This cherry-pick
    lets 22.1.x carry the fix without waiting for a 22.1.9 or 22.2.0 point
    release upstream.
    
    Closes: #1141868
    Gbp-Dch: full

commit fa997e39aca00815da81e28c31dd231eb3c74f5f
Author: Sylvestre Ledru <sylvestre@debian.org>
Date:   Wed Jun 24 12:34:29 2026 +0200

    d/rules: keep sccache server alive (SCCACHE_IDLE_TIMEOUT=0) + use GCS cache on arm64
    
    The libfuzzer step calls sccache directly with no backend env, long after the
    server was started in configure. With the default 600s idle timeout the server
    dies during the sccache-quiet final-link phase, so libfuzzer auto-spawns a
    default-config server that writes to $HOME/.cache/sccache and fails with
    HOME=/nonexistent. Set SCCACHE_IDLE_TIMEOUT=0 on both --start-server calls so
    the configured server (GCS or local) stays up for the whole build. Also add
    arm64 to the GCP arch filter so it uses the shared GCS cache like amd64.