Upgrade llama.cpp from b10938 to b10948 - #436
Merged
Merged
Conversation
Upstream merged this project's own PR ggml-org/llama.cpp#28775 ("ggml-cpu(s390x): guard VXE-only repack helpers", commit 6978052), first tagged at b10948, so patches/0013 is now redundant: pristine b10948 already wraps vxe_dot_acc / vxe_splat_granule / vxe_fold in the #if defined(__VXE__) || defined(__VXE2__) guard the patch added, and `git apply` of the patch fails "does not apply" exactly as the fail-loud applier is designed to signal. Dropped, not refreshed, per the 0009 precedent at b10280. The rationale that outlives the patch -- why build-linux-s390x is a scalar (non-VXE) cross build, and why -DGGML_VXE=ON is the wrong response to a future VXE compile error -- moves into a "0013 was dropped at the b10948 bump" note in CLAUDE.md; the s390x job's own comment in publish.yml now points there. Verified against pristine b10948: patches 0001-0012 apply clean, 0013 fails; the three standing drop-checks still say "still required" (0001 no common_params_parse_main in common/arg.h; 0010 vocab_type still uncast at server-context.cpp:4554; 0012 bare splits[i] /= split_sum at llama-model.cpp:1491). A fresh `cmake -B build -DBUILD_TESTING=ON` configures clean through the real FetchContent path, stamping nine patches at head 5f436dddb (= b10948), with extraction unchanged at 138 CLI / 57 request / 15 trainer names. With the real s390x-linux-gnu-g++ cross toolchain, upstream's guarded repack.cpp compiles clean both with the CI job's scalar flags and with -mvx -mzvector -march=z15. The rest of the range touches no file on the API-compatibility review list: no common/, include/, tools/server/ or tools/mtmd/ changes, so the three mechanical server-contract greps have no input to compare. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SuJySZsHGCaxFq5Jo5cEtc
…g-changes history Two rows, per the version-bump runbook's step 4: what moved in the range (nothing on the priority API-compatibility review list -- no common/, include/, tools/server/ or tools/mtmd/ files, so the three mechanical server-contract greps have no input to compare), and the patch/upstream-verification row for the drop of patches/0013. The verification row records what was actually run rather than inferred: 0001-0012 apply at pristine b10948 and 0013 fails; the three standing drop-checks still say "still required"; the s390x half re-checked with the real s390x-linux-gnu-g++ cross toolchain (upstream's guarded repack.cpp compiles clean under both the job's scalar flags and -mvx -mzvector -march=z15); a fresh configure stamping nine patches at head 5f436dddb; extraction unchanged at 138/57/15 names; ctest 537/537; 40 Java_* exports, 0 mangled; and the NativeLibraryLoadSmokeTest run (4/4, 0 skipped) that cross-validates the bumped LLAMA_CPP_VERSION constant against the linked build-info. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01SuJySZsHGCaxFq5Jo5cEtc
bernardladenthin
had a problem deploying
to
maven-central
September 13, 2026 21:47 — with 
GitHub Actions
Failure
GitHub Actions
Failure
bernardladenthin
had a problem deploying
to
maven-central
September 13, 2026 21:47 — with
GitHub Actions
Failure
GitHub Actions
Failure
bernardladenthin
had a problem deploying
to
startgate
September 13, 2026 21:47 — with
GitHub Actions
Error
GitHub Actions
Error
|
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
0013-s390x-repack-guard-vxe-only-helpers.patch— upstream merged the equivalent fix as ggml-org/llama.cpp#28775 (commit6978052, first tagged at b10948)Details
Patch removal: The s390x repack helper guard that
0013added is now present upstream, byte-identical except for a trailing comment. The patch applier correctly fails with "does not apply cleanly" at configure time, as designed. Per the documented precedent (0009), the patch is dropped rather than refreshed.s390x configuration note: The underlying configuration (scalar s390x binary via
-DGGML_NATIVE=OFFin the cross-build job) remains unchanged and correct. CLAUDE.md now documents this configuration and warns against the tempting but incorrect fix of adding-DGGML_VXE=ON(which would cause dozens of z14+ builtin errors). Future VXE-related breaks should be checked against this description before reintroducing a local patch.Version updates: All references to b10938 updated to b10948:
llama/CMakeLists.txt— FetchContent GIT_TAGllama/src/main/java/net/ladenthin/llama/value/LlamaCppVersion.java— compile-time constantREADME.md— badge link.github/workflows/publish.yml— s390x build commentdocs/history/llama-cpp-breaking-changes.md— history entryCLAUDE.md— multiple references in documentation and examplesTest plan
Related issues / PRs
Upstream: ggml-org/llama.cpp#28775
Checklist
CONTRIBUTING.mdandCODE_OF_CONDUCT.mdhttps://claude.ai/code/session_01SuJySZsHGCaxFq5Jo5cEtc