Upgrade llama.cpp from b9842 to b9859

CLAUDE.md

@@ -6,7 +6,7 @@ This file provides guidance to Claude Code (claude.ai/code) when working with co
 
 Java bindings for [llama.cpp](https://github.com/ggerganov/llama.cpp) via JNI, providing a high-level API for LLM inference in Java. The Java layer communicates with a native C++ library through JNI.
 
-Current llama.cpp pinned version: **b9842**
+Current llama.cpp pinned version: **b9859**
 
 ## Upgrading CUDA Version
 
@@ -286,7 +286,7 @@ needs no extra step here, `build-webui` re-reads the tag and rebuilds the matchi
 ships no UI):
 ```bash
 # needs node/npm + network; embed.cpp is plain C++17 (no npm)
-git clone --depth 1 --branch b9842 https://github.com/ggml-org/llama.cpp /tmp/lc
+git clone --depth 1 --branch b9859 https://github.com/ggml-org/llama.cpp /tmp/lc
 ( cd /tmp/lc/tools/ui && npm ci && npm run build \
   && ( cd dist && find . -type f -not -path './_gzip/*' \
        | while read -r f; do mkdir -p "_gzip/$(dirname "$f")"; gzip -9 -c "$f" > "_gzip/$f"; done ) \
@@ -320,7 +320,7 @@ plus a cache token are present, `build.sh` adds
 - `SCCACHE_WEBDAV_TOKEN: ${{ secrets.DEPOT_TOKEN }}` — a Depot **organization** token, stored
   as the repo secret **`DEPOT_TOKEN`**.
 
-Because `sccache` is **content-addressed** and llama.cpp is pinned (`GIT_TAG b9842`), the
+Because `sccache` is **content-addressed** and llama.cpp is pinned (`GIT_TAG b9859`), the
 ~280 upstream object files are byte-identical every run, so a warm cache recompiles only the
 *changed* files. Depot's cache is **shared across all branches** (unlike GitHub's
 per-branch `actions/cache`), so every branch builds incrementally; a `b<nnnn>` version bump
@@ -1027,7 +1027,7 @@ ctest --test-dir build --output-on-failure -R "ResultsToJson"
 
 #### Upstream source location (in CMake build tree)
 
-llama.cpp is fetched via CMake FetchContent, pinned to `GIT_TAG b9842`.
+llama.cpp is fetched via CMake FetchContent, pinned to `GIT_TAG b9859`.
 
 **GoogleTest** is a separate `BUILD_TESTING`-only FetchContent (`GIT_TAG v1.17.0`), used solely
 by the `jllama_test` C++ unit-test binary — not by the shipped library, and not coupled to the

README.md

@@ -7,7 +7,7 @@
 **Build:**  
 ![Java 8+](https://img.shields.io/badge/Java-8%2B-informational)  
 ![Platform](https://img.shields.io/badge/Platform-Linux%20%7C%20macOS%20%7C%20Windows%20%7C%20Android-lightgrey)  
-[![llama.cpp b9842](https://img.shields.io/badge/llama.cpp-%23b9842-informational)](https://github.com/ggml-org/llama.cpp/releases/tag/b9842)  
+[![llama.cpp b9859](https://img.shields.io/badge/llama.cpp-%23b9859-informational)](https://github.com/ggml-org/llama.cpp/releases/tag/b9859)  
 [![JPMS](https://img.shields.io/badge/JPMS-modular%20JAR-25A162)](https://openjdk.org/projects/jigsaw/)  
 ![JUnit](https://img.shields.io/badge/tested%20with-JUnit6-25A162)  
 [![JSpecify](https://img.shields.io/badge/JSpecify-1.0.0%20%40NullMarked-25A162)](https://jspecify.dev)  

docs/history/llama-cpp-breaking-changes.md

@@ -409,3 +409,6 @@ Used during `llama.cpp` version bumps: when upgrading, scan this file from the r
 | b9839–b9840 | upstream verification (sandbox) | All four patches (`0001`–`0004`) re-verified to **apply cleanly** against b9840 via `git apply --check` over the actual b9840 sources fetched from `raw.githubusercontent.com` (github.com git-clone is blocked in this sandbox, so a full `FetchContent` build could not run — exit 0 for all four). The b9839→b9840 range touches **no** patch-target file (`common/arg.{cpp,h}`, `tools/server/server-context.{cpp,h}`, `server-common.cpp`, `test-chat.cpp`, the ~34 standalone mains) — it is entirely additive DeepSeek-V4 code — so the patch hunks and offsets are byte-identical to b9839. OuteTTS generator anchors hold (upstream `tools/tts/tts.cpp` unchanged in this range). Full build + `ctest` (target 459/459) to be confirmed by the CI pipeline |
 | b9840–b9842 | `common/preset.cpp` + `ggml/src/ggml-vulkan/ggml-vulkan.cpp` | Internal-only, no API surface. (1) `common/preset.cpp` adds a `canonical_tag()` helper and **canonicalizes the tag suffix** of INI preset section names (everything after the last `:` is upper-cased / normalized via a `<regex>`), so `[model:q4_k_m]` and `[model:Q4_K_M]` resolve to one preset. Preset loading is a CLI/`common` feature; the project's C++ never calls `common_preset*` — verified `grep -rn "common_preset\|canonical_tag\|load_from_ini" src/main/cpp src/test/cpp` &#x2192; zero matches. (2) `ggml-vulkan.cpp` reworks the per-graph **command-buffer submission heuristic** from "weight-matrix bytes per submit" to "estimated FLOPs per submit" (new `ggml_vk_get_node_flops()` over MUL_MAT/CONV/FLASH_ATTN nodes; `last_total_mul_mat_bytes` → `last_total_flops`; submit every ~200 GFLOP, still bounded by `max_nodes_per_submit`). Pure Vulkan-backend perf tuning, behaviour-neutral to callers; only affects the `vulkan-windows-x86-64` classifier. No project source changes required. |
 | b9840–b9842 | upstream verification (sandbox) | All four patches (`0001`–`0004`) re-verified to **apply cleanly** against b9842 via `git apply --check` over the actual b9842 sources (github.com git-clone is blocked in this sandbox, so a full `FetchContent` build could not run). The b9840→b9842 range touches **no** patch-target file (`common/arg.{cpp,h}`, `tools/server/server-context.{cpp,h}`, `server-common.cpp`, `test-chat.cpp`, the ~34 standalone mains) and **no** OuteTTS generator anchor (`tools/tts/tts.cpp` unchanged) — the only edits are `common/preset.cpp` and `ggml-vulkan.cpp` — so all patch hunks/offsets are byte-identical to b9840. Full build + `ctest` (target 459/459) to be confirmed by the CI pipeline. |
+| b9842–b9859 | `common/arg.cpp` + `common/http.h` + `tools/server/server-{http,models}.cpp` + `tools/server/server-cors-proxy.h` | **IPv6 URL handling + hf-split primary fix**, all inside upstream-compiled TUs the project already builds. (1) `common/http.h` gains a `common_http_format_host()` helper that brackets an IPv6 literal host (`[::1]`) per RFC 3986, and `common_http_parse_url` now splits the authority so a bracketed IPv6 literal keeps its inner colons; `server-http.cpp` (listening-address string), `server-models.cpp` (proxy `Host` header) and `server-cors-proxy.h` (proxy log) each `#include "http.h"` and route the host through it. `server-http.cpp`/`server-models.cpp`/`server-cors-proxy.h` are already compiled into `jllama`; the project binds none of these symbols and passes host/port as plain params, so behaviour is unchanged for localhost binds. (2) `common/arg.cpp` `common_models_handler_apply` now threads a `primary` hf-split file (the `00001-of` part) through the `add_tasks` lambda instead of assuming index 0 — internal to the `--hf`/`--hf-repo-v`/`--spec-draft-hf` download planner, which the project never calls (`grep -rn "common_models_handler\|common_http_format_host" src/main/cpp src/test/cpp` → zero matches). No project source changes required. |
+| b9842–b9859 | `ggml/src/ggml-cpu/` + `ggml/src/ggml-cuda/` + `ggml/src/ggml-opencl/` + `ggml/src/ggml-vulkan/` + `ggml/src/ggml-webgpu/` + `ggml/src/ggml-hexagon/` + `ggml/src/ggml-backend.cpp` + `src/models/qwen3next.cpp` + `tools/ui/**` | Backend-internal only, no API surface visible to `jllama.cpp`. CPU adds an AVX2/AVX `ggml_vec_dot_nvfp4_q8_0` + a UE4M3 lookup table (`kvalues_mxfp4` renamed to shared `kvalues_fp4`); CUDA adds head-dim-512 flash-attention MMA/tile instances, a strided `get_rows_back` grid-clamp fix (new `test-backend-ops` case for row count > 65535), a gfx900 MMQ gate, and drops the CPU→CUDA async-copy path (scheduler now copies inputs synchronously); OpenCL adds full Q1_0 mul_mat/mul_mv + a `GGML_OPENCL_USE_ADRENO_BIN_KERNELS` prebuilt-binary-kernel loader (OFF by default; affects only the `opencl-*` classifiers); Vulkan rolls the mul_mm BK loop on Asahi/Honeykrisp; WebGPU adds NVFP4 support; Hexagon reworks HVX/HMX flash-attention (new `flash-attn-ops.h`/`hmx-fa-kernels.h`, MUL_MAT_ADD fusion). `qwen3next.cpp` records `t_layer_inp[il]` for MTP. All internal to upstream-compiled `libllama`/`ggml`/backends; the WebUI **auto-follows** the pinned `GIT_TAG` (the `build-webui` CI job rebuilds it), so its edits (PWA navigate-fallback, chat-store foreign-conversation guards) need no manual step. No project source changes required. |
+| b9842–b9859 | upstream verification (sandbox) | All four patches (`0001`–`0004`) re-verified to **apply cleanly** against b9859 via `git apply --check` over the actual b9859 sources fetched from `raw.githubusercontent.com` (github.com git-clone is blocked in this sandbox, so a full `FetchContent` build could not run — exit 0 for `common/arg.{cpp,h}`, `tests/test-arg-parser.cpp`, `tools/server/server-context.{cpp,h}`, `server-common.cpp`, `tests/test-chat.cpp`). The only patch-target file that changed in this range is `common/arg.cpp`, whose b9859 edit is in `common_models_handler_apply` (~L496) — disjoint from patch 0001's `make_utf8_argv`/`common_params_parse` hunks (~L931/L971) and the ~34 standalone-main flips (unchanged in this range), so patch 0001 still applies. Patches 0002/0003/0004 target files untouched in b9842→b9859, so their hunks are byte-identical to b9842. OuteTTS generator anchors hold (`tools/tts/tts.cpp` unchanged). Full build + `ctest` (target 459/459) to be confirmed by the CI pipeline. |

llama/CMakeLists.txt

@@ -143,7 +143,7 @@ set(LLAMA_BUILD_APP OFF CACHE BOOL "" FORCE)
 FetchContent_Declare(
 	llama.cpp
 	GIT_REPOSITORY https://github.com/ggerganov/llama.cpp.git
-	GIT_TAG        b9842
+	GIT_TAG        b9859
 	PATCH_COMMAND  ${CMAKE_COMMAND}
 		-DPATCH_DIR=${CMAKE_CURRENT_SOURCE_DIR}/patches
 		-DLLAMA_SRC=<SOURCE_DIR>
@@ -166,7 +166,7 @@ execute_process(
     COMMAND ${CMAKE_COMMAND}
         -DTTS_SRC=${llama.cpp_SOURCE_DIR}/tools/tts/tts.cpp
         -DOUT_CPP=${JLLAMA_TTS_GEN_CPP}
-        -DLLAMA_TAG=b9842
+        -DLLAMA_TAG=b9859
         -P ${CMAKE_CURRENT_SOURCE_DIR}/cmake/generate-tts-upstream.cmake
     RESULT_VARIABLE JLLAMA_TTS_GEN_RESULT
 )