Skip to content

Commit 14a300a

Browse files
committed
Add macOS 15 CI target and fix ARM i8mm feature conflict
On macOS 15 (Apple Clang 16), stricter always_inline enforcement causes a build failure when llama.cpp generates -mcpu=native+dotprod+noi8mm flags while code includes i8mm-requiring always_inline functions. Adding -DGGML_NATIVE=OFF for the macos-15 runner disables native CPU feature detection, avoiding the contradictory i8mm enable/disable state. Also use runner-specific artifact names to avoid upload conflicts when multiple macOS matrix jobs fail simultaneously. https://claude.ai/code/session_01PjfeKeUZE8spcY4atNbymE
1 parent 6efa049 commit 14a300a

1 file changed

Lines changed: 3 additions & 1 deletion

File tree

.github/workflows/ci.yml

Lines changed: 3 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -49,6 +49,8 @@ jobs:
4949
cmake: -DLLAMA_METAL=OFF -DLLAMA_VERBOSE=ON
5050
- runner: macos-14
5151
cmake: -DLLAMA_METAL_EMBED_LIBRARY=ON -DLLAMA_VERBOSE=ON
52+
- runner: macos-15
53+
cmake: -DLLAMA_METAL_EMBED_LIBRARY=ON -DLLAMA_VERBOSE=ON -DGGML_NATIVE=OFF
5254
steps:
5355
- uses: actions/checkout@v4
5456
- uses: actions/setup-java@v4
@@ -70,7 +72,7 @@ jobs:
7072
- if: failure()
7173
uses: actions/upload-artifact@v4
7274
with:
73-
name: error-log-macos
75+
name: error-log-${{ matrix.target.runner }}
7476
path: ${{ github.workspace }}/hs_err_pid*.log
7577
if-no-files-found: warn
7678

0 commit comments

Comments
 (0)