Compile bug:

[EAddario]

Git commit

2bf0415

Operating systems

Mac

GGML backends

Metal

Problem description & steps to reproduce

When building #27815 with BUILD_SHARED_LIBS=OFF ggml-rpc is a static archive and never links so librdma symbols stay undefined.

Fix is simple, in ggml/src/ggml-rpc/CMakeLists.txt change target_link_options(ggml-rpc PRIVATE "LINKER:-weak_library,${RDMA_LIB}") with target_link_options(ggml-rpc PUBLIC "LINKER:-weak_library,${RDMA_LIB}")

First Bad Commit

2bf0415

Compile command

cmake --build build --config Release

Relevant log output

-- llama.cpp version: 0.3.0-dev
CMAKE_BUILD_TYPE=Release
-- ccache found, compilation results will be cached. Disable with GGML_CCACHE=OFF.
-- CMAKE_SYSTEM_PROCESSOR: arm64
-- GGML_SYSTEM_ARCH: ARM
-- Including CPU backend
-- Accelerate framework found
-- ARM detected
CMake Warning at ggml/src/ggml-cpu/CMakeLists.txt:138 (message):
  ARM -march/-mcpu not found, -mcpu=native will be used
Call Stack (most recent call first):
  ggml/src/CMakeLists.txt:584 (ggml_add_cpu_backend_variant_impl)


-- Checking for ARM features using flags:
--   -U__ARM_FEATURE_SVE
--   -mcpu=native+dotprod+i8mm+nosve+sme
-- Adding CPU backend variant ggml-cpu: -U__ARM_FEATURE_SVE;-mcpu=native+dotprod+i8mm+nosve+sme
-- BLAS found, Libraries: /Library/Developer/CommandLineTools/SDKs/MacOSX.sdk/System/Library/Frameworks/Accelerate.framework
-- BLAS found, Includes:
-- Including BLAS backend
-- Metal framework found
-- Including METAL backend
-- Using RPC backend
--   RDMA transport enabled (Apple RDMA-over-Thunderbolt, UC)
-- Including RPC backend
-- ggml version: 0.22.0
-- ggml commit:  2bf041515
-- OpenSSL found: 3.6.4
-- Generating embedded license file for target: llama-app
-- Configuring done (0.2s)
-- Generating done (0.3s)
-- Build files have been written to: /Users/ed/Development/C++/llama.cpp/build
[  1%] Built target llama-gemma3-cli
[  2%] Building CXX object common/CMakeFiles/llama-common-base.dir/build-info.cpp.o

...

[ 18%] Linking CXX executable ../../bin/llama-gguf
[ 18%] Linking CXX executable ../../bin/llama-gguf-hash
[ 19%] Linking CXX executable ../../bin/ggml-rpc-server
[ 20%] Linking CXX executable ../../bin/ggml-metal-tuning
Undefined symbols for architecture arm64:
  "_ibv_alloc_pd", referenced from:
      apple_rdma::probe(int, unsigned char const*, unsigned char*) in libggml-rpc.a[4](transport-apple.cpp.o)
  "_ibv_close_device", referenced from:
      apple_rdma::probe(int, unsigned char const*, unsigned char*) in libggml-rpc.a[4](transport-apple.cpp.o)
      apple_rdma::impl::~impl() in libggml-rpc.a[4](transport-apple.cpp.o)

...

ld: symbol(s) not found for architecture arm64
clang++: error: linker command failed with exit code 1 (use -v to see invocation)

...

[ 50%] Linking CXX static library libllama.a
[ 50%] Built target llama
-- UI: archive verified and extracted
-- UI: HF download succeeded, stamp updated (b10694)
-- UI: gzip compression applied (/Users/ed/Development/C++/llama.cpp/build/tools/ui/dist/_gzip)
embed: write output file /Users/ed/Development/C++/llama.cpp/build/tools/ui/ui.cpp
[ 50%] Built target llama-ui-assets
make: *** [all] Error 2