Misc. bug: All llama executables exit immediately without console output

[Ikaron]

Name and Version

Multiple.
SYCL build as well as CPU only build from git rev-parse head eb5c3dc
Also the prebuilt Windows SYCL binary 3bcd40b tag b4040
NOTE:
Prebuilt binary fb76ec3 tag b3038 WORKS!

Operating systems

Windows

Which llama.cpp modules do you know to be affected?

llama-cli, llama-server

Problem description & steps to reproduce

I originally thought this was a problem with the SYCL builds, but I also compiled CPU only with the same result. Note that "main.exe" from the old SYCL prebuilt works as expected.

Not sure if relevant but I don't recognise the OpenMP installs that were found.

Steps to reproduce:

1. Install or build llama-cpp (CPU/SYCL)
   Build via any:
   ./examples/sycl/win-build-sycl.bat

cmake -B build -G "Ninja" -DGGML_SYCL=ON -DCMAKE_C_COMPILER=cl -DCMAKE_CXX_COMPILER=icx -DCMAKE_BUILD_TYPE=Release -DGGML_SYCL_F16=ON
cmake --build build --config Release -j

(Also, build warning:)
icx: warning: unknown argument ignored in clang-cl: '-machine:x64' [-Wunknown-argument]

2. Run any llama executable (llama-cli, llama-ls-sycl-device, llama-server, etc.)
   Use the "Intel oneAPI command prompt for Intel 64 for Visual Studio 2022" command prompt, then run any llama exe
   OR
   Execute "C:\Program Files (x86)\Intel\oneAPI\setvars.bat" intel64, then run any llama exe
   OR
   Run examples\sycl\win-run-llama2.bat

Build config (CPU):

B:\LLM\llama-src\llama.cpp>cmake -B build
-- Building for: Visual Studio 17 2022
-- Selecting Windows SDK version 10.0.22621.0 to target Windows 10.0.22631.
-- The C compiler identification is MSVC 19.41.34120.0
-- The CXX compiler identification is MSVC 19.41.34120.0
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Check for working C compiler: C:/Program Files/Microsoft Visual Studio/2022/Community/VC/Tools/MSVC/14.41.34120/bin/Hostx64/x64/cl.exe - skipped
-- Detecting C compile features
-- Detecting C compile features - done
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Check for working CXX compiler: C:/Program Files/Microsoft Visual Studio/2022/Community/VC/Tools/MSVC/14.41.34120/bin/Hostx64/x64/cl.exe - skipped
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- Found Git: C:/Program Files/Git/cmd/git.exe (found version "2.47.1.windows.1")
-- Performing Test CMAKE_HAVE_LIBC_PTHREAD
-- Performing Test CMAKE_HAVE_LIBC_PTHREAD - Failed
-- Looking for pthread_create in pthreads
-- Looking for pthread_create in pthreads - not found
-- Looking for pthread_create in pthread
-- Looking for pthread_create in pthread - not found
-- Found Threads: TRUE
-- Warning: ccache not found - consider installing it for faster compilation or disable this warning with GGML_CCACHE=OFF
-- CMAKE_SYSTEM_PROCESSOR: AMD64
-- CMAKE_GENERATOR_PLATFORM:
-- Including CPU backend
-- Found OpenMP_C: -openmp (found version "2.0")
-- Found OpenMP_CXX: -openmp (found version "2.0")
-- Found OpenMP: TRUE (found version "2.0")
-- x86 detected
-- Performing Test HAS_AVX_1
-- Performing Test HAS_AVX_1 - Success
-- Performing Test HAS_AVX2_1
-- Performing Test HAS_AVX2_1 - Success
-- Performing Test HAS_FMA_1
-- Performing Test HAS_FMA_1 - Success
-- Performing Test HAS_AVX512_1
-- Performing Test HAS_AVX512_1 - Failed
-- Performing Test HAS_AVX512_2
-- Performing Test HAS_AVX512_2 - Failed
-- Adding CPU backend variant ggml-cpu: /arch:AVX2 GGML_AVX2;GGML_FMA;GGML_F16C
-- Configuring done (12.9s)
-- Generating done (0.9s)
-- Build files have been written to: B:/LLM/llama-src/llama.cpp/build

Build config (SYCL):

B:\LLM\llama-src\llama.cpp>cmake -B build -G "Ninja" -DGGML_SYCL=ON -DCMAKE_C_COMPILER=cl -DCMAKE_CXX_COMPILER=icx  -DCMAKE_BUILD_TYPE=Release -DGGML_SYCL_F16=ON
-- The C compiler identification is MSVC 19.41.34120.0
-- The CXX compiler identification is IntelLLVM 2025.0.4 with MSVC-like command-line
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Check for working C compiler: C:/Program Files/Microsoft Visual Studio/2022/Community/VC/Tools/MSVC/14.41.34120/bin/Hostx64/x64/cl.exe - skipped
-- Detecting C compile features
-- Detecting C compile features - done
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Check for working CXX compiler: C:/Program Files (x86)/Intel/oneAPI/compiler/latest/bin/icx.exe - skipped
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- Found Git: C:/Program Files/Git/cmd/git.exe (found version "2.47.1.windows.1")
-- Performing Test CMAKE_HAVE_LIBC_PTHREAD
-- Performing Test CMAKE_HAVE_LIBC_PTHREAD - Failed
-- Looking for pthread_create in pthreads
-- Looking for pthread_create in pthreads - not found
-- Looking for pthread_create in pthread
-- Looking for pthread_create in pthread - not found
-- Found Threads: TRUE
-- Warning: ccache not found - consider installing it for faster compilation or disable this warning with GGML_CCACHE=OFF
-- CMAKE_SYSTEM_PROCESSOR: AMD64
-- CMAKE_GENERATOR_PLATFORM:
-- Including CPU backend
-- Found OpenMP_C: -openmp (found version "2.0")
-- Found OpenMP_CXX: -Qiopenmp (found version "5.1")
-- Found OpenMP: TRUE (found version "2.0")
-- x86 detected
-- Performing Test HAS_AVX_1
-- Performing Test HAS_AVX_1 - Success
-- Performing Test HAS_AVX2_1
-- Performing Test HAS_AVX2_1 - Success
-- Performing Test HAS_FMA_1
-- Performing Test HAS_FMA_1 - Success
-- Performing Test HAS_AVX512_1
-- Performing Test HAS_AVX512_1 - Failed
-- Performing Test HAS_AVX512_2
-- Performing Test HAS_AVX512_2 - Failed
-- Adding CPU backend variant ggml-cpu: /arch:AVX2 GGML_AVX2;GGML_FMA;GGML_F16C
-- Performing Test SUPPORTS_SYCL
-- Performing Test SUPPORTS_SYCL - Success
-- Using oneAPI Release SYCL compiler (icpx).
-- SYCL found
-- DNNL found:1
-- Found IntelSYCL: C:/Program Files (x86)/Intel/oneAPI/compiler/latest/include (found version "202001")
-- MKL_VERSION: 2025.0.1
-- MKL_ROOT: C:/Program Files (x86)/Intel/oneAPI/mkl/latest
-- MKL_ARCH: intel64
-- MKL_SYCL_LINK: None, set to ` dynamic` by default
-- MKL_LINK: None, set to ` dynamic` by default
-- MKL_SYCL_INTERFACE_FULL: None, set to ` intel_ilp64` by default
-- MKL_INTERFACE_FULL: None, set to ` intel_ilp64` by default
-- MKL_SYCL_THREADING: None, set to ` tbb_thread` by default
-- MKL_THREADING: None, set to ` intel_thread` by default
-- MKL_MPI: None, set to ` intelmpi` by default
-- Found C:/Program Files (x86)/Intel/oneAPI/mkl/latest/lib/mkl_scalapack_ilp64_dll.lib
-- Found DLL: C:/Program Files (x86)/Intel/oneAPI/mkl/latest/bin/mkl_scalapack_ilp64.2.dll
-- Found C:/Program Files (x86)/Intel/oneAPI/mkl/latest/lib/mkl_cdft_core_dll.lib
-- Found DLL: C:/Program Files (x86)/Intel/oneAPI/mkl/latest/bin/mkl_cdft_core.2.dll
-- Found C:/Program Files (x86)/Intel/oneAPI/mkl/latest/lib/mkl_intel_ilp64_dll.lib
-- Found C:/Program Files (x86)/Intel/oneAPI/mkl/latest/lib/mkl_intel_thread_dll.lib
-- Found DLL: C:/Program Files (x86)/Intel/oneAPI/mkl/latest/bin/mkl_intel_thread.2.dll
-- Found C:/Program Files (x86)/Intel/oneAPI/mkl/latest/lib/mkl_core_dll.lib
-- Found DLL: C:/Program Files (x86)/Intel/oneAPI/mkl/latest/bin/mkl_core.2.dll
-- Found C:/Program Files (x86)/Intel/oneAPI/mkl/latest/lib/mkl_blacs_ilp64_dll.lib
-- Found DLL: C:/Program Files (x86)/Intel/oneAPI/mkl/latest/bin/mkl_blacs_ilp64.2.dll
-- Found C:/Program Files (x86)/Intel/oneAPI/mkl/latest/lib/mkl_sycl_blas_dll.lib
-- Found DLL: C:/Program Files (x86)/Intel/oneAPI/mkl/latest/bin/mkl_sycl_blas.5.dll
-- Found C:/Program Files (x86)/Intel/oneAPI/mkl/latest/lib/mkl_sycl_lapack_dll.lib
-- Found DLL: C:/Program Files (x86)/Intel/oneAPI/mkl/latest/bin/mkl_sycl_lapack.5.dll
-- Found C:/Program Files (x86)/Intel/oneAPI/mkl/latest/lib/mkl_sycl_dft_dll.lib
-- Found DLL: C:/Program Files (x86)/Intel/oneAPI/mkl/latest/bin/mkl_sycl_dft.5.dll
-- Found C:/Program Files (x86)/Intel/oneAPI/mkl/latest/lib/mkl_sycl_sparse_dll.lib
-- Found DLL: C:/Program Files (x86)/Intel/oneAPI/mkl/latest/bin/mkl_sycl_sparse.5.dll
-- Found C:/Program Files (x86)/Intel/oneAPI/mkl/latest/lib/mkl_sycl_data_fitting_dll.lib
-- Found DLL: C:/Program Files (x86)/Intel/oneAPI/mkl/latest/bin/mkl_sycl_data_fitting.5.dll
-- Found C:/Program Files (x86)/Intel/oneAPI/mkl/latest/lib/mkl_sycl_rng_dll.lib
-- Found DLL: C:/Program Files (x86)/Intel/oneAPI/mkl/latest/bin/mkl_sycl_rng.5.dll
-- Found C:/Program Files (x86)/Intel/oneAPI/mkl/latest/lib/mkl_sycl_stats_dll.lib
-- Found DLL: C:/Program Files (x86)/Intel/oneAPI/mkl/latest/bin/mkl_sycl_stats.5.dll
-- Found C:/Program Files (x86)/Intel/oneAPI/mkl/latest/lib/mkl_sycl_vm_dll.lib
-- Found DLL: C:/Program Files (x86)/Intel/oneAPI/mkl/latest/bin/mkl_sycl_vm.5.dll
-- Found C:/Program Files (x86)/Intel/oneAPI/mkl/latest/lib/mkl_tbb_thread_dll.lib
-- Found DLL: C:/Program Files (x86)/Intel/oneAPI/mkl/latest/bin/mkl_tbb_thread.2.dll
-- Found C:/Program Files (x86)/Intel/oneAPI/compiler/latest/lib/libiomp5md.lib
-- Including SYCL backend
-- Configuring done (10.0s)
-- Generating done (0.2s)
-- Build files have been written to: B:/LLM/llama-src/llama.cpp/build

First Bad Commit

No response

Relevant log output

B:\LLM\llama-src\llama.cpp\build\bin> llama-cli --version

B:\LLM\llama-src\llama.cpp\build\bin>

B:\LLM\llama-src\llama.cpp\build\bin> llama-ls-sycl-device

B:\LLM\llama-src\llama.cpp\build\bin>

etc.

[qnixsynapse]

What happens when you append --log-verbose to llama-server?

Unfortunately, I don't have a windows machine to test.

[waltersamson]

What happens when you append --log-verbose to llama-server?

Unfortunately, I don't have a windows machine to test.

Sorry for the duplicated issue #10944

--log-verbose has no effect. I don't think the code advanced to the point that it can take cmdline params. It fails much earlier.

Also, some binaries do work, such as llama-gguf.exe.

[qnixsynapse]

If you are able to share a stack trace, it would be very helpful. The stack trace would allow us to pinpoint the issue.

[easyfab]

I had a similar problem a while ago if I remember correctly I had to reinstall Microsoft Visual C++ redistributable.
Not sure it's the same here but you can try.

[slaren]

Can you check if #10960 solves this problem?

[waltersamson]

Can you check if #10960 solves this problem?

Now that I see b4388 contains this commit, I tried llama-b4388-bin-win-avx2-x64.zip, and unfortunately no, it still doesn't work.

I tried running b4388 llama-server.exe with NTtrace, and got something, not sure if it'll help:
nttrace.txt

[NeoZhangJianyu]

@Ikaron
Maybe you missed to activate oneAPI running time.

Could you run following cmd before execute the tool?

call "c:\Program Files (x86)\Intel\oneAPI\setvars.bat"

[hajimeg3]

I had a similar problem a while ago if I remember correctly I had to reinstall Microsoft Visual C++ redistributable. Not sure it's the same here but you can try.

I had the same situation too, and reinstall MSVC++ redist resolved this
though I had installed Visual Studio 2022 community.