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.

[waltersamson]

I updated vcredist 2015-2022 to version 14.42.34433.0 and it works for me now.

[bqhuyy]

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

I updated vcredist 2015-2022 to version 14.42.34433.0 and it works for me now.

This works like a charm. Thanks a lot

[Ikaron]

Hey, I tried updating the vcredist but I've also gone through the dump (it is a crash) and found the offending line to be ggml/src/ggml-sycl/dpct/helper.hpp line 953
if (result.empty()) throw std::runtime_error("can not find preferred GPU platform");
Basically, no sycl compatible GPU is found, but this doesn't cause any command line output, leading to the behaviour. Some text like "No GPU found" would be very helpful here. The runtime error thrown never ends up showing in the console.