Eval bug: Unresolved Symbol <__memcpy_chk> when running (any?) model

[rubin55]

Name and Version

version: 8541 (ded446b)
built with IntelLLVM 2025.3.2 for Linux x86_64

Operating systems

Linux

GGML backends

SYCL

Hardware

Intel 1370p, Intel Xe

Models

HauhauCS/Qwen3.5-9B-Uncensored-HauhauCS-Aggressive:Q4_K_M

Problem description & steps to reproduce

When I attempt to run the model (probably any model), I get the following error: Unresolved Symbol <__memcpy_chk>.

How I've built llama.cpp:

  source /opt/intel/oneapi/setvars.sh
 
  local _cmake_options=(
    -B build
    -S .
    -DCMAKE_BUILD_TYPE=Release
    -DCMAKE_INSTALL_PREFIX='/usr'
    -DBUILD_SHARED_LIBS=ON
    -DLLAMA_BUILD_TESTS=OFF
    -DLLAMA_USE_SYSTEM_GGML=OFF
    -DGGML_ALL_WARNINGS=OFF
    -DGGML_ALL_WARNINGS_3RD_PARTY=OFF
    -DGGML_BUILD_EXAMPLES=OFF
    -DGGML_BUILD_TESTS=OFF
    -DGGML_OPENMP=ON
    -DGGML_LTO=ON
    -DGGML_RPC=ON
    -DCMAKE_C_COMPILER=icx
    -DCMAKE_CXX_COMPILER=icpx
    -DGGML_SYCL=ON
    -DGGML_SYCL_F16=ON
    -DLLAMA_BUILD_SERVER=ON
    -DLLAMA_OPENSSL=ON
    -DGGML_NATIVE=ON
    -Wno-dev
  )

  cmake "${_cmake_options[@]}"
  cmake --build build

First Bad Commit

No response

Relevant log output

$ source /opt/intel/oneapi/setvars.sh

:: initializing oneAPI environment ...
   -bash: BASH_VERSION = 5.3.9(1)-release
   args: Using "$@" for setvars.sh arguments:
:: advisor -- latest
:: ccl -- latest
:: compiler -- latest
:: dal -- latest
:: debugger -- latest
:: dev-utilities -- latest
:: dnnl -- latest
:: dpcpp-ct -- latest
:: dpl -- latest
:: ipp -- latest
:: ippcp -- latest
:: mkl -- latest
:: mpi -- latest
:: tbb -- latest
:: umf -- latest
:: vtune -- latest
:: oneAPI environment initialized ::

$ llama-cli -hf HauhauCS/Qwen3.5-9B-Uncensored-HauhauCS-Aggressive:Q4_K_M
[==================================================] 100%  (5366 MB / 5366 MB)
[==================================================] 100%  (879 MB / 879 MB)

Loading model... -The program was built for 1 devices
Build program log for 'Intel(R) Iris(R) Xe Graphics':
Module <0x118f1270>:  Unresolved Symbol <__memcpy_chk>
Module <0x118f1270>:  Unresolved Symbol <__memcpy_chk>
Exception caught at file:/tmp/makepkg.builddir/llama.cpp-sycl-git/src/llama.cpp/ggml/src/ggml-sycl/ggml-sycl.cpp, line:2794, func:operator()
SYCL error: CHECK_TRY_ERROR(op(ctx, src0, src1, dst, src0_dd_i, src1_ddf_i, src1_ddq_i, dst_dd_i, dev[i].row_low, dev[i].row_high, src1_ncols, src1_padded_col_size, stream)): Exception caught in this line of code.
  in function ggml_sycl_op_mul_mat at /tmp/makepkg.builddir/llama.cpp-sycl-git/src/llama.cpp/ggml/src/ggml-sycl/ggml-sycl.cpp:2794
/tmp/makepkg.builddir/llama.cpp-sycl-git/src/llama.cpp/ggml/src/ggml-sycl/../ggml-sycl/common.hpp:134: SYCL error
|[New LWP 985807]
[New LWP 985806]
[New LWP 984137]
[New LWP 984136]
⚠️ warning: File "/opt/intel/oneapi/compiler/2025.3/lib/libsycl.so.8.0.0-gdb.py" auto-loading has been declined by your `auto-load safe-path' set to "$debugdir:$datadir/auto-load".
To enable execution of this file add
	add-auto-load-safe-path /opt/intel/oneapi/compiler/2025.3/lib/libsycl.so.8.0.0-gdb.py
line to your configuration file "/home/rubin/.gdbinit".
To completely disable this security protection add
	set auto-load safe-path /
line to your configuration file "/home/rubin/.gdbinit".
For more information about this security protection see the
"Auto-loading safe path" section in the GDB manual.  E.g., run from the shell:
	info "(gdb)Auto-loading safe path"
Downloading 20.51 M separate debug info for /usr/lib/libze_loader.so.1...
Downloading 6.06 M separate debug info for /usr/lib/libspdlog.so.1.17...
Downloading 1.12 M separate debug info for /usr/lib/libfmt.so.12...
Downloading 329.14 M separate debug info for /usr/lib/libze_intel_gpu.so.1...
Downloading 2.77 M separate debug info for /usr/lib/libigdgmm.so.12...
Downloading 17.13 M separate debug info for /usr/lib/libze_tracing_layer.so.1...
Downloading 24.42 M separate debug info for /usr/lib/libigdfcl.so.2...
Downloading 2.81 G separate debug info for /usr/lib/libopencl-clang.so.16...
Downloading 5.92 M separate debug info for /usr/lib/libzstd.so.1...
Downloading 2.12 G separate debug info for /usr/lib/libigc.so.2...
Downloading 201.79 M separate debug info for /usr/lib/intel-opencl/libigdrcl.so...
Downloading 2.05 M separate debug info for /usr/lib/libnss_mymachines.so.2...
Downloading 1.56 M separate debug info for /usr/lib/libnss_resolve.so.2...
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/usr/lib/libthread_db.so.1".
Downloading 1.79 K source file /usr/src/debug/glibc/glibc/nptl/../sysdeps/unix/sysv/linux/x86_64/syscall_cancel.S...
__syscall_cancel_arch () at ../sysdeps/unix/sysv/linux/x86_64/syscall_cancel.S:56
56		ret
#0  __syscall_cancel_arch () at ../sysdeps/unix/sysv/linux/x86_64/syscall_cancel.S:56
56		ret
#1  0x00007f59bb0a239c in __internal_syscall_cancel (a1=<optimized out>, a2=<optimized out>, a3=<optimized out>, a4=<optimized out>, a5=a5@entry=0, a6=a6@entry=0, nr=61) at cancellation.c:49
49	  result = __syscall_cancel_arch (&pd->cancelhandling, nr, a1, a2, a3, a4, a5,
#2  0x00007f59bb0a23e4 in __syscall_cancel (a1=<optimized out>, a2=<optimized out>, a3=<optimized out>, a4=<optimized out>, a5=a5@entry=0, a6=a6@entry=0, nr=61) at cancellation.c:75
75	  long int r = __internal_syscall_cancel (a1, a2, a3, a4, a5, a6,
#3  0x00007f59bb11267f in __GI___wait4 (pid=<optimized out>, stat_loc=<optimized out>, options=<optimized out>, usage=<optimized out>) at ../sysdeps/unix/sysv/linux/wait4.c:30
30	  return SYSCALL_CANCEL (wait4, pid, stat_loc, options, usage);
#4  0x00007f59bb92279f in ggml_print_backtrace () from /usr/lib/libggml-base.so.0
#5  0x00007f59bb921f46 in ggml_abort () from /usr/lib/libggml-base.so.0
#6  0x00007f59b723a1fc in ?? () from /usr/lib/libggml-sycl.so.0
#7  0x00007f59b728490f in ?? () from /usr/lib/libggml-sycl.so.0
#8  0x00007f59b726abb4 in ?? () from /usr/lib/libggml-sycl.so.0
#9  0x00007f59b72692a4 in ?? () from /usr/lib/libggml-sycl.so.0
#10 0x00007f59b7267d6e in ?? () from /usr/lib/libggml-sycl.so.0
#11 0x00007f59bb94dd59 in ggml_backend_sched_graph_compute_async () from /usr/lib/libggml-base.so.0
#12 0x00007f59bb669a75 in llama_context::graph_compute(ggml_cgraph*, bool) () from /usr/lib/libllama.so.0
#13 0x00007f59bb669830 in llama_context::process_ubatch(llama_ubatch const&, llm_graph_type, llama_memory_context_i*, ggml_status&) () from /usr/lib/libllama.so.0
#14 0x00007f59bb66b991 in llama_context::decode(llama_batch const&) () from /usr/lib/libllama.so.0
#15 0x00007f59bb6720e3 in llama_decode () from /usr/lib/libllama.so.0
#16 0x0000000000569cf9 in ?? ()
#17 0x00000000004a605a in ?? ()
#18 0x0000000000426234 in ?? ()
#19 0x00007f59bb0366c1 in __libc_start_call_main (main=main@entry=0x423730, argc=argc@entry=3, argv=argv@entry=0x7fffbb43f208) at ../sysdeps/nptl/libc_start_call_main.h:59
59	      result = main (argc, argv, __environ MAIN_AUXVEC_PARAM);
#20 0x00007f59bb0367f9 in __libc_start_main_impl (main=0x423730, argc=3, argv=0x7fffbb43f208, init=<optimized out>, fini=<optimized out>, rtld_fini=<optimized out>, stack_end=0x7fffbb43f1f8) at ../csu/libc-start.c:360
360	  __libc_start_call_main (main, argc, argv MAIN_AUXVEC_PARAM);
#21 0x00000000004234c5 in ?? ()
/[Inferior 1 (process 984135) detached]
/Aborted                    (core dumped) llama-cli -hf HauhauCS/Qwen3.5-9B-Uncensored-HauhauCS-Aggressive:Q4_K_M

[rubin55]

This most likely had to do with _FORTIFY_SOURCE which Arch linux enables by default when building packages. When I unset CFLAGS CXXFLAGS LDFLAGS right before I do the cmake steps, I end up with a working llama.cpp.