Eval bug: CANNOT LINK EXECUTABLE "./llama-cli": library "libomp.so" not found: needed by main executable

[Krallbe68]

Name and Version

./llama-cli --version
CANNOT LINK EXECUTABLE "./llama-cli": library "libomp.so" not found: needed by main executable

Operating systems

Other? (Please let us know in description)

GGML backends

Vulkan

Hardware

Qualcomm snapdragon 8 gen 1
I use opencl!!!
I am using termux

Models

Llama 3.21b q4

Problem description & steps to reproduce

./llama-cli -m /sdcard/download/0.gguf CANNOT LINK EXECUTABLE "./llama-cli": library "libomp.so" not found: needed by main executable when I try to use any binary package I'm getting this error

First Bad Commit

No response

Relevant log output

ls
android-ndk-r27b  android-ndk-r27b-aarch64.zip  dev  llama.cpp  st.sh
~ $ cd dev
~/dev $ cd llm
~/dev/llm $ ls
OpenCL-Headers  OpenCL-ICD-Loader  llama.cpp
~/dev/llm $ cd llama.cpp
~/dev/llm/llama.cpp $ l
No command l found, did you mean:
 Command ld in package binutils
 Command ld in package binutils-is-llvm
 Command c in package c-script
 Command [ in package coreutils
 Command lp in package cups
 Command dl in package gatling
 Command k in package kona
 Command lf in package lf
 Command pl in package libgnustep-base
 Command lr in package lr
 Command ml in package mercury
 Command al in package mono
 Command lz in package mtools
 Command o in package o-editor
 Command ol in package ol
 Command q in package q-dns-client
 Command sl in package sl
 Command ul in package util-linux
 Command X in package xorg-server
~/dev/llm/llama.cpp $ ls
AUTHORS            build-android                  flake.nix    pyproject.toml
CMakeLists.txt     ci                             ggml         pyrightconfig.json
CMakePresets.json  cmake                          gguf-py      requirements
CODEOWNERS         common                         grammars     requirements.txt
CONTRIBUTING.md    convert_hf_to_gguf.py          include      scripts
LICENSE            convert_hf_to_gguf_update.py   media        spm-headers
Makefile           convert_llama_ggml_to_gguf.py  models       src
Package.swift      convert_lora_to_gguf.py        mypy.ini     tests
README.md          docs                           pocs
SECURITY.md        examples                       poetry.lock
Sources            flake.lock                     prompts
~/dev/llm/llama.cpp $ build-android
build-android: command not found
~/dev/llm/llama.cpp $ cd build-android
~/.../llama.cpp/build-android $ ls
CMakeCache.txt         bin                    ggml                  pocs
CMakeFiles             build.ninja            install_manifest.txt  src
CTestTestfile.cmake    cmake_install.cmake    lib                   tests               DartConfiguration.tcl  common                 llama-config.cmake
Testing                compile_commands.json  llama-version.cmake
autogenerated          examples               llama.pc                                  ~/.../llama.cpp/build-android $ cd bi
bash: cd: bi: No such file or directory
~/.../llama.cpp/build-android $ cd bin
~/.../build-android/bin $ ls
llama-batched                  llama-lookup-merge        test-autorelease               llama-batched-bench            llama-lookup-stats        test-backend-ops
llama-bench                    llama-minicpmv-cli        test-barrier
llama-cli                      llama-parallel            test-c
llama-convert-llama2c-to-ggml  llama-passkey             test-chat
llama-cvector-generator        llama-perplexity          test-chat-template
llama-embedding                llama-q8dot               test-gguf
llama-eval-callback            llama-quantize            test-grammar-integration
llama-export-lora              llama-quantize-stats      test-grammar-parser
llama-gbnf-validator           llama-qwen2vl-cli         test-json-schema-to-grammar
llama-gen-docs                 llama-retrieval           test-llama-grammar
llama-gguf                     llama-run                 test-log
llama-gguf-hash                llama-save-load-state     test-model-load-cancel
llama-gguf-split               llama-server              test-quantize-fns
llama-gritlm                   llama-simple              test-quantize-perf
llama-imatrix                  llama-simple-chat         test-rope
llama-infill                   llama-speculative         test-sampling
llama-llava-cli                llama-speculative-simple  test-tokenizer-0
llama-llava-clip-quantize-cli  llama-tokenize            test-tokenizer-1-bpe
llama-lookahead                llama-tts                 test-tokenizer-1-spm
llama-lookup                   llama-vdot
llama-lookup-create            test-arg-parser
~/.../build-android/bin $ ./llama-cli -m /sdcard/download/0.gguf
CANNOT LINK EXECUTABLE "./llama-cli": library "libomp.so" not found: needed by main executable
~/.../build-android/bin $

[ngxson]

According to https://github.com/ggml-org/llama.cpp/blob/master/docs/android.md

You can turn it off by -DGGML_OPENMP=OFF

$ cmake \
  -DCMAKE_TOOLCHAIN_FILE=$ANDROID_NDK/build/cmake/android.toolchain.cmake \
  -DANDROID_ABI=arm64-v8a \
  -DANDROID_PLATFORM=android-28 \
  -DCMAKE_C_FLAGS="-march=armv8.7a" \
  -DCMAKE_CXX_FLAGS="-march=armv8.7a" \
  -DGGML_OPENMP=OFF \
  -DGGML_LLAMAFILE=OFF \
  -B build-android

[Krallbe68]

cmake .. -G Ninja
-DCMAKE_TOOLCHAIN_FILE=$ANDROID_NDK/build/cmake/android.toolchain.cmake
-DANDROID_ABI=arm64-v8a
-DANDROID_PLATFORM=android-28
-DGGML_OPENMP=OFF
-DGGML_LLAMAFILE=OFF
-DGGML_OPENCL=ON

I used this code to compile again and this time I get an error like this: ~/.../build-android/bin $ ./llama-cli CANNOT LINK EXECUTABLE "./llama-cli": library "libOpenCL.so.1" not found: needed by main executable

[belog2867]

If you use termux

LD_LIBRARY_PATH=/vendor/lib64:$PREFIX/lib ./llama-cli

[belog2867]

If it still doesn't work, maybe your mobile phone doesn't support opencl.

[Krallbe68]

My device is Xiaomi Mi 10t Pro, it has Snapdragon 865 processor and Adreno 650 graphics unit. According to what I looked on the internet, it supports opencl 2.0, but there is no libopencl.co in vendor/lib64.