Fix cuda compilation

[slaren]

Continued from #1127. The -Wno-pedantic is necessary to avoid a warning: style of line directive is a GCC extension for every single line in the compilation.

[hlhr202]

Hi @slaren is there any workaround for cmake cublas fpic?
I m compiling through rust and got such error:
relocation R_X86_64_PC32 against symbol `stderr@@GLIBC_2.2.5' can not be used when making a shared object; recompile with -fPIC

[Green-Sky]

@hlhr202 did you set LLAMA_SHARED ?

[hlhr202]

LLAMA_SHARED

@Green-Sky this will result to another error

/usr/bin/ld: /usr/lib/gcc/x86_64-linux-gnu/11/crtbeginT.o: relocation R_X86_64_32 against hidden symbol `__TMC_END__' can not be used when making a shared object
  /usr/bin/ld: failed to set dynamic section sizes: bad value
  collect2: error: ld returned 1 exit status
  gmake[2]: *** [CMakeFiles/llama.dir/build.make:106: libllama.so] Error 1
  gmake[1]: *** [CMakeFiles/Makefile2:111: CMakeFiles/llama.dir/all] Error 2
  gmake: *** [Makefile:91: all] Error 2
  thread 'main' panicked at 'Failed to build lib', packages/llama-cpp/llama-sys/build.rs:106:9

[Green-Sky]

what if you have BUILD_SHARED_LIBS ON and LLAMA_STATIC OFF ?

[hlhr202]

what if you have BUILD_SHARED_LIBS ON and LLAMA_STATIC OFF ?

I may try dynamically linking later since I am statically linking llama to rust program, so currently I cant turn off LLAMA_STATIC...

[hlhr202]

@Green-Sky Thanks for your suggestion. I v tried shared lib and it works properly. but still if there is a statically linking method, it would be convinient for us to distribute downstream applications.

[tchereau]

Hi @slaren is there any workaround for cmake cublas fpic? I m compiling through rust and got such error: relocation R_X86_64_PC32 against symbol `stderr@@GLIBC_2.2.5' can not be used when making a shared object; recompile with -fPIC

line 65
.arg("-DCMAKE_POSITION_INDEPENDENT_CODE=ON");
work for me