ggml : fix build broken with -march=armv9-a on MacOS

[DamonFool]

llama.cpp fails to build with -march=armv9-a on MacOS.

cmake .. -DCMAKE_C_FLAGS=-march=armv9-a \
         -DCMAKE_CXX_FLAGS=-march=armv9-a

The failure is caused by 'sys/prctl.h' file not found.

In file included from /Users/jiefu/llama.cpp/ggml/src/ggml-quants.c:6:
/Users/jiefu/llama.cpp/ggml/src/ggml-cpu/ggml-cpu-impl.h:72:10: fatal error: 'sys/prctl.h' file not found
   72 | #include <sys/prctl.h>
      |          ^~~~~~~~~~~~~
1 error generated.

Similar issue had been reported before: sysown/proxysql#1920 .
Reason: sys/prctl.h does not exist on Mac OS X, only on Linux.

The fix follows what has been done in ggml by adding defined(__linux__) : https://github.com/ggml-org/llama.cpp/blob/master/ggml/src/ggml.c#L72

[slaren]

To use SVE ggml_init_arm_arch_features needs to be implemented to initialize sve_cnt.

[DamonFool]

To use SVE ggml_init_arm_arch_features needs to be implemented to initialize sve_cnt.

Good catch.

But it is hard for me to implement the logic since I don't have a Mac which supports sve.

I would suggest fixing the build failure as soon as possible.
And then implement ggml_init_arm_arch_features in another PR by some sve expert.

Maybe, we can add a #pragma message just like https://github.com/ggml-org/llama.cpp/blob/master/ggml/src/ggml-opencl/ggml-opencl.cpp#L2801

What do you think?
Thanks.

[slaren]

A message would not be enough, it would need to be an #error to stop the build. Enabling SVE without implementing this function will likely lead to errors.

[DamonFool]

A message would not be enough, it would need to be an #error to stop the build.

Done.
Thanks @slaren .

[DamonFool]

Thanks @slaren and @ggerganov .