Build error with GCC12: /usr/lib/gcc/aarch64-linux-gnu/12/include/arm_neon.h:29182:1: error: inlining failed in call to ‘always_inline’ ‘vfmaq_f16’: target specific option mismatch

[Li Yi]

When I build Docker image with Apple M1 Silicon. The base image includes with GCC 12.

And I got some build error as following.

In file included from /ctransformers/models/ggml/ggml.c:274:
/usr/lib/gcc/aarch64-linux-gnu/12/include/arm_neon.h: In function ‘ggml_vec_dot_f16’:
/usr/lib/gcc/aarch64-linux-gnu/12/include/arm_neon.h:29182:1: error: inlining failed in call to ‘always_inline’ ‘vfmaq_f16’: target specific option mismatch
29182 | vfmaq_f16 (float16x8_t __a, float16x8_t __b, float16x8_t __c)
      | ^~~~~~~~~
/ctransformers/models/ggml/ggml.c:1884:37: note: called from here
 1884 |     #define GGML_F16x8_FMA(a, b, c) vfmaq_f16(a, b, c)
      |                                     ^~~~~~~~~~~~~~~~~~
/ctransformers/models/ggml/ggml.c:1911:41: note: in expansion of macro ‘GGML_F16x8_FMA’
 1911 |     #define GGML_F16_VEC_FMA            GGML_F16x8_FMA
      |                                         ^~~~~~~~~~~~~~
/ctransformers/models/ggml/ggml.c:2408:22: note: in expansion of macro ‘GGML_F16_VEC_FMA’
 2408 |             sum[j] = GGML_F16_VEC_FMA(sum[j], ax[j], ay[j]);
      |                      ^~~~~~~~~~~~~~~~
/usr/lib/gcc/aarch64-linux-gnu/12/include/arm_neon.h:29182:1: error: inlining failed in call to ‘always_inline’ ‘vfmaq_f16’: target specific option mismatch
29182 | vfmaq_f16 (float16x8_t __a, float16x8_t __b, float16x8_t __c)
      | ^~~~~~~~~
/ctransformers/models/ggml/ggml.c:1884:37: note: called from here
 1884 |     #define GGML_F16x8_FMA(a, b, c) vfmaq_f16(a, b, c)
      |                                     ^~~~~~~~~~~~~~~~~~
/ctransformers/models/ggml/ggml.c:1911:41: note: in expansion of macro ‘GGML_F16x8_FMA’
 1911 |     #define GGML_F16_VEC_FMA            GGML_F16x8_FMA
      |                                         ^~~~~~~~~~~~~~
/ctransformers/models/ggml/ggml.c:2408:22: note: in expansion of macro ‘GGML_F16_VEC_FMA’
 2408 |             sum[j] = GGML_F16_VEC_FMA(sum[j], ax[j], ay[j]);
      |                      ^~~~~~~~~~~~~~~~
/usr/lib/gcc/aarch64-linux-gnu/12/include/arm_neon.h:28760:1: error: inlining failed in call to ‘always_inline’ ‘vaddq_f16’: target specific option mismatch
28760 | vaddq_f16 (float16x8_t __a, float16x8_t __b)
      | ^~~~~~~~~
/ctransformers/models/ggml/ggml.c:1891:20: note: called from here
 1891 |             x[i] = vaddq_f16(x[i], x[offset+i]);                  \
      |                    ^~~~~~~~~~~~~~~~~~~~~~~~~~~~
/ctransformers/models/ggml/ggml.c:1914:41: note: in expansion of macro ‘GGML_F16x8_REDUCE’
 1914 |     #define GGML_F16_VEC_REDUCE         GGML_F16x8_REDUCE
      |                                         ^~~~~~~~~~~~~~~~~
/ctransformers/models/ggml/ggml.c:2413:5: note: in expansion of macro ‘GGML_F16_VEC_REDUCE’
 2413 |     GGML_F16_VEC_REDUCE(sumf, sum);
      |     ^~~~~~~~~~~~~~~~~~~
/usr/lib/gcc/aarch64-linux-gnu/12/include/arm_neon.h:28760:1: error: inlining failed in call to ‘always_inline’ ‘vaddq_f16’: target specific option mismatch
28760 | vaddq_f16 (float16x8_t __a, float16x8_t __b)
      | ^~~~~~~~~
/ctransformers/models/ggml/ggml.c:1895:20: note: called from here
 1895 |             x[i] = vaddq_f16(x[i], x[offset+i]);                  \
      |                    ^~~~~~~~~~~~~~~~~~~~~~~~~~~~
/ctransformers/models/ggml/ggml.c:1914:41: note: in expansion of macro ‘GGML_F16x8_REDUCE’
 1914 |     #define GGML_F16_VEC_REDUCE         GGML_F16x8_REDUCE
      |                                         ^~~~~~~~~~~~~~~~~
/ctransformers/models/ggml/ggml.c:2413:5: note: in expansion of macro ‘GGML_F16_VEC_REDUCE’
 2413 |     GGML_F16_VEC_REDUCE(sumf, sum);
      |     ^~~~~~~~~~~~~~~~~~~
/usr/lib/gcc/aarch64-linux-gnu/12/include/arm_neon.h:28760:1: error: inlining failed in call to ‘always_inline’ ‘vaddq_f16’: target specific option mismatch
28760 | vaddq_f16 (float16x8_t __a, float16x8_t __b)
      | ^~~~~~~~~
/ctransformers/models/ggml/ggml.c:1899:20: note: called from here
 1899 |             x[i] = vaddq_f16(x[i], x[offset+i]);                  \
      |                    ^~~~~~~~~~~~~~~~~~~~~~~~~~~~
/ctransformers/models/ggml/ggml.c:1914:41: note: in expansion of macro ‘GGML_F16x8_REDUCE’
 1914 |     #define GGML_F16_VEC_REDUCE         GGML_F16x8_REDUCE
      |                                         ^~~~~~~~~~~~~~~~~
/ctransformers/models/ggml/ggml.c:2413:5: note: in expansion of macro ‘GGML_F16_VEC_REDUCE’
 2413 |     GGML_F16_VEC_REDUCE(sumf, sum);
      |     ^~~~~~~~~~~~~~~~~~~
make[2]: *** [CMakeFiles/ctransformers.dir/build.make:90: CMakeFiles/ctransformers.dir/models/ggml/ggml.c.o] Error 1
make[1]: *** [CMakeFiles/Makefile2:83: CMakeFiles/ctransformers.dir/all] Error 2

It is similar to serge-chat/serge#64

And after I roll back the GCC to GCC11, it works