Maratyszcza / NNPACK

Acceleration package for neural networks on multi-core CPUs
BSD 2-Clause "Simplified" License
1.68k stars 317 forks source link

Build error with ndk-build #92

Closed ilisc2 closed 7 years ago

ilisc2 commented 7 years ago

I am trying to build with ndk, but got the following errors.

In file included from jni/../include/nnpack/activations.h:25: jni/../include/nnpack/arm_neon.h:26:10: warning: implicit declaration of function 'vcvt_f32_f16' is invalid in C99 [-Wimplicit-function-declaration] return vcvt_f32_f16(vld1_f16((const fp16*) address)); ^ jni/../include/nnpack/arm_neon.h:26:10: error: returning 'int' from a function with incompatible result type 'float32x4_t' (vector of 4 'float32_t' values) return vcvt_f32_f16(vld1_f16((const fp16) address)); ^~~~~~~~~~~ jni/../include/nnpack/arm_neon.h:30:10: error: returning 'int' from a function with incompatible result type 'float32x4_t' (vector of 4 'float32_t' values) return vcvt_f32_f16(vld1_f16((const __fp16) ^~~~~~~~~ jni/../include/nnpack/arm_neon.h:35:31: warning: implicit declaration of function 'vcvt_f16_f32' is invalid in C99 [-Wimplicit-function-declaration] vst1_f16((fp16*) address, vcvt_f16_f32(vector)); ^ jni/../include/nnpack/arm_neon.h:35:3: error: initializing 'float16x4_t' (vector of 4 'float16_t' values) with an expression of incompatible type 'int' vst1_f16((fp16) address, vcvt_f16_f32(vector)); ^ ~~~~ /Users/ivy.li/dev/tools/android-ndk-r15b/toolchains/llvm/prebuilt/darwin-x86_64/lib64/clang/5.0.300080/include/arm_neon.h:25224:15: note: expanded from macro 'vst1_f16' float16x4_t s1 = p1; \ ^ ~~~~ In file included from jni/../jni/../src/psimd/2d-fourier-8x8.c:6: In file included from jni/../include/nnpack/activations.h:25: jni/../include/nnpack/arm_neon.h:39:3: error: initializing 'float16x4_t' (vector of 4 'float16_t' values) with an expression of incompatible type 'int' vst1_f16((__fp16) __builtin_assume_aligned(address, sizeof(float16x4_t)), ^ /Users/ivy.li/dev/tools/android-ndk-r15b/toolchains/llvm/prebuilt/darwin-x86_64/lib64/clang/5.0.300080/include/arm_neon.h:25224:15: note: expanded from macro 'vst1_f16' float16x4_t s1 = p1; \