Extended Description

It seems that __ARM_NEON is not defined by clang even when -mfpu=neon and the correct target is used. This generate error when including <arm_neon.h> but if I define __ARM_NEON manually, the program compiles and generate correct code.

Native compiling on ARM seems to be fine.

Comparing to arm-linux-gnueabihf-gcc and native ARM clang, it seems that at least __ARM_NEON__ is also missing.

Cross compiling clang

% clang --target=arm-linux-gnueabihf -mfpu=neon -dM -E - < /dev/null | grep -i neon

GCC

% arm-linux-gnueabihf-gcc -mfpu=neon -dM -E - < /dev/null | grep -i neon 
#define __ARM_NEON_FP 4
#define __ARM_NEON__ 1
#define __ARM_NEON 1

Native clang

$ clang -dM -E - < /dev/null | grep -i neon
#define __ARM_NEON 1
#define __ARM_NEON__ 1

Example code (taken from stackoverflow)

#include <arm_neon.h>

void
runTest(const float vector[], float result[])
{
    float32x4_t vA = vld1q_f32(vector);
    vA = vmulq_f32(vA, vA);
    vst1q_f32(result, vA);
}

A single line #include <arm_neon.h> is also enough to trigger the issue.

System:
clang 3.6.0 (and 3.5.0) from official ArchLinux x86_64 repo.