Merge pull request #3384 from ilya-lavrenov:neon_new
This commit is contained in:
@@ -605,6 +605,48 @@ inline uint32x4_t cv_vrndq_u32_f32(float32x4_t v)
|
||||
return vcvtq_u32_f32(vaddq_f32(v, v_05));
|
||||
}
|
||||
|
||||
inline float32x4_t cv_vrecpq_f32(float32x4_t val)
|
||||
{
|
||||
float32x4_t reciprocal = vrecpeq_f32(val);
|
||||
reciprocal = vmulq_f32(vrecpsq_f32(val, reciprocal), reciprocal);
|
||||
reciprocal = vmulq_f32(vrecpsq_f32(val, reciprocal), reciprocal);
|
||||
return reciprocal;
|
||||
}
|
||||
|
||||
inline float32x2_t cv_vrecp_f32(float32x2_t val)
|
||||
{
|
||||
float32x2_t reciprocal = vrecpe_f32(val);
|
||||
reciprocal = vmul_f32(vrecps_f32(val, reciprocal), reciprocal);
|
||||
reciprocal = vmul_f32(vrecps_f32(val, reciprocal), reciprocal);
|
||||
return reciprocal;
|
||||
}
|
||||
|
||||
inline float32x4_t cv_vrsqrtq_f32(float32x4_t val)
|
||||
{
|
||||
float32x4_t e = vrsqrteq_f32(val);
|
||||
e = vmulq_f32(vrsqrtsq_f32(vmulq_f32(e, e), val), e);
|
||||
e = vmulq_f32(vrsqrtsq_f32(vmulq_f32(e, e), val), e);
|
||||
return e;
|
||||
}
|
||||
|
||||
inline float32x2_t cv_vrsqrt_f32(float32x2_t val)
|
||||
{
|
||||
float32x2_t e = vrsqrte_f32(val);
|
||||
e = vmul_f32(vrsqrts_f32(vmul_f32(e, e), val), e);
|
||||
e = vmul_f32(vrsqrts_f32(vmul_f32(e, e), val), e);
|
||||
return e;
|
||||
}
|
||||
|
||||
inline float32x4_t cv_vsqrtq_f32(float32x4_t val)
|
||||
{
|
||||
return cv_vrecpq_f32(cv_vrsqrtq_f32(val));
|
||||
}
|
||||
|
||||
inline float32x2_t cv_vsqrt_f32(float32x2_t val)
|
||||
{
|
||||
return cv_vrecp_f32(cv_vrsqrt_f32(val));
|
||||
}
|
||||
|
||||
#endif
|
||||
|
||||
} // cv
|
||||
|
Reference in New Issue
Block a user