generic-library/opencv
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Fixed NEON compilation issue

Browse Source
This commit is contained in:
Maksim Shabunin
2015-12-03 15:40:55 +03:00
parent 5473dbebed
commit 5c5d0e6743
3 changed files with 125 additions and 80 deletions
Download Patch File Download Diff File Expand all files Collapse all files

81
modules/core/include/opencv2/core/base.hpp
View File

@@ -679,86 +679,7 @@ CV_EXPORTS void setUseIPP(bool flag);
//! @} core_utils
//! @addtogroup core_utils_neon
//! @{
#if CV_NEON
inline int32x2_t cv_vrnd_s32_f32(float32x2_t v)
{
static int32x2_t v_sign = vdup_n_s32(1 << 31),
v_05 = vreinterpret_s32_f32(vdup_n_f32(0.5f));
int32x2_t v_addition = vorr_s32(v_05, vand_s32(v_sign, vreinterpret_s32_f32(v)));
return vcvt_s32_f32(vadd_f32(v, vreinterpret_f32_s32(v_addition)));
}
inline int32x4_t cv_vrndq_s32_f32(float32x4_t v)
{
static int32x4_t v_sign = vdupq_n_s32(1 << 31),
v_05 = vreinterpretq_s32_f32(vdupq_n_f32(0.5f));
int32x4_t v_addition = vorrq_s32(v_05, vandq_s32(v_sign, vreinterpretq_s32_f32(v)));
return vcvtq_s32_f32(vaddq_f32(v, vreinterpretq_f32_s32(v_addition)));
}
inline uint32x2_t cv_vrnd_u32_f32(float32x2_t v)
{
static float32x2_t v_05 = vdup_n_f32(0.5f);
return vcvt_u32_f32(vadd_f32(v, v_05));
}
inline uint32x4_t cv_vrndq_u32_f32(float32x4_t v)
{
static float32x4_t v_05 = vdupq_n_f32(0.5f);
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
//! @} core_utils_neon
#include "opencv2/hal/neon_utils.hpp"
} // cv