third_party/opencv
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Merge pull request #8900 from alalek:update_android_build

Browse Source
This commit is contained in:
Vadim Pisarevsky
2017-06-23 10:58:52 +00:00
parent 259147e32c 3e3e2dd512
commit fa7e7e0ff9
24 changed files with 240 additions and 155 deletions
Download Patch File Download Diff File Expand all files Collapse all files
modules/core/include/opencv2/core/cv_cpu_dispatch.h
+5 -1
View File
@@ -71,7 +71,11 @@
# define CV_AVX 1
#endif
#ifdef CV_CPU_COMPILE_FP16
# include <immintrin.h>
# if defined(__arm__) || defined(__aarch64__) || defined(_M_ARM)
# include <arm_neon.h>
# else
# include <immintrin.h>
# endif
# define CV_FP16 1
#endif
#ifdef CV_CPU_COMPILE_AVX2
modules/core/include/opencv2/core/hal/intrin_neon.hpp
+20 -8
View File
@@ -279,15 +279,27 @@ struct v_float64x2
#endif
#if CV_FP16
// Workaround for old comiplers
// Workaround for old compilers
template <typename T> static inline int16x4_t vreinterpret_s16_f16(T a)
{ return (int16x4_t)a; }
template <typename T> static inline float16x4_t vreinterpret_f16_s16(T a)
{ return (float16x4_t)a; }
template <typename T> static inline float16x4_t vld1_f16(const T* ptr)
{ return vreinterpret_f16_s16(vld1_s16((const short*)ptr)); }
template <typename T> static inline void vst1_f16(T* ptr, float16x4_t a)
{ vst1_s16((short*)ptr, vreinterpret_s16_f16(a)); }
template <typename T> static inline float16x4_t cv_vld1_f16(const T* ptr)
{
#ifndef vld1_f16 // APPLE compiler defines vld1_f16 as macro
return vreinterpret_f16_s16(vld1_s16((const short*)ptr));
#else
return vld1_f16((const __fp16*)ptr);
#endif
}
template <typename T> static inline void cv_vst1_f16(T* ptr, float16x4_t a)
{
#ifndef vst1_f16 // APPLE compiler defines vst1_f16 as macro
vst1_s16((short*)ptr, vreinterpret_s16_f16(a));
#else
vst1_f16((__fp16*)ptr, a);
#endif
}
struct v_float16x4
{
@@ -299,7 +311,7 @@ struct v_float16x4
v_float16x4(short v0, short v1, short v2, short v3)
{
short v[] = {v0, v1, v2, v3};
val = vld1_f16(v);
val = cv_vld1_f16(v);
}
short get0() const
{
@@ -778,9 +790,9 @@ OPENCV_HAL_IMPL_NEON_LOADSTORE_OP(v_float64x2, double, f64)
#if CV_FP16
// Workaround for old comiplers
inline v_float16x4 v_load_f16(const short* ptr)
{ return v_float16x4(vld1_f16(ptr)); }
{ return v_float16x4(cv_vld1_f16(ptr)); }
inline void v_store_f16(short* ptr, v_float16x4& a)
{ vst1_f16(ptr, a.val); }
{ cv_vst1_f16(ptr, a.val); }
#endif
#define OPENCV_HAL_IMPL_NEON_REDUCE_OP_8(_Tpvec, _Tpnvec, scalartype, func, vectorfunc, suffix) \