potok/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2026-07-30 07:43:03 +04:00
Code Packages 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/CMakeLists.txt
+5 -1
View File
@@ -32,7 +32,11 @@ source_group("Src" FILES "${OPENCV_MODULE_opencv_core_BINARY_DIR}/version_string
ocv_glob_module_sources(SOURCES "${OPENCV_MODULE_opencv_core_BINARY_DIR}/version_string.inc"
HEADERS ${lib_cuda_hdrs} ${lib_cuda_hdrs_detail})
ocv_module_include_directories(${the_module} ${ZLIB_INCLUDE_DIRS} ${OPENCL_INCLUDE_DIRS} ${CPUFEATURES_INCLUDE_DIRS})
ocv_module_include_directories(${the_module} ${ZLIB_INCLUDE_DIRS} ${OPENCL_INCLUDE_DIRS})
if(ANDROID AND HAVE_CPUFEATURES)
ocv_append_sourge_file_compile_definitions(${CMAKE_CURRENT_SOURCE_DIR}/src/system.cpp "HAVE_CPUFEATURES=1")
ocv_module_include_directories(${CPUFEATURES_INCLUDE_DIRS})
endif()
ocv_create_module(${extra_libs})
ocv_target_link_libraries(${the_module} ${ZLIB_LIBRARIES} "${OPENCL_LIBRARIES}" "${VA_LIBRARIES}" "${LAPACK_LIBRARIES}" "${CPUFEATURES_LIBRARIES}" "${HALIDE_LIBRARIES}")
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) \
modules/core/src/convert.fp16.cpp
+2 -7
View File
@@ -108,11 +108,6 @@ void cvtScaleHalf_SIMD16f32f( const short* src, size_t sstep, float* dst, size_t
#elif CV_NEON
const static int cVectorWidth = 4;
template <typename T> static inline float16x4_t vld1_f16(const T* ptr)
{ return (float16x4_t)vld1_s16((const short*)ptr); }
template <typename T> static inline void vst1_f16(T* ptr, float16x4_t a)
{ vst1_s16((short*)ptr, a); }
void cvtScaleHalf_SIMD32f16f( const float* src, size_t sstep, short* dst, size_t dstep, cv::Size size )
{
CV_INSTRUMENT_REGION()
@@ -129,7 +124,7 @@ void cvtScaleHalf_SIMD32f16f( const float* src, size_t sstep, short* dst, size_t
float16x4_t v_dst = vcvt_f16_f32(v_src);
vst1_f16((__fp16*)dst + x, v_dst);
cv_vst1_f16((__fp16*)dst + x, v_dst);
}
for ( ; x < size.width; x++ )
@@ -151,7 +146,7 @@ void cvtScaleHalf_SIMD16f32f( const short* src, size_t sstep, float* dst, size_t
int x = 0;
for ( ; x <= size.width - cVectorWidth ; x += cVectorWidth )
{
float16x4_t v_src = vld1_f16((__fp16*)src + x);
float16x4_t v_src = cv_vld1_f16((__fp16*)src + x);
float32x4_t v_dst = vcvt_f32_f16(v_src);
modules/core/src/convert.hpp
+4 -4
View File
@@ -13,7 +13,7 @@ const unsigned int kBiasFp32Exponent = 127;
#endif
#if CV_FP16_TYPE
float convertFp16SW(short fp16)
inline float convertFp16SW(short fp16)
{
// Fp16 -> Fp32
Cv16suf a;
@@ -21,7 +21,7 @@ float convertFp16SW(short fp16)
return (float)a.h;
}
#else
float convertFp16SW(short fp16)
inline float convertFp16SW(short fp16)
{
// Fp16 -> Fp32
Cv16suf b;
@@ -75,7 +75,7 @@ float convertFp16SW(short fp16)
#endif
#if CV_FP16_TYPE
short convertFp16SW(float fp32)
inline short convertFp16SW(float fp32)
{
// Fp32 -> Fp16
Cv16suf a;
@@ -83,7 +83,7 @@ short convertFp16SW(float fp32)
return a.i;
}
#else
short convertFp16SW(float fp32)
inline short convertFp16SW(float fp32)
{
// Fp32 -> Fp16
Cv32suf a;
modules/core/src/system.cpp
+17 -1
View File
@@ -73,7 +73,7 @@ Mutex* __initialization_mutex_initializer = &getInitializationMutex();
#endif
#endif
#if defined ANDROID
#if defined ANDROID && defined HAVE_CPUFEATURES
# include <cpu-features.h>
#endif
@@ -450,11 +450,27 @@ struct HWFeatures
have[CV_CPU_NEON] = true;
have[CV_CPU_FP16] = true;
#elif defined __arm__ && defined __ANDROID__
#if defined HAVE_CPUFEATURES
__android_log_print(ANDROID_LOG_INFO, "OpenCV", "calling android_getCpuFeatures() ...");
uint64_t features = android_getCpuFeatures();
__android_log_print(ANDROID_LOG_INFO, "OpenCV", "calling android_getCpuFeatures() ... Done (%llx)", features);
have[CV_CPU_NEON] = (features & ANDROID_CPU_ARM_FEATURE_NEON) != 0;
have[CV_CPU_FP16] = (features & ANDROID_CPU_ARM_FEATURE_VFP_FP16) != 0;
#else
__android_log_print(ANDROID_LOG_INFO, "OpenCV", "cpufeatures library is not avaialble for CPU detection");
#if CV_NEON
__android_log_print(ANDROID_LOG_INFO, "OpenCV", "- NEON instructions is enabled via build flags");
have[CV_CPU_NEON] = true;
#else
__android_log_print(ANDROID_LOG_INFO, "OpenCV", "- NEON instructions is NOT enabled via build flags");
#endif
#if CV_FP16
__android_log_print(ANDROID_LOG_INFO, "OpenCV", "- FP16 instructions is enabled via build flags");
have[CV_CPU_FP16] = true;
#else
__android_log_print(ANDROID_LOG_INFO, "OpenCV", "- FP16 instructions is NOT enabled via build flags");
#endif
#endif
#elif defined __arm__
int cpufile = open("/proc/self/auxv", O_RDONLY);