Skip to content

O
opencv
Commit b4bcdd10 authored by Maksim Shabunin's avatar Maksim Shabunin
Browse files
Options

HAL: improvements 

- added new functions from core module: split, merge, add, sub, mul, div, ...
- added function replacement mechanism
- added example of HAL replacement library
parent e8742be3
Expand all Hide whitespace changes
Inline Side-by-side
Showing with 468 additions and 312 deletions
CMakeLists.txt
View file @ b4bcdd10
......@@ -587,6 +587,11 @@ include(cmake/OpenCVFindMatlab.cmake)
include(cmake/OpenCVDetectVTK.cmake)
if (OPENCV_HAL_HEADERS AND OPENCV_HAL_LIBS)
get_filename_component(OPENCV_HAL_HEADERS "${OPENCV_HAL_HEADERS}" ABSOLUTE)
get_filename_component(OPENCV_HAL_LIBS "${OPENCV_HAL_LIBS}" ABSOLUTE)
endif()
# ----------------------------------------------------------------------------
# Add CUDA libraries (needed for apps/tools, samples)
# ----------------------------------------------------------------------------
......
cmake/templates/custom_hal.hpp.in 0 → 100644
View file @ b4bcdd10
#ifndef _CUSTOM_HAL_INCLUDED_
#define _CUSTOM_HAL_INCLUDED_
@OPENCV_HAL_HEADERS_INCLUDES@
#endif
modules/core/include/opencv2/core/base.hpp
View file @ b4bcdd10
......@@ -762,6 +762,4 @@ inline float32x2_t cv_vsqrt_f32(float32x2_t val)
} // cv
#include "sse_utils.hpp"
#endif //__OPENCV_CORE_BASE_HPP__
modules/core/include/opencv2/core/utility.hpp
View file @ b4bcdd10
......@@ -277,37 +277,6 @@ execution time.
*/
CV_EXPORTS_W int64 getCPUTickCount();
/** @brief Available CPU features.
remember to keep this list identical to the one in cvdef.h
*/
enum CpuFeatures {
CPU_MMX = 1,
CPU_SSE = 2,
CPU_SSE2 = 3,
CPU_SSE3 = 4,
CPU_SSSE3 = 5,
CPU_SSE4_1 = 6,
CPU_SSE4_2 = 7,
CPU_POPCNT = 8,
CPU_AVX = 10,
CPU_AVX2 = 11,
CPU_FMA3 = 12,
CPU_AVX_512F = 13,
CPU_AVX_512BW = 14,
CPU_AVX_512CD = 15,
CPU_AVX_512DQ = 16,
CPU_AVX_512ER = 17,
CPU_AVX_512IFMA512 = 18,
CPU_AVX_512PF = 19,
CPU_AVX_512VBMI = 20,
CPU_AVX_512VL = 21,
CPU_NEON = 100
};
/** @brief Returns true if the specified feature is supported by the host hardware.
The function returns true if the host hardware supports the specified feature. When user calls
......
modules/core/src/arithm.cpp
View file @ b4bcdd10
This diff is collapsed.
modules/core/src/convert.cpp
View file @ b4bcdd10
This diff is collapsed.
modules/core/src/precomp.hpp
View file @ b4bcdd10
......@@ -83,6 +83,11 @@ typedef void (*BinaryFunc)(const uchar* src1, size_t step1,
uchar* dst, size_t step, Size sz,
void*);
typedef void (*BinaryFuncC)(const uchar* src1, size_t step1,
const uchar* src2, size_t step2,
uchar* dst, size_t step, int width, int height,
void*);
BinaryFunc getConvertFunc(int sdepth, int ddepth);
BinaryFunc getCopyMaskFunc(size_t esz);
......@@ -114,46 +119,6 @@ extern const uchar g_Saturate8u[];
void deleteThreadAllocData();
#endif
template<typename T1, typename T2=T1, typename T3=T1> struct OpAdd
{
typedef T1 type1;
typedef T2 type2;
typedef T3 rtype;
T3 operator ()(const T1 a, const T2 b) const { return saturate_cast<T3>(a + b); }
};
template<typename T1, typename T2=T1, typename T3=T1> struct OpSub
{
typedef T1 type1;
typedef T2 type2;
typedef T3 rtype;
T3 operator ()(const T1 a, const T2 b) const { return saturate_cast<T3>(a - b); }
};
template<typename T1, typename T2=T1, typename T3=T1> struct OpRSub
{
typedef T1 type1;
typedef T2 type2;
typedef T3 rtype;
T3 operator ()(const T1 a, const T2 b) const { return saturate_cast<T3>(b - a); }
};
template<typename T> struct OpMin
{
typedef T type1;
typedef T type2;
typedef T rtype;
T operator ()(const T a, const T b) const { return std::min(a, b); }
};
template<typename T> struct OpMax
{
typedef T type1;
typedef T type2;
typedef T rtype;
T operator ()(const T a, const T b) const { return std::max(a, b); }
};
inline Size getContinuousSize_( int flags, int cols, int rows, int widthScale )
{
int64 sz = (int64)cols * rows * widthScale;
......@@ -201,11 +166,6 @@ struct NoVec
size_t operator()(const void*, const void*, void*, size_t) const { return 0; }
};
extern volatile bool USE_SSE2;
extern volatile bool USE_SSE4_2;
extern volatile bool USE_AVX;
extern volatile bool USE_AVX2;
enum { BLOCK_SIZE = 1024 };
#if defined HAVE_IPP && (IPP_VERSION_X100 >= 700)
......
modules/core/src/system.cpp
View file @ b4bcdd10
......@@ -86,45 +86,6 @@ Mutex* __initialization_mutex_initializer = &getInitializationMutex();
#undef max
#undef abs
#include <tchar.h>
#if defined _MSC_VER
#if _MSC_VER >= 1400
#include <intrin.h>
#elif defined _M_IX86
static void __cpuid(int* cpuid_data, int)
{
__asm
{
push ebx
push edi
mov edi, cpuid_data
mov eax, 1
cpuid
mov [edi], eax
mov [edi + 4], ebx
mov [edi + 8], ecx
mov [edi + 12], edx
pop edi
pop ebx
}
}
static void __cpuidex(int* cpuid_data, int, int)
{
__asm
{
push edi
mov edi, cpuid_data
mov eax, 7
mov ecx, 0
cpuid
mov [edi], eax
mov [edi + 4], ebx
mov [edi + 8], ecx
mov [edi + 12], edx
pop edi
}
}
#endif
#endif
#ifdef WINRT
#include <wrl/client.h>
......@@ -237,160 +198,15 @@ void Exception::formatMessage()
msg = format("%s:%d: error: (%d) %s\n", file.c_str(), line, code, err.c_str());
}
struct HWFeatures
{
enum { MAX_FEATURE = CV_HARDWARE_MAX_FEATURE };
HWFeatures(void)
{
memset( have, 0, sizeof(have) );
x86_family = 0;
}
static HWFeatures initialize(void)
{
HWFeatures f;
int cpuid_data[4] = { 0, 0, 0, 0 };
#if defined _MSC_VER && (defined _M_IX86 || defined _M_X64)
__cpuid(cpuid_data, 1);
#elif defined __GNUC__ && (defined __i386__ || defined __x86_64__)
#ifdef __x86_64__
asm __volatile__
(
"movl $1, %%eax\n\t"
"cpuid\n\t"
:[eax]"=a"(cpuid_data[0]),[ebx]"=b"(cpuid_data[1]),[ecx]"=c"(cpuid_data[2]),[edx]"=d"(cpuid_data[3])
:
: "cc"
);
#else
asm volatile
(
"pushl %%ebx\n\t"
"movl $1,%%eax\n\t"
"cpuid\n\t"
"popl %%ebx\n\t"
: "=a"(cpuid_data[0]), "=c"(cpuid_data[2]), "=d"(cpuid_data[3])
:
: "cc"
);
#endif
#endif
f.x86_family = (cpuid_data[0] >> 8) & 15;
if( f.x86_family >= 6 )
{
f.have[CV_CPU_MMX] = (cpuid_data[3] & (1 << 23)) != 0;
f.have[CV_CPU_SSE] = (cpuid_data[3] & (1<<25)) != 0;
f.have[CV_CPU_SSE2] = (cpuid_data[3] & (1<<26)) != 0;
f.have[CV_CPU_SSE3] = (cpuid_data[2] & (1<<0)) != 0;
f.have[CV_CPU_SSSE3] = (cpuid_data[2] & (1<<9)) != 0;
f.have[CV_CPU_FMA3] = (cpuid_data[2] & (1<<12)) != 0;
f.have[CV_CPU_SSE4_1] = (cpuid_data[2] & (1<<19)) != 0;
f.have[CV_CPU_SSE4_2] = (cpuid_data[2] & (1<<20)) != 0;
f.have[CV_CPU_POPCNT] = (cpuid_data[2] & (1<<23)) != 0;
f.have[CV_CPU_AVX] = (((cpuid_data[2] & (1<<28)) != 0)&&((cpuid_data[2] & (1<<27)) != 0));//OS uses XSAVE_XRSTORE and CPU support AVX
// make the second call to the cpuid command in order to get
// information about extended features like AVX2
#if defined _MSC_VER && (defined _M_IX86 || defined _M_X64)
__cpuidex(cpuid_data, 7, 0);
#elif defined __GNUC__ && (defined __i386__ || defined __x86_64__)
#ifdef __x86_64__
asm __volatile__
(
"movl $7, %%eax\n\t"
"movl $0, %%ecx\n\t"
"cpuid\n\t"
:[eax]"=a"(cpuid_data[0]),[ebx]"=b"(cpuid_data[1]),[ecx]"=c"(cpuid_data[2]),[edx]"=d"(cpuid_data[3])
:
: "cc"
);
#else
asm volatile
(
"pushl %%ebx\n\t"
"movl $7,%%eax\n\t"
"movl $0,%%ecx\n\t"
"cpuid\n\t"
"movl %%ebx, %0\n\t"
"popl %%ebx\n\t"
: "=r"(cpuid_data[1]), "=c"(cpuid_data[2])
:
: "cc"
);
#endif
#endif
f.have[CV_CPU_AVX2] = (cpuid_data[1] & (1<<5)) != 0;
f.have[CV_CPU_AVX_512F] = (cpuid_data[1] & (1<<16)) != 0;
f.have[CV_CPU_AVX_512DQ] = (cpuid_data[1] & (1<<17)) != 0;
f.have[CV_CPU_AVX_512IFMA512] = (cpuid_data[1] & (1<<21)) != 0;
f.have[CV_CPU_AVX_512PF] = (cpuid_data[1] & (1<<26)) != 0;
f.have[CV_CPU_AVX_512ER] = (cpuid_data[1] & (1<<27)) != 0;
f.have[CV_CPU_AVX_512CD] = (cpuid_data[1] & (1<<28)) != 0;
f.have[CV_CPU_AVX_512BW] = (cpuid_data[1] & (1<<30)) != 0;
f.have[CV_CPU_AVX_512VL] = (cpuid_data[1] & (1<<31)) != 0;
f.have[CV_CPU_AVX_512VBMI] = (cpuid_data[2] & (1<<1)) != 0;
}
#if defined ANDROID || defined __linux__
#ifdef __aarch64__
f.have[CV_CPU_NEON] = true;
#else
int cpufile = open("/proc/self/auxv", O_RDONLY);
if (cpufile >= 0)
{
Elf32_auxv_t auxv;
const size_t size_auxv_t = sizeof(auxv);
while ((size_t)read(cpufile, &auxv, size_auxv_t) == size_auxv_t)
{
if (auxv.a_type == AT_HWCAP)
{
f.have[CV_CPU_NEON] = (auxv.a_un.a_val & 4096) != 0;
break;
}
}
close(cpufile);
}
#endif
#elif (defined __clang__ || defined __APPLE__) && (defined __ARM_NEON__ || (defined __ARM_NEON && defined __aarch64__))
f.have[CV_CPU_NEON] = true;
#endif
return f;
}
int x86_family;
bool have[MAX_FEATURE+1];
};
static HWFeatures featuresEnabled = HWFeatures::initialize(), featuresDisabled = HWFeatures();
static HWFeatures* currentFeatures = &featuresEnabled;
bool checkHardwareSupport(int feature)
{
CV_DbgAssert( 0 <= feature && feature <= CV_HARDWARE_MAX_FEATURE );
return currentFeatures->have[feature];
return cv::hal::checkHardwareSupport(feature);
}
volatile bool useOptimizedFlag = true;
volatile bool USE_SSE2 = featuresEnabled.have[CV_CPU_SSE2];
volatile bool USE_SSE4_2 = featuresEnabled.have[CV_CPU_SSE4_2];
volatile bool USE_AVX = featuresEnabled.have[CV_CPU_AVX];
volatile bool USE_AVX2 = featuresEnabled.have[CV_CPU_AVX2];
void setUseOptimized( bool flag )
{
useOptimizedFlag = flag;
currentFeatures = flag ? &featuresEnabled : &featuresDisabled;
USE_SSE2 = currentFeatures->have[CV_CPU_SSE2];
cv::hal::setUseOptimized(flag);
ipp::setUseIPP(flag);
#ifdef HAVE_OPENCL
......@@ -403,7 +219,7 @@ void setUseOptimized( bool flag )
bool useOptimized(void)
{
return useOptimizedFlag;
return cv::hal::useOptimized();
}
int64 getTickCount(void)
......@@ -683,12 +499,12 @@ redirectError( CvErrorCallback errCallback, void* userdata, void** prevUserdata)
CV_IMPL int cvCheckHardwareSupport(int feature)
{
CV_DbgAssert( 0 <= feature && feature <= CV_HARDWARE_MAX_FEATURE );
return cv::currentFeatures->have[feature];
return cv::hal::checkHardwareSupport(feature);
}
CV_IMPL int cvUseOptimized( int flag )
{
int prevMode = cv::useOptimizedFlag;
int prevMode = cv::useOptimized();
cv::setUseOptimized( flag != 0 );
return prevMode;
}
......
modules/hal/CMakeLists.txt
View file @ b4bcdd10
......@@ -2,10 +2,20 @@ set(the_description "The Hardware Acceleration Layer (HAL) module")
set(OPENCV_MODULE_TYPE STATIC)
if(OPENCV_HAL_HEADERS AND OPENCV_HAL_LIBS)
set(OPENCV_HAL_HEADERS_INCLUDES "#include \"${OPENCV_HAL_HEADERS}\"")
set(DEPS "${OPENCV_HAL_LIBS}")
else()
set(OPENCV_HAL_HEADERS_INCLUDES "// using default HAL")
set(DEPS "")
endif()
configure_file("${OpenCV_SOURCE_DIR}/cmake/templates/custom_hal.hpp.in" "${CMAKE_BINARY_DIR}/custom_hal.hpp" @ONLY)
if(UNIX)
if(CMAKE_COMPILER_IS_GNUCXX OR CV_ICC)
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fPIC")
endif()
endif()
ocv_define_module(hal)
ocv_define_module(hal ${DEPS})
modules/hal/include/opencv2/hal.hpp
View file @ b4bcdd10
This diff is collapsed.
modules/hal/include/opencv2/hal/defs.h
View file @ b4bcdd10
......@@ -53,6 +53,7 @@
#endif
#include <limits.h>
#include "opencv2/hal/interface.hpp"
#if defined __ICL
# define CV_ICC __ICL
......@@ -117,9 +118,38 @@
#define CV_CPU_NEON 100
// when adding to this list remember to update the enum in core/utility.cpp
// when adding to this list remember to update the following enum
#define CV_HARDWARE_MAX_FEATURE 255
/** @brief Available CPU features.
*/
enum CpuFeatures {
CPU_MMX = 1,
CPU_SSE = 2,
CPU_SSE2 = 3,
CPU_SSE3 = 4,
CPU_SSSE3 = 5,
CPU_SSE4_1 = 6,
CPU_SSE4_2 = 7,
CPU_POPCNT = 8,
CPU_AVX = 10,
CPU_AVX2 = 11,
CPU_FMA3 = 12,
CPU_AVX_512F = 13,
CPU_AVX_512BW = 14,
CPU_AVX_512CD = 15,
CPU_AVX_512DQ = 16,
CPU_AVX_512ER = 17,
CPU_AVX_512IFMA512 = 18,
CPU_AVX_512PF = 19,
CPU_AVX_512VBMI = 20,
CPU_AVX_512VL = 21,
CPU_NEON = 100
};
// do not include SSE/AVX/NEON headers for NVCC compiler
#ifndef __CUDACC__
......@@ -257,49 +287,6 @@
# define CV_VFP 0
#endif
/* primitive types */
/*
schar - signed 1 byte integer
uchar - unsigned 1 byte integer
short - signed 2 byte integer
ushort - unsigned 2 byte integer
int - signed 4 byte integer
uint - unsigned 4 byte integer
int64 - signed 8 byte integer
uint64 - unsigned 8 byte integer
*/
#if !defined _MSC_VER && !defined __BORLANDC__
# if defined __cplusplus && __cplusplus >= 201103L && !defined __APPLE__
# include <cstdint>
typedef std::uint32_t uint;
# else
# include <stdint.h>
typedef uint32_t uint;
# endif
#else
typedef unsigned uint;
#endif
typedef signed char schar;
#ifndef __IPL_H__
typedef unsigned char uchar;
typedef unsigned short ushort;
#endif
#if defined _MSC_VER || defined __BORLANDC__
typedef __int64 int64;
typedef unsigned __int64 uint64;
# define CV_BIG_INT(n) n##I64
# define CV_BIG_UINT(n) n##UI64
#else
typedef int64_t int64;
typedef uint64_t uint64;
# define CV_BIG_INT(n) n##LL
# define CV_BIG_UINT(n) n##ULL
#endif
/* fundamental constants */
#define CV_PI 3.1415926535897932384626433832795
#define CV_2PI 6.283185307179586476925286766559
......@@ -321,6 +308,19 @@ typedef union Cv64suf
}
Cv64suf;
namespace cv { namespace hal {
bool checkHardwareSupport(int feature);
void setUseOptimized(bool onoff);
bool useOptimized();
}}
#define USE_SSE2 (cv::hal::checkHardwareSupport(CV_CPU_SSE))
#define USE_SSE4_2 (cv::hal::checkHardwareSupport(CV_CPU_SSE4_2))
#define USE_AVX (cv::hal::checkHardwareSupport(CV_CPU_AVX))
#define USE_AVX2 (cv::hal::checkHardwareSupport(CV_CPU_AVX2))
/****************************************************************************************\
* fast math *
......
modules/hal/include/opencv2/hal/interface.hpp 0 → 100644
View file @ b4bcdd10
#ifndef _HAL_INTERFACE_HPP_INCLUDED_
#define _HAL_INTERFACE_HPP_INCLUDED_
#define CV_HAL_ERROR_OK 0
#define CV_HAL_ERROR_NI 1
#define CV_HAL_ERROR_UNKNOWN -1
#define CV_HAL_CMP_EQ 0
#define CV_HAL_CMP_GT 1
#define CV_HAL_CMP_GE 2
#define CV_HAL_CMP_LT 3
#define CV_HAL_CMP_LE 4
#define CV_HAL_CMP_NE 5
#ifdef __cplusplus
namespace cv { namespace hal {
namespace Error {
enum
{
Ok = 0,
NotImplemented = 1,
Unknown = -1
};
}
enum
{
CMP_EQ = 0,
CMP_GT = 1,
CMP_GE = 2,
CMP_LT = 3,
CMP_LE = 4,
CMP_NE = 5
};
}}
#endif
#ifdef __cplusplus
#include <cstddef>
#else
#include <stddef.h>
#endif
/* primitive types */
/*
schar - signed 1 byte integer
uchar - unsigned 1 byte integer
short - signed 2 byte integer
ushort - unsigned 2 byte integer
int - signed 4 byte integer
uint - unsigned 4 byte integer
int64 - signed 8 byte integer
uint64 - unsigned 8 byte integer
*/
#if !defined _MSC_VER && !defined __BORLANDC__
# if defined __cplusplus && __cplusplus >= 201103L && !defined __APPLE__
# include <cstdint>
typedef std::uint32_t uint;
# else
# include <stdint.h>
typedef uint32_t uint;
# endif
#else
typedef unsigned uint;
#endif
typedef signed char schar;
#ifndef __IPL_H__
typedef unsigned char uchar;
typedef unsigned short ushort;
#endif
#if defined _MSC_VER || defined __BORLANDC__
typedef __int64 int64;
typedef unsigned __int64 uint64;
# define CV_BIG_INT(n) n##I64
# define CV_BIG_UINT(n) n##UI64
#else
typedef int64_t int64;
typedef uint64_t uint64;
# define CV_BIG_INT(n) n##LL
# define CV_BIG_UINT(n) n##ULL
#endif
#endif
modules/core/include/opencv2/core/sse_utils.hpp → modules/hal/include/opencv2/hal/sse_utils.hpp
View file @ b4bcdd10
......@@ -46,6 +46,8 @@
# error sse_utils.hpp header must be compiled as C++
#endif
#include "opencv2/hal/defs.h"
#if CV_SSE2
inline void _mm_deinterleave_epi8(__m128i & v_r0, __m128i & v_r1, __m128i & v_g0, __m128i & v_g1)
......
modules/hal/samples/simple_hal/CMakeLists.txt 0 → 100644
View file @ b4bcdd10
cmake_minimum_required(VERSION 2.8.8 FATAL_ERROR)
if(UNIX)
if(CMAKE_COMPILER_IS_GNUCXX OR CV_ICC)
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fPIC")
endif()
endif()
add_library(simple_hal simple.cpp)
set(OPENCV_HAL_DIR "${CMAKE_CURRENT_SOURCE_DIR}/../..")
target_include_directories(simple_hal PUBLIC ${CMAKE_CURRENT_SOURCE_DIR} ${OPENCV_HAL_DIR}/include)
modules/hal/samples/simple_hal/simple.cpp 0 → 100644
View file @ b4bcdd10
#include "simple.hpp"
int slow_and8u(const uchar* src1, size_t step1, const uchar* src2, size_t step2, uchar* dst, size_t step, int width, int height)
{
for(; height--; src1 = src1 + step1, src2 = src2 + step2, dst = dst + step)
for(int x = 0 ; x < width; x++ )
dst[x] = src1[x] & src2[x];
return cv::hal::Error::Ok;
}
int slow_or8u(const uchar* src1, size_t step1, const uchar* src2, size_t step2, uchar* dst, size_t step, int width, int height)
{
for(; height--; src1 = src1 + step1, src2 = src2 + step2, dst = dst + step)
for(int x = 0 ; x < width; x++ )
dst[x] = src1[x] | src2[x];
return cv::hal::Error::Ok;
}
int slow_xor8u(const uchar* src1, size_t step1, const uchar* src2, size_t step2, uchar* dst, size_t step, int width, int height)
{
for(; height--; src1 = src1 + step1, src2 = src2 + step2, dst = dst + step)
for(int x = 0 ; x < width; x++ )
dst[x] = src1[x] ^ src2[x];
return cv::hal::Error::Ok;
}
int slow_not8u(const uchar* src1, size_t step1, const uchar* src2, size_t step2, uchar* dst, size_t step, int width, int height)
{
for(; height--; src1 = src1 + step1, src2 = src2 + step2, dst = dst + step)
for(int x = 0 ; x < width; x++ )
dst[x] = ~src1[x];
return cv::hal::Error::Ok;
}
modules/hal/samples/simple_hal/simple.hpp 0 → 100644
View file @ b4bcdd10
#ifndef _SIMPLE_HPP_INCLUDED_
#define _SIMPLE_HPP_INCLUDED_
#include "opencv2/hal/interface.hpp"
int slow_and8u(const uchar* src1, size_t step1, const uchar* src2, size_t step2, uchar* dst, size_t step, int width, int height);
int slow_or8u(const uchar* src1, size_t step1, const uchar* src2, size_t step2, uchar* dst, size_t step, int width, int height);
int slow_xor8u(const uchar* src1, size_t step1, const uchar* src2, size_t step2, uchar* dst, size_t step, int width, int height);
int slow_not8u(const uchar* src1, size_t step1, const uchar* src2, size_t step2, uchar* dst, size_t step, int width, int height);
#undef hal_and8u
#define hal_and8u slow_and8u
#undef hal_or8u
#define hal_or8u slow_or8u
#undef hal_xor8u
#define hal_xor8u slow_xor8u
#undef hal_not8u
#define hal_not8u slow_not8u
#endif
modules/hal/src/arithm.cpp
View file @ b4bcdd10
This diff is collapsed.
modules/hal/src/arithm_core.hpp 0 → 100644
View file @ b4bcdd10
This diff is collapsed.
modules/hal/src/arithm_simd.hpp 0 → 100644
View file @ b4bcdd10
This diff is collapsed.
modules/hal/src/hardware.cpp 0 → 100644
View file @ b4bcdd10
#include "precomp.hpp"
#if defined WIN32 || defined _WIN32 || defined WINCE
#include <windows.h>
#if defined _MSC_VER
#if _MSC_VER >= 1400
#include <intrin.h>
#elif defined _M_IX86
static void __cpuid(int* cpuid_data, int)
{
__asm
{
push ebx
push edi
mov edi, cpuid_data
mov eax, 1
cpuid
mov [edi], eax
mov [edi + 4], ebx
mov [edi + 8], ecx
mov [edi + 12], edx
pop edi
pop ebx
}
}
static void __cpuidex(int* cpuid_data, int, int)
{
__asm
{
push edi
mov edi, cpuid_data
mov eax, 7
mov ecx, 0
cpuid
mov [edi], eax
mov [edi + 4], ebx
mov [edi + 8], ecx
mov [edi + 12], edx
pop edi
}
}
#endif
#endif
#endif
#if defined ANDROID || defined __linux__
# include <unistd.h>
# include <fcntl.h>
# include <elf.h>
# include <linux/auxvec.h>
#endif
#if defined __linux__ || defined __APPLE__ || defined __EMSCRIPTEN__
#include <unistd.h>
#include <stdio.h>
#include <sys/types.h>
#if defined ANDROID
#include <sys/sysconf.h>
#endif
#endif
#ifdef ANDROID
# include <android/log.h>
#endif
struct HWFeatures
{
enum { MAX_FEATURE = CV_HARDWARE_MAX_FEATURE };
HWFeatures(void)
{
memset( have, 0, sizeof(have) );
x86_family = 0;
}
static HWFeatures initialize(void)
{
HWFeatures f;
int cpuid_data[4] = { 0, 0, 0, 0 };
#if defined _MSC_VER && (defined _M_IX86 || defined _M_X64)
__cpuid(cpuid_data, 1);
#elif defined __GNUC__ && (defined __i386__ || defined __x86_64__)
#ifdef __x86_64__
asm __volatile__
(
"movl $1, %%eax\n\t"
"cpuid\n\t"
:[eax]"=a"(cpuid_data[0]),[ebx]"=b"(cpuid_data[1]),[ecx]"=c"(cpuid_data[2]),[edx]"=d"(cpuid_data[3])
:
: "cc"
);
#else
asm volatile
(
"pushl %%ebx\n\t"
"movl $1,%%eax\n\t"
"cpuid\n\t"
"popl %%ebx\n\t"
: "=a"(cpuid_data[0]), "=c"(cpuid_data[2]), "=d"(cpuid_data[3])
:
: "cc"
);
#endif
#endif
f.x86_family = (cpuid_data[0] >> 8) & 15;
if( f.x86_family >= 6 )
{
f.have[CV_CPU_MMX] = (cpuid_data[3] & (1 << 23)) != 0;
f.have[CV_CPU_SSE] = (cpuid_data[3] & (1<<25)) != 0;
f.have[CV_CPU_SSE2] = (cpuid_data[3] & (1<<26)) != 0;
f.have[CV_CPU_SSE3] = (cpuid_data[2] & (1<<0)) != 0;
f.have[CV_CPU_SSSE3] = (cpuid_data[2] & (1<<9)) != 0;
f.have[CV_CPU_FMA3] = (cpuid_data[2] & (1<<12)) != 0;
f.have[CV_CPU_SSE4_1] = (cpuid_data[2] & (1<<19)) != 0;
f.have[CV_CPU_SSE4_2] = (cpuid_data[2] & (1<<20)) != 0;
f.have[CV_CPU_POPCNT] = (cpuid_data[2] & (1<<23)) != 0;
f.have[CV_CPU_AVX] = (((cpuid_data[2] & (1<<28)) != 0)&&((cpuid_data[2] & (1<<27)) != 0));//OS uses XSAVE_XRSTORE and CPU support AVX
// make the second call to the cpuid command in order to get
// information about extended features like AVX2
#if defined _MSC_VER && (defined _M_IX86 || defined _M_X64)
__cpuidex(cpuid_data, 7, 0);
#elif defined __GNUC__ && (defined __i386__ || defined __x86_64__)
#ifdef __x86_64__
asm __volatile__
(
"movl $7, %%eax\n\t"
"movl $0, %%ecx\n\t"
"cpuid\n\t"
:[eax]"=a"(cpuid_data[0]),[ebx]"=b"(cpuid_data[1]),[ecx]"=c"(cpuid_data[2]),[edx]"=d"(cpuid_data[3])
:
: "cc"
);
#else
asm volatile
(
"pushl %%ebx\n\t"
"movl $7,%%eax\n\t"
"movl $0,%%ecx\n\t"
"cpuid\n\t"
"movl %%ebx, %0\n\t"
"popl %%ebx\n\t"
: "=r"(cpuid_data[1]), "=c"(cpuid_data[2])
:
: "cc"
);
#endif
#endif
f.have[CV_CPU_AVX2] = (cpuid_data[1] & (1<<5)) != 0;
f.have[CV_CPU_AVX_512F] = (cpuid_data[1] & (1<<16)) != 0;
f.have[CV_CPU_AVX_512DQ] = (cpuid_data[1] & (1<<17)) != 0;
f.have[CV_CPU_AVX_512IFMA512] = (cpuid_data[1] & (1<<21)) != 0;
f.have[CV_CPU_AVX_512PF] = (cpuid_data[1] & (1<<26)) != 0;
f.have[CV_CPU_AVX_512ER] = (cpuid_data[1] & (1<<27)) != 0;
f.have[CV_CPU_AVX_512CD] = (cpuid_data[1] & (1<<28)) != 0;
f.have[CV_CPU_AVX_512BW] = (cpuid_data[1] & (1<<30)) != 0;
f.have[CV_CPU_AVX_512VL] = (cpuid_data[1] & (1<<31)) != 0;
f.have[CV_CPU_AVX_512VBMI] = (cpuid_data[2] & (1<<1)) != 0;
}
#if defined ANDROID || defined __linux__
#ifdef __aarch64__
f.have[CV_CPU_NEON] = true;
#else
int cpufile = open("/proc/self/auxv", O_RDONLY);
if (cpufile >= 0)
{
Elf32_auxv_t auxv;
const size_t size_auxv_t = sizeof(auxv);
while ((size_t)read(cpufile, &auxv, size_auxv_t) == size_auxv_t)
{
if (auxv.a_type == AT_HWCAP)
{
f.have[CV_CPU_NEON] = (auxv.a_un.a_val & 4096) != 0;
break;
}
}
close(cpufile);
}
#endif
#elif (defined __clang__ || defined __APPLE__) && (defined __ARM_NEON__ || (defined __ARM_NEON && defined __aarch64__))
f.have[CV_CPU_NEON] = true;
#endif
return f;
}
int x86_family;
bool have[MAX_FEATURE+1];
};
static HWFeatures featuresEnabled = HWFeatures::initialize(), featuresDisabled = HWFeatures();
static HWFeatures* currentFeatures = &featuresEnabled;
volatile bool useOptimizedFlag = true;
namespace cv { namespace hal {
bool checkHardwareSupport(int feature)
{
// CV_DbgAssert( 0 <= feature && feature <= CV_HARDWARE_MAX_FEATURE );
return currentFeatures->have[feature];
}
void setUseOptimized( bool flag )
{
useOptimizedFlag = flag;
currentFeatures = flag ? &featuresEnabled : &featuresDisabled;
}
bool useOptimized(void)
{
return useOptimizedFlag;
}
}}
modules/hal/src/merge.cpp 0 → 100644
View file @ b4bcdd10
This diff is collapsed.
modules/hal/src/precomp.hpp
View file @ b4bcdd10
......@@ -47,3 +47,13 @@
#include <cstdlib>
#include <limits>
#include <float.h>
#include <cstring>
#include <cassert>
#include "opencv2/hal/sse_utils.hpp"
#if defined HAVE_IPP && (IPP_VERSION_X100 >= 700)
#define ARITHM_USE_IPP 1
#else
#define ARITHM_USE_IPP 0
#endif
modules/hal/src/replacement.hpp 0 → 100644
View file @ b4bcdd10
This diff is collapsed.
modules/hal/src/split.cpp 0 → 100644
View file @ b4bcdd10
This diff is collapsed.
modules/imgproc/src/precomp.hpp
View file @ b4bcdd10
......@@ -94,4 +94,6 @@ extern const float icv8x32fSqrTab[];
#include "_geom.h"
#include "filterengine.hpp"
#include "opencv2/hal/sse_utils.hpp"
#endif /*__OPENCV_CV_INTERNAL_H_*/
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment