/*M/////////////////////////////////////////////////////////////////////////////////////// // // IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING. // // By downloading, copying, installing or using the software you agree to this license. // If you do not agree to this license, do not download, install, // copy or use the software. // // // Intel License Agreement // For Open Source Computer Vision Library // // Copyright (C) 2000, Intel Corporation, all rights reserved. // Third party copyrights are property of their respective owners. // // Redistribution and use in source and binary forms, with or without modification, // are permitted provided that the following conditions are met: // // * Redistribution's of source code must retain the above copyright notice, // this list of conditions and the following disclaimer. // // * Redistribution's in binary form must reproduce the above copyright notice, // this list of conditions and the following disclaimer in the documentation // and/or other materials provided with the distribution. // // * The name of Intel Corporation may not be used to endorse or promote products // derived from this software without specific prior written permission. // // This software is provided by the copyright holders and contributors "as is" and // any express or implied warranties, including, but not limited to, the implied // warranties of merchantability and fitness for a particular purpose are disclaimed. // In no event shall the Intel Corporation or contributors be liable for any direct, // indirect, incidental, special, exemplary, or consequential damages // (including, but not limited to, procurement of substitute goods or services; // loss of use, data, or profits; or business interruption) however caused // and on any theory of liability, whether in contract, strict liability, // or tort (including negligence or otherwise) arising in any way out of // the use of this software, even if advised of the possibility of such damage. // //M*/ /****************************************************************************************/ /* Dynamic detection and loading of IPP modules */ /****************************************************************************************/ #include "_cxcore.h" #if defined _MSC_VER && _MSC_VER >= 1200 #pragma warning( disable: 4115 ) /* type definition in () */ #endif #if defined _MSC_VER && defined WIN64 && !defined EM64T #pragma optimize( "", off ) #endif #if defined WIN32 || defined WIN64 #include #else #include #include #endif #include #include #include #define CV_PROC_GENERIC 0 #define CV_PROC_SHIFT 10 #define CV_PROC_ARCH_MASK ((1 << CV_PROC_SHIFT) - 1) #define CV_PROC_IA32_GENERIC 1 #define CV_PROC_IA32_WITH_MMX (CV_PROC_IA32_GENERIC|(2 << CV_PROC_SHIFT)) #define CV_PROC_IA32_WITH_SSE (CV_PROC_IA32_GENERIC|(3 << CV_PROC_SHIFT)) #define CV_PROC_IA32_WITH_SSE2 (CV_PROC_IA32_GENERIC|(4 << CV_PROC_SHIFT)) #define CV_PROC_IA64 2 #define CV_PROC_EM64T 3 #define CV_GET_PROC_ARCH(model) ((model) & CV_PROC_ARCH_MASK) typedef struct CvProcessorInfo { int model; int count; double frequency; // clocks per microsecond } CvProcessorInfo; #undef MASM_INLINE_ASSEMBLY #if defined WIN32 && !defined WIN64 #if defined _MSC_VER #define MASM_INLINE_ASSEMBLY 1 #elif defined __BORLANDC__ #if __BORLANDC__ >= 0x560 #define MASM_INLINE_ASSEMBLY 1 #endif #endif #endif /* determine processor type */ static void icvInitProcessorInfo( CvProcessorInfo* cpu_info ) { memset( cpu_info, 0, sizeof(*cpu_info) ); cpu_info->model = CV_PROC_GENERIC; #if defined WIN32 || defined WIN64 #ifndef PROCESSOR_ARCHITECTURE_AMD64 #define PROCESSOR_ARCHITECTURE_AMD64 9 #endif #ifndef PROCESSOR_ARCHITECTURE_IA32_ON_WIN64 #define PROCESSOR_ARCHITECTURE_IA32_ON_WIN64 10 #endif SYSTEM_INFO sys; LARGE_INTEGER freq; GetSystemInfo( &sys ); if( sys.wProcessorArchitecture == PROCESSOR_ARCHITECTURE_INTEL && sys.dwProcessorType == PROCESSOR_INTEL_PENTIUM && sys.wProcessorLevel >= 6 ) { int version = 0, features = 0, family = 0; int id = 0; HKEY key = 0; cpu_info->count = (int)sys.dwNumberOfProcessors; unsigned long val = 0, sz = sizeof(val); if( RegOpenKeyEx( HKEY_LOCAL_MACHINE, "HARDWARE\\DESCRIPTION\\SYSTEM\\CentralProcessor\\0\\", 0, KEY_QUERY_VALUE, &key ) >= 0 ) { if( RegQueryValueEx( key, "~MHz", 0, 0, (uchar*)&val, &sz ) >= 0 ) cpu_info->frequency = (double)val; RegCloseKey( key ); } #ifdef MASM_INLINE_ASSEMBLY __asm { /* use CPUID to determine the features supported */ pushfd mov eax, 1 push ebx push esi push edi #ifdef __BORLANDC__ db 0fh db 0a2h #else _emit 0x0f _emit 0xa2 #endif pop edi pop esi pop ebx mov version, eax mov features, edx popfd } #elif defined WIN32 && __GNUC__ > 2 asm volatile ( "movl $1,%%eax\n\t" ".byte 0x0f; .byte 0xa2\n\t" "movl %%eax, %0\n\t" "movl %%edx, %1\n\t" : "=r"(version), "=r" (features) : : "%ebx", "%esi", "%edi" ); #else { static const char cpuid_code[] = "\x53\x56\x57\xb8\x01\x00\x00\x00\x0f\xa2\x5f\x5e\x5b\xc3"; typedef int64 (CV_CDECL * func_ptr)(void); func_ptr cpuid = (func_ptr)(void*)cpuid_code; int64 cpuid_val = cpuid(); version = (int)cpuid_val; features = (int)(cpuid_val >> 32); } #endif #define ICV_CPUID_M6 ((1<<15)|(1<<23)) /* cmov + MMX */ #define ICV_CPUID_A6 ((1<<25)|ICV_CPUID_M6) /* + SSE */ #define ICV_CPUID_W7 ((1<<26)|ICV_CPUID_A6) /* + SSE2 */ family = (version >> 8) & 15; if( family >= 6 && (features & ICV_CPUID_M6) != 0 ) /* Pentium II or higher */ id = features & ICV_CPUID_W7; cpu_info->model = id == ICV_CPUID_W7 ? CV_PROC_IA32_WITH_SSE2 : id == ICV_CPUID_A6 ? CV_PROC_IA32_WITH_SSE : id == ICV_CPUID_M6 ? CV_PROC_IA32_WITH_MMX : CV_PROC_IA32_GENERIC; } else { #if defined EM64T if( sys.wProcessorArchitecture == PROCESSOR_ARCHITECTURE_AMD64 ) cpu_info->model = CV_PROC_EM64T; #elif defined WIN64 if( sys.wProcessorArchitecture == PROCESSOR_ARCHITECTURE_IA64 ) cpu_info->model = CV_PROC_IA64; #endif if( QueryPerformanceFrequency( &freq ) ) cpu_info->frequency = (double)freq.QuadPart; } #else cpu_info->frequency = 1; #ifdef __x86_64__ cpu_info->model = CV_PROC_EM64T; #elif defined __ia64__ cpu_info->model = CV_PROC_IA64; #elif !defined __i386__ cpu_info->model = CV_PROC_GENERIC; #else cpu_info->model = CV_PROC_IA32_GENERIC; // reading /proc/cpuinfo file (proc file system must be supported) FILE *file = fopen( "/proc/cpuinfo", "r" ); if( file ) { char buffer[1024]; int max_size = sizeof(buffer)-1; for(;;) { const char* ptr = fgets( buffer, max_size, file ); if( !ptr ) break; if( strncmp( buffer, "flags", 5 ) == 0 ) { if( strstr( buffer, "mmx" ) && strstr( buffer, "cmov" )) { cpu_info->model = CV_PROC_IA32_WITH_MMX; if( strstr( buffer, "xmm" ) || strstr( buffer, "sse" )) { cpu_info->model = CV_PROC_IA32_WITH_SSE; if( strstr( buffer, "emm" )) cpu_info->model = CV_PROC_IA32_WITH_SSE2; } } } else if( strncmp( buffer, "cpu MHz", 7 ) == 0 ) { char* pos = strchr( buffer, ':' ); if( pos ) cpu_info->frequency = strtod( pos + 1, &pos ); } } fclose( file ); if( CV_GET_PROC_ARCH(cpu_info->model) != CV_PROC_IA32_GENERIC ) cpu_info->frequency = 1; else assert( cpu_info->frequency > 1 ); } #endif #endif } CV_INLINE const CvProcessorInfo* icvGetProcessorInfo() { static CvProcessorInfo cpu_info; static int init_cpu_info = 0; if( !init_cpu_info ) { icvInitProcessorInfo( &cpu_info ); init_cpu_info = 1; } return &cpu_info; } /****************************************************************************************/ /* Make functions descriptions */ /****************************************************************************************/ #undef IPCVAPI_EX #define IPCVAPI_EX(type,func_name,names,modules,arg) \ { (void**)&func_name##_p, (void*)(size_t)-1, names, modules, 0 }, #undef IPCVAPI_C_EX #define IPCVAPI_C_EX(type,func_name,names,modules,arg) \ { (void**)&func_name##_p, (void*)(size_t)-1, names, modules, 0 }, static CvPluginFuncInfo cxcore_ipp_tab[] = { #undef _CXCORE_IPP_H_ #include "_cxipp.h" #undef _CXCORE_IPP_H_ {0, 0, 0, 0, 0} }; /* determine processor type, load appropriate dll and initialize all function pointers */ #if defined WIN32 || defined WIN64 #define DLL_PREFIX "" #define DLL_SUFFIX ".dll" #else #define DLL_PREFIX "lib" #define DLL_SUFFIX ".so" #define LoadLibrary(name) dlopen(name, RTLD_LAZY) #define FreeLibrary(name) dlclose(name) #define GetProcAddress dlsym typedef void* HMODULE; #endif #if 0 /*def _DEBUG*/ #define DLL_DEBUG_FLAG "d" #else #define DLL_DEBUG_FLAG "" #endif #define VERBOSE_LOADING 0 #if VERBOSE_LOADING #define ICV_PRINTF(args) printf args; fflush(stdout) #else #define ICV_PRINTF(args) #endif typedef struct CvPluginInfo { const char* basename; HMODULE handle; char name[100]; } CvPluginInfo; static CvPluginInfo plugins[CV_PLUGIN_MAX]; static CvModuleInfo cxcore_info = { 0, "cxcore", CV_VERSION, cxcore_ipp_tab }; CvModuleInfo *CvModule::first = 0, *CvModule::last = 0; CvModule::CvModule( CvModuleInfo* _info ) { cvRegisterModule( _info ); info = last; } CvModule::~CvModule() { if( info ) { CvModuleInfo* p = first; for( ; p != 0 && p->next != info; p = p->next ) ; if( p ) p->next = info->next; if( first == info ) first = info->next; if( last == info ) last = p; cvFree( &info ); info = 0; } } static int icvUpdatePluginFuncTab( CvPluginFuncInfo* func_tab ) { int i, loaded_functions = 0; // 1. reset pointers for( i = 0; func_tab[i].func_addr != 0; i++ ) { if( func_tab[i].default_func_addr == (void*)(size_t)-1 ) func_tab[i].default_func_addr = *func_tab[i].func_addr; else *func_tab[i].func_addr = func_tab[i].default_func_addr; func_tab[i].loaded_from = 0; } // ippopencv substitutes all the other IPP modules if( plugins[CV_PLUGIN_OPTCV].handle != 0 ) { for( i = 2; i < CV_PLUGIN_MKL; i++ ) { assert( plugins[i].handle == 0 ); plugins[i].handle = plugins[CV_PLUGIN_OPTCV].handle; } } // 2. try to find corresponding functions in ipp* and reassign pointers to them for( i = 0; func_tab[i].func_addr != 0; i++ ) { #if defined _MSC_VER && _MSC_VER >= 1200 #pragma warning( disable: 4054 4055 ) /* converting pointers to code<->data */ #endif char name[100]; int j = 0, idx = 0; assert( func_tab[i].loaded_from == 0 ); if( func_tab[i].search_modules ) { uchar* addr = 0; const char* name_ptr = func_tab[i].func_names; for( ; j < 10 && name_ptr; j++ ) { const char* name_start = name_ptr; const char* name_end; while( !isalpha(name_start[0]) && name_start[0] != '\0' ) name_start++; if( !name_start[0] ) name_start = 0; name_end = name_start ? strchr( name_start, ',' ) : 0; idx = (func_tab[i].search_modules / (1<= 1200 #pragma warning( default: 4054 4055 ) #endif } } #if VERBOSE_LOADING { int not_loaded = 0; ICV_PRINTF(("\nTotal loaded: %d\n\n", loaded_functions )); printf( "***************************************************\nNot loaded ...\n\n" ); for( i = 0; func_tab[i].func_addr != 0; i++ ) if( !func_tab[i].loaded_from ) { ICV_PRINTF(( "%s\n", func_tab[i].func_names )); not_loaded++; } ICV_PRINTF(("\nTotal: %d\n", not_loaded )); } #endif if( plugins[CV_PLUGIN_OPTCV].handle != 0 ) { for( i = 2; i < CV_PLUGIN_MKL; i++ ) plugins[i].handle = 0; } return loaded_functions; } CV_IMPL int cvRegisterModule( const CvModuleInfo* module ) { CvModuleInfo* module_copy = 0; CV_FUNCNAME( "cvRegisterModule" ); __BEGIN__; size_t name_len, version_len; CV_ASSERT( module != 0 && module->name != 0 && module->version != 0 ); name_len = strlen(module->name); version_len = strlen(module->version); CV_CALL( module_copy = (CvModuleInfo*)cvAlloc( sizeof(*module_copy) + name_len + 1 + version_len + 1 )); *module_copy = *module; module_copy->name = (char*)(module_copy + 1); module_copy->version = (char*)(module_copy + 1) + name_len + 1; memcpy( (void*)module_copy->name, module->name, name_len + 1 ); memcpy( (void*)module_copy->version, module->version, version_len + 1 ); module_copy->next = 0; if( CvModule::first == 0 ) CvModule::first = module_copy; else CvModule::last->next = module_copy; CvModule::last = module_copy; if( CvModule::first == CvModule::last ) { CV_CALL( cvUseOptimized(1)); } else { CV_CALL( icvUpdatePluginFuncTab( module_copy->func_tab )); } __END__; if( cvGetErrStatus() < 0 && module_copy ) cvFree( &module_copy ); return module_copy ? 0 : -1; } CV_IMPL int cvUseOptimized( int load_flag ) { int i, loaded_modules = 0, loaded_functions = 0; CvModuleInfo* module; const CvProcessorInfo* cpu_info = icvGetProcessorInfo(); int arch = CV_GET_PROC_ARCH(cpu_info->model); // TODO: implement some more elegant way // to find the latest and the greatest IPP/MKL libraries static const char* opencv_sfx[] = { "100", "099", "097", 0 }; static const char* ipp_sfx_ia32[] = { "-6.1", "-6.0", "-5.3", "-5.2", "-5.1", "", 0 }; static const char* ipp_sfx_ia64[] = { "64-6.1", "64-6.0", "64-5.3", "64-5.2", "64-5.1", "64", 0 }; static const char* ipp_sfx_em64t[] = { "em64t-6.1", "em64t-6.0", "em64t-5.3", "em64t-5.2", "em64t-5.1", "em64t", 0 }; static const char* mkl_sfx_ia32[] = { "p4", "p3", "def", 0 }; static const char* mkl_sfx_ia64[] = { "i2p", "itp", 0 }; static const char* mkl_sfx_em64t[] = { "def", 0 }; const char** ipp_suffix = arch == CV_PROC_IA64 ? ipp_sfx_ia64 : arch == CV_PROC_EM64T ? ipp_sfx_em64t : ipp_sfx_ia32; const char** mkl_suffix = arch == CV_PROC_IA64 ? mkl_sfx_ia64 : arch == CV_PROC_EM64T ? mkl_sfx_em64t : mkl_sfx_ia32; for( i = 0; i < CV_PLUGIN_MAX; i++ ) plugins[i].basename = 0; plugins[CV_PLUGIN_NONE].basename = 0; plugins[CV_PLUGIN_NONE].name[0] = '\0'; plugins[CV_PLUGIN_OPTCV].basename = "ippopencv"; plugins[CV_PLUGIN_IPPCV].basename = "ippcv"; plugins[CV_PLUGIN_IPPI].basename = "ippi"; plugins[CV_PLUGIN_IPPS].basename = "ipps"; plugins[CV_PLUGIN_IPPVM].basename = "ippvm"; plugins[CV_PLUGIN_IPPCC].basename = "ippcc"; plugins[CV_PLUGIN_MKL].basename = "mkl_"; // try to load optimized dlls for( i = 1; i < CV_PLUGIN_MAX; i++ ) { // unload previously loaded optimized modules if( plugins[i].handle ) { FreeLibrary( plugins[i].handle ); plugins[i].handle = 0; } // do not load regular IPP modules if the custom merged IPP module is already found. if( i < CV_PLUGIN_MKL && load_flag && plugins[CV_PLUGIN_OPTCV].handle != 0 ) continue; if( load_flag && plugins[i].basename && (arch == CV_PROC_IA32_GENERIC || arch == CV_PROC_IA64 || arch == CV_PROC_EM64T) ) { const char** suffix = i == CV_PLUGIN_OPTCV ? opencv_sfx : i < CV_PLUGIN_MKL ? ipp_suffix : mkl_suffix; if( suffix == mkl_sfx_ia32 ) { if( !(cpu_info->model & CV_PROC_IA32_WITH_SSE2) ) suffix++; if( !(cpu_info->model & CV_PROC_IA32_WITH_SSE) ) suffix++; } for( ; *suffix != 0; suffix++ ) { sprintf( plugins[i].name, DLL_PREFIX "%s%s" DLL_DEBUG_FLAG DLL_SUFFIX, plugins[i].basename, *suffix ); ICV_PRINTF(("loading %s...\n", plugins[i].name )); plugins[i].handle = LoadLibrary( plugins[i].name ); if( plugins[i].handle != 0 ) { ICV_PRINTF(("%s loaded\n", plugins[i].name )); loaded_modules++; break; } #ifndef WIN32 // temporary workaround for MacOSX sprintf( plugins[i].name, DLL_PREFIX "%s%s" DLL_DEBUG_FLAG ".dylib", plugins[i].basename, *suffix ); ICV_PRINTF(("loading %s...\n", plugins[i].name )); plugins[i].handle = LoadLibrary( plugins[i].name ); if( plugins[i].handle != 0 ) { ICV_PRINTF(("%s loaded\n", plugins[i].name )); loaded_modules++; break; } #endif } } } for( module = CvModule::first; module != 0; module = module->next ) loaded_functions += icvUpdatePluginFuncTab( module->func_tab ); return loaded_functions; } CvModule cxcore_module( &cxcore_info ); CV_IMPL void cvGetModuleInfo( const char* name, const char **version, const char **plugin_list ) { static char joint_verinfo[1024] = ""; static char plugin_list_buf[1024] = ""; CV_FUNCNAME( "cvGetLibraryInfo" ); if( version ) *version = 0; if( plugin_list ) *plugin_list = 0; __BEGIN__; CvModuleInfo* module; if( version ) { if( name ) { size_t i, name_len = strlen(name); for( module = CvModule::first; module != 0; module = module->next ) { if( strlen(module->name) == name_len ) { for( i = 0; i < name_len; i++ ) { int c0 = toupper(module->name[i]), c1 = toupper(name[i]); if( c0 != c1 ) break; } if( i == name_len ) break; } } if( !module ) CV_ERROR( CV_StsObjectNotFound, "The module is not found" ); *version = module->version; } else { char* ptr = joint_verinfo; for( module = CvModule::first; module != 0; module = module->next ) { sprintf( ptr, "%s: %s%s", module->name, module->version, module->next ? ", " : "" ); ptr += strlen(ptr); } *version = joint_verinfo; } } if( plugin_list ) { char* ptr = plugin_list_buf; int i; for( i = 0; i < CV_PLUGIN_MAX; i++ ) if( plugins[i].handle != 0 ) { sprintf( ptr, "%s, ", plugins[i].name ); ptr += strlen(ptr); } if( ptr > plugin_list_buf ) { ptr[-2] = '\0'; *plugin_list = plugin_list_buf; } else *plugin_list = ""; } __END__; } typedef int64 (CV_CDECL * rdtsc_func)(void); /* helper functions for RNG initialization and accurate time measurement */ CV_IMPL int64 cvGetTickCount( void ) { const CvProcessorInfo* cpu_info = icvGetProcessorInfo(); if( cpu_info->frequency > 1 && CV_GET_PROC_ARCH(cpu_info->model) == CV_PROC_IA32_GENERIC ) { #ifdef MASM_INLINE_ASSEMBLY #ifdef __BORLANDC__ __asm db 0fh __asm db 31h #else __asm _emit 0x0f; __asm _emit 0x31; #endif #elif (defined __GNUC__ || defined CV_ICC) && defined __i386__ int64 t; asm volatile (".byte 0xf; .byte 0x31" /* "rdtsc" */ : "=A" (t)); return t; #else static const char code[] = "\x0f\x31\xc3"; rdtsc_func func = (rdtsc_func)(void*)code; return func(); #endif } else { #if defined WIN32 || defined WIN64 LARGE_INTEGER counter; QueryPerformanceCounter( &counter ); return (int64)counter.QuadPart; #else struct timeval tv; struct timezone tz; gettimeofday( &tv, &tz ); return (int64)tv.tv_sec*1000000 + tv.tv_usec; #endif } } CV_IMPL double cvGetTickFrequency() { return icvGetProcessorInfo()->frequency; } static int icvNumThreads = 0; static int icvNumProcs = 0; CV_IMPL int cvGetNumThreads(void) { if( !icvNumProcs ) cvSetNumThreads(0); return icvNumThreads; } CV_IMPL void cvSetNumThreads( int threads ) { if( !icvNumProcs ) { #ifdef _OPENMP icvNumProcs = omp_get_num_procs(); icvNumProcs = MIN( icvNumProcs, CV_MAX_THREADS ); #else icvNumProcs = 1; #endif } #ifdef _OPENMP if( threads <= 0 ) threads = icvNumProcs; //else // threads = MIN( threads, icvNumProcs ); icvNumThreads = threads; #else icvNumThreads = 1; #endif } CV_IMPL int cvGetThreadNum(void) { #ifdef _OPENMP return omp_get_thread_num(); #else return 0; #endif } /* End of file. */