float_cast.h (5955B)
1 /* Copyright (C) 2001 Erik de Castro Lopo <erikd AT mega-nerd DOT com> */ 2 /* 3 Redistribution and use in source and binary forms, with or without 4 modification, are permitted provided that the following conditions 5 are met: 6 7 - Redistributions of source code must retain the above copyright 8 notice, this list of conditions and the following disclaimer. 9 10 - Redistributions in binary form must reproduce the above copyright 11 notice, this list of conditions and the following disclaimer in the 12 documentation and/or other materials provided with the distribution. 13 14 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 15 ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 16 LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 17 A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER 18 OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 19 EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 20 PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 21 PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 22 LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 23 NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 24 SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 25 */ 26 27 /* Version 1.1 */ 28 29 #ifndef FLOAT_CAST_H 30 #define FLOAT_CAST_H 31 32 33 #include "arch.h" 34 35 /*============================================================================ 36 ** On Intel Pentium processors (especially PIII and probably P4), converting 37 ** from float to int is very slow. To meet the C specs, the code produced by 38 ** most C compilers targeting Pentium needs to change the FPU rounding mode 39 ** before the float to int conversion is performed. 40 ** 41 ** Changing the FPU rounding mode causes the FPU pipeline to be flushed. It 42 ** is this flushing of the pipeline which is so slow. 43 ** 44 ** Fortunately the ISO C99 specifications define the functions lrint, lrintf, 45 ** llrint and llrintf which fix this problem as a side effect. 46 ** 47 ** On Unix-like systems, the configure process should have detected the 48 ** presence of these functions. If they weren't found we have to replace them 49 ** here with a standard C cast. 50 */ 51 52 /* 53 ** The C99 prototypes for lrint and lrintf are as follows: 54 ** 55 ** long int lrintf (float x) ; 56 ** long int lrint (double x) ; 57 */ 58 59 /* The presence of the required functions are detected during the configure 60 ** process and the values HAVE_LRINT and HAVE_LRINTF are set accordingly in 61 ** the config.h file. 62 */ 63 64 /* With GCC, when SSE is available, the fastest conversion is cvtss2si. */ 65 #if defined(__GNUC__) && defined(__SSE__) 66 67 #include <xmmintrin.h> 68 static OPUS_INLINE opus_int32 float2int(float x) {return _mm_cvt_ss2si(_mm_set_ss(x));} 69 70 #elif (defined(_MSC_VER) && _MSC_VER >= 1400) && (defined(_M_X64) || (defined(_M_IX86_FP) && _M_IX86_FP >= 1)) 71 72 #include <xmmintrin.h> 73 static OPUS_INLINE opus_int32 float2int(float value) 74 { 75 /* _mm_load_ss will generate same code as _mm_set_ss 76 ** in _MSC_VER >= 1914 /02 so keep __mm_load__ss 77 ** for backward compatibility. 78 */ 79 return _mm_cvtss_si32(_mm_load_ss(&value)); 80 } 81 82 #elif (defined(_MSC_VER) && _MSC_VER >= 1400) && defined (_M_IX86) 83 84 #include <math.h> 85 86 /* Win32 doesn't seem to have these functions. 87 ** Therefore implement OPUS_INLINE versions of these functions here. 88 */ 89 90 static OPUS_INLINE opus_int32 91 float2int (float flt) 92 { int intgr; 93 94 _asm 95 { fld flt 96 fistp intgr 97 } ; 98 99 return intgr ; 100 } 101 #elif defined(__aarch64__) 102 103 #include <arm_neon.h> 104 static OPUS_INLINE opus_int32 float2int(float flt) 105 { 106 return vcvtns_s32_f32(flt); 107 } 108 109 #elif defined(HAVE_LRINTF) && defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L 110 111 /* These defines enable functionality introduced with the 1999 ISO C 112 ** standard. They must be defined before the inclusion of math.h to 113 ** engage them. If optimisation is enabled, these functions will be 114 ** inlined. With optimisation switched off, you have to link in the 115 ** maths library using -lm. 116 */ 117 118 #define _ISOC9X_SOURCE 1 119 #define _ISOC99_SOURCE 1 120 121 #define __USE_ISOC9X 1 122 #define __USE_ISOC99 1 123 124 #include <math.h> 125 #define float2int(x) lrintf(x) 126 127 #elif defined(HAVE_LRINT) && defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L 128 129 #define _ISOC9X_SOURCE 1 130 #define _ISOC99_SOURCE 1 131 132 #define __USE_ISOC9X 1 133 #define __USE_ISOC99 1 134 135 #include <math.h> 136 #define float2int(x) lrint(x) 137 138 #else 139 140 #if (defined(__GNUC__) && defined(__STDC__) && __STDC__ && __STDC_VERSION__ >= 199901L) 141 /* supported by gcc in C99 mode, but not by all other compilers */ 142 #warning "Don't have the functions lrint() and lrintf ()." 143 #warning "Replacing these functions with a standard C cast." 144 #endif /* __STDC_VERSION__ >= 199901L */ 145 #include <math.h> 146 #define float2int(flt) ((int)(floor(.5+flt))) 147 #endif 148 149 #ifndef DISABLE_FLOAT_API 150 static OPUS_INLINE opus_int16 FLOAT2INT16(float x) 151 { 152 x = x*CELT_SIG_SCALE; 153 x = MAX32(x, -32768); 154 x = MIN32(x, 32767); 155 return (opus_int16)float2int(x); 156 } 157 158 static OPUS_INLINE opus_int32 FLOAT2INT24(float x) 159 { 160 x = x*(CELT_SIG_SCALE*256.f); 161 x = MAX32(x, -16777216); 162 x = MIN32(x, 16777216); 163 return float2int(x); 164 } 165 #ifdef FIXED_POINT 166 static OPUS_INLINE opus_int32 FLOAT2SIG(float x) 167 { 168 x = x*((opus_int32)32768<<SIG_SHIFT); 169 x = MAX32(x, -(65536<<SIG_SHIFT)); 170 x = MIN32(x, 65536<<SIG_SHIFT); 171 return float2int(x); 172 } 173 #endif 174 #endif /* DISABLE_FLOAT_API */ 175 176 #endif /* FLOAT_CAST_H */