1 /*! 2 * \copy 3 * Copyright (c) 2013, Cisco Systems 4 * All rights reserved. 5 * 6 * Redistribution and use in source and binary forms, with or without 7 * modification, are permitted provided that the following conditions 8 * are met: 9 * 10 * * Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 13 * * Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in 15 * the documentation and/or other materials provided with the 16 * distribution. 17 * 18 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 19 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 20 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 21 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 22 * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 23 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, 24 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 25 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER 26 * CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN 28 * ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 * 31 */ 32 33 #ifndef ENCODE_MB_AUX_H 34 #define ENCODE_MB_AUX_H 35 36 #include "typedefs.h" 37 #include "wels_func_ptr_def.h" 38 #include "copy_mb.h" 39 40 namespace WelsEnc { 41 void WelsInitEncodingFuncs (SWelsFuncPtrList* pFuncList, uint32_t uiCpuFlag); 42 int32_t WelsGetNoneZeroCount_c (int16_t* pLevel); 43 44 /**************************************************************************** 45 * Scan and Score functions 46 ****************************************************************************/ 47 void WelsScan4x4Ac_c (int16_t* pZigValue, int16_t* pDct); 48 void WelsScan4x4Dc (int16_t* pLevel, int16_t* pDct); 49 void WelsScan4x4DcAc_c (int16_t* pLevel, int16_t* pDct); 50 int32_t WelsCalculateSingleCtr4x4_c (int16_t* pDct); 51 52 /**************************************************************************** 53 * HDM and Quant functions 54 ****************************************************************************/ 55 void WelsHadamardT4Dc_c (int16_t* pLumaDc, int16_t* pDct); 56 int32_t WelsHadamardQuant2x2_c (int16_t* pRes, const int16_t kiFF, int16_t iMF, int16_t* pDct, int16_t* pBlock); 57 int32_t WelsHadamardQuant2x2Skip_c (int16_t* pRes, int16_t iFF, int16_t iMF); 58 59 void WelsQuant4x4_c (int16_t* pDct, const int16_t* pFF, const int16_t* pMF); 60 void WelsQuant4x4Dc_c (int16_t* pDct, int16_t iFF, int16_t iMF); 61 void WelsQuantFour4x4_c (int16_t* pDct, const int16_t* pFF, const int16_t* pQpTable); 62 void WelsQuantFour4x4Max_c (int16_t* pDct, const int16_t* pF, const int16_t* pQpTable, int16_t* pMax); 63 64 65 /**************************************************************************** 66 * DCT functions 67 ****************************************************************************/ 68 void WelsDctT4_c (int16_t* pDct, uint8_t* pPixel1, int32_t iStride1, uint8_t* pPixel2, int32_t iStride2); 69 // dct_data is no-use here, just for the same interface with dct_save functions 70 void WelsDctFourT4_c (int16_t* pDct, uint8_t* pPixel1, int32_t iStride1, uint8_t* pPixel2, int32_t iStride2); 71 72 #if defined(__cplusplus) 73 extern "C" { 74 #endif//__cplusplus 75 76 #ifdef X86_ASM 77 78 int32_t WelsGetNoneZeroCount_sse2 (int16_t* pLevel); 79 int32_t WelsGetNoneZeroCount_sse42 (int16_t* pLevel); 80 81 /**************************************************************************** 82 * Scan and Score functions 83 ****************************************************************************/ 84 void WelsScan4x4Ac_sse2 (int16_t* zig_value, int16_t* pDct); 85 void WelsScan4x4DcAc_ssse3 (int16_t* pLevel, int16_t* pDct); 86 void WelsScan4x4DcAc_sse2 (int16_t* pLevel, int16_t* pDct); 87 int32_t WelsCalculateSingleCtr4x4_sse2 (int16_t* pDct); 88 89 /**************************************************************************** 90 * DCT functions 91 ****************************************************************************/ 92 void WelsDctT4_mmx (int16_t* pDct, uint8_t* pPixel1, int32_t iStride1, uint8_t* pPixel2, int32_t iStride2); 93 void WelsDctT4_sse2 (int16_t* pDct, uint8_t* pPixel1, int32_t iStride1, uint8_t* pPixel2, int32_t iStride2); 94 void WelsDctFourT4_sse2 (int16_t* pDct, uint8_t* pPixel1, int32_t iStride1, uint8_t* pPixel2, int32_t iStride2); 95 void WelsDctT4_avx2 (int16_t* pDct, uint8_t* pPixel1, int32_t iStride1, uint8_t* pPixel2, int32_t iStride2); 96 void WelsDctFourT4_avx2 (int16_t* pDct, uint8_t* pPixel1, int32_t iStride1, uint8_t* pPixel2, int32_t iStride2); 97 98 /**************************************************************************** 99 * HDM and Quant functions 100 ****************************************************************************/ 101 int32_t WelsHadamardQuant2x2_mmx (int16_t* pRes, const int16_t kiFF, int16_t iMF, int16_t* pDct, int16_t* pBlock); 102 void WelsHadamardT4Dc_sse2 (int16_t* pLumaDc, int16_t* pDct); 103 int32_t WelsHadamardQuant2x2Skip_mmx (int16_t* pRes, int16_t iFF, int16_t iMF); 104 105 void WelsQuant4x4_sse2 (int16_t* pDct, const int16_t* pFF, const int16_t* pMF); 106 void WelsQuant4x4Dc_sse2 (int16_t* pDct, int16_t iFF, int16_t iMF); 107 void WelsQuantFour4x4_sse2 (int16_t* pDct, const int16_t* pFF, const int16_t* pMF); 108 void WelsQuantFour4x4Max_sse2 (int16_t* pDct, const int16_t* pFF, const int16_t* pMF, int16_t* pMax); 109 110 void WelsQuant4x4_avx2 (int16_t* pDct, const int16_t* pFF, const int16_t* pMF); 111 void WelsQuant4x4Dc_avx2 (int16_t* pDct, int16_t iFF, int16_t iMF); 112 void WelsQuantFour4x4_avx2 (int16_t* pDct, const int16_t* pFF, const int16_t* pMF); 113 void WelsQuantFour4x4Max_avx2 (int16_t* pDct, const int16_t* pFF, const int16_t* pMF, int16_t* pMax); 114 115 #endif 116 117 #ifdef HAVE_NEON 118 void WelsHadamardT4Dc_neon (int16_t* pLumaDc, int16_t* pDct); 119 int32_t WelsHadamardQuant2x2_neon (int16_t* pRes, const int16_t kiFF, int16_t iMF, int16_t* pDct, int16_t* pBlock); 120 int32_t WelsHadamardQuant2x2Skip_neon (int16_t* pRes, int16_t iFF, int16_t iMF); 121 int32_t WelsHadamardQuant2x2SkipKernel_neon (int16_t* pRes, int16_t iThreshold); // avoid divide operator 122 123 void WelsDctT4_neon (int16_t* pDct, uint8_t* pPixel1, int32_t iStride1, uint8_t* pPixel2, int32_t iStride2); 124 void WelsDctFourT4_neon (int16_t* pDct, uint8_t* pPixel1, int32_t iStride1, uint8_t* pPixel2, int32_t iStride2); 125 126 int32_t WelsGetNoneZeroCount_neon (int16_t* pLevel); 127 128 void WelsQuant4x4_neon (int16_t* pDct, const int16_t* pFF, const int16_t* pMF); 129 void WelsQuant4x4Dc_neon (int16_t* pDct, int16_t iFF, int16_t iMF); 130 void WelsQuantFour4x4_neon (int16_t* pDct, const int16_t* pFF, const int16_t* pMF); 131 void WelsQuantFour4x4Max_neon (int16_t* pDct, const int16_t* pFF, const int16_t* pMF, int16_t* pMax); 132 #endif 133 134 #ifdef HAVE_NEON_AARCH64 135 void WelsHadamardT4Dc_AArch64_neon (int16_t* pLumaDc, int16_t* pDct); 136 int32_t WelsHadamardQuant2x2_AArch64_neon (int16_t* pRes, const int16_t kiFF, int16_t iMF, int16_t* pDct, int16_t* pBlock); 137 int32_t WelsHadamardQuant2x2Skip_AArch64_neon (int16_t* pRes, int16_t iFF, int16_t iMF); 138 int32_t WelsHadamardQuant2x2SkipKernel_AArch64_neon (int16_t* pRes, int16_t iThreshold); // avoid divide operator 139 140 void WelsDctT4_AArch64_neon (int16_t* pDct, uint8_t* pPixel1, int32_t iStride1, uint8_t* pPixel2, int32_t iStride2); 141 void WelsDctFourT4_AArch64_neon (int16_t* pDct, uint8_t* pPixel1, int32_t iStride1, uint8_t* pPixel2, int32_t iStride2); 142 143 int32_t WelsGetNoneZeroCount_AArch64_neon (int16_t* pLevel); 144 145 void WelsQuant4x4_AArch64_neon (int16_t* pDct, const int16_t* pFF, const int16_t* pMF); 146 void WelsQuant4x4Dc_AArch64_neon (int16_t* pDct, int16_t iFF, int16_t iMF); 147 void WelsQuantFour4x4_AArch64_neon (int16_t* pDct, const int16_t* pFF, const int16_t* pMF); 148 void WelsQuantFour4x4Max_AArch64_neon (int16_t* pDct, const int16_t* pFF, const int16_t* pMF, int16_t* pMax); 149 #endif 150 151 #ifdef HAVE_MMI 152 int32_t WelsGetNoneZeroCount_mmi (int16_t* pLevel); 153 154 /**************************************************************************** 155 * * Scan and Score functions 156 * ****************************************************************************/ 157 void WelsScan4x4Ac_mmi (int16_t* zig_value, int16_t* pDct); 158 void WelsScan4x4DcAc_mmi (int16_t* pLevel, int16_t* pDct); 159 int32_t WelsCalculateSingleCtr4x4_mmi (int16_t* pDct); 160 161 /**************************************************************************** 162 * * DCT functions 163 * ****************************************************************************/ 164 void WelsDctT4_mmi (int16_t* pDct, uint8_t* pPixel1, int32_t iStride1, uint8_t* pPixel2, int32_t iStride2); 165 void WelsDctFourT4_mmi (int16_t* pDct, uint8_t* pPixel1, int32_t iStride1, uint8_t* pPixel2, int32_t iStride2); 166 167 /**************************************************************************** 168 * * HDM and Quant functions 169 * ****************************************************************************/ 170 void WelsHadamardT4Dc_mmi (int16_t* pLumaDc, int16_t* pDct); 171 172 void WelsQuant4x4_mmi (int16_t* pDct, const int16_t* pFF, const int16_t* pMF); 173 void WelsQuant4x4Dc_mmi (int16_t* pDct, int16_t iFF, int16_t iMF); 174 void WelsQuantFour4x4_mmi (int16_t* pDct, const int16_t* pFF, const int16_t* pMF); 175 void WelsQuantFour4x4Max_mmi (int16_t* pDct, const int16_t* pFF, const int16_t* pMF, int16_t* pMax); 176 #endif//HAVE_MMI 177 #if defined(__cplusplus) 178 } 179 #endif//__cplusplus 180 181 ALIGNED_DECLARE (extern const int16_t, g_kiQuantInterFF[58][8], 16); 182 #define g_iQuantIntraFF (g_kiQuantInterFF +6 ) 183 ALIGNED_DECLARE (extern const int16_t, g_kiQuantMF[52][8], 16); 184 } 185 #endif//ENCODE_MB_AUX_H 186