Showing the changes to codec/decoder/core/src/decode_mb_aux.cpp (1 of 25 files). Show all
Unified · Split
| @@ -76,6 +76,96 @@ void IdctResAddPred_c (uint8_t* pPred, const int32_t kiStride, int16_t* pRs) { · 75 unmodified lines | |||
| 76 | } | 76 | } |
| 77 | } | 77 | } |
| 78 | 78 | ||
| 79 | void IdctResAddPred8x8_c (uint8_t* pPred, const int32_t kiStride, int16_t* pRs) { | ||
| 80 | // To make the ASM code easy to write, should using one funciton to apply hor and ver together, such as we did on HEVC | ||
| 81 | // Ugly code, just for easy debug, the final version need optimization | ||
| 82 | int16_t p[8], b[8]; | ||
| 83 | int16_t a[4]; | ||
| 84 | |||
| 85 | int16_t iTmp[64]; | ||
| 86 | int16_t iRes[64]; | ||
| 87 | |||
| 88 | // Horizontal | ||
| 89 | for (int i = 0; i < 8; i++) { | ||
| 90 | for (int j = 0; j < 8; j++) { | ||
| 91 | p[j] = pRs[j + (i << 3)]; | ||
| 92 | } | ||
| 93 | a[0] = p[0] + p[4]; | ||
| 94 | a[1] = p[0] - p[4]; | ||
| 95 | a[2] = p[6] - (p[2] >> 1); | ||
| 96 | a[3] = p[2] + (p[6] >> 1); | ||
| 97 | |||
| 98 | b[0] = a[0] + a[3]; | ||
| 99 | b[2] = a[1] - a[2]; | ||
| 100 | b[4] = a[1] + a[2]; | ||
| 101 | b[6] = a[0] - a[3]; | ||
| 102 | |||
| 103 | a[0] = -p[3] + p[5] - p[7] - (p[7] >> 1); | ||
| 104 | a[1] = p[1] + p[7] - p[3] - (p[3] >> 1); | ||
| 105 | a[2] = -p[1] + p[7] + p[5] + (p[5] >> 1); | ||
| 106 | a[3] = p[3] + p[5] + p[1] + (p[1] >> 1); | ||
| 107 | |||
| 108 | b[1] = a[0] + (a[3] >> 2); | ||
| 109 | b[3] = a[1] + (a[2] >> 2); | ||
| 110 | b[5] = a[2] - (a[1] >> 2); | ||
| 111 | b[7] = a[3] - (a[0] >> 2); | ||
| 112 | |||
| 113 | iTmp[0 + (i << 3)] = b[0] + b[7]; | ||
| 114 | iTmp[1 + (i << 3)] = b[2] - b[5]; | ||
| 115 | iTmp[2 + (i << 3)] = b[4] + b[3]; | ||
| 116 | iTmp[3 + (i << 3)] = b[6] + b[1]; | ||
| 117 | iTmp[4 + (i << 3)] = b[6] - b[1]; | ||
| 118 | iTmp[5 + (i << 3)] = b[4] - b[3]; | ||
| 119 | iTmp[6 + (i << 3)] = b[2] + b[5]; | ||
| 120 | iTmp[7 + (i << 3)] = b[0] - b[7]; | ||
| 121 | } | ||
| 122 | |||
| 123 | //Vertical | ||
| 124 | for (int i = 0; i < 8; i++) { | ||
| 125 | for (int j = 0; j < 8; j++) { | ||
| 126 | p[j] = iTmp[i + (j << 3)]; | ||
| 127 | } | ||
| 128 | |||
| 129 | a[0] = p[0] + p[4]; | ||
| 130 | a[1] = p[0] - p[4]; | ||
| 131 | a[2] = p[6] - (p[2] >> 1); | ||
| 132 | a[3] = p[2] + (p[6] >> 1); | ||
| 133 | |||
| 134 | b[0] = a[0] + a[3]; | ||
| 135 | b[2] = a[1] - a[2]; | ||
| 136 | b[4] = a[1] + a[2]; | ||
| 137 | b[6] = a[0] - a[3]; | ||
| 138 | |||
| 139 | a[0] = -p[3] + p[5] - p[7] - (p[7] >> 1); | ||
| 140 | a[1] = p[1] + p[7] - p[3] - (p[3] >> 1); | ||
| 141 | a[2] = -p[1] + p[7] + p[5] + (p[5] >> 1); | ||
| 142 | a[3] = p[3] + p[5] + p[1] + (p[1] >> 1); | ||
| 143 | |||
| 144 | |||
| 145 | b[1] = a[0] + (a[3] >> 2); | ||
| 146 | b[7] = a[3] - (a[0] >> 2); | ||
| 147 | b[3] = a[1] + (a[2] >> 2); | ||
| 148 | b[5] = a[2] - (a[1] >> 2); | ||
| 149 | |||
| 150 | iRes[ (0 << 3) + i] = b[0] + b[7]; | ||
| 151 | iRes[ (1 << 3) + i] = b[2] - b[5]; | ||
| 152 | iRes[ (2 << 3) + i] = b[4] + b[3]; | ||
| 153 | iRes[ (3 << 3) + i] = b[6] + b[1]; | ||
| 154 | iRes[ (4 << 3) + i] = b[6] - b[1]; | ||
| 155 | iRes[ (5 << 3) + i] = b[4] - b[3]; | ||
| 156 | iRes[ (6 << 3) + i] = b[2] + b[5]; | ||
| 157 | iRes[ (7 << 3) + i] = b[0] - b[7]; | ||
| 158 | } | ||
| 159 | |||
| 160 | uint8_t* pDst = pPred; | ||
| 161 | for (int i = 0; i < 8; i++) { | ||
| 162 | for (int j = 0; j < 8; j++) { | ||
| 163 | pDst[i * kiStride + j] = WelsClip1 (((32 + iRes[ (i << 3) + j]) >> 6) + pDst[i * kiStride + j]); | ||
| 164 | } | ||
| 165 | } | ||
| 166 | |||
| 167 | } | ||
| 168 | |||
| 79 | void GetI4LumaIChromaAddrTable (int32_t* pBlockOffset, const int32_t kiYStride, const int32_t kiUVStride) { | 169 | void GetI4LumaIChromaAddrTable (int32_t* pBlockOffset, const int32_t kiYStride, const int32_t kiUVStride) { |
| 80 | int32_t* pOffset = pBlockOffset; | 170 | int32_t* pOffset = pBlockOffset; |
| 81 | int32_t i; | 171 | int32_t i; |