1 /* 2 * Copyright (c) 2019, Intel Corporation 3 * 4 * Permission is hereby granted, free of charge, to any person obtaining a 5 * copy of this software and associated documentation files (the "Software"), 6 * to deal in the Software without restriction, including without limitation 7 * the rights to use, copy, modify, merge, publish, distribute, sublicense, 8 * and/or sell copies of the Software, and to permit persons to whom the 9 * Software is furnished to do so, subject to the following conditions: 10 * 11 * The above copyright notice and this permission notice shall be included 12 * in all copies or substantial portions of the Software. 13 * 14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS 15 * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 17 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR 18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, 19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR 20 * OTHER DEALINGS IN THE SOFTWARE. 21 */ 22 23 { 24 /* 25 Buffer layout after shuffle 26 _________________________________________________ 27 |_______Block0__________|_______Block1__________| 28 |_______Block2__________|_______Block3__________| 29 |_______Block4__________|_______Block5__________| 30 |_______Block6__________|_______Block7__________| 31 32 Write back buffer layout correlate to the block number#, each box stands for 1 GRF 33 _______________________________________________ 34 |____R0_________R1_____|____R2_________R3_____| 35 |____G0_________G1_____|____G2_________G3_____| 36 |____B0_________B1_____|____B2_________B3_____| 37 |____A0_________A1_____|____A2_________A3_____| 38 |____R4_________R5_____|____R6_________R7_____| 39 |____G4_________G5_____|____G6_________G7_____| 40 |____B4_________B5_____|____B6_________B7_____| 41 |____A4_________A5_____|____A6_________A7_____| 42 */ 43 44 matrix_ref<uchar, 16, 16> Result_Y = DataBuffer.format<uchar, 192, 16>().select<16, 1, 16, 1>(128, 0); 45 matrix_ref<uchar, 8, 16> Result_UV = DataBuffer.format<uchar, 192, 16>().select<8, 1, 16, 1>(160, 0); 46 matrix_ref<ushort, 2, 16> Result_Temp_V = DataBuffer.format<ushort, 96, 16>().select<2, 1, 16, 1>(88, 0); 47 matrix_ref<ushort, 2, 16> Result_Temp_U = DataBuffer.format<ushort, 96, 16>().select<2, 1, 16, 1>(90, 0); 48 49 SurfaceIndex Dst_Surface_Y(MDF_FC_OUTPUT_BTI_START); 50 SurfaceIndex Dst_Surface_UV(MDF_FC_OUTPUT_BTI_START + MDF_FC_UV_PLANE_BTI_OFFSET); 51 52 #pragma unroll 53 // Rounding Y plane 54 for (uchar i = 0; i < 4; i++) 55 { 56 DataBuffer.format<ushort, 96, 16>().select<1, 1, 16, 1>(16 * i + 2, 0) = cm_add<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 16, 1>(16 * i + 2, 0), 0x80, SAT); 57 DataBuffer.format<ushort, 96, 16>().select<1, 1, 16, 1>(16 * i + 3, 0) = cm_add<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 16, 1>(16 * i + 3, 0), 0x80, SAT); 58 DataBuffer.format<ushort, 96, 16>().select<1, 1, 16, 1>(16 * i + 10, 0) = cm_add<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 16, 1>(16 * i + 10, 0), 0x80, SAT); 59 DataBuffer.format<ushort, 96, 16>().select<1, 1, 16, 1>(16 * i + 11, 0) = cm_add<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 16, 1>(16 * i + 11, 0), 0x80, SAT); 60 } 61 62 #pragma unroll 63 // Write Y plane 64 for (uchar i = 0; i < 4; i++) 65 { 66 Result_Y.select<1, 1, 8, 1>(4 * i, 0) = DataBuffer.format<uchar, 96, 32>().select<1, 1, 8, 2>(16 * i + 2, 1); 67 Result_Y.select<1, 1, 8, 1>(4 * i, 8) = DataBuffer.format<uchar, 96, 32>().select<1, 1, 8, 2>(16 * i + 10, 1); 68 Result_Y.select<1, 1, 8, 1>(4 * i + 1, 0) = DataBuffer.format<uchar, 96, 32>().select<1, 1, 8, 2>(16 * i + 2, 17); 69 Result_Y.select<1, 1, 8, 1>(4 * i + 1, 8) = DataBuffer.format<uchar, 96, 32>().select<1, 1, 8, 2>(16 * i + 10, 17); 70 Result_Y.select<1, 1, 8, 1>(4 * i + 2, 0) = DataBuffer.format<uchar, 96, 32>().select<1, 1, 8, 2>(16 * i + 3, 1); 71 Result_Y.select<1, 1, 8, 1>(4 * i + 2, 8) = DataBuffer.format<uchar, 96, 32>().select<1, 1, 8, 2>(16 * i + 11, 1); 72 Result_Y.select<1, 1, 8, 1>(4 * i + 3, 0) = DataBuffer.format<uchar, 96, 32>().select<1, 1, 8, 2>(16 * i + 3, 17); 73 Result_Y.select<1, 1, 8, 1>(4 * i + 3, 8) = DataBuffer.format<uchar, 96, 32>().select<1, 1, 8, 2>(16 * i + 11, 17); 74 } 75 76 write(Dst_Surface_Y, DstX, DstY, Result_Y); 77 78 #pragma unroll 79 for (uchar i = 0; i < 4; i++) 80 { 81 // Write UV plane 82 switch (RotationChromaSitingFlag & 0x07000000) 83 { 84 case CHROMA_SUBSAMPLING_TOP_CENTER: 85 Result_Temp_V.select<1, 1, 16, 1>(0, 0).format<ushort, 2, 8>().select<1, 1, 4, 1>(0, 0) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i, 0), DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i, 1)); 86 Result_Temp_V.select<1, 1, 16, 1>(0, 0).format<ushort, 2, 8>().select<1, 1, 4, 1>(0, 4) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 8, 0), DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 8, 1)); 87 Result_Temp_V.select<1, 1, 16, 1>(1, 0).format<ushort, 2, 8>().select<1, 1, 4, 1>(0, 0) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 1, 0), DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 1, 1)); 88 Result_Temp_V.select<1, 1, 16, 1>(1, 0).format<ushort, 2, 8>().select<1, 1, 4, 1>(0, 4) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 9, 0), DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 9, 1)); 89 90 Result_Temp_U.select<1, 1, 16, 1>(0, 0).format<ushort, 2, 8>().select<1, 1, 4, 1>(0, 0) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 4, 0), DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 4, 1)); 91 Result_Temp_U.select<1, 1, 16, 1>(0, 0).format<ushort, 2, 8>().select<1, 1, 4, 1>(0, 4) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 12, 0), DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 12, 1)); 92 Result_Temp_U.select<1, 1, 16, 1>(1, 0).format<ushort, 2, 8>().select<1, 1, 4, 1>(0, 0) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 5, 0), DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 5, 1)); 93 Result_Temp_U.select<1, 1, 16, 1>(1, 0).format<ushort, 2, 8>().select<1, 1, 4, 1>(0, 4) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 13, 0), DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 13, 1)); 94 95 Result_Temp_U = cm_add<ushort>(Result_Temp_U, 0x80, SAT); 96 Result_Temp_V = cm_add<ushort>(Result_Temp_V, 0x80, SAT); 97 98 Result_UV.select<2, 1, 8, 2>(2 * i, 0) = Result_Temp_U.format<uchar, 4, 16>().select<2, 2, 8, 2>(0, 1); 99 Result_UV.select<2, 1, 8, 2>(2 * i, 1) = Result_Temp_V.format<uchar, 4, 16>().select<2, 2, 8, 2>(0, 1); 100 break; 101 case CHROMA_SUBSAMPLING_CENTER_CENTER: 102 Result_Temp_V.select<1, 1, 16, 1>(0, 0).format<ushort, 2, 8>().select<1, 1, 4, 1>(0, 0) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i, 0), DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i, 1)); 103 Result_Temp_V.select<1, 1, 16, 1>(0, 0).format<ushort, 2, 8>().select<1, 1, 4, 1>(0, 4) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 8, 0), DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 8, 1)); 104 Result_Temp_V.select<1, 1, 16, 1>(0, 0).format<ushort, 2, 8>().select<1, 1, 4, 1>(1, 0) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i, 8), DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i, 9)); 105 Result_Temp_V.select<1, 1, 16, 1>(0, 0).format<ushort, 2, 8>().select<1, 1, 4, 1>(1, 4) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 8, 8), DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 8, 9)); 106 Result_Temp_V.select<1, 1, 16, 1>(1, 0).format<ushort, 2, 8>().select<1, 1, 4, 1>(0, 0) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 1, 0), DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 1, 1)); 107 Result_Temp_V.select<1, 1, 16, 1>(1, 0).format<ushort, 2, 8>().select<1, 1, 4, 1>(0, 4) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 9, 0), DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 9, 1)); 108 Result_Temp_V.select<1, 1, 16, 1>(1, 0).format<ushort, 2, 8>().select<1, 1, 4, 1>(1, 0) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 1, 8), DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 1, 9)); 109 Result_Temp_V.select<1, 1, 16, 1>(1, 0).format<ushort, 2, 8>().select<1, 1, 4, 1>(1, 4) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 9, 8), DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 9, 9)); 110 111 Result_Temp_U.select<1, 1, 16, 1>(0, 0).format<ushort, 2, 8>().select<1, 1, 4, 1>(0, 0) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 4, 0), DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 4, 1)); 112 Result_Temp_U.select<1, 1, 16, 1>(0, 0).format<ushort, 2, 8>().select<1, 1, 4, 1>(0, 4) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 12, 0), DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 12, 1)); 113 Result_Temp_U.select<1, 1, 16, 1>(0, 0).format<ushort, 2, 8>().select<1, 1, 4, 1>(1, 0) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 4, 8), DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 4, 9)); 114 Result_Temp_U.select<1, 1, 16, 1>(0, 0).format<ushort, 2, 8>().select<1, 1, 4, 1>(1, 4) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 12, 8), DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 12, 9)); 115 Result_Temp_U.select<1, 1, 16, 1>(1, 0).format<ushort, 2, 8>().select<1, 1, 4, 1>(0, 0) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 5, 0), DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 5, 1)); 116 Result_Temp_U.select<1, 1, 16, 1>(1, 0).format<ushort, 2, 8>().select<1, 1, 4, 1>(0, 4) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 13, 0), DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 13, 1)); 117 Result_Temp_U.select<1, 1, 16, 1>(1, 0).format<ushort, 2, 8>().select<1, 1, 4, 1>(1, 0) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 5, 8), DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 5, 9)); 118 Result_Temp_U.select<1, 1, 16, 1>(1, 0).format<ushort, 2, 8>().select<1, 1, 4, 1>(1, 4) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 13, 8), DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 13, 9)); 119 120 Result_Temp_V.format<ushort, 4, 8>().select<2, 2, 8, 1>(0, 0) = cm_avg<ushort>(Result_Temp_V.format<ushort, 4, 8>().select<2, 2, 8, 1>(0, 0), Result_Temp_V.format<ushort, 4, 8>().select<2, 2, 8, 1>(1, 0)); 121 Result_Temp_U.format<ushort, 4, 8>().select<2, 2, 8, 1>(0, 0) = cm_avg<ushort>(Result_Temp_U.format<ushort, 4, 8>().select<2, 2, 8, 1>(0, 0), Result_Temp_U.format<ushort, 4, 8>().select<2, 2, 8, 1>(1, 0)); 122 123 Result_Temp_U = cm_add<ushort>(Result_Temp_U, 0x80, SAT); 124 Result_Temp_V = cm_add<ushort>(Result_Temp_V, 0x80, SAT); 125 126 Result_UV.select<2, 1, 8, 2>(2 * i, 0) = Result_Temp_U.format<uchar, 4, 16>().select<2, 2, 8, 2>(0, 1); 127 Result_UV.select<2, 1, 8, 2>(2 * i, 1) = Result_Temp_V.format<uchar, 4, 16>().select<2, 2, 8, 2>(0, 1); 128 129 break; 130 case CHROMA_SUBSAMPLING_BOTTOM_CENTER: 131 Result_Temp_V.select<1, 1, 16, 1>(0, 0).format<ushort, 2, 8>().select<1, 1, 4, 1>(1, 0) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i, 8), DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i, 9)); 132 Result_Temp_V.select<1, 1, 16, 1>(0, 0).format<ushort, 2, 8>().select<1, 1, 4, 1>(1, 4) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 8, 8), DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 8, 9)); 133 Result_Temp_V.select<1, 1, 16, 1>(1, 0).format<ushort, 2, 8>().select<1, 1, 4, 1>(1, 0) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 1, 8), DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 1, 9)); 134 Result_Temp_V.select<1, 1, 16, 1>(1, 0).format<ushort, 2, 8>().select<1, 1, 4, 1>(1, 4) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 9, 8), DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 9, 9)); 135 136 Result_Temp_U.select<1, 1, 16, 1>(0, 0).format<ushort, 2, 8>().select<1, 1, 4, 1>(1, 0) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 4, 8), DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 4, 9)); 137 Result_Temp_U.select<1, 1, 16, 1>(0, 0).format<ushort, 2, 8>().select<1, 1, 4, 1>(1, 4) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 12, 8), DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 12, 9)); 138 Result_Temp_U.select<1, 1, 16, 1>(1, 0).format<ushort, 2, 8>().select<1, 1, 4, 1>(1, 0) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 5, 8), DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 5, 9)); 139 Result_Temp_U.select<1, 1, 16, 1>(1, 0).format<ushort, 2, 8>().select<1, 1, 4, 1>(1, 4) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 13, 8), DataBuffer.format<ushort, 96, 16>().select<1, 1, 4, 2>(16 * i + 13, 9)); 140 141 Result_Temp_U = cm_add<ushort>(Result_Temp_U, 0x80, SAT); 142 Result_Temp_V = cm_add<ushort>(Result_Temp_V, 0x80, SAT); 143 144 Result_UV.select<2, 1, 8, 2>(2 * i, 0) = Result_Temp_U.format<uchar, 4, 16>().select<2, 2, 8, 2>(1, 1); 145 Result_UV.select<2, 1, 8, 2>(2 * i, 1) = Result_Temp_V.format<uchar, 4, 16>().select<2, 2, 8, 2>(1, 1); 146 break; 147 case CHROMA_SUBSAMPLING_TOP_LEFT: 148 DataBuffer.format<ushort, 96, 16>().select<2, 1, 16, 1>(16 * i, 0) = cm_add<ushort>(DataBuffer.format<ushort, 96, 16>().select<2, 1, 16, 1>(16 * i, 0), 0x80, SAT); 149 DataBuffer.format<ushort, 96, 16>().select<2, 1, 16, 1>(16 * i + 8, 0) = cm_add<ushort>(DataBuffer.format<ushort, 96, 16>().select<2, 1, 16, 1>(16 * i + 8, 0), 0x80, SAT); 150 151 DataBuffer.format<ushort, 96, 16>().select<2, 1, 16, 1>(16 * i + 4, 0) = cm_add<ushort>(DataBuffer.format<ushort, 96, 16>().select<2, 1, 16, 1>(16 * i + 4, 0), 0x80, SAT); 152 DataBuffer.format<ushort, 96, 16>().select<2, 1, 16, 1>(16 * i + 12, 0) = cm_add<ushort>(DataBuffer.format<ushort, 96, 16>().select<2, 1, 16, 1>(16 * i + 12, 0), 0x80, SAT); 153 154 Result_UV.select<2, 1, 4, 2>(2 * i, 1) = DataBuffer.format<uchar, 96, 32>().select<2, 1, 4, 4>(16 * i, 1); 155 Result_UV.select<2, 1, 4, 2>(2 * i, 9) = DataBuffer.format<uchar, 96, 32>().select<2, 1, 4, 4>(16 * i + 8, 1); 156 157 Result_UV.select<2, 1, 4, 2>(2 * i, 0) = DataBuffer.format<uchar, 96, 32>().select<2, 1, 4, 4>(16 * i + 4, 1); 158 Result_UV.select<2, 1, 4, 2>(2 * i, 8) = DataBuffer.format<uchar, 96, 32>().select<2, 1, 4, 4>(16 * i + 12, 1); 159 160 break; 161 case CHROMA_SUBSAMPLING_CENTER_LEFT: 162 Result_Temp_V.select<1, 1, 8, 1>(0, 0) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 8, 1>(16 * i, 0), DataBuffer.format<ushort, 96, 16>().select<1, 1, 8, 1>(16 * i, 8)); 163 Result_Temp_V.select<1, 1, 8, 1>(0, 8) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 8, 1>(16 * i + 8, 0), DataBuffer.format<ushort, 96, 16>().select<1, 1, 8, 1>(16 * i + 8, 8)); 164 Result_Temp_V.select<1, 1, 8, 1>(1, 0) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 8, 1>(16 * i + 1, 0), DataBuffer.format<ushort, 96, 16>().select<1, 1, 8, 1>(16 * i + 1, 8)); 165 Result_Temp_V.select<1, 1, 8, 1>(1, 8) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 8, 1>(16 * i + 9, 0), DataBuffer.format<ushort, 96, 16>().select<1, 1, 8, 1>(16 * i + 9, 8)); 166 167 Result_Temp_U.select<1, 1, 8, 1>(0, 0) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 8, 1>(16 * i + 4, 0), DataBuffer.format<ushort, 96, 16>().select<1, 1, 8, 1>(16 * i + 4, 8)); 168 Result_Temp_U.select<1, 1, 8, 1>(0, 8) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 8, 1>(16 * i + 12, 0), DataBuffer.format<ushort, 96, 16>().select<1, 1, 8, 1>(16 * i + 12, 8)); 169 Result_Temp_U.select<1, 1, 8, 1>(1, 0) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 8, 1>(16 * i + 5, 0), DataBuffer.format<ushort, 96, 16>().select<1, 1, 8, 1>(16 * i + 5, 8)); 170 Result_Temp_U.select<1, 1, 8, 1>(1, 8) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 8, 1>(16 * i + 13, 0), DataBuffer.format<ushort, 96, 16>().select<1, 1, 8, 1>(16 * i + 13, 8)); 171 172 Result_Temp_U = cm_add<ushort>(Result_Temp_U, 0x80, SAT); 173 Result_Temp_V = cm_add<ushort>(Result_Temp_V, 0x80, SAT); 174 175 Result_UV.select<2, 1, 8, 2>(2 * i, 0) = Result_Temp_U.format<uchar, 2, 32>().select<2, 1, 8, 4>(0, 1); 176 Result_UV.select<2, 1, 8, 2>(2 * i, 1) = Result_Temp_V.format<uchar, 2, 32>().select<2, 1, 8, 4>(0, 1); 177 178 break; 179 case CHROMA_SUBSAMPLING_BOTTOM_LEFT: 180 DataBuffer.format<ushort, 96, 16>().select<2, 1, 16, 1>(16 * i, 0) = cm_add<ushort>(DataBuffer.format<ushort, 96, 16>().select<2, 1, 16, 1>(16 * i, 0), 0x80, SAT); 181 DataBuffer.format<ushort, 96, 16>().select<2, 1, 16, 1>(16 * i + 8, 0) = cm_add<ushort>(DataBuffer.format<ushort, 96, 16>().select<2, 1, 16, 1>(16 * i + 8, 0), 0x80, SAT); 182 183 DataBuffer.format<ushort, 96, 16>().select<2, 1, 16, 1>(16 * i + 4, 0) = cm_add<ushort>(DataBuffer.format<ushort, 96, 16>().select<2, 1, 16, 1>(16 * i + 4, 0), 0x80, SAT); 184 DataBuffer.format<ushort, 96, 16>().select<2, 1, 16, 1>(16 * i + 12, 0) = cm_add<ushort>(DataBuffer.format<ushort, 96, 16>().select<2, 1, 16, 1>(16 * i + 12, 0), 0x80, SAT); 185 186 Result_UV.select<2, 1, 4, 2>(2 * i, 1) = DataBuffer.format<uchar, 96, 32>().select<2, 1, 4, 4>(16 * i, 17); 187 Result_UV.select<2, 1, 4, 2>(2 * i, 9) = DataBuffer.format<uchar, 96, 32>().select<2, 1, 4, 4>(16 * i + 8, 17); 188 189 Result_UV.select<2, 1, 4, 2>(2 * i, 0) = DataBuffer.format<uchar, 96, 32>().select<2, 1, 4, 4>(16 * i + 4, 17); 190 Result_UV.select<2, 1, 4, 2>(2 * i, 8) = DataBuffer.format<uchar, 96, 32>().select<2, 1, 4, 4>(16 * i + 12, 17); 191 192 break; 193 default: 194 Result_Temp_V.select<1, 1, 8, 1>(0, 0) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 8, 2>(16 * i, 0), DataBuffer.format<ushort, 96, 16>().select<1, 1, 8, 2>(16 * i, 1)); 195 Result_Temp_V.select<1, 1, 8, 1>(0, 8) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 8, 2>(16 * i + 8, 0), DataBuffer.format<ushort, 96, 16>().select<1, 1, 8, 2>(16 * i + 8, 1)); 196 Result_Temp_V.select<1, 1, 8, 1>(1, 0) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 8, 2>(16 * i + 1, 0), DataBuffer.format<ushort, 96, 16>().select<1, 1, 8, 2>(16 * i + 1, 1)); 197 Result_Temp_V.select<1, 1, 8, 1>(1, 8) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 8, 2>(16 * i + 9, 0), DataBuffer.format<ushort, 96, 16>().select<1, 1, 8, 2>(16 * i + 9, 1)); 198 199 Result_Temp_U.select<1, 1, 8, 1>(0, 0) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 8, 2>(16 * i + 4, 0), DataBuffer.format<ushort, 96, 16>().select<1, 1, 8, 2>(16 * i + 4, 1)); 200 Result_Temp_U.select<1, 1, 8, 1>(0, 8) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 8, 2>(16 * i + 12, 0), DataBuffer.format<ushort, 96, 16>().select<1, 1, 8, 2>(16 * i + 12, 1)); 201 Result_Temp_U.select<1, 1, 8, 1>(1, 0) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 8, 2>(16 * i + 5, 0), DataBuffer.format<ushort, 96, 16>().select<1, 1, 8, 2>(16 * i + 5, 1)); 202 Result_Temp_U.select<1, 1, 8, 1>(1, 8) = cm_avg<ushort>(DataBuffer.format<ushort, 96, 16>().select<1, 1, 8, 2>(16 * i + 13, 0), DataBuffer.format<ushort, 96, 16>().select<1, 1, 8, 2>(16 * i + 13, 1)); 203 204 Result_Temp_U = cm_add<ushort>(Result_Temp_U, 0x80, SAT); 205 Result_Temp_V = cm_add<ushort>(Result_Temp_V, 0x80, SAT); 206 207 Result_UV.select<2, 1, 8, 2>(2 * i, 0) = Result_Temp_U.format<uchar, 2, 32>().select<2, 1, 8, 4>(0, 1); 208 Result_UV.select<2, 1, 8, 2>(2 * i, 1) = Result_Temp_V.format<uchar, 2, 32>().select<2, 1, 8, 4>(0, 1); 209 break; 210 } 211 } 212 213 write(Dst_Surface_UV, DstX, DstY >> 1, Result_UV); 214 }