1 /* 2 * Copyright (c) 2010 The WebM project authors. All Rights Reserved. 3 * 4 * Use of this source code is governed by a BSD-style license 5 * that can be found in the LICENSE file in the root of the source 6 * tree. An additional intellectual property rights grant can be found 7 * in the file PATENTS. All contributing project authors may 8 * be found in the AUTHORS file in the root of the source tree. 9 */ 10 11 12 #include "vpx_ports/config.h" 13 #include <math.h> 14 #include "subpixel.h" 15 #include "vpx_ports/mem.h" 16 17 #define BLOCK_HEIGHT_WIDTH 4 18 #define VP8_FILTER_WEIGHT 128 19 #define VP8_FILTER_SHIFT 7 20 21 DECLARE_ALIGNED(16, static const short, sub_pel_filters[8][6]) = 22 { 23 { 0, 0, 128, 0, 0, 0 }, // note that 1/8 pel positions are just as per alpha -0.5 bicubic 24 { 0, -6, 123, 12, -1, 0 }, 25 { 2, -11, 108, 36, -8, 1 }, // New 1/4 pel 6 tap filter 26 { 0, -9, 93, 50, -6, 0 }, 27 { 3, -16, 77, 77, -16, 3 }, // New 1/2 pel 6 tap filter 28 { 0, -6, 50, 93, -9, 0 }, 29 { 1, -8, 36, 108, -11, 2 }, // New 1/4 pel 6 tap filter 30 { 0, -1, 12, 123, -6, 0 }, 31 }; 32 33 34 extern void vp8_filter_block2d_first_pass_armv6 35 ( 36 unsigned char *src_ptr, 37 short *output_ptr, 38 unsigned int src_pixels_per_line, 39 unsigned int output_width, 40 unsigned int output_height, 41 const short *vp8_filter 42 ); 43 44 extern void vp8_filter_block2d_second_pass_armv6 45 ( 46 short *src_ptr, 47 unsigned char *output_ptr, 48 unsigned int output_pitch, 49 unsigned int cnt, 50 const short *vp8_filter 51 ); 52 53 extern void vp8_filter_block2d_first_pass_only_armv6 54 ( 55 unsigned char *src_ptr, 56 unsigned char *output_ptr, 57 unsigned int src_pixels_per_line, 58 unsigned int cnt, 59 unsigned int output_pitch, 60 const short *vp8_filter 61 ); 62 63 64 extern void vp8_filter_block2d_second_pass_only_armv6 65 ( 66 unsigned char *src_ptr, 67 unsigned char *output_ptr, 68 unsigned int src_pixels_per_line, 69 unsigned int cnt, 70 unsigned int output_pitch, 71 const short *vp8_filter 72 ); 73 74 #if HAVE_ARMV6 75 void vp8_sixtap_predict_armv6 76 ( 77 unsigned char *src_ptr, 78 int src_pixels_per_line, 79 int xoffset, 80 int yoffset, 81 unsigned char *dst_ptr, 82 int dst_pitch 83 ) 84 { 85 const short *HFilter; 86 const short *VFilter; 87 DECLARE_ALIGNED_ARRAY(4, short, FData, 12*4); // Temp data bufffer used in filtering 88 89 90 HFilter = sub_pel_filters[xoffset]; // 6 tap 91 VFilter = sub_pel_filters[yoffset]; // 6 tap 92 93 // Vfilter is null. First pass only 94 if (xoffset && !yoffset) 95 { 96 //vp8_filter_block2d_first_pass_armv6 ( src_ptr, FData+2, src_pixels_per_line, 4, 4, HFilter ); 97 //vp8_filter_block2d_second_pass_armv6 ( FData+2, dst_ptr, dst_pitch, 4, VFilter ); 98 99 vp8_filter_block2d_first_pass_only_armv6(src_ptr, dst_ptr, src_pixels_per_line, 4, dst_pitch, HFilter); 100 } 101 // Hfilter is null. Second pass only 102 else if (!xoffset && yoffset) 103 { 104 vp8_filter_block2d_second_pass_only_armv6(src_ptr, dst_ptr, src_pixels_per_line, 4, dst_pitch, VFilter); 105 } 106 else 107 { 108 // Vfilter is a 4 tap filter 109 if (yoffset & 0x1) 110 vp8_filter_block2d_first_pass_armv6(src_ptr - src_pixels_per_line, FData + 1, src_pixels_per_line, 4, 7, HFilter); 111 // Vfilter is 6 tap filter 112 else 113 vp8_filter_block2d_first_pass_armv6(src_ptr - (2 * src_pixels_per_line), FData, src_pixels_per_line, 4, 9, HFilter); 114 115 vp8_filter_block2d_second_pass_armv6(FData + 2, dst_ptr, dst_pitch, 4, VFilter); 116 } 117 } 118 119 /* 120 void vp8_sixtap_predict8x4_armv6 121 ( 122 unsigned char *src_ptr, 123 int src_pixels_per_line, 124 int xoffset, 125 int yoffset, 126 unsigned char *dst_ptr, 127 int dst_pitch 128 ) 129 { 130 const short *HFilter; 131 const short *VFilter; 132 DECLARE_ALIGNED_ARRAY(4, short, FData, 16*8); // Temp data bufffer used in filtering 133 134 HFilter = sub_pel_filters[xoffset]; // 6 tap 135 VFilter = sub_pel_filters[yoffset]; // 6 tap 136 137 138 // if (xoffset && !yoffset) 139 // { 140 // vp8_filter_block2d_first_pass_only_armv6 ( src_ptr, dst_ptr, src_pixels_per_line, 8, dst_pitch, HFilter ); 141 // } 142 // Hfilter is null. Second pass only 143 // else if (!xoffset && yoffset) 144 // { 145 // vp8_filter_block2d_second_pass_only_armv6 ( src_ptr, dst_ptr, src_pixels_per_line, 8, dst_pitch, VFilter ); 146 // } 147 // else 148 // { 149 // if (yoffset & 0x1) 150 // vp8_filter_block2d_first_pass_armv6 ( src_ptr-src_pixels_per_line, FData+1, src_pixels_per_line, 8, 7, HFilter ); 151 // else 152 153 vp8_filter_block2d_first_pass_armv6 ( src_ptr-(2*src_pixels_per_line), FData, src_pixels_per_line, 8, 9, HFilter ); 154 155 vp8_filter_block2d_second_pass_armv6 ( FData+2, dst_ptr, dst_pitch, 4, 8, VFilter ); 156 // } 157 } 158 */ 159 160 void vp8_sixtap_predict8x8_armv6 161 ( 162 unsigned char *src_ptr, 163 int src_pixels_per_line, 164 int xoffset, 165 int yoffset, 166 unsigned char *dst_ptr, 167 int dst_pitch 168 ) 169 { 170 const short *HFilter; 171 const short *VFilter; 172 DECLARE_ALIGNED_ARRAY(4, short, FData, 16*8); // Temp data bufffer used in filtering 173 174 HFilter = sub_pel_filters[xoffset]; // 6 tap 175 VFilter = sub_pel_filters[yoffset]; // 6 tap 176 177 if (xoffset && !yoffset) 178 { 179 vp8_filter_block2d_first_pass_only_armv6(src_ptr, dst_ptr, src_pixels_per_line, 8, dst_pitch, HFilter); 180 } 181 // Hfilter is null. Second pass only 182 else if (!xoffset && yoffset) 183 { 184 vp8_filter_block2d_second_pass_only_armv6(src_ptr, dst_ptr, src_pixels_per_line, 8, dst_pitch, VFilter); 185 } 186 else 187 { 188 if (yoffset & 0x1) 189 vp8_filter_block2d_first_pass_armv6(src_ptr - src_pixels_per_line, FData + 1, src_pixels_per_line, 8, 11, HFilter); 190 else 191 vp8_filter_block2d_first_pass_armv6(src_ptr - (2 * src_pixels_per_line), FData, src_pixels_per_line, 8, 13, HFilter); 192 193 vp8_filter_block2d_second_pass_armv6(FData + 2, dst_ptr, dst_pitch, 8, VFilter); 194 } 195 } 196 197 198 void vp8_sixtap_predict16x16_armv6 199 ( 200 unsigned char *src_ptr, 201 int src_pixels_per_line, 202 int xoffset, 203 int yoffset, 204 unsigned char *dst_ptr, 205 int dst_pitch 206 ) 207 { 208 const short *HFilter; 209 const short *VFilter; 210 DECLARE_ALIGNED_ARRAY(4, short, FData, 24*16); // Temp data bufffer used in filtering 211 212 HFilter = sub_pel_filters[xoffset]; // 6 tap 213 VFilter = sub_pel_filters[yoffset]; // 6 tap 214 215 if (xoffset && !yoffset) 216 { 217 vp8_filter_block2d_first_pass_only_armv6(src_ptr, dst_ptr, src_pixels_per_line, 16, dst_pitch, HFilter); 218 } 219 // Hfilter is null. Second pass only 220 else if (!xoffset && yoffset) 221 { 222 vp8_filter_block2d_second_pass_only_armv6(src_ptr, dst_ptr, src_pixels_per_line, 16, dst_pitch, VFilter); 223 } 224 else 225 { 226 if (yoffset & 0x1) 227 vp8_filter_block2d_first_pass_armv6(src_ptr - src_pixels_per_line, FData + 1, src_pixels_per_line, 16, 19, HFilter); 228 else 229 vp8_filter_block2d_first_pass_armv6(src_ptr - (2 * src_pixels_per_line), FData, src_pixels_per_line, 16, 21, HFilter); 230 231 vp8_filter_block2d_second_pass_armv6(FData + 2, dst_ptr, dst_pitch, 16, VFilter); 232 } 233 234 } 235 #endif 236
