Parent Directory
| 
Revision Log
Revision 1988 - (view) (download)
| 1 : | edgomez | 1609 | /***************************************************************************** |
| 2 : | * | ||
| 3 : | * XVID MPEG-4 VIDEO CODEC | ||
| 4 : | * - MPEG4 Quantization MPEG implementation with altivec optimization - | ||
| 5 : | * | ||
| 6 : | * Copyright(C) 2005 Christoph Naegeli <chn@kbw.ch> | ||
| 7 : | * | ||
| 8 : | * This program is free software ; you can redistribute it and/or modify | ||
| 9 : | * it under the terms of the GNU General Public License as published by | ||
| 10 : | * the Free Software Foundation ; either version 2 of the License, or | ||
| 11 : | * (at your option) any later version. | ||
| 12 : | * | ||
| 13 : | * This program is distributed in the hope that it will be useful, | ||
| 14 : | * but WITHOUT ANY WARRANTY ; without even the implied warranty of | ||
| 15 : | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
| 16 : | * GNU General Public License for more details. | ||
| 17 : | * | ||
| 18 : | * You should have received a copy of the GNU General Public License | ||
| 19 : | * along with this program ; if not, write to the Free Software | ||
| 20 : | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | ||
| 21 : | * | ||
| 22 : | Isibaar | 1988 | * $Id$ |
| 23 : | edgomez | 1609 | * |
| 24 : | ****************************************************************************/ | ||
| 25 : | |||
| 26 : | #ifdef HAVE_ALTIVEC_H | ||
| 27 : | #include <altivec.h> | ||
| 28 : | #endif | ||
| 29 : | |||
| 30 : | #include "../../portab.h" | ||
| 31 : | #include "../../global.h" | ||
| 32 : | |||
| 33 : | #include "../quant.h" | ||
| 34 : | #include "../quant_matrix.h" | ||
| 35 : | |||
| 36 : | #undef DEBUG | ||
| 37 : | #include <stdio.h> | ||
| 38 : | |||
| 39 : | /* Some useful typedefs */ | ||
| 40 : | typedef vector unsigned char vec_uint8_t; | ||
| 41 : | typedef vector unsigned short vec_uint16_t; | ||
| 42 : | typedef vector unsigned int vec_uint32_t; | ||
| 43 : | |||
| 44 : | typedef vector signed char vec_sint8_t; | ||
| 45 : | typedef vector signed short vec_sint16_t; | ||
| 46 : | typedef vector signed int vec_sint32_t; | ||
| 47 : | |||
| 48 : | /***************************************************************************** | ||
| 49 : | * Local data | ||
| 50 : | ****************************************************************************/ | ||
| 51 : | |||
| 52 : | #define VM18P 3 | ||
| 53 : | #define VM18Q 4 | ||
| 54 : | |||
| 55 : | /* divide-by-multiply table | ||
| 56 : | * needs 17 bit shift (16 causes slight errors when q > 19) */ | ||
| 57 : | |||
| 58 : | #define SCALEBITS 17 | ||
| 59 : | |||
| 60 : | #define DEQUANT_MPEG_INTRA() \ | ||
| 61 : | level = vec_perm(vec_ld(0,coeff_ptr),vec_ld(16,coeff_ptr),vec_lvsl(0,coeff_ptr));\ | ||
| 62 : | zero_less = vec_cmplt(level,ox00);\ | ||
| 63 : | level = vec_abs(level);\ | ||
| 64 : | vintra = vec_perm(vec_ld(0,intra_matrix),vec_ld(16,intra_matrix),vec_lvsl(0,intra_matrix));\ | ||
| 65 : | even = vec_mule((vec_uint16_t)level,vintra);\ | ||
| 66 : | odd = vec_mulo((vec_uint16_t)level,vintra);\ | ||
| 67 : | t = vec_splat_u32(-16);\ | ||
| 68 : | et = vec_msum( (vec_uint16_t)even, (vec_uint16_t)swap, (vec_uint32_t)ox00);\ | ||
| 69 : | ot = vec_msum( (vec_uint16_t)odd, (vec_uint16_t)swap, (vec_uint32_t)ox00);\ | ||
| 70 : | et = vec_sl(et, t);\ | ||
| 71 : | ot = vec_sl(ot, t);\ | ||
| 72 : | even = vec_mulo( (vec_uint16_t)even, (vec_uint16_t)vquant);\ | ||
| 73 : | odd = vec_mulo( (vec_uint16_t)odd, (vec_uint16_t)vquant);\ | ||
| 74 : | t = vec_splat_u32(3);\ | ||
| 75 : | even = vec_add(even,et);\ | ||
| 76 : | odd = vec_add(odd,ot);\ | ||
| 77 : | even = vec_sr(even,t);\ | ||
| 78 : | odd = vec_sr(odd,t);\ | ||
| 79 : | /* Pack & Clamp to [-2048,2047] */\ | ||
| 80 : | level = vec_packs( (vec_sint32_t)vec_mergeh(even,odd), (vec_sint32_t)vec_mergel(even,odd) );\ | ||
| 81 : | t = (vec_uint32_t)vec_splat_s16(-1);\ | ||
| 82 : | overflow = vec_cmpgt(level, vec_add(vec_2048, (vec_sint16_t)t));\ | ||
| 83 : | level = vec_sel(level, vec_2048, overflow);\ | ||
| 84 : | overflow = (vector bool short)vec_and(overflow, vec_xor(zero_less, (vec_sint16_t)t));\ | ||
| 85 : | t = (vec_uint32_t)vec_splat_s16(1);\ | ||
| 86 : | overflow = (vector bool short)vec_and(overflow, (vec_sint16_t)t);\ | ||
| 87 : | level = vec_sub(level, (vec_sint16_t)overflow);\ | ||
| 88 : | /* Invert the negative ones */\ | ||
| 89 : | level = vec_xor(level, zero_less);\ | ||
| 90 : | level = vec_add(level, (vec_sint16_t)vec_and(zero_less, (vec_uint16_t)t));\ | ||
| 91 : | /* Save & Advance Pointers */\ | ||
| 92 : | vec_st(level,0,data_ptr);\ | ||
| 93 : | data_ptr += 8;\ | ||
| 94 : | intra_matrix += 8;\ | ||
| 95 : | coeff_ptr += 8 | ||
| 96 : | |||
| 97 : | |||
| 98 : | /* This function assuems: | ||
| 99 : | * data: 16 Byte aligned | ||
| 100 : | */ | ||
| 101 : | |||
| 102 : | uint32_t | ||
| 103 : | dequant_mpeg_intra_altivec_c(int16_t * data, | ||
| 104 : | const int16_t * coeff, | ||
| 105 : | const uint32_t quant, | ||
| 106 : | const uint32_t dcscalar, | ||
| 107 : | const uint16_t * mpeg_quant_matrices) | ||
| 108 : | { | ||
| 109 : | register const uint16_t *intra_matrix = get_intra_matrix(mpeg_quant_matrices); | ||
| 110 : | register const int16_t *coeff_ptr = coeff; | ||
| 111 : | register int16_t *data_ptr = data; | ||
| 112 : | |||
| 113 : | register vec_sint16_t ox00; | ||
| 114 : | register vec_sint16_t level; | ||
| 115 : | register vec_sint16_t vec_2048; | ||
| 116 : | register vec_uint16_t vintra; | ||
| 117 : | register vec_uint32_t swap; | ||
| 118 : | register vec_uint32_t even,odd; | ||
| 119 : | register vec_uint32_t et,ot,t; | ||
| 120 : | |||
| 121 : | vec_uint32_t vquant; | ||
| 122 : | vector bool short zero_less; | ||
| 123 : | vector bool short overflow; | ||
| 124 : | |||
| 125 : | #ifdef DEBUG | ||
| 126 : | if((long)data & 0xf) | ||
| 127 : | fprintf(stderr, "xvidcore: error in dequant_mpeg_intra_altivec_c, incorrect align: %x\n", data); | ||
| 128 : | #endif | ||
| 129 : | |||
| 130 : | /* Initialize */ | ||
| 131 : | ox00 = vec_splat_s16(0); | ||
| 132 : | *((uint32_t*)&vquant) = quant; | ||
| 133 : | vquant = vec_splat(vquant,0); | ||
| 134 : | |||
| 135 : | swap = vec_rl(vquant, vec_splat_u32(-16)); | ||
| 136 : | vec_2048 = (vec_sint16_t)vec_rl(vec_splat_u16(8),vec_splat_u16(8)); | ||
| 137 : | |||
| 138 : | DEQUANT_MPEG_INTRA(); | ||
| 139 : | DEQUANT_MPEG_INTRA(); | ||
| 140 : | DEQUANT_MPEG_INTRA(); | ||
| 141 : | DEQUANT_MPEG_INTRA(); | ||
| 142 : | |||
| 143 : | DEQUANT_MPEG_INTRA(); | ||
| 144 : | DEQUANT_MPEG_INTRA(); | ||
| 145 : | DEQUANT_MPEG_INTRA(); | ||
| 146 : | DEQUANT_MPEG_INTRA(); | ||
| 147 : | |||
| 148 : | /* Process the first */ | ||
| 149 : | data[0] = coeff[0] * dcscalar; | ||
| 150 : | if (data[0] < -2048) { | ||
| 151 : | data[0] = -2048; | ||
| 152 : | } else if (data[0] > 2047) { | ||
| 153 : | data[0] = 2047; | ||
| 154 : | } | ||
| 155 : | |||
| 156 : | return 0; | ||
| 157 : | } | ||
| 158 : | |||
| 159 : | |||
| 160 : | |||
| 161 : | #define DEQUANT_MPEG_INTER() \ | ||
| 162 : | level = vec_perm(vec_ld(0,coeff),vec_ld(16,coeff),vec_lvsl(0,coeff));\ | ||
| 163 : | zero_eq = vec_cmpeq(level,ox00);\ | ||
| 164 : | zero_less = vec_cmplt(level,ox00);\ | ||
| 165 : | level = vec_abs(level);\ | ||
| 166 : | vinter = vec_perm(vec_ld(0,inter_matrix),vec_ld(16,inter_matrix),vec_lvsl(0,inter_matrix));\ | ||
| 167 : | t = vec_splat_u32(1);\ | ||
| 168 : | hi = (vec_uint32_t)vec_unpackh(level);\ | ||
| 169 : | lo = (vec_uint32_t)vec_unpackl(level);\ | ||
| 170 : | hi = vec_sl(hi, t);\ | ||
| 171 : | lo = vec_sl(lo, t);\ | ||
| 172 : | hi = vec_add(hi, t);\ | ||
| 173 : | lo = vec_add(lo, t);\ | ||
| 174 : | /* Multiplication with vinter */\ | ||
| 175 : | sw_hi = vec_rl(hi, v16);\ | ||
| 176 : | sw_lo = vec_rl(lo, v16);\ | ||
| 177 : | hi = vec_mulo((vec_uint16_t)hi, vec_mergeh((vec_uint16_t)ox00,vinter));\ | ||
| 178 : | lo = vec_mulo((vec_uint16_t)lo, vec_mergel((vec_uint16_t)ox00,vinter));\ | ||
| 179 : | sw_hi = vec_mulo((vec_uint16_t)sw_hi, vec_mergeh((vec_uint16_t)ox00,vinter));\ | ||
| 180 : | sw_lo = vec_mulo((vec_uint16_t)sw_lo, vec_mergeh((vec_uint16_t)ox00,vinter));\ | ||
| 181 : | hi = vec_add(hi, vec_sl(sw_hi,v16));\ | ||
| 182 : | lo = vec_add(lo, vec_sl(sw_lo,v16));\ | ||
| 183 : | /* Multiplication with Quant */\ | ||
| 184 : | t = vec_splat_u32(4);\ | ||
| 185 : | sw_hi = vec_msum( (vec_uint16_t)hi, (vec_uint16_t)swap, (vec_uint32_t)ox00 );\ | ||
| 186 : | sw_lo = vec_msum( (vec_uint16_t)lo, (vec_uint16_t)swap, (vec_uint32_t)ox00 );\ | ||
| 187 : | hi = vec_mulo( (vec_uint16_t)hi, (vec_uint16_t)vquant );\ | ||
| 188 : | lo = vec_mulo( (vec_uint16_t)lo, (vec_uint16_t)vquant );\ | ||
| 189 : | hi = vec_add(hi, vec_sl(sw_hi,v16));\ | ||
| 190 : | lo = vec_add(lo, vec_sl(sw_lo,v16));\ | ||
| 191 : | hi = vec_sr(hi, t);\ | ||
| 192 : | lo = vec_sr(lo, t);\ | ||
| 193 : | /* Pack & Clamp to [-2048,2047] */\ | ||
| 194 : | t = (vec_uint32_t)vec_splat_s16(-1);\ | ||
| 195 : | level = vec_packs((vec_sint32_t)hi, (vec_sint32_t)lo);\ | ||
| 196 : | overflow = vec_cmpgt(level, vec_add(v2048, (vec_sint16_t)t));\ | ||
| 197 : | level = vec_sel(level, v2048, overflow);\ | ||
| 198 : | overflow = (vector bool short)vec_and(overflow, vec_xor(zero_less, (vec_sint16_t)t));\ | ||
| 199 : | t = (vec_uint32_t)vec_splat_s16(1);\ | ||
| 200 : | overflow = (vector bool short)vec_and(overflow, (vec_sint16_t)t);\ | ||
| 201 : | level = vec_sub(level, (vec_sint16_t)overflow);\ | ||
| 202 : | level = vec_sel(level, ox00, zero_eq);\ | ||
| 203 : | level = vec_xor(level, zero_less);\ | ||
| 204 : | level = vec_add(level, (vec_sint16_t)vec_and(zero_less, (vec_uint16_t)t));\ | ||
| 205 : | /* Get vsum */\ | ||
| 206 : | vsum = vec_xor(vsum, (vec_uint32_t)vec_unpackh(level));\ | ||
| 207 : | vsum = vec_xor(vsum, (vec_uint32_t)vec_unpackl(level));\ | ||
| 208 : | /* Save & Advance Pointers */\ | ||
| 209 : | vec_st(level,0,data);\ | ||
| 210 : | data+=8;\ | ||
| 211 : | inter_matrix+=8;\ | ||
| 212 : | coeff+=8 | ||
| 213 : | |||
| 214 : | |||
| 215 : | /* This function assumes: | ||
| 216 : | * data: 16 Byte aligned | ||
| 217 : | */ | ||
| 218 : | |||
| 219 : | uint32_t | ||
| 220 : | dequant_mpeg_inter_altivec_c(int16_t * data, | ||
| 221 : | const int16_t * coeff, | ||
| 222 : | const uint32_t quant, | ||
| 223 : | const uint16_t * mpeg_quant_matrices) | ||
| 224 : | { | ||
| 225 : | register uint32_t sum; | ||
| 226 : | register const uint16_t *inter_matrix = get_inter_matrix(mpeg_quant_matrices); | ||
| 227 : | |||
| 228 : | register vec_sint16_t ox00; | ||
| 229 : | register vec_sint16_t v2048; | ||
| 230 : | register vec_sint16_t level; | ||
| 231 : | register vec_uint16_t vinter; | ||
| 232 : | register vec_uint32_t hi,lo; | ||
| 233 : | register vec_uint32_t sw_hi,sw_lo; | ||
| 234 : | register vec_uint32_t swap; | ||
| 235 : | register vec_uint32_t t,v16; | ||
| 236 : | |||
| 237 : | vec_uint32_t vsum; | ||
| 238 : | vec_uint32_t vquant; | ||
| 239 : | vector bool short zero_eq; | ||
| 240 : | vector bool short zero_less; | ||
| 241 : | vector bool short overflow; | ||
| 242 : | |||
| 243 : | #ifdef DEBUG | ||
| 244 : | if((long)data & 0xf) | ||
| 245 : | fprintf(stderr, "xvidcore: error in dequant_mpeg_inter_altivec_c, incorrect align: %x\n", data); | ||
| 246 : | #endif | ||
| 247 : | |||
| 248 : | /* Initialization */ | ||
| 249 : | ox00 = vec_splat_s16(0); | ||
| 250 : | v16 = vec_splat_u32(-16); | ||
| 251 : | v2048 = vec_rl(vec_splat_s16(8),vec_splat_u16(8)); | ||
| 252 : | |||
| 253 : | vsum = (vec_uint32_t)ox00; | ||
| 254 : | |||
| 255 : | *((uint32_t*)&vquant) = quant; | ||
| 256 : | vquant = vec_splat(vquant,0); | ||
| 257 : | swap = vec_rl(vquant,v16); | ||
| 258 : | |||
| 259 : | DEQUANT_MPEG_INTER(); | ||
| 260 : | DEQUANT_MPEG_INTER(); | ||
| 261 : | DEQUANT_MPEG_INTER(); | ||
| 262 : | DEQUANT_MPEG_INTER(); | ||
| 263 : | |||
| 264 : | DEQUANT_MPEG_INTER(); | ||
| 265 : | DEQUANT_MPEG_INTER(); | ||
| 266 : | DEQUANT_MPEG_INTER(); | ||
| 267 : | DEQUANT_MPEG_INTER(); | ||
| 268 : | |||
| 269 : | sum = ((uint32_t*)&vsum)[0]; | ||
| 270 : | sum ^= ((uint32_t*)&vsum)[1]; | ||
| 271 : | sum ^= ((uint32_t*)&vsum)[2]; | ||
| 272 : | sum ^= ((uint32_t*)&vsum)[3]; | ||
| 273 : | |||
| 274 : | /* mismatch control */ | ||
| 275 : | if((sum & 1) == 0) { | ||
| 276 : | data -= 1; | ||
| 277 : | *data ^= 1; | ||
| 278 : | } | ||
| 279 : | |||
| 280 : | return 0; | ||
| 281 : | } |
| No admin address has been configured | ViewVC Help |
| Powered by ViewVC 1.0.4 |