|
MythTV
0.26-pre
|
|
MythTV
0.26-pre
|
00001 /* 00002 * idct_altivec.c 00003 * Copyright (C) 2000-2003 Michel Lespinasse <walken@zoy.org> 00004 * Copyright (C) 1999-2000 Aaron Holtzman <aholtzma@ess.engr.uvic.ca> 00005 * 00006 * This file is part of mpeg2dec, a free MPEG-2 video stream decoder. 00007 * See http://libmpeg2.sourceforge.net/ for updates. 00008 * 00009 * mpeg2dec is free software; you can redistribute it and/or modify 00010 * it under the terms of the GNU General Public License as published by 00011 * the Free Software Foundation; either version 2 of the License, or 00012 * (at your option) any later version. 00013 * 00014 * mpeg2dec is distributed in the hope that it will be useful, 00015 * but WITHOUT ANY WARRANTY; without even the implied warranty of 00016 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the 00017 * GNU General Public License for more details. 00018 * 00019 * You should have received a copy of the GNU General Public License 00020 * along with this program; if not, write to the Free Software 00021 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA 00022 */ 00023 00024 #include "mpeg2config.h" 00025 00026 #if ARCH_PPC 00027 00028 #if HAVE_ALTIVEC_H 00029 #include <altivec.h> 00030 #endif 00031 #include <inttypes.h> 00032 00033 #include "mpeg2.h" 00034 #include "attributes.h" 00035 #include "mpeg2_internal.h" 00036 00037 typedef vector signed char vector_s8_t; 00038 typedef vector unsigned char vector_u8_t; 00039 typedef vector signed short vector_s16_t; 00040 typedef vector unsigned short vector_u16_t; 00041 typedef vector signed int vector_s32_t; 00042 typedef vector unsigned int vector_u32_t; 00043 00044 #if HAVE_ALTIVEC_H && (__GNUC__ * 100 + __GNUC_MINOR__ < 303) 00045 /* work around gcc <3.3 vec_mergel bug */ 00046 static inline vector_s16_t my_vec_mergel (vector_s16_t const A, 00047 vector_s16_t const B) 00048 { 00049 static const vector_u8_t mergel = { 00050 0x08, 0x09, 0x18, 0x19, 0x0a, 0x0b, 0x1a, 0x1b, 00051 0x0c, 0x0d, 0x1c, 0x1d, 0x0e, 0x0f, 0x1e, 0x1f 00052 }; 00053 return vec_perm (A, B, mergel); 00054 } 00055 #undef vec_mergel 00056 #define vec_mergel my_vec_mergel 00057 #endif 00058 00059 #if HAVE_ALTIVEC_H /* gnu */ 00060 #define VEC_S16(a,b,c,d,e,f,g,h) {a, b, c, d, e, f, g, h} 00061 #else /* apple */ 00062 #define VEC_S16(a,b,c,d,e,f,g,h) (vector_s16_t) (a, b, c, d, e, f, g, h) 00063 #endif 00064 00065 static const vector_s16_t constants ATTR_ALIGN(16) = 00066 VEC_S16 (23170, 13573, 6518, 21895, -23170, -21895, 32, 31); 00067 static const vector_s16_t constants_1 ATTR_ALIGN(16) = 00068 VEC_S16 (16384, 22725, 21407, 19266, 16384, 19266, 21407, 22725); 00069 static const vector_s16_t constants_2 ATTR_ALIGN(16) = 00070 VEC_S16 (16069, 22289, 20995, 18895, 16069, 18895, 20995, 22289); 00071 static const vector_s16_t constants_3 ATTR_ALIGN(16) = 00072 VEC_S16 (21407, 29692, 27969, 25172, 21407, 25172, 27969, 29692); 00073 static const vector_s16_t constants_4 ATTR_ALIGN(16) = 00074 VEC_S16 (13623, 18895, 17799, 16019, 13623, 16019, 17799, 18895); 00075 00076 #define IDCT \ 00077 vector_s16_t vx0, vx1, vx2, vx3, vx4, vx5, vx6, vx7; \ 00078 vector_s16_t vy0, vy1, vy2, vy3, vy4, vy5, vy6, vy7; \ 00079 vector_s16_t a0, a1, a2, ma2, c4, mc4, zero, bias; \ 00080 vector_s16_t t0, t1, t2, t3, t4, t5, t6, t7, t8; \ 00081 vector_u16_t shift; \ 00082 \ 00083 c4 = vec_splat (constants, 0); \ 00084 a0 = vec_splat (constants, 1); \ 00085 a1 = vec_splat (constants, 2); \ 00086 a2 = vec_splat (constants, 3); \ 00087 mc4 = vec_splat (constants, 4); \ 00088 ma2 = vec_splat (constants, 5); \ 00089 bias = (vector_s16_t)vec_splat ((vector_s32_t)constants, 3); \ 00090 \ 00091 zero = vec_splat_s16 (0); \ 00092 \ 00093 vx0 = vec_adds (block[0], block[4]); \ 00094 vx4 = vec_subs (block[0], block[4]); \ 00095 t5 = vec_mradds (vx0, constants_1, zero); \ 00096 t0 = vec_mradds (vx4, constants_1, zero); \ 00097 \ 00098 vx1 = vec_mradds (a1, block[7], block[1]); \ 00099 vx7 = vec_mradds (a1, block[1], vec_subs (zero, block[7])); \ 00100 t1 = vec_mradds (vx1, constants_2, zero); \ 00101 t8 = vec_mradds (vx7, constants_2, zero); \ 00102 \ 00103 vx2 = vec_mradds (a0, block[6], block[2]); \ 00104 vx6 = vec_mradds (a0, block[2], vec_subs (zero, block[6])); \ 00105 t2 = vec_mradds (vx2, constants_3, zero); \ 00106 t4 = vec_mradds (vx6, constants_3, zero); \ 00107 \ 00108 vx3 = vec_mradds (block[3], constants_4, zero); \ 00109 vx5 = vec_mradds (block[5], constants_4, zero); \ 00110 t7 = vec_mradds (a2, vx5, vx3); \ 00111 t3 = vec_mradds (ma2, vx3, vx5); \ 00112 \ 00113 t6 = vec_adds (t8, t3); \ 00114 t3 = vec_subs (t8, t3); \ 00115 t8 = vec_subs (t1, t7); \ 00116 t1 = vec_adds (t1, t7); \ 00117 t6 = vec_mradds (a0, t6, t6); /* a0+1 == 2*c4 */ \ 00118 t1 = vec_mradds (a0, t1, t1); /* a0+1 == 2*c4 */ \ 00119 \ 00120 t7 = vec_adds (t5, t2); \ 00121 t2 = vec_subs (t5, t2); \ 00122 t5 = vec_adds (t0, t4); \ 00123 t0 = vec_subs (t0, t4); \ 00124 t4 = vec_subs (t8, t3); \ 00125 t3 = vec_adds (t8, t3); \ 00126 \ 00127 vy0 = vec_adds (t7, t1); \ 00128 vy7 = vec_subs (t7, t1); \ 00129 vy1 = vec_adds (t5, t3); \ 00130 vy6 = vec_subs (t5, t3); \ 00131 vy2 = vec_adds (t0, t4); \ 00132 vy5 = vec_subs (t0, t4); \ 00133 vy3 = vec_adds (t2, t6); \ 00134 vy4 = vec_subs (t2, t6); \ 00135 \ 00136 vx0 = vec_mergeh (vy0, vy4); \ 00137 vx1 = vec_mergel (vy0, vy4); \ 00138 vx2 = vec_mergeh (vy1, vy5); \ 00139 vx3 = vec_mergel (vy1, vy5); \ 00140 vx4 = vec_mergeh (vy2, vy6); \ 00141 vx5 = vec_mergel (vy2, vy6); \ 00142 vx6 = vec_mergeh (vy3, vy7); \ 00143 vx7 = vec_mergel (vy3, vy7); \ 00144 \ 00145 vy0 = vec_mergeh (vx0, vx4); \ 00146 vy1 = vec_mergel (vx0, vx4); \ 00147 vy2 = vec_mergeh (vx1, vx5); \ 00148 vy3 = vec_mergel (vx1, vx5); \ 00149 vy4 = vec_mergeh (vx2, vx6); \ 00150 vy5 = vec_mergel (vx2, vx6); \ 00151 vy6 = vec_mergeh (vx3, vx7); \ 00152 vy7 = vec_mergel (vx3, vx7); \ 00153 \ 00154 vx0 = vec_mergeh (vy0, vy4); \ 00155 vx1 = vec_mergel (vy0, vy4); \ 00156 vx2 = vec_mergeh (vy1, vy5); \ 00157 vx3 = vec_mergel (vy1, vy5); \ 00158 vx4 = vec_mergeh (vy2, vy6); \ 00159 vx5 = vec_mergel (vy2, vy6); \ 00160 vx6 = vec_mergeh (vy3, vy7); \ 00161 vx7 = vec_mergel (vy3, vy7); \ 00162 \ 00163 vx0 = vec_adds (vx0, bias); \ 00164 t5 = vec_adds (vx0, vx4); \ 00165 t0 = vec_subs (vx0, vx4); \ 00166 \ 00167 t1 = vec_mradds (a1, vx7, vx1); \ 00168 t8 = vec_mradds (a1, vx1, vec_subs (zero, vx7)); \ 00169 \ 00170 t2 = vec_mradds (a0, vx6, vx2); \ 00171 t4 = vec_mradds (a0, vx2, vec_subs (zero, vx6)); \ 00172 \ 00173 t7 = vec_mradds (a2, vx5, vx3); \ 00174 t3 = vec_mradds (ma2, vx3, vx5); \ 00175 \ 00176 t6 = vec_adds (t8, t3); \ 00177 t3 = vec_subs (t8, t3); \ 00178 t8 = vec_subs (t1, t7); \ 00179 t1 = vec_adds (t1, t7); \ 00180 \ 00181 t7 = vec_adds (t5, t2); \ 00182 t2 = vec_subs (t5, t2); \ 00183 t5 = vec_adds (t0, t4); \ 00184 t0 = vec_subs (t0, t4); \ 00185 t4 = vec_subs (t8, t3); \ 00186 t3 = vec_adds (t8, t3); \ 00187 \ 00188 vy0 = vec_adds (t7, t1); \ 00189 vy7 = vec_subs (t7, t1); \ 00190 vy1 = vec_mradds (c4, t3, t5); \ 00191 vy6 = vec_mradds (mc4, t3, t5); \ 00192 vy2 = vec_mradds (c4, t4, t0); \ 00193 vy5 = vec_mradds (mc4, t4, t0); \ 00194 vy3 = vec_adds (t2, t6); \ 00195 vy4 = vec_subs (t2, t6); \ 00196 \ 00197 shift = vec_splat_u16 (6); \ 00198 vx0 = vec_sra (vy0, shift); \ 00199 vx1 = vec_sra (vy1, shift); \ 00200 vx2 = vec_sra (vy2, shift); \ 00201 vx3 = vec_sra (vy3, shift); \ 00202 vx4 = vec_sra (vy4, shift); \ 00203 vx5 = vec_sra (vy5, shift); \ 00204 vx6 = vec_sra (vy6, shift); \ 00205 vx7 = vec_sra (vy7, shift); 00206 00207 void mpeg2_idct_copy_altivec (int16_t * const _block, uint8_t * dest, 00208 const int stride) 00209 { 00210 vector_s16_t * const block = (vector_s16_t *)_block; 00211 vector_u8_t tmp; 00212 00213 IDCT 00214 00215 #define COPY(dest,src) \ 00216 tmp = vec_packsu (src, src); \ 00217 vec_ste ((vector_u32_t)tmp, 0, (unsigned int *)dest); \ 00218 vec_ste ((vector_u32_t)tmp, 4, (unsigned int *)dest); 00219 00220 COPY (dest, vx0) dest += stride; 00221 COPY (dest, vx1) dest += stride; 00222 COPY (dest, vx2) dest += stride; 00223 COPY (dest, vx3) dest += stride; 00224 COPY (dest, vx4) dest += stride; 00225 COPY (dest, vx5) dest += stride; 00226 COPY (dest, vx6) dest += stride; 00227 COPY (dest, vx7) 00228 00229 block[0] = block[1] = block[2] = block[3] = zero; 00230 block[4] = block[5] = block[6] = block[7] = zero; 00231 } 00232 00233 void mpeg2_idct_add_altivec (const int last, int16_t * const _block, 00234 uint8_t * dest, const int stride) 00235 { 00236 vector_s16_t * const block = (vector_s16_t *)_block; 00237 vector_u8_t tmp; 00238 vector_s16_t tmp2, tmp3; 00239 vector_u8_t perm0; 00240 vector_u8_t perm1; 00241 vector_u8_t p0, p1, p; 00242 00243 IDCT 00244 00245 p0 = vec_lvsl (0, dest); 00246 p1 = vec_lvsl (stride, dest); 00247 p = vec_splat_u8 (-1); 00248 perm0 = vec_mergeh (p, p0); 00249 perm1 = vec_mergeh (p, p1); 00250 00251 #define ADD(dest,src,perm) \ 00252 /* *(uint64_t *)&tmp = *(uint64_t *)dest; */ \ 00253 tmp = vec_ld (0, dest); \ 00254 tmp2 = (vector_s16_t)vec_perm (tmp, (vector_u8_t)zero, perm); \ 00255 tmp3 = vec_adds (tmp2, src); \ 00256 tmp = vec_packsu (tmp3, tmp3); \ 00257 vec_ste ((vector_u32_t)tmp, 0, (unsigned int *)dest); \ 00258 vec_ste ((vector_u32_t)tmp, 4, (unsigned int *)dest); 00259 00260 ADD (dest, vx0, perm0) dest += stride; 00261 ADD (dest, vx1, perm1) dest += stride; 00262 ADD (dest, vx2, perm0) dest += stride; 00263 ADD (dest, vx3, perm1) dest += stride; 00264 ADD (dest, vx4, perm0) dest += stride; 00265 ADD (dest, vx5, perm1) dest += stride; 00266 ADD (dest, vx6, perm0) dest += stride; 00267 ADD (dest, vx7, perm1) 00268 00269 block[0] = block[1] = block[2] = block[3] = zero; 00270 block[4] = block[5] = block[6] = block[7] = zero; 00271 } 00272 00273 void mpeg2_idct_altivec_init (void) 00274 { 00275 extern uint8_t mpeg2_scan_norm[64]; 00276 extern uint8_t mpeg2_scan_alt[64]; 00277 int i, j; 00278 00279 /* the altivec idct uses a transposed input, so we patch scan tables */ 00280 for (i = 0; i < 64; i++) { 00281 j = mpeg2_scan_norm[i]; 00282 mpeg2_scan_norm[i] = (j >> 3) | ((j & 7) << 3); 00283 j = mpeg2_scan_alt[i]; 00284 mpeg2_scan_alt[i] = (j >> 3) | ((j & 7) << 3); 00285 } 00286 } 00287 00288 #endif
1.7.6.1
