Cinelerra: hvirtual/quicktime/ffmpeg/libavcodec/ppc/idct

00001 /*
00002  * Copyright (c) 2001 Michel Lespinasse
00003  *
00004  * This library is free software; you can redistribute it and/or
00005  * modify it under the terms of the GNU Lesser General Public
00006  * License as published by the Free Software Foundation; either
00007  * version 2 of the License, or (at your option) any later version.
00008  *
00009  * This library is distributed in the hope that it will be useful,
00010  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00011  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00012  * Lesser General Public License for more details.
00013  *
00014  * You should have received a copy of the GNU Lesser General Public
00015  * License along with this library; if not, write to the Free Software
00016  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
00017  *
00018  */
00019 
00020 /*
00021  * NOTE: This code is based on GPL code from the libmpeg2 project.  The
00022  * author, Michel Lespinasses, has given explicit permission to release
00023  * under LGPL as part of ffmpeg.
00024  *
00025  */
00026 
00027 /*
00028  * FFMpeg integration by Dieter Shirley
00029  *
00030  * This file is a direct copy of the altivec idct module from the libmpeg2
00031  * project.  I've deleted all of the libmpeg2 specific code, renamed the functions and
00032  * re-ordered the function parameters.  The only change to the IDCT function
00033  * itself was to factor out the partial transposition, and to perform a full
00034  * transpose at the end of the function.
00035  */
00036 
00037 
00038 #include <stdlib.h>                                      /* malloc(), free() */
00039 #include <string.h>
00040 #include "../dsputil.h"
00041 
00042 #include "gcc_fixes.h"
00043 
00044 #include "dsputil_altivec.h"
00045 
00046 #define vector_s16_t vector signed short
00047 #define const_vector_s16_t const_vector signed short
00048 #define vector_u16_t vector unsigned short
00049 #define vector_s8_t vector signed char
00050 #define vector_u8_t vector unsigned char
00051 #define vector_s32_t vector signed int
00052 #define vector_u32_t vector unsigned int
00053 
00054 #define IDCT_HALF                                       \
00055     /* 1st stage */                                     \
00056     t1 = vec_mradds (a1, vx7, vx1 );                    \
00057     t8 = vec_mradds (a1, vx1, vec_subs (zero, vx7));    \
00058     t7 = vec_mradds (a2, vx5, vx3);                     \
00059     t3 = vec_mradds (ma2, vx3, vx5);                    \
00060                                                         \
00061     /* 2nd stage */                                     \
00062     t5 = vec_adds (vx0, vx4);                           \
00063     t0 = vec_subs (vx0, vx4);                           \
00064     t2 = vec_mradds (a0, vx6, vx2);                     \
00065     t4 = vec_mradds (a0, vx2, vec_subs (zero, vx6));    \
00066     t6 = vec_adds (t8, t3);                             \
00067     t3 = vec_subs (t8, t3);                             \
00068     t8 = vec_subs (t1, t7);                             \
00069     t1 = vec_adds (t1, t7);                             \
00070                                                         \
00071     /* 3rd stage */                                     \
00072     t7 = vec_adds (t5, t2);                             \
00073     t2 = vec_subs (t5, t2);                             \
00074     t5 = vec_adds (t0, t4);                             \
00075     t0 = vec_subs (t0, t4);                             \
00076     t4 = vec_subs (t8, t3);                             \
00077     t3 = vec_adds (t8, t3);                             \
00078                                                         \
00079     /* 4th stage */                                     \
00080     vy0 = vec_adds (t7, t1);                            \
00081     vy7 = vec_subs (t7, t1);                            \
00082     vy1 = vec_mradds (c4, t3, t5);                      \
00083     vy6 = vec_mradds (mc4, t3, t5);                     \
00084     vy2 = vec_mradds (c4, t4, t0);                      \
00085     vy5 = vec_mradds (mc4, t4, t0);                     \
00086     vy3 = vec_adds (t2, t6);                            \
00087     vy4 = vec_subs (t2, t6);
00088 
00089         
00090 #define IDCT                                                            \
00091     vector_s16_t vx0, vx1, vx2, vx3, vx4, vx5, vx6, vx7;                \
00092     vector_s16_t vy0, vy1, vy2, vy3, vy4, vy5, vy6, vy7;                \
00093     vector_s16_t a0, a1, a2, ma2, c4, mc4, zero, bias;                  \
00094     vector_s16_t t0, t1, t2, t3, t4, t5, t6, t7, t8;                    \
00095     vector_u16_t shift;                                                 \
00096                                                                         \
00097     c4 = vec_splat (constants[0], 0);                                   \
00098     a0 = vec_splat (constants[0], 1);                                   \
00099     a1 = vec_splat (constants[0], 2);                                   \
00100     a2 = vec_splat (constants[0], 3);                                   \
00101     mc4 = vec_splat (constants[0], 4);                                  \
00102     ma2 = vec_splat (constants[0], 5);                                  \
00103     bias = (vector_s16_t)vec_splat ((vector_s32_t)constants[0], 3);     \
00104                                                                         \
00105     zero = vec_splat_s16 (0);                                           \
00106     shift = vec_splat_u16 (4);                                          \
00107                                                                         \
00108     vx0 = vec_mradds (vec_sl (block[0], shift), constants[1], zero);    \
00109     vx1 = vec_mradds (vec_sl (block[1], shift), constants[2], zero);    \
00110     vx2 = vec_mradds (vec_sl (block[2], shift), constants[3], zero);    \
00111     vx3 = vec_mradds (vec_sl (block[3], shift), constants[4], zero);    \
00112     vx4 = vec_mradds (vec_sl (block[4], shift), constants[1], zero);    \
00113     vx5 = vec_mradds (vec_sl (block[5], shift), constants[4], zero);    \
00114     vx6 = vec_mradds (vec_sl (block[6], shift), constants[3], zero);    \
00115     vx7 = vec_mradds (vec_sl (block[7], shift), constants[2], zero);    \
00116                                                                         \
00117     IDCT_HALF                                                           \
00118                                                                         \
00119     vx0 = vec_mergeh (vy0, vy4);                                        \
00120     vx1 = vec_mergel (vy0, vy4);                                        \
00121     vx2 = vec_mergeh (vy1, vy5);                                        \
00122     vx3 = vec_mergel (vy1, vy5);                                        \
00123     vx4 = vec_mergeh (vy2, vy6);                                        \
00124     vx5 = vec_mergel (vy2, vy6);                                        \
00125     vx6 = vec_mergeh (vy3, vy7);                                        \
00126     vx7 = vec_mergel (vy3, vy7);                                        \
00127                                                                         \
00128     vy0 = vec_mergeh (vx0, vx4);                                        \
00129     vy1 = vec_mergel (vx0, vx4);                                        \
00130     vy2 = vec_mergeh (vx1, vx5);                                        \
00131     vy3 = vec_mergel (vx1, vx5);                                        \
00132     vy4 = vec_mergeh (vx2, vx6);                                        \
00133     vy5 = vec_mergel (vx2, vx6);                                        \
00134     vy6 = vec_mergeh (vx3, vx7);                                        \
00135     vy7 = vec_mergel (vx3, vx7);                                        \
00136                                                                         \
00137     vx0 = vec_adds (vec_mergeh (vy0, vy4), bias);                       \
00138     vx1 = vec_mergel (vy0, vy4);                                        \
00139     vx2 = vec_mergeh (vy1, vy5);                                        \
00140     vx3 = vec_mergel (vy1, vy5);                                        \
00141     vx4 = vec_mergeh (vy2, vy6);                                        \
00142     vx5 = vec_mergel (vy2, vy6);                                        \
00143     vx6 = vec_mergeh (vy3, vy7);                                        \
00144     vx7 = vec_mergel (vy3, vy7);                                        \
00145                                                                         \
00146     IDCT_HALF                                                           \
00147                                                                         \
00148     shift = vec_splat_u16 (6);                                          \
00149     vx0 = vec_sra (vy0, shift);                                         \
00150     vx1 = vec_sra (vy1, shift);                                         \
00151     vx2 = vec_sra (vy2, shift);                                         \
00152     vx3 = vec_sra (vy3, shift);                                         \
00153     vx4 = vec_sra (vy4, shift);                                         \
00154     vx5 = vec_sra (vy5, shift);                                         \
00155     vx6 = vec_sra (vy6, shift);                                         \
00156     vx7 = vec_sra (vy7, shift);
00157 
00158 
00159 static const_vector_s16_t constants[5] = {
00160     (vector_s16_t) AVV(23170, 13573, 6518, 21895, -23170, -21895, 32, 31),
00161     (vector_s16_t) AVV(16384, 22725, 21407, 19266, 16384, 19266, 21407, 22725),
00162     (vector_s16_t) AVV(22725, 31521, 29692, 26722, 22725, 26722, 29692, 31521),
00163     (vector_s16_t) AVV(21407, 29692, 27969, 25172, 21407, 25172, 27969, 29692),
00164     (vector_s16_t) AVV(19266, 26722, 25172, 22654, 19266, 22654, 25172, 26722)
00165 };
00166 
00167 void idct_put_altivec(uint8_t* dest, int stride, vector_s16_t* block)
00168 {
00169 POWERPC_PERF_DECLARE(altivec_idct_put_num, 1);
00170 #ifdef ALTIVEC_USE_REFERENCE_C_CODE
00171 POWERPC_PERF_START_COUNT(altivec_idct_put_num, 1);
00172     void simple_idct_put(uint8_t *dest, int line_size, int16_t *block);
00173     simple_idct_put(dest, stride, (int16_t*)block);
00174 POWERPC_PERF_STOP_COUNT(altivec_idct_put_num, 1);
00175 #else /* ALTIVEC_USE_REFERENCE_C_CODE */
00176     vector_u8_t tmp;
00177 
00178 #ifdef POWERPC_PERFORMANCE_REPORT
00179 POWERPC_PERF_START_COUNT(altivec_idct_put_num, 1);
00180 #endif
00181     IDCT
00182 
00183 #define COPY(dest,src)                                          \
00184     tmp = vec_packsu (src, src);                                \
00185     vec_ste ((vector_u32_t)tmp, 0, (unsigned int *)dest);       \
00186     vec_ste ((vector_u32_t)tmp, 4, (unsigned int *)dest);
00187 
00188     COPY (dest, vx0)    dest += stride;
00189     COPY (dest, vx1)    dest += stride;
00190     COPY (dest, vx2)    dest += stride;
00191     COPY (dest, vx3)    dest += stride;
00192     COPY (dest, vx4)    dest += stride;
00193     COPY (dest, vx5)    dest += stride;
00194     COPY (dest, vx6)    dest += stride;
00195     COPY (dest, vx7)
00196 
00197 POWERPC_PERF_STOP_COUNT(altivec_idct_put_num, 1);
00198 #endif /* ALTIVEC_USE_REFERENCE_C_CODE */
00199 }
00200 
00201 void idct_add_altivec(uint8_t* dest, int stride, vector_s16_t* block)
00202 {
00203 POWERPC_PERF_DECLARE(altivec_idct_add_num, 1);
00204 #ifdef ALTIVEC_USE_REFERENCE_C_CODE
00205 POWERPC_PERF_START_COUNT(altivec_idct_add_num, 1);
00206     void simple_idct_add(uint8_t *dest, int line_size, int16_t *block);
00207     simple_idct_add(dest, stride, (int16_t*)block);
00208 POWERPC_PERF_STOP_COUNT(altivec_idct_add_num, 1);
00209 #else /* ALTIVEC_USE_REFERENCE_C_CODE */
00210     vector_u8_t tmp;
00211     vector_s16_t tmp2, tmp3;
00212     vector_u8_t perm0;
00213     vector_u8_t perm1;
00214     vector_u8_t p0, p1, p;
00215 
00216 #ifdef POWERPC_PERFORMANCE_REPORT
00217 POWERPC_PERF_START_COUNT(altivec_idct_add_num, 1);
00218 #endif
00219 
00220     IDCT
00221 
00222     p0 = vec_lvsl (0, dest);
00223     p1 = vec_lvsl (stride, dest);
00224     p = vec_splat_u8 (-1);
00225     perm0 = vec_mergeh (p, p0);
00226     perm1 = vec_mergeh (p, p1);
00227 
00228 #define ADD(dest,src,perm)                                              \
00229     /* *(uint64_t *)&tmp = *(uint64_t *)dest; */                        \
00230     tmp = vec_ld (0, dest);                                             \
00231     tmp2 = (vector_s16_t)vec_perm (tmp, (vector_u8_t)zero, perm);       \
00232     tmp3 = vec_adds (tmp2, src);                                        \
00233     tmp = vec_packsu (tmp3, tmp3);                                      \
00234     vec_ste ((vector_u32_t)tmp, 0, (unsigned int *)dest);               \
00235     vec_ste ((vector_u32_t)tmp, 4, (unsigned int *)dest);
00236 
00237     ADD (dest, vx0, perm0)      dest += stride;
00238     ADD (dest, vx1, perm1)      dest += stride;
00239     ADD (dest, vx2, perm0)      dest += stride;
00240     ADD (dest, vx3, perm1)      dest += stride;
00241     ADD (dest, vx4, perm0)      dest += stride;
00242     ADD (dest, vx5, perm1)      dest += stride;
00243     ADD (dest, vx6, perm0)      dest += stride;
00244     ADD (dest, vx7, perm1)
00245 
00246 POWERPC_PERF_STOP_COUNT(altivec_idct_add_num, 1);
00247 #endif /* ALTIVEC_USE_REFERENCE_C_CODE */
00248 }
00249
hvirtual/quicktime/ffmpeg/libavcodec/ppc/idct_altivec.c
