Cinelerra: hvirtual/quicktime/ffmpeg/libavcodec/i386/mpegvideo_mmx

00001 /*
00002  * MPEG video MMX templates
00003  *
00004  * Copyright (c) 2002 Michael Niedermayer <michaelni@gmx.at>
00005  *
00006  * This library is free software; you can redistribute it and/or
00007  * modify it under the terms of the GNU Lesser General Public
00008  * License as published by the Free Software Foundation; either
00009  * version 2 of the License, or (at your option) any later version.
00010  *
00011  * This library is distributed in the hope that it will be useful,
00012  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00013  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00014  * Lesser General Public License for more details.
00015  *
00016  * You should have received a copy of the GNU Lesser General Public
00017  * License along with this library; if not, write to the Free Software
00018  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
00019  */
00020 #undef SPREADW
00021 #undef PMAXW
00022 #ifdef HAVE_MMX2
00023 #define SPREADW(a) "pshufw $0, " #a ", " #a " \n\t"
00024 #define PMAXW(a,b) "pmaxsw " #a ", " #b " \n\t"
00025 
00026 #else
00027 #define SPREADW(a) \
00028         "punpcklwd " #a ", " #a " \n\t"\
00029         "punpcklwd " #a ", " #a " \n\t"
00030 #define PMAXW(a,b) \
00031         "psubusw " #a ", " #b " \n\t"\
00032         "paddw " #a ", " #b " \n\t"
00033 #endif
00034 
00035 static int RENAME(dct_quantize)(MpegEncContext *s,
00036                             DCTELEM *block, int n,
00037                             int qscale, int *overflow)
00038 {
00039     long last_non_zero_p1;
00040     int level=0, q; //=0 is cuz gcc says uninitalized ...
00041     const uint16_t *qmat, *bias;
00042     __align8 int16_t temp_block[64];
00043     
00044     assert((7&(int)(&temp_block[0])) == 0); //did gcc align it correctly?
00045 
00046     //s->fdct (block);
00047     RENAMEl(ff_fdct) (block); //cant be anything else ...
00048 
00049     if(s->dct_error_sum)
00050         s->denoise_dct(s, block);
00051 
00052     if (s->mb_intra) {
00053         int dummy;
00054         if (n < 4)
00055             q = s->y_dc_scale;
00056         else
00057             q = s->c_dc_scale;
00058         /* note: block[0] is assumed to be positive */
00059         if (!s->h263_aic) {
00060 #if 1
00061         asm volatile (
00062                 "mul %%ecx              \n\t"
00063                 : "=d" (level), "=a"(dummy)
00064                 : "a" ((block[0]>>2) + q), "c" (inverse[q<<1])
00065         );
00066 #else
00067         asm volatile (
00068                 "xorl %%edx, %%edx      \n\t"
00069                 "divw %%cx              \n\t"
00070                 "movzwl %%ax, %%eax     \n\t"
00071                 : "=a" (level)
00072                 : "a" ((block[0]>>2) + q), "c" (q<<1)
00073                 : "%edx"
00074         );
00075 #endif
00076         } else
00077             /* For AIC we skip quant/dequant of INTRADC */
00078             level = (block[0] + 4)>>3;
00079             
00080         block[0]=0; //avoid fake overflow
00081 //        temp_block[0] = (block[0] + (q >> 1)) / q;
00082         last_non_zero_p1 = 1;
00083         bias = s->q_intra_matrix16[qscale][1];
00084         qmat = s->q_intra_matrix16[qscale][0];
00085     } else {
00086         last_non_zero_p1 = 0;
00087         bias = s->q_inter_matrix16[qscale][1];
00088         qmat = s->q_inter_matrix16[qscale][0];
00089     }
00090 
00091     if((s->out_format == FMT_H263 || s->out_format == FMT_H261) && s->mpeg_quant==0){
00092     
00093         asm volatile(
00094             "movd %%"REG_a", %%mm3              \n\t" // last_non_zero_p1
00095             SPREADW(%%mm3)
00096             "pxor %%mm7, %%mm7                  \n\t" // 0
00097             "pxor %%mm4, %%mm4                  \n\t" // 0
00098             "movq (%2), %%mm5                   \n\t" // qmat[0]
00099             "pxor %%mm6, %%mm6                  \n\t"
00100             "psubw (%3), %%mm6                  \n\t" // -bias[0]
00101             "mov $-128, %%"REG_a"               \n\t"
00102             ".balign 16                         \n\t"
00103             "1:                                 \n\t"
00104             "pxor %%mm1, %%mm1                  \n\t" // 0
00105             "movq (%1, %%"REG_a"), %%mm0        \n\t" // block[i]
00106             "pcmpgtw %%mm0, %%mm1               \n\t" // block[i] <= 0 ? 0xFF : 0x00
00107             "pxor %%mm1, %%mm0                  \n\t" 
00108             "psubw %%mm1, %%mm0                 \n\t" // ABS(block[i])
00109             "psubusw %%mm6, %%mm0               \n\t" // ABS(block[i]) + bias[0]
00110             "pmulhw %%mm5, %%mm0                \n\t" // (ABS(block[i])*qmat[0] - bias[0]*qmat[0])>>16
00111             "por %%mm0, %%mm4                   \n\t" 
00112             "pxor %%mm1, %%mm0                  \n\t" 
00113             "psubw %%mm1, %%mm0                 \n\t" // out=((ABS(block[i])*qmat[0] - bias[0]*qmat[0])>>16)*sign(block[i])
00114             "movq %%mm0, (%5, %%"REG_a")        \n\t"
00115             "pcmpeqw %%mm7, %%mm0               \n\t" // out==0 ? 0xFF : 0x00
00116             "movq (%4, %%"REG_a"), %%mm1        \n\t" 
00117             "movq %%mm7, (%1, %%"REG_a")        \n\t" // 0
00118             "pandn %%mm1, %%mm0                 \n\t"
00119             PMAXW(%%mm0, %%mm3)
00120             "add $8, %%"REG_a"                  \n\t"
00121             " js 1b                             \n\t"
00122             "movq %%mm3, %%mm0                  \n\t"
00123             "psrlq $32, %%mm3                   \n\t"
00124             PMAXW(%%mm0, %%mm3)
00125             "movq %%mm3, %%mm0                  \n\t"
00126             "psrlq $16, %%mm3                   \n\t"
00127             PMAXW(%%mm0, %%mm3)
00128             "movd %%mm3, %%"REG_a"              \n\t"
00129             "movzb %%al, %%"REG_a"              \n\t" // last_non_zero_p1
00130             : "+a" (last_non_zero_p1)
00131             : "r" (block+64), "r" (qmat), "r" (bias),
00132               "r" (inv_zigzag_direct16+64), "r" (temp_block+64)
00133         );
00134         // note the asm is split cuz gcc doesnt like that many operands ...
00135         asm volatile(
00136             "movd %1, %%mm1                     \n\t" // max_qcoeff
00137             SPREADW(%%mm1)
00138             "psubusw %%mm1, %%mm4               \n\t" 
00139             "packuswb %%mm4, %%mm4              \n\t"
00140             "movd %%mm4, %0                     \n\t" // *overflow
00141         : "=g" (*overflow)
00142         : "g" (s->max_qcoeff)
00143         );
00144     }else{ // FMT_H263
00145         asm volatile(
00146             "movd %%"REG_a", %%mm3              \n\t" // last_non_zero_p1
00147             SPREADW(%%mm3)
00148             "pxor %%mm7, %%mm7                  \n\t" // 0
00149             "pxor %%mm4, %%mm4                  \n\t" // 0
00150             "mov $-128, %%"REG_a"               \n\t"
00151             ".balign 16                         \n\t"
00152             "1:                                 \n\t"
00153             "pxor %%mm1, %%mm1                  \n\t" // 0
00154             "movq (%1, %%"REG_a"), %%mm0        \n\t" // block[i]
00155             "pcmpgtw %%mm0, %%mm1               \n\t" // block[i] <= 0 ? 0xFF : 0x00
00156             "pxor %%mm1, %%mm0                  \n\t" 
00157             "psubw %%mm1, %%mm0                 \n\t" // ABS(block[i])
00158             "movq (%3, %%"REG_a"), %%mm6        \n\t" // bias[0]
00159             "paddusw %%mm6, %%mm0               \n\t" // ABS(block[i]) + bias[0]
00160             "movq (%2, %%"REG_a"), %%mm5                \n\t" // qmat[i]
00161             "pmulhw %%mm5, %%mm0                \n\t" // (ABS(block[i])*qmat[0] + bias[0]*qmat[0])>>16
00162             "por %%mm0, %%mm4                   \n\t" 
00163             "pxor %%mm1, %%mm0                  \n\t" 
00164             "psubw %%mm1, %%mm0                 \n\t" // out=((ABS(block[i])*qmat[0] - bias[0]*qmat[0])>>16)*sign(block[i])
00165             "movq %%mm0, (%5, %%"REG_a")        \n\t"
00166             "pcmpeqw %%mm7, %%mm0               \n\t" // out==0 ? 0xFF : 0x00
00167             "movq (%4, %%"REG_a"), %%mm1                \n\t" 
00168             "movq %%mm7, (%1, %%"REG_a")                \n\t" // 0
00169             "pandn %%mm1, %%mm0                 \n\t"
00170             PMAXW(%%mm0, %%mm3)
00171             "add $8, %%"REG_a"                  \n\t"
00172             " js 1b                             \n\t"
00173             "movq %%mm3, %%mm0                  \n\t"
00174             "psrlq $32, %%mm3                   \n\t"
00175             PMAXW(%%mm0, %%mm3)
00176             "movq %%mm3, %%mm0                  \n\t"
00177             "psrlq $16, %%mm3                   \n\t"
00178             PMAXW(%%mm0, %%mm3)
00179             "movd %%mm3, %%"REG_a"              \n\t"
00180             "movzb %%al, %%"REG_a"              \n\t" // last_non_zero_p1
00181             : "+a" (last_non_zero_p1)
00182             : "r" (block+64), "r" (qmat+64), "r" (bias+64),
00183               "r" (inv_zigzag_direct16+64), "r" (temp_block+64)
00184         );
00185         // note the asm is split cuz gcc doesnt like that many operands ...
00186         asm volatile(
00187             "movd %1, %%mm1                     \n\t" // max_qcoeff
00188             SPREADW(%%mm1)
00189             "psubusw %%mm1, %%mm4               \n\t" 
00190             "packuswb %%mm4, %%mm4              \n\t"
00191             "movd %%mm4, %0                     \n\t" // *overflow
00192         : "=g" (*overflow)
00193         : "g" (s->max_qcoeff)
00194         );
00195     }
00196 
00197     if(s->mb_intra) block[0]= level;
00198     else            block[0]= temp_block[0];
00199 
00200     if(s->dsp.idct_permutation_type == FF_SIMPLE_IDCT_PERM){
00201         if(last_non_zero_p1 <= 1) goto end;
00202         block[0x08] = temp_block[0x01]; block[0x10] = temp_block[0x08]; 
00203         block[0x20] = temp_block[0x10]; 
00204         if(last_non_zero_p1 <= 4) goto end;
00205         block[0x18] = temp_block[0x09]; block[0x04] = temp_block[0x02]; 
00206         block[0x09] = temp_block[0x03]; 
00207         if(last_non_zero_p1 <= 7) goto end;
00208         block[0x14] = temp_block[0x0A]; block[0x28] = temp_block[0x11]; 
00209         block[0x12] = temp_block[0x18]; block[0x02] = temp_block[0x20]; 
00210         if(last_non_zero_p1 <= 11) goto end;
00211         block[0x1A] = temp_block[0x19]; block[0x24] = temp_block[0x12]; 
00212         block[0x19] = temp_block[0x0B]; block[0x01] = temp_block[0x04]; 
00213         block[0x0C] = temp_block[0x05]; 
00214         if(last_non_zero_p1 <= 16) goto end;
00215         block[0x11] = temp_block[0x0C]; block[0x29] = temp_block[0x13]; 
00216         block[0x16] = temp_block[0x1A]; block[0x0A] = temp_block[0x21]; 
00217         block[0x30] = temp_block[0x28]; block[0x22] = temp_block[0x30]; 
00218         block[0x38] = temp_block[0x29]; block[0x06] = temp_block[0x22]; 
00219         if(last_non_zero_p1 <= 24) goto end;
00220         block[0x1B] = temp_block[0x1B]; block[0x21] = temp_block[0x14]; 
00221         block[0x1C] = temp_block[0x0D]; block[0x05] = temp_block[0x06]; 
00222         block[0x0D] = temp_block[0x07]; block[0x15] = temp_block[0x0E]; 
00223         block[0x2C] = temp_block[0x15]; block[0x13] = temp_block[0x1C]; 
00224         if(last_non_zero_p1 <= 32) goto end;
00225         block[0x0B] = temp_block[0x23]; block[0x34] = temp_block[0x2A]; 
00226         block[0x2A] = temp_block[0x31]; block[0x32] = temp_block[0x38]; 
00227         block[0x3A] = temp_block[0x39]; block[0x26] = temp_block[0x32]; 
00228         block[0x39] = temp_block[0x2B]; block[0x03] = temp_block[0x24]; 
00229         if(last_non_zero_p1 <= 40) goto end;
00230         block[0x1E] = temp_block[0x1D]; block[0x25] = temp_block[0x16]; 
00231         block[0x1D] = temp_block[0x0F]; block[0x2D] = temp_block[0x17]; 
00232         block[0x17] = temp_block[0x1E]; block[0x0E] = temp_block[0x25]; 
00233         block[0x31] = temp_block[0x2C]; block[0x2B] = temp_block[0x33]; 
00234         if(last_non_zero_p1 <= 48) goto end;
00235         block[0x36] = temp_block[0x3A]; block[0x3B] = temp_block[0x3B]; 
00236         block[0x23] = temp_block[0x34]; block[0x3C] = temp_block[0x2D]; 
00237         block[0x07] = temp_block[0x26]; block[0x1F] = temp_block[0x1F]; 
00238         block[0x0F] = temp_block[0x27]; block[0x35] = temp_block[0x2E]; 
00239         if(last_non_zero_p1 <= 56) goto end;
00240         block[0x2E] = temp_block[0x35]; block[0x33] = temp_block[0x3C]; 
00241         block[0x3E] = temp_block[0x3D]; block[0x27] = temp_block[0x36]; 
00242         block[0x3D] = temp_block[0x2F]; block[0x2F] = temp_block[0x37]; 
00243         block[0x37] = temp_block[0x3E]; block[0x3F] = temp_block[0x3F];
00244     }else if(s->dsp.idct_permutation_type == FF_LIBMPEG2_IDCT_PERM){
00245         if(last_non_zero_p1 <= 1) goto end;
00246         block[0x04] = temp_block[0x01]; 
00247         block[0x08] = temp_block[0x08]; block[0x10] = temp_block[0x10]; 
00248         if(last_non_zero_p1 <= 4) goto end;
00249         block[0x0C] = temp_block[0x09]; block[0x01] = temp_block[0x02]; 
00250         block[0x05] = temp_block[0x03]; 
00251         if(last_non_zero_p1 <= 7) goto end;
00252         block[0x09] = temp_block[0x0A]; block[0x14] = temp_block[0x11]; 
00253         block[0x18] = temp_block[0x18]; block[0x20] = temp_block[0x20]; 
00254         if(last_non_zero_p1 <= 11) goto end;
00255         block[0x1C] = temp_block[0x19]; 
00256         block[0x11] = temp_block[0x12]; block[0x0D] = temp_block[0x0B]; 
00257         block[0x02] = temp_block[0x04]; block[0x06] = temp_block[0x05]; 
00258         if(last_non_zero_p1 <= 16) goto end;
00259         block[0x0A] = temp_block[0x0C]; block[0x15] = temp_block[0x13]; 
00260         block[0x19] = temp_block[0x1A]; block[0x24] = temp_block[0x21]; 
00261         block[0x28] = temp_block[0x28]; block[0x30] = temp_block[0x30]; 
00262         block[0x2C] = temp_block[0x29]; block[0x21] = temp_block[0x22]; 
00263         if(last_non_zero_p1 <= 24) goto end;
00264         block[0x1D] = temp_block[0x1B]; block[0x12] = temp_block[0x14]; 
00265         block[0x0E] = temp_block[0x0D]; block[0x03] = temp_block[0x06]; 
00266         block[0x07] = temp_block[0x07]; block[0x0B] = temp_block[0x0E]; 
00267         block[0x16] = temp_block[0x15]; block[0x1A] = temp_block[0x1C]; 
00268         if(last_non_zero_p1 <= 32) goto end;
00269         block[0x25] = temp_block[0x23]; block[0x29] = temp_block[0x2A]; 
00270         block[0x34] = temp_block[0x31]; block[0x38] = temp_block[0x38]; 
00271         block[0x3C] = temp_block[0x39]; block[0x31] = temp_block[0x32]; 
00272         block[0x2D] = temp_block[0x2B]; block[0x22] = temp_block[0x24]; 
00273         if(last_non_zero_p1 <= 40) goto end;
00274         block[0x1E] = temp_block[0x1D]; block[0x13] = temp_block[0x16]; 
00275         block[0x0F] = temp_block[0x0F]; block[0x17] = temp_block[0x17]; 
00276         block[0x1B] = temp_block[0x1E]; block[0x26] = temp_block[0x25]; 
00277         block[0x2A] = temp_block[0x2C]; block[0x35] = temp_block[0x33]; 
00278         if(last_non_zero_p1 <= 48) goto end;
00279         block[0x39] = temp_block[0x3A]; block[0x3D] = temp_block[0x3B]; 
00280         block[0x32] = temp_block[0x34]; block[0x2E] = temp_block[0x2D]; 
00281             block[0x23] = temp_block[0x26]; block[0x1F] = temp_block[0x1F]; 
00282         block[0x27] = temp_block[0x27]; block[0x2B] = temp_block[0x2E]; 
00283         if(last_non_zero_p1 <= 56) goto end;
00284         block[0x36] = temp_block[0x35]; block[0x3A] = temp_block[0x3C]; 
00285         block[0x3E] = temp_block[0x3D]; block[0x33] = temp_block[0x36]; 
00286         block[0x2F] = temp_block[0x2F]; block[0x37] = temp_block[0x37]; 
00287         block[0x3B] = temp_block[0x3E]; block[0x3F] = temp_block[0x3F];
00288     }else{
00289         if(last_non_zero_p1 <= 1) goto end;
00290         block[0x01] = temp_block[0x01]; 
00291         block[0x08] = temp_block[0x08]; block[0x10] = temp_block[0x10]; 
00292         if(last_non_zero_p1 <= 4) goto end;
00293         block[0x09] = temp_block[0x09]; block[0x02] = temp_block[0x02]; 
00294         block[0x03] = temp_block[0x03]; 
00295         if(last_non_zero_p1 <= 7) goto end;
00296         block[0x0A] = temp_block[0x0A]; block[0x11] = temp_block[0x11]; 
00297         block[0x18] = temp_block[0x18]; block[0x20] = temp_block[0x20]; 
00298         if(last_non_zero_p1 <= 11) goto end;
00299         block[0x19] = temp_block[0x19]; 
00300         block[0x12] = temp_block[0x12]; block[0x0B] = temp_block[0x0B]; 
00301         block[0x04] = temp_block[0x04]; block[0x05] = temp_block[0x05]; 
00302         if(last_non_zero_p1 <= 16) goto end;
00303         block[0x0C] = temp_block[0x0C]; block[0x13] = temp_block[0x13]; 
00304         block[0x1A] = temp_block[0x1A]; block[0x21] = temp_block[0x21]; 
00305         block[0x28] = temp_block[0x28]; block[0x30] = temp_block[0x30]; 
00306         block[0x29] = temp_block[0x29]; block[0x22] = temp_block[0x22]; 
00307         if(last_non_zero_p1 <= 24) goto end;
00308         block[0x1B] = temp_block[0x1B]; block[0x14] = temp_block[0x14]; 
00309         block[0x0D] = temp_block[0x0D]; block[0x06] = temp_block[0x06]; 
00310         block[0x07] = temp_block[0x07]; block[0x0E] = temp_block[0x0E]; 
00311         block[0x15] = temp_block[0x15]; block[0x1C] = temp_block[0x1C]; 
00312         if(last_non_zero_p1 <= 32) goto end;
00313         block[0x23] = temp_block[0x23]; block[0x2A] = temp_block[0x2A]; 
00314         block[0x31] = temp_block[0x31]; block[0x38] = temp_block[0x38]; 
00315         block[0x39] = temp_block[0x39]; block[0x32] = temp_block[0x32]; 
00316         block[0x2B] = temp_block[0x2B]; block[0x24] = temp_block[0x24]; 
00317         if(last_non_zero_p1 <= 40) goto end;
00318         block[0x1D] = temp_block[0x1D]; block[0x16] = temp_block[0x16]; 
00319         block[0x0F] = temp_block[0x0F]; block[0x17] = temp_block[0x17]; 
00320         block[0x1E] = temp_block[0x1E]; block[0x25] = temp_block[0x25]; 
00321         block[0x2C] = temp_block[0x2C]; block[0x33] = temp_block[0x33]; 
00322         if(last_non_zero_p1 <= 48) goto end;
00323         block[0x3A] = temp_block[0x3A]; block[0x3B] = temp_block[0x3B]; 
00324         block[0x34] = temp_block[0x34]; block[0x2D] = temp_block[0x2D]; 
00325         block[0x26] = temp_block[0x26]; block[0x1F] = temp_block[0x1F]; 
00326         block[0x27] = temp_block[0x27]; block[0x2E] = temp_block[0x2E]; 
00327         if(last_non_zero_p1 <= 56) goto end;
00328         block[0x35] = temp_block[0x35]; block[0x3C] = temp_block[0x3C]; 
00329         block[0x3D] = temp_block[0x3D]; block[0x36] = temp_block[0x36]; 
00330         block[0x2F] = temp_block[0x2F]; block[0x37] = temp_block[0x37]; 
00331         block[0x3E] = temp_block[0x3E]; block[0x3F] = temp_block[0x3F];
00332     }
00333     end:
00334 /*
00335     for(i=0; i<last_non_zero_p1; i++)
00336     {
00337        int j= zigzag_direct_noperm[i];
00338        block[block_permute_op(j)]= temp_block[j];
00339     }
00340 */
00341 
00342     return last_non_zero_p1 - 1;
00343 }
hvirtual/quicktime/ffmpeg/libavcodec/i386/mpegvideo_mmx_template.c
