MythTV: zoom_filter_xmmx.c Source File

00001 #include <mythtv/mythconfig.h>
00002 
00003 #if defined(MMX) && !defined(ARCH_X86_64)
00004 /* a definir pour avoir exactement le meme resultat que la fonction C
00005  * (un chouillat plus lent)
00006  */
00007 #define STRICT_COMPAT
00008 //#define HAVE_ATHLON
00009 
00010 #define BUFFPOINTNB 16
00011 #define BUFFPOINTMASK 0xffff
00012 #define BUFFINCR 0xff
00013 
00014 #define sqrtperte 16
00015 // faire : a % sqrtperte <=> a & pertemask
00016 #define PERTEMASK 0xf
00017 // faire : a / sqrtperte <=> a >> PERTEDEC
00018 #define PERTEDEC 4
00019 
00020 
00021 //#define MMX_TRACE
00022 #include "mmx.h"
00023 
00024 int zoom_filter_xmmx_supported () {
00025         return (mm_support()&0x8)>>3;
00026 }
00027 
00028 void zoom_filter_xmmx (int prevX, int prevY,
00029                                                                                          unsigned int *expix1, unsigned int *expix2,
00030                                                                                          int *lbruS, int *lbruD, int buffratio,
00031                                                                                          int precalCoef[16][16])
00032 {
00033   int bufsize = prevX * prevY; /* taille du buffer */
00034   volatile int loop;                    /* variable de boucle */
00035 
00036         mmx_t *brutS = (mmx_t*)lbruS; /* buffer de transformation source */
00037         mmx_t *brutD = (mmx_t*)lbruD; /* buffer de transformation dest */
00038 
00039   volatile mmx_t prevXY;
00040         volatile mmx_t ratiox;
00041         //volatile mmx_t interpix;
00042 
00043         expix1[0]=expix1[prevX-1]=expix1[prevX*prevY-1]=expix1[prevX*prevY-prevX]=0;
00044 
00045         prevXY.ud[0] = (prevX-1)<<PERTEDEC;
00046         prevXY.ud[1] = (prevY-1)<<PERTEDEC;
00047         
00048         ratiox.d[0] = buffratio;
00049         ratiox.d[1] = buffratio;
00050         movq_m2r (ratiox, mm6);
00051         pslld_i2r (16,mm6);
00052 
00053   pxor_r2r (mm7,mm7); /* mise a zero de mm7 */
00054                 
00055   loop=0;
00056 
00057         /*
00058          * NOTE : mm6 et mm7 ne sont pas modifies dans la boucle.
00059          */
00060         while (loop < bufsize)
00061                 {
00062                         /*
00063                          * pre : mm6 = [buffratio<<16|buffratio<<16]
00064                          * post : mm0 = S + ((D-S)*buffratio)>>16 format [X|Y]
00065                          * modified = mm0,mm1,mm2
00066                          */
00067                         
00068                         __asm__ __volatile__ (
00069                                 "movq %0,%%mm0\n"
00070                                 "movq %1,%%mm1\n"
00071                                 : :"m"(brutS[loop]),"m"(brutD[loop])
00072                                 );               /* mm0 = S */
00073                         
00074                         psubd_r2r (mm0,mm1);           /* mm1 = D - S */
00075                         movq_r2r (mm1, mm2);           /* mm2 = D - S */
00076                         
00077                         pslld_i2r (16,mm1);
00078                         mmx_r2r (pmulhuw, mm6, mm1);   /* mm1 = ?? */
00079                         pmullw_r2r (mm6, mm2);
00080                 
00081                         paddd_r2r (mm2, mm1);     /* mm1 = (D - S) * buffratio >> 16 */
00082                         pslld_i2r (16,mm0);
00083                         
00084                         paddd_r2r (mm1, mm0);     /* mm0 = S + mm1 */
00085                         psrld_i2r (16, mm0);
00086                         
00087                         /*
00088                          * pre : mm0 : position vector on screen
00089                          *       prevXY : coordinate of the lower-right point on screen
00090                          * post : clipped mm0
00091                          * modified : mm0,mm1,mm2
00092                          */
00093                         movq_m2r (prevXY,mm1);
00094                         pcmpgtd_r2r (mm0, mm1); /* mm0 en X contient :
00095                                                                                                                                  1111 si prevXY > px
00096                                                                                                                                  0000 si prevXY <= px
00097                                                                                                                                  (idem pour y) */
00098 #ifdef STRICT_COMPAT
00099                         movq_r2r (mm1,mm2);
00100                         punpckhdq_r2r (mm2,mm2);
00101                         punpckldq_r2r (mm1,mm1);
00102                         pand_r2r (mm2, mm0);
00103 #endif
00104                         pand_r2r (mm1, mm0);    /* on met a zero la partie qui deborde */
00105                         
00106                         /*
00107                          * pre : mm0 : clipped position on screen
00108                          *
00109                          * post : mm3 & mm4 : coefs for this position
00110                          *              mm1 : X vector [0|X]
00111                          *
00112                          * modif : eax,ecx
00113                          */
00114                         __asm__ __volatile__ (
00115                                 "movd %%mm0,%%ecx\n"
00116                                 "movq %%mm0,%%mm1\n"
00117 
00118                                 "andl $15,%%ecx\n"
00119                                 "psrlq $32,%%mm1\n"
00120 
00121                                 "shll $6,%%ecx\n"
00122                                 "movd %%mm1,%%eax\n"
00123 
00124                                 "addl %0,%%ecx\n"
00125                                 "andl $15,%%eax\n"
00126 
00127                                 "movd (%%ecx,%%eax,4),%%mm3\n"
00128                                 /* ::"X"(precalCoef):"eax","ecx"); */
00129                                 ::"m"(precalCoef):"eax","ecx");
00130                                 
00131 
00132                         /*
00133                          * extraction des coefficients...
00134                          *
00135                          * pre : coef dans mm3
00136                          *
00137                          * post : coef extraits dans mm3 (c1 & c2)
00138                          *                        et mm4 (c3 & c4)
00139                          *
00140                          * modif : mm5
00141                          */
00142 // entrelace avec portion d'apres (cf les '^')
00143 //                      movq_r2r (mm3, mm5);            /* ??-??-??-??-c4-c3-c2-c1 */
00144 //                      punpcklbw_r2r (mm5, mm3);       /* c4-c4-c3-c3-c2-c2-c1-c1 */
00145 //                      movq_r2r (mm3, mm4);            /* c4-c4-c3-c3-c2-c2-c1-c1 */
00146 //                      movq_r2r (mm3, mm5);            /* c4-c4-c3-c3-c2-c2-c1-c1 */
00147                         
00148 //                      punpcklbw_r2r (mm5, mm3);       /* c2-c2-c2-c2-c1-c1-c1-c1 */
00149 //                      punpckhbw_r2r (mm5, mm4);       /* c4-c4-c4-c4-c3-c3-c3-c3 */
00150                         
00151                         /*
00152                          * pre : mm0 : Y pos [*|Y]
00153                          *       mm1 : X pos [*|X]
00154                          *
00155                          * post : mm0 : expix1[position]
00156                          *        mm2 : expix1[position+largeur]
00157                          *
00158                          * modif : eax,ecx
00159                          */
00160                         psrld_i2r (PERTEDEC,mm0);
00161                         psrld_i2r (PERTEDEC,mm1);
00162                         __asm__ __volatile__ (
00163                                 "movd %%mm1,%%eax\n"
00164                                 /*^*/ "movq %%mm3,%%mm5\n"       /*^*/
00165 
00166                                 "mull %1\n"
00167                                 "movd %%mm0,%%ecx\n"
00168                                 /*^*/ "punpcklbw %%mm5, %%mm3\n" /*^*/
00169 
00170                                 "addl %%ecx,%%eax\n"
00171                                 /*^*/ "movq %%mm3,%%mm4\n"       /*^*/
00172                                 /*^*/ "movq %%mm3,%%mm5\n"       /*^*/
00173 
00174                                 "movl %0,%%ecx\n"
00175                                 /*^*/ "punpcklbw %%mm5,%%mm3\n"  /*^*/
00176 
00177                                 "movq (%%ecx,%%eax,4),%%mm0\n"
00178                                 /*^*/ "punpckhbw %%mm5,%%mm4\n"  /*^*/
00179 
00180                                 "addl %1,%%eax\n"
00181                                 "movq (%%ecx,%%eax,4),%%mm2\n"
00182                                 
00183                                 : : "X"(expix1), "X"(prevX):"eax","ecx"
00184                                 );
00185 
00186                         /*
00187                          * pre :       mm0 : expix1[position]
00188                          *             mm2 : expix1[position+largeur]
00189                          *       mm3 & mm4 : coefs
00190                          */
00191 
00192                         /* recopie des deux premiers pixels dans mm0 et mm1 */
00193                         movq_r2r (mm0, mm1);            /* b1-v1-r1-a1-b2-v2-r2-a2 */
00194 
00195                         /* depackage du premier pixel */
00196                         punpcklbw_r2r (mm7, mm0);       /* 00-b2-00-v2-00-r2-00-a2 */
00197 
00198                         /* extraction des coefficients... */
00199 
00200                         movq_r2r (mm3, mm5);      /* c2-c2-c2-c2-c1-c1-c1-c1 */
00201 
00202                         /*^en parrallele^*/ /* depackage du 2ieme pixel */
00203                         /*^*/ punpckhbw_r2r (mm7, mm1); /* 00-b1-00-v1-00-r1-00-a1 */
00204 
00205                         punpcklbw_r2r (mm7, mm5);       /* 00-c1-00-c1-00-c1-00-c1 */
00206                         punpckhbw_r2r (mm7, mm3);       /* 00-c2-00-c2-00-c2-00-c2 */
00207                         
00208                         /* multiplication des pixels par les coefficients */
00209                         pmullw_r2r (mm5, mm0);          /* c1*b2-c1*v2-c1*r2-c1*a2 */
00210                         pmullw_r2r (mm3, mm1);          /* c2*b1-c2*v1-c2*r1-c2*a1 */
00211                         paddw_r2r (mm1, mm0);
00212                                 
00213                         /* ...extraction des 2 derniers coefficients */
00214                         movq_r2r (mm4, mm5);                    /* c4-c4-c4-c4-c3-c3-c3-c3 */
00215                         punpcklbw_r2r (mm7, mm4);       /* 00-c3-00-c3-00-c3-00-c3 */
00216                         punpckhbw_r2r (mm7, mm5);       /* 00-c4-00-c4-00-c4-00-c4 */
00217                         
00218                         /* recuperation des 2 derniers pixels */
00219                         movq_r2r (mm2, mm1);
00220                         
00221                         /* depackage des pixels */
00222                         punpcklbw_r2r (mm7, mm1);
00223                         punpckhbw_r2r (mm7, mm2);
00224                         
00225                         /* multiplication pas les coeffs */
00226                         pmullw_r2r (mm4, mm1);
00227                         pmullw_r2r (mm5, mm2);
00228                            
00229                         /* ajout des valeurs obtenues � la valeur finale */
00230                         paddw_r2r (mm1, mm0);
00231                         paddw_r2r (mm2, mm0);
00232                            
00233                         /* division par 256 = 16+16+16+16, puis repackage du pixel final */
00234                         psrlw_i2r (8, mm0);
00235                         packuswb_r2r (mm7, mm0);
00236 
00237                         movd_r2m (mm0,expix2[loop]);
00238                         
00239                         ++loop;
00240                 }
00241 #ifdef HAVE_ATHLON
00242         __asm__ __volatile__ ("femms\n");
00243 #else
00244         emms();
00245 #endif
00246 }
00247 #else
00248 int zoom_filter_xmmx_supported () {
00249         return 0;
00250 }
00251 void zoom_filter_xmmx (int prevX, int prevY,
00252                                                                                          unsigned int *expix1, unsigned int *expix2,
00253                                                                                          int *lbruS, int *lbruD, int buffratio,
00254                                                                                          int precalCoef[16][16])
00255 {
00256     (void) prevX;     (void) prevY;
00257     (void) expix1;    (void) expix2;
00258     (void) lbruS;     (void) lbruD;
00259     (void) buffratio; (void) precalCoef;
00260         return;
00261 }
00262 #endif