Cinelerra: hvirtual/quicktime/ffmpeg/libavcodec/ps2/idct

00001 /*
00002   Originally provided by Intel at AP-922
00003   http://developer.intel.com/vtune/cbts/strmsimd/922down.htm
00004   (See more app notes at http://developer.intel.com/vtune/cbts/strmsimd/appnotes.htm)
00005   but in a limited edition.
00006 
00007   column code adapted from peter gubanov
00008   Copyright (c) 2000-2001 Peter Gubanov <peter@elecard.net.ru>
00009   http://www.elecard.com/peter/idct.shtml
00010   Rounding trick Copyright (c) 2000 Michel Lespinasse <walken@zoy.org>
00011 
00012   MMI port by Leon van Stuivenberg
00013 */  
00014 #include "../common.h"
00015 #include "../dsputil.h"
00016 #include "mmi.h"
00017 
00018 #define BITS_INV_ACC    5       // 4 or 5 for IEEE
00019 #define SHIFT_INV_ROW   (16 - BITS_INV_ACC)
00020 #define SHIFT_INV_COL   (1 + BITS_INV_ACC)
00021 
00022 #define TG1     6518
00023 #define TG2     13573
00024 #define TG3     21895
00025 #define CS4     23170
00026 
00027 #define ROUNDER_0       0
00028 #define ROUNDER_1       16
00029 
00030 #define TAB_i_04        (32+0)
00031 #define TAB_i_17        (32+64)
00032 #define TAB_i_26        (32+128)
00033 #define TAB_i_35        (32+192)
00034 
00035 #define TG_1_16         (32+256+0)
00036 #define TG_2_16         (32+256+16)
00037 #define TG_3_16         (32+256+32)
00038 #define COS_4_16        (32+256+48)
00039 
00040 #define CLIPMAX         (32+256+64+0)
00041 
00042 static short consttable[] align16 = {
00043 /* rounder 0*/  // assume SHIFT_INV_ROW == 11
00044  0x3ff, 1, 0x3ff, 1, 0x3ff, 1, 0x3ff, 1,
00045 /* rounder 1*/
00046  0x3ff, 0, 0x3ff, 0, 0x3ff, 0, 0x3ff, 0,
00047 /* row 0/4*/
00048  16384,  21407, -16384, -21407,  22725,  19266, -22725, -12873, 
00049   8867,  16384,   8867,  16384,   4520,  12873,  -4520,  19266, 
00050  16384,  -8867,  16384,  -8867,  12873, -22725,  19266, -22725, 
00051  21407, -16384, -21407,  16384,  19266,   4520, -12873,   4520, 
00052 /* row 1/7*/
00053  22725,  29692, -22725, -29692,  31521,  26722, -31521, -17855, 
00054  12299,  22725,  12299,  22725,   6270,  17855,  -6270,  26722, 
00055  22725, -12299,  22725, -12299,  17855, -31521,  26722, -31521, 
00056  29692, -22725, -29692,  22725,  26722,   6270, -17855,   6270, 
00057 /* row 2/6*/
00058  21407,  27969, -21407, -27969,  29692,  25172, -29692, -16819, 
00059  11585,  21407,  11585,  21407,   5906,  16819,  -5906,  25172, 
00060  21407, -11585,  21407, -11585,  16819, -29692,  25172, -29692, 
00061  27969, -21407, -27969,  21407,  25172,   5906, -16819,   5906, 
00062 /*row 3/5*/
00063  19266,  25172, -19266, -25172,  26722,  22654, -26722, -15137, 
00064  10426,  19266,  10426,  19266,   5315,  15137,  -5315,  22654, 
00065  19266, -10426,  19266, -10426,  15137, -26722,  22654, -26722, 
00066  25172, -19266, -25172,  19266,  22654,   5315, -15137,   5315,
00067 /*column constants*/
00068  TG1, TG1, TG1, TG1, TG1, TG1, TG1, TG1,
00069  TG2, TG2, TG2, TG2, TG2, TG2, TG2, TG2,
00070  TG3, TG3, TG3, TG3, TG3, TG3, TG3, TG3,
00071  CS4, CS4, CS4, CS4, CS4, CS4, CS4, CS4,
00072 /* clamp */
00073  255, 255, 255, 255, 255, 255, 255, 255
00074 };
00075 
00076 
00077 #define DCT_8_INV_ROW1(blk, rowoff, taboff, rnd, outreg) { \
00078         lq(blk, rowoff, $16);   /* r16 = x7  x5  x3  x1  x6  x4  x2  x0 */ \
00079         /*slot*/ \
00080         lq($24, 0+taboff, $17); /* r17 = w */ \
00081         /*delay slot $16*/ \
00082         lq($24, 16+taboff, $18);/* r18 = w */ \
00083         prevh($16, $2);         /* r2  = x1  x3  x5  x7  x0  x2  x4  x6 */ \
00084         lq($24, 32+taboff, $19);/* r19 = w */ \
00085         phmadh($17, $16, $17);  /* r17 = b1"b0'a1"a0' */ \
00086         lq($24, 48+taboff, $20);/* r20 = w */ \
00087         phmadh($18, $2, $18);   /* r18 = b1'b0"a1'a0" */ \
00088         phmadh($19, $16, $19);  /* r19 = b3"b2'a3"a2' */ \
00089         phmadh($20, $2, $20);   /* r20 = b3'b2"a3'a2" */ \
00090         paddw($17, $18, $17);   /* r17 = (b1)(b0)(a1)(a0) */ \
00091         paddw($19, $20, $19);   /* r19 = (b3)(b2)(a3)(a2) */ \
00092         pcpyld($19, $17, $18);  /* r18 = (a3)(a2)(a1)(a0) */ \
00093         pcpyud($17, $19, $20);  /* r20 = (b3)(b2)(b1)(b0) */ \
00094         paddw($18, rnd, $18);   /* r18 = (a3)(a2)(a1)(a0) */\
00095         paddw($18, $20, $17);   /* r17 = ()()()(a0+b0) */ \
00096         psubw($18, $20, $20);   /* r20 = ()()()(a0-b0) */ \
00097         psraw($17, SHIFT_INV_ROW, $17); /* r17 = (y3 y2 y1 y0) */ \
00098         psraw($20, SHIFT_INV_ROW, $20); /* r20 = (y4 y5 y6 y7) */ \
00099         ppach($20, $17, outreg);/* out = y4 y5 y6 y7 y3 y2 y1 y0  Note order */ \
00100 \
00101         prevh(outreg, $2);      \
00102         pcpyud($2, $2, $2);     \
00103         pcpyld($2, outreg, outreg);     \
00104 }
00105 
00106 
00107 #define DCT_8_INV_COL8() \
00108 \
00109         lq($24, TG_3_16, $2);   /* r2  = tn3 */ \
00110 \
00111         pmulth($11, $2, $17);   /* r17 = x3 * tn3 (6420) */ \
00112         psraw($17, 15, $17);    \
00113         pmfhl_uw($3);           /* r3  = 7531 */        \
00114         psraw($3, 15, $3);      \
00115         pinteh($3, $17, $17);   /* r17 = x3 * tn3 */ \
00116         psubh($17, $13, $17);   /* r17 = tm35 */        \
00117 \
00118         pmulth($13, $2, $18);   /* r18 = x5 * tn3 (6420) */ \
00119         psraw($18, 15, $18);    \
00120         pmfhl_uw($3);           /* r3  = 7531 */        \
00121         psraw($3, 15, $3);      \
00122         pinteh($3, $18, $18);   /* r18 = x5 * tn3 */ \
00123         paddh($18, $11, $18);   /* r18 = tp35 */        \
00124 \
00125         lq($24, TG_1_16, $2);   /* r2  = tn1 */ \
00126 \
00127         pmulth($15, $2, $19);   /* r19 = x7 * tn1 (6420) */ \
00128         psraw($19, 15, $19);    \
00129         pmfhl_uw($3);           /* r3  = 7531 */        \
00130         psraw($3, 15, $3);      \
00131         pinteh($3, $19, $19);   /* r19 = x7 * tn1 */ \
00132         paddh($19, $9, $19);    /* r19 = tp17 */        \
00133 \
00134         pmulth($9, $2, $20);    /* r20 = x1 * tn1 (6420) */ \
00135         psraw($20, 15, $20);    \
00136         pmfhl_uw($3);           /* r3  = 7531 */        \
00137         psraw($3, 15, $3);      \
00138         pinteh($3, $20, $20);   /* r20 = x1 * tn1 */ \
00139         psubh($20, $15, $20);   /* r20 = tm17 */        \
00140 \
00141         psubh($19, $18, $3);    /* r3  = t1 */  \
00142         paddh($20, $17, $16);   /* r16 = t2 */  \
00143         psubh($20, $17, $23);   /* r23 = b3 */  \
00144         paddh($19, $18, $20);   /* r20 = b0 */  \
00145 \
00146         lq($24, COS_4_16, $2);  /* r2  = cs4 */ \
00147 \
00148         paddh($3, $16, $21);    /* r21 = t1+t2 */       \
00149         psubh($3, $16, $22);    /* r22 = t1-t2 */       \
00150 \
00151         pmulth($21, $2, $21);   /* r21 = cs4 * (t1+t2) 6420 */ \
00152         psraw($21, 15, $21);    \
00153         pmfhl_uw($3);           /* r3  = 7531 */        \
00154         psraw($3, 15, $3);      \
00155         pinteh($3, $21, $21);   /* r21 = b1 */  \
00156 \
00157         pmulth($22, $2, $22);   /* r22 = cs4 * (t1-t2) 6420 */ \
00158         psraw($22, 15, $22);    \
00159         pmfhl_uw($3);           /* r3  = 7531 */        \
00160         psraw($3, 15, $3);      \
00161         pinteh($3, $22, $22);   /* r22 = b2 */  \
00162 \
00163         lq($24, TG_2_16, $2);   /* r2  = tn2 */ \
00164 \
00165         pmulth($10, $2, $17);   /* r17 = x2 * tn2 (6420) */ \
00166         psraw($17, 15, $17);    \
00167         pmfhl_uw($3);           /* r3  = 7531 */        \
00168         psraw($3, 15, $3);      \
00169         pinteh($3, $17, $17);   /* r17 = x3 * tn3 */ \
00170         psubh($17, $14, $17);   /* r17 = tm26 */        \
00171 \
00172         pmulth($14, $2, $18);   /* r18 = x6 * tn2 (6420) */ \
00173         psraw($18, 15, $18);    \
00174         pmfhl_uw($3);           /* r3  = 7531 */        \
00175         psraw($3, 15, $3);      \
00176         pinteh($3, $18, $18);   /* r18 = x6 * tn2 */ \
00177         paddh($18, $10, $18);   /* r18 = tp26 */        \
00178 \
00179         paddh($8, $12, $2);     /* r2  = tp04 */        \
00180         psubh($8, $12, $3);     /* r3  = tm04 */        \
00181 \
00182         paddh($2, $18, $16);    /* r16 = a0 */                  \
00183         psubh($2, $18, $19);    /* r19 = a3 */                  \
00184         psubh($3, $17, $18);    /* r18 = a2 */                  \
00185         paddh($3, $17, $17);    /* r17 = a1 */
00186 
00187 
00188 #define DCT_8_INV_COL8_STORE(blk) \
00189 \
00190         paddh($16, $20, $2);    /* y0  a0+b0 */         \
00191         psubh($16, $20, $16);   /* y7  a0-b0 */         \
00192         psrah($2, SHIFT_INV_COL, $2);           \
00193         psrah($16, SHIFT_INV_COL, $16);         \
00194         sq($2, 0, blk);                         \
00195         sq($16, 112, blk);                      \
00196 \
00197         paddh($17, $21, $3);    /* y1  a1+b1 */         \
00198         psubh($17, $21, $17);   /* y6  a1-b1 */         \
00199         psrah($3, SHIFT_INV_COL, $3);           \
00200         psrah($17, SHIFT_INV_COL, $17);         \
00201         sq($3, 16, blk);                        \
00202         sq($17, 96, blk);                       \
00203 \
00204         paddh($18, $22, $2);    /* y2  a2+b2 */ \
00205         psubh($18, $22, $18);   /* y5  a2-b2 */ \
00206         psrah($2, SHIFT_INV_COL, $2);   \
00207         psrah($18, SHIFT_INV_COL, $18); \
00208         sq($2, 32, blk);                        \
00209         sq($18, 80, blk);                       \
00210 \
00211         paddh($19, $23, $3);    /* y3  a3+b3 */ \
00212         psubh($19, $23, $19);   /* y4  a3-b3 */ \
00213         psrah($3, SHIFT_INV_COL, $3);   \
00214         psrah($19, SHIFT_INV_COL, $19); \
00215         sq($3, 48, blk);                        \
00216         sq($19, 64, blk);
00217 
00218 
00219 
00220 #define DCT_8_INV_COL8_PMS() \
00221         paddh($16, $20, $2);    /* y0  a0+b0 */         \
00222         psubh($16, $20, $20);   /* y7  a0-b0 */         \
00223         psrah($2, SHIFT_INV_COL, $16);          \
00224         psrah($20, SHIFT_INV_COL, $20);         \
00225 \
00226         paddh($17, $21, $3);    /* y1  a1+b1 */         \
00227         psubh($17, $21, $21);   /* y6  a1-b1 */         \
00228         psrah($3, SHIFT_INV_COL, $17);          \
00229         psrah($21, SHIFT_INV_COL, $21);         \
00230 \
00231         paddh($18, $22, $2);    /* y2  a2+b2 */ \
00232         psubh($18, $22, $22);   /* y5  a2-b2 */ \
00233         psrah($2, SHIFT_INV_COL, $18);  \
00234         psrah($22, SHIFT_INV_COL, $22); \
00235 \
00236         paddh($19, $23, $3);    /* y3  a3+b3 */ \
00237         psubh($19, $23, $23);   /* y4  a3-b3 */ \
00238         psrah($3, SHIFT_INV_COL, $19);  \
00239         psrah($23, SHIFT_INV_COL, $23);
00240 
00241 #define PUT(rs) \
00242         pminh(rs, $11, $2);     \
00243         pmaxh($2, $0, $2);      \
00244         ppacb($0, $2, $2); \
00245         sd3(2, 0, 4); \
00246         __asm__ __volatile__ ("add $4, $5, $4");
00247 
00248 #define DCT_8_INV_COL8_PUT() \
00249         PUT($16);               \
00250         PUT($17);               \
00251         PUT($18);               \
00252         PUT($19);               \
00253         PUT($23);               \
00254         PUT($22);               \
00255         PUT($21);               \
00256         PUT($20);
00257 
00258 #define ADD(rs) \
00259     ld3(4, 0, 2); \
00260     pextlb($0, $2, $2); \
00261     paddh($2, rs, $2); \
00262     pminh($2, $11, $2); \
00263     pmaxh($2, $0, $2);  \
00264     ppacb($0, $2, $2); \
00265     sd3(2, 0, 4); \
00266     __asm__ __volatile__ ("add $4, $5, $4");
00267 
00268 /*fixme: schedule*/
00269 #define DCT_8_INV_COL8_ADD() \
00270         ADD($16);               \
00271         ADD($17);               \
00272         ADD($18);               \
00273         ADD($19);               \
00274         ADD($23);               \
00275         ADD($22);               \
00276         ADD($21);               \
00277         ADD($20);
00278 
00279 
00280 void ff_mmi_idct(int16_t * block)
00281 {
00282     /* $4 = block */
00283     __asm__ __volatile__("la $24, %0"::"m"(consttable[0]));
00284     lq($24, ROUNDER_0, $8);
00285     lq($24, ROUNDER_1, $7);
00286     DCT_8_INV_ROW1($4, 0, TAB_i_04, $8, $8);
00287     DCT_8_INV_ROW1($4, 16, TAB_i_17, $7, $9);
00288     DCT_8_INV_ROW1($4, 32, TAB_i_26, $7, $10);
00289     DCT_8_INV_ROW1($4, 48, TAB_i_35, $7, $11);
00290     DCT_8_INV_ROW1($4, 64, TAB_i_04, $7, $12);
00291     DCT_8_INV_ROW1($4, 80, TAB_i_35, $7, $13);
00292     DCT_8_INV_ROW1($4, 96, TAB_i_26, $7, $14);
00293     DCT_8_INV_ROW1($4, 112, TAB_i_17, $7, $15);
00294     DCT_8_INV_COL8();
00295     DCT_8_INV_COL8_STORE($4);
00296  
00297     //let savedtemp regs be saved
00298     __asm__ __volatile__(" ":::"$16", "$17", "$18", "$19", "$20", "$21", "$22", "$23");
00299 }
00300 
00301 
00302 void ff_mmi_idct_put(uint8_t *dest, int line_size, DCTELEM *block)
00303 {
00304     /* $4 = dest, $5 = line_size, $6 = block */
00305     __asm__ __volatile__("la $24, %0"::"m"(consttable[0]));
00306     lq($24, ROUNDER_0, $8);
00307     lq($24, ROUNDER_1, $7);
00308     DCT_8_INV_ROW1($6, 0, TAB_i_04, $8, $8);
00309     DCT_8_INV_ROW1($6, 16, TAB_i_17, $7, $9);
00310     DCT_8_INV_ROW1($6, 32, TAB_i_26, $7, $10);
00311     DCT_8_INV_ROW1($6, 48, TAB_i_35, $7, $11);
00312     DCT_8_INV_ROW1($6, 64, TAB_i_04, $7, $12);
00313     DCT_8_INV_ROW1($6, 80, TAB_i_35, $7, $13);
00314     DCT_8_INV_ROW1($6, 96, TAB_i_26, $7, $14);
00315     DCT_8_INV_ROW1($6, 112, TAB_i_17, $7, $15);
00316     DCT_8_INV_COL8();
00317     lq($24, CLIPMAX, $11);
00318     DCT_8_INV_COL8_PMS();
00319     DCT_8_INV_COL8_PUT();
00320 
00321     //let savedtemp regs be saved
00322     __asm__ __volatile__(" ":::"$16", "$17", "$18", "$19", "$20", "$21", "$22", "$23");
00323 }
00324 
00325 
00326 void ff_mmi_idct_add(uint8_t *dest, int line_size, DCTELEM *block)
00327 {
00328     /* $4 = dest, $5 = line_size, $6 = block */
00329     __asm__ __volatile__("la $24, %0"::"m"(consttable[0]));
00330     lq($24, ROUNDER_0, $8);
00331     lq($24, ROUNDER_1, $7);
00332     DCT_8_INV_ROW1($6, 0, TAB_i_04, $8, $8);
00333     DCT_8_INV_ROW1($6, 16, TAB_i_17, $7, $9);
00334     DCT_8_INV_ROW1($6, 32, TAB_i_26, $7, $10);
00335     DCT_8_INV_ROW1($6, 48, TAB_i_35, $7, $11);
00336     DCT_8_INV_ROW1($6, 64, TAB_i_04, $7, $12);
00337     DCT_8_INV_ROW1($6, 80, TAB_i_35, $7, $13);
00338     DCT_8_INV_ROW1($6, 96, TAB_i_26, $7, $14);
00339     DCT_8_INV_ROW1($6, 112, TAB_i_17, $7, $15);
00340     DCT_8_INV_COL8();
00341     lq($24, CLIPMAX, $11);
00342     DCT_8_INV_COL8_PMS();
00343     DCT_8_INV_COL8_ADD();
00344 
00345     //let savedtemp regs be saved
00346     __asm__ __volatile__(" ":::"$16", "$17", "$18", "$19", "$20", "$21", "$22", "$23");
00347 }
00348
hvirtual/quicktime/ffmpeg/libavcodec/ps2/idct_mmi.c
