ps2 idct bugfix patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
[ffmpeg.git] / libavcodec / ps2 / idct_mmi.c
1 /*
2   Originally provided by Intel at AP-922
3   http://developer.intel.com/vtune/cbts/strmsimd/922down.htm
4   (See more app notes at http://developer.intel.com/vtune/cbts/strmsimd/appnotes.htm)
5   but in a limited edition.
6
7   column code adapted from peter gubanov
8   Copyright (c) 2000-2001 Peter Gubanov <peter@elecard.net.ru>
9   http://www.elecard.com/peter/idct.shtml
10   Rounding trick Copyright (c) 2000 Michel Lespinasse <walken@zoy.org>
11
12   mmi port by leonvs@iae.nl
13 */  
14 #include "../common.h"
15 #include "mmi.h"
16
17 #define BITS_INV_ACC    5       // 4 or 5 for IEEE
18 #define SHIFT_INV_ROW   (16 - BITS_INV_ACC)
19 #define SHIFT_INV_COL   (1 + BITS_INV_ACC)
20
21 // assume SHIFT_INV_ROW == 11
22 static int roundertable[8][4] align16 = {
23     { 0x103ff, 0x103ff, 0x103ff, 0x103ff },
24     { 1023,  1023,  1023,  1023}
25 };
26
27
28 #define TAB_i_04        0
29 #define TAB_i_17        64
30 #define TAB_i_26        128
31 #define TAB_i_35        192
32
33 static short rowtable[4][32] align16 = {
34     {
35      16384, 16384, 22725, 12873, 21407, 8867, 19266, 4520,
36      16384, -16383, 19266, -22724, 8867, -21406, -4519, -12872,
37      16384, -16383, 12873, 4520, -8866, 21407, -22724, 19266,
38      16384, 16384, 4520, 19266, -21406, -8866, -12872, -22724},
39     {
40      22725, 22725, 31521, 17855, 29692, 12299, 26722, 6270,
41      22725, -22724, 26722, -31520, 12299, -29691, -6269, -17854,
42      22725, -22724, 17855, 6270, -12298, 29692, -31520, 26722,
43      22725, 22725, 6270, 26722, -29691, -12298, -17854, -31520},
44     {
45      21407, 21407, 29692, 16819, 27969, 11585, 25172, 5906,
46      21407, -21406, 25172, -29691, 11585, -27968, -5905, -16818,
47      21407, -21406, 16819, 5906, -11584, 27969, -29691, 25172,
48      21407, 21407, 5906, 25172, -27968, -11584, -16818, -29691},
49     {
50      19266, 19266, 26722, 15137, 25172, 10426, 22654, 5315,
51      19266, -19265, 22654, -26721, 10426, -25171, -5314, -15136,
52      19266, -19265, 15137, 5315, -10425, 25172, -26721, 22654,
53      19266, 19266, 5315, 22654, -25171, -10425, -15136, -26721}
54 };
55
56
57 #define TG_3_16_minus_one       0
58 #define ONE_plus_tg_3_16        16
59 #define ONE_plus_tg_1_16        32
60 #define TG_1_16_minus_one       48
61 #define TG_2_16_minus_one       64
62 #define ONE_plus_tg_2_16        80
63 #define ZERO_ocos_4_16          96
64
65 #define TG1     6518
66 #define TG2     13573
67 #define TG3     21895
68 #define MN1     -32768
69 #define PL1     32768
70 #define CS4     23170
71
72 static short coltable[7][8] align16 = {
73     { MN1,  TG3,  MN1,  TG3,  MN1,  TG3,  MN1,  TG3},
74     {-TG3, -PL1, -TG3, -PL1, -TG3, -PL1, -TG3, -PL1},
75     {-TG1, -PL1, -TG1, -PL1, -TG1, -PL1, -TG1, -PL1},
76     { MN1,  TG1,  MN1,  TG1,  MN1,  TG1,  MN1,  TG1},
77     { MN1,  TG2,  MN1,  TG2,  MN1,  TG2,  MN1,  TG2},
78     {-TG2, -PL1, -TG2, -PL1, -TG2, -PL1, -TG2, -PL1},
79     { CS4,    0,  CS4,    0,  CS4,    0,  CS4,    0}
80 };
81
82 #define noprevh(rt, rd)
83
84
85 #define DCT_8_INV_ROW1(rowoff, taboff, rnd, outreg) { \
86 \
87         lq($4, rowoff, $16);    /* r16 = x7  x6  x5  x4  x3  x2  x1  x0 */ \
88         lq($24, 0+taboff, $17); /* r17 = w19 w17 w3  w1  w18 w16 w2  w0 */ \
89         pinth($16, $16, $16);   /* r16 = x7  x3  x6  x2  x5  x1  x4  x0 */ \
90         phmadh($17, $16, $17);  /* r17 = (x7*w19+x3*w17)b0'' (x6*w3+x2*w1)a0'' (x5*w18+x1*w16)b0' (x4*w2+x0*w0)a0' */ \
91         lq($24, 16+taboff, $18);/* r18 = w23 w21 w7  w5  w22 w20 w6  w4 */ \
92         lq($24, 32+taboff, $19);/* r19 = w27 w25 w11 w9  w26 w24 w10 w8 */ \
93         lq($24, 48+taboff, $20);/* r20 = w31 w29 w15 w13 w30 w28 w14 w12 */ \
94         phmadh($18, $16, $18);  /* r18 = (b1'')(a1'')(b1')(a1') */ \
95         pcpyud($17, $17, $21);  /* r21 = (b0'')(a0'')(b0'')(a0'') */ \
96         paddw($17, $21, $17);   /* r17 = (--)(--)(b0)(a0) */ \
97         phmadh($19, $16, $19);  /* r19 = (b2'')(a2'')(b2')(a2') */ \
98         pcpyud($18, $18, $21);  /* r21 = (b1'')(a1'')(b1'')(a1'') */ \
99         paddw($18, $21, $18);   /* r18 = (--)(--)(b1)(a1) */ \
100         pcpyud($19, $19, $21);  \
101         phmadh($20, $16, $20);  /* r12 = (b3'')(a3'')(b3')(a3') */ \
102         paddw($19, $21, $19);   /* r19 = (--)(--)(b2)(a2) */ \
103         pextlw($19, $17, $16);  /* r16 = (b2)(b0)(a2)(a0) */ \
104         pcpyud($20, $20, $21);  \
105         paddw($20, $21, $20);   /* r20 = (--)(--)(b3)(a3) */ \
106         pextlw($20, $18, $17);  /* r17 = (b3)(b1)(a3)(a1) */ \
107         pextlw($17, $16, $20);  /* r20 = (a3)(a2)(a1)(a0)" */ \
108         pextuw($17, $16, $21);  /* r21 = (b3)(b2)(b1)(b0) */ \
109         paddw($20, rnd, $20);   /* r20 = (a3)(a2)(a1)(a0) */\
110         paddw($20, $21, $17);   /* r17 = ()()()(a0+b0) */ \
111         psubw($20, $21, $18);   /* r18 = ()()()(a0-b0) */ \
112         psraw($17, SHIFT_INV_ROW, $17); /* r17 = (y3 y2 y1 y0) */ \
113         psraw($18, SHIFT_INV_ROW, $18); /* r18 = (y4 y5 y6 y7) */ \
114         ppach($18, $17, outreg);/* out = y4 y5 y6 y7 y3 y2 y1 y0  Note order */ \
115 }
116
117
118 #define DCT_8_INV_COL4(pextop, blkoff, revop) { \
119         lq($24, TG_3_16_minus_one, $2); /* r2  = (tn3)(-1) x 4 */       \
120         pextop($11, $13, $3);           /* r3  = (x3)(x5) x 4 */        \
121         lq($24, ONE_plus_tg_3_16, $16); /* r16 = -((+1)(tn3)) x 4 */    \
122         phmadh($3, $2, $17);            /* r17 = (tm35) x 4 */          \
123         lq($24, ONE_plus_tg_1_16, $2);  /* r2  = -((+1)(tn1)) x 4 */    \
124         phmadh($3, $16, $18);           /* r18 = -(tp35) x 4 */         \
125         lq($24, TG_1_16_minus_one, $16);/* r16 = (tn1)(-1) x 4 */       \
126         pextop($9, $15, $3);            /* r3  = (x1)(x7) x 4 */        \
127         phmadh($3, $2, $19);                                            \
128         lq($24, ZERO_ocos_4_16, $2);    /* r2  = (0)(cos4) x 4 */       \
129         phmadh($3, $16, $20);           /* r20 = (tm17) x 4 */          \
130         psubw($0, $19, $19);            /* r19 = (tp17) x 4 */          \
131         paddw($19, $18, $3);            /* r3  = t1 */                  \
132         paddw($20, $17, $16);           /* r16 = t2 */                  \
133         psubw($20, $17, $23);           /* r23 = b3 */                  \
134         psubw($19, $18, $20);           /* r20 = b0 */                  \
135         paddw($3, $16, $17);            /* (t1+t2) */                   \
136         psubw($3, $16, $18);            /* (t1-t2) */                   \
137         psraw($17, 15, $17);                                            \
138         lq($24, TG_2_16_minus_one, $3); /* r3  = (tn2)(-1) x 4 */       \
139         pmulth($17, $2, $21);           /* r21 = b1 */                  \
140         psraw($18, 15, $18);                                            \
141         lq($24, ONE_plus_tg_2_16, $16); /* r16 = -((+1)(tn2)) x 4 */    \
142         pmulth($18, $2, $22);           /* r22 = b2 */                  \
143 \
144         pextop($10, $14, $2);           /* r2  = (x2)(x6) x 4 */        \
145         phmadh($2, $3, $18);            /* r18 = (tm26) x 4 */          \
146         phmadh($2, $16, $19);           /* r19 = -(tp26) x 4 */         \
147         pextop($8, $0, $17);            /* r17 = (x0)(0) x 4 */         \
148         psraw($17, 1, $17);                                             \
149         pextop($12, $0, $16);           /* r16 = (x4)(0) x 4 */         \
150         psraw($16, 1, $16);                                             \
151         paddw($17, $16, $2);            /* r2  = tp04 */                \
152         psubw($17, $16, $3);            /* r3  = tm04 */                \
153         psubw($2, $19, $16);            /* r16 = a0 */                  \
154         paddw($3, $18, $17);            /* r17 = a1 */                  \
155         psubw($3, $18, $18);            /* r18 = a2 */                  \
156         paddw($2, $19, $19);            /* r19 = a3 */                  \
157 \
158         paddw($16, $20, $2);    /* y0  a0+b0 */         \
159         psubw($16, $20, $16);   /* y7  a0-b0 */         \
160         psraw($2, SHIFT_INV_COL+15, $2);                \
161         psraw($16, SHIFT_INV_COL+15, $16);              \
162         ppach($0, $2, $2);                              \
163         ppach($0, $16, $16);                            \
164         revop($2, $2);                                  \
165         revop($16, $16);                                \
166         sd3(2, 0+blkoff, 4);                            \
167         sd3(16, 112+blkoff, 4);                         \
168 \
169         paddw($17, $21, $3);    /* y1  a1+b1 */         \
170         psubw($17, $21, $17);   /* y6  a1-b1 */         \
171         psraw($3, SHIFT_INV_COL+15, $3);                \
172         psraw($17, SHIFT_INV_COL+15, $17);              \
173         ppach($0, $3, $3);                              \
174         ppach($0, $17, $17);                    \
175         revop($3, $3);                          \
176         revop($17, $17);                        \
177         sd3(3, 16+blkoff, 4);                   \
178         sd3(17, 96+blkoff, 4);                  \
179 \
180         paddw($18, $22, $2);    /* y2  a2+b2 */ \
181         psubw($18, $22, $18);   /* y5  a2-b2 */ \
182         psraw($2, SHIFT_INV_COL+15, $2);        \
183         psraw($18, SHIFT_INV_COL+15, $18);      \
184         ppach($0, $2, $2);                      \
185         ppach($0, $18, $18);                    \
186         revop($2, $2);                          \
187         revop($18, $18);                        \
188         sd3(2, 32+blkoff, 4);                   \
189         sd3(18, 80+blkoff, 4);                  \
190 \
191         paddw($19, $23, $3);    /* y3  a3+b3 */ \
192         psubw($19, $23, $19);   /* y4  a3-b3 */ \
193         psraw($3, SHIFT_INV_COL+15, $3);        \
194         psraw($19, SHIFT_INV_COL+15, $19);      \
195         ppach($0, $3, $3);                      \
196         ppach($0, $19, $19);                    \
197         revop($3, $3);                          \
198         revop($19, $19);                        \
199         sd3(3, 48+blkoff, 4);                   \
200         sd3(19, 64+blkoff, 4);                  \
201 }
202
203
204 void ff_mmi_idct(int16_t * block)
205 {
206     /* $4 = block */
207     __asm__ __volatile__("la $7, %0"::"m"(roundertable[0][0]));
208     __asm__ __volatile__("la $24, %0"::"m"(rowtable[0][0]));
209     lq($7, 0, $6);
210     lq($7, 16, $7);
211     DCT_8_INV_ROW1(0, TAB_i_04, $6, $8);
212     DCT_8_INV_ROW1(16, TAB_i_17, $7, $9);
213     DCT_8_INV_ROW1(32, TAB_i_26, $7, $10);
214     DCT_8_INV_ROW1(48, TAB_i_35, $7, $11);
215     DCT_8_INV_ROW1(64, TAB_i_04, $7, $12);
216     DCT_8_INV_ROW1(80, TAB_i_35, $7, $13);
217     DCT_8_INV_ROW1(96, TAB_i_26, $7, $14);
218     DCT_8_INV_ROW1(112, TAB_i_17, $7, $15);
219
220     __asm__ __volatile__("la $24, %0"::"m"(coltable[0][0]));
221     DCT_8_INV_COL4(pextlh, 0, noprevh);
222     DCT_8_INV_COL4(pextuh, 8, prevh);
223
224     //let savedtemp regs be saved
225     __asm__ __volatile__(" ":::"$16", "$17", "$18", "$19", "$20", "$21",
226                          "$22", "$23", "$6", "$7");
227 }
228