credit update
[libav.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 Leon van Stuivenberg
13 */
14 #include "../common.h"
15 #include "../dsputil.h"
16 #include "mmi.h"
17
18 #define BITS_INV_ACC 5 // 4 or 5 for IEEE
19 #define SHIFT_INV_ROW (16 - BITS_INV_ACC)
20 #define SHIFT_INV_COL (1 + BITS_INV_ACC)
21
22 #define TG1 6518
23 #define TG2 13573
24 #define TG3 21895
25 #define CS4 23170
26
27 #define ROUNDER_0 0
28 #define ROUNDER_1 16
29
30 #define TAB_i_04 (32+0)
31 #define TAB_i_17 (32+64)
32 #define TAB_i_26 (32+128)
33 #define TAB_i_35 (32+192)
34
35 #define TG_1_16 (32+256+0)
36 #define TG_2_16 (32+256+16)
37 #define TG_3_16 (32+256+32)
38 #define COS_4_16 (32+256+48)
39
40 #define CLIPMAX (32+256+64+0)
41
42 static short consttable[] align16 = {
43 /* rounder 0*/ // assume SHIFT_INV_ROW == 11
44 0x3ff, 1, 0x3ff, 1, 0x3ff, 1, 0x3ff, 1,
45 /* rounder 1*/
46 0x3ff, 0, 0x3ff, 0, 0x3ff, 0, 0x3ff, 0,
47 /* row 0/4*/
48 16384, 21407, -16384, -21407, 22725, 19266, -22725, -12873,
49 8867, 16384, 8867, 16384, 4520, 12873, -4520, 19266,
50 16384, -8867, 16384, -8867, 12873, -22725, 19266, -22725,
51 21407, -16384, -21407, 16384, 19266, 4520, -12873, 4520,
52 /* row 1/7*/
53 22725, 29692, -22725, -29692, 31521, 26722, -31521, -17855,
54 12299, 22725, 12299, 22725, 6270, 17855, -6270, 26722,
55 22725, -12299, 22725, -12299, 17855, -31521, 26722, -31521,
56 29692, -22725, -29692, 22725, 26722, 6270, -17855, 6270,
57 /* row 2/6*/
58 21407, 27969, -21407, -27969, 29692, 25172, -29692, -16819,
59 11585, 21407, 11585, 21407, 5906, 16819, -5906, 25172,
60 21407, -11585, 21407, -11585, 16819, -29692, 25172, -29692,
61 27969, -21407, -27969, 21407, 25172, 5906, -16819, 5906,
62 /*row 3/5*/
63 19266, 25172, -19266, -25172, 26722, 22654, -26722, -15137,
64 10426, 19266, 10426, 19266, 5315, 15137, -5315, 22654,
65 19266, -10426, 19266, -10426, 15137, -26722, 22654, -26722,
66 25172, -19266, -25172, 19266, 22654, 5315, -15137, 5315,
67 /*column constants*/
68 TG1, TG1, TG1, TG1, TG1, TG1, TG1, TG1,
69 TG2, TG2, TG2, TG2, TG2, TG2, TG2, TG2,
70 TG3, TG3, TG3, TG3, TG3, TG3, TG3, TG3,
71 CS4, CS4, CS4, CS4, CS4, CS4, CS4, CS4,
72 /* clamp */
73 255, 255, 255, 255, 255, 255, 255, 255
74 };
75
76
77 #define DCT_8_INV_ROW1(blk, rowoff, taboff, rnd, outreg) { \
78 lq(blk, rowoff, $16); /* r16 = x7 x5 x3 x1 x6 x4 x2 x0 */ \
79 /*slot*/ \
80 lq($24, 0+taboff, $17); /* r17 = w */ \
81 /*delay slot $16*/ \
82 lq($24, 16+taboff, $18);/* r18 = w */ \
83 prevh($16, $2); /* r2 = x1 x3 x5 x7 x0 x2 x4 x6 */ \
84 lq($24, 32+taboff, $19);/* r19 = w */ \
85 phmadh($17, $16, $17); /* r17 = b1"b0'a1"a0' */ \
86 lq($24, 48+taboff, $20);/* r20 = w */ \
87 phmadh($18, $2, $18); /* r18 = b1'b0"a1'a0" */ \
88 phmadh($19, $16, $19); /* r19 = b3"b2'a3"a2' */ \
89 phmadh($20, $2, $20); /* r20 = b3'b2"a3'a2" */ \
90 paddw($17, $18, $17); /* r17 = (b1)(b0)(a1)(a0) */ \
91 paddw($19, $20, $19); /* r19 = (b3)(b2)(a3)(a2) */ \
92 pcpyld($19, $17, $18); /* r18 = (a3)(a2)(a1)(a0) */ \
93 pcpyud($17, $19, $20); /* r20 = (b3)(b2)(b1)(b0) */ \
94 paddw($18, rnd, $18); /* r18 = (a3)(a2)(a1)(a0) */\
95 paddw($18, $20, $17); /* r17 = ()()()(a0+b0) */ \
96 psubw($18, $20, $20); /* r20 = ()()()(a0-b0) */ \
97 psraw($17, SHIFT_INV_ROW, $17); /* r17 = (y3 y2 y1 y0) */ \
98 psraw($20, SHIFT_INV_ROW, $20); /* r20 = (y4 y5 y6 y7) */ \
99 ppach($20, $17, outreg);/* out = y4 y5 y6 y7 y3 y2 y1 y0 Note order */ \
100 \
101 prevh(outreg, $2); \
102 pcpyud($2, $2, $2); \
103 pcpyld($2, outreg, outreg); \
104 }
105
106
107 #define DCT_8_INV_COL8() \
108 \
109 lq($24, TG_3_16, $2); /* r2 = tn3 */ \
110 \
111 pmulth($11, $2, $17); /* r17 = x3 * tn3 (6420) */ \
112 psraw($17, 15, $17); \
113 pmfhl_uw($3); /* r3 = 7531 */ \
114 psraw($3, 15, $3); \
115 pinteh($3, $17, $17); /* r17 = x3 * tn3 */ \
116 psubh($17, $13, $17); /* r17 = tm35 */ \
117 \
118 pmulth($13, $2, $18); /* r18 = x5 * tn3 (6420) */ \
119 psraw($18, 15, $18); \
120 pmfhl_uw($3); /* r3 = 7531 */ \
121 psraw($3, 15, $3); \
122 pinteh($3, $18, $18); /* r18 = x5 * tn3 */ \
123 paddh($18, $11, $18); /* r18 = tp35 */ \
124 \
125 lq($24, TG_1_16, $2); /* r2 = tn1 */ \
126 \
127 pmulth($15, $2, $19); /* r19 = x7 * tn1 (6420) */ \
128 psraw($19, 15, $19); \
129 pmfhl_uw($3); /* r3 = 7531 */ \
130 psraw($3, 15, $3); \
131 pinteh($3, $19, $19); /* r19 = x7 * tn1 */ \
132 paddh($19, $9, $19); /* r19 = tp17 */ \
133 \
134 pmulth($9, $2, $20); /* r20 = x1 * tn1 (6420) */ \
135 psraw($20, 15, $20); \
136 pmfhl_uw($3); /* r3 = 7531 */ \
137 psraw($3, 15, $3); \
138 pinteh($3, $20, $20); /* r20 = x1 * tn1 */ \
139 psubh($20, $15, $20); /* r20 = tm17 */ \
140 \
141 psubh($19, $18, $3); /* r3 = t1 */ \
142 paddh($20, $17, $16); /* r16 = t2 */ \
143 psubh($20, $17, $23); /* r23 = b3 */ \
144 paddh($19, $18, $20); /* r20 = b0 */ \
145 \
146 lq($24, COS_4_16, $2); /* r2 = cs4 */ \
147 \
148 paddh($3, $16, $21); /* r21 = t1+t2 */ \
149 psubh($3, $16, $22); /* r22 = t1-t2 */ \
150 \
151 pmulth($21, $2, $21); /* r21 = cs4 * (t1+t2) 6420 */ \
152 psraw($21, 15, $21); \
153 pmfhl_uw($3); /* r3 = 7531 */ \
154 psraw($3, 15, $3); \
155 pinteh($3, $21, $21); /* r21 = b1 */ \
156 \
157 pmulth($22, $2, $22); /* r22 = cs4 * (t1-t2) 6420 */ \
158 psraw($22, 15, $22); \
159 pmfhl_uw($3); /* r3 = 7531 */ \
160 psraw($3, 15, $3); \
161 pinteh($3, $22, $22); /* r22 = b2 */ \
162 \
163 lq($24, TG_2_16, $2); /* r2 = tn2 */ \
164 \
165 pmulth($10, $2, $17); /* r17 = x2 * tn2 (6420) */ \
166 psraw($17, 15, $17); \
167 pmfhl_uw($3); /* r3 = 7531 */ \
168 psraw($3, 15, $3); \
169 pinteh($3, $17, $17); /* r17 = x3 * tn3 */ \
170 psubh($17, $14, $17); /* r17 = tm26 */ \
171 \
172 pmulth($14, $2, $18); /* r18 = x6 * tn2 (6420) */ \
173 psraw($18, 15, $18); \
174 pmfhl_uw($3); /* r3 = 7531 */ \
175 psraw($3, 15, $3); \
176 pinteh($3, $18, $18); /* r18 = x6 * tn2 */ \
177 paddh($18, $10, $18); /* r18 = tp26 */ \
178 \
179 paddh($8, $12, $2); /* r2 = tp04 */ \
180 psubh($8, $12, $3); /* r3 = tm04 */ \
181 \
182 paddh($2, $18, $16); /* r16 = a0 */ \
183 psubh($2, $18, $19); /* r19 = a3 */ \
184 psubh($3, $17, $18); /* r18 = a2 */ \
185 paddh($3, $17, $17); /* r17 = a1 */
186
187
188 #define DCT_8_INV_COL8_STORE(blk) \
189 \
190 paddh($16, $20, $2); /* y0 a0+b0 */ \
191 psubh($16, $20, $16); /* y7 a0-b0 */ \
192 psrah($2, SHIFT_INV_COL, $2); \
193 psrah($16, SHIFT_INV_COL, $16); \
194 sq($2, 0, blk); \
195 sq($16, 112, blk); \
196 \
197 paddh($17, $21, $3); /* y1 a1+b1 */ \
198 psubh($17, $21, $17); /* y6 a1-b1 */ \
199 psrah($3, SHIFT_INV_COL, $3); \
200 psrah($17, SHIFT_INV_COL, $17); \
201 sq($3, 16, blk); \
202 sq($17, 96, blk); \
203 \
204 paddh($18, $22, $2); /* y2 a2+b2 */ \
205 psubh($18, $22, $18); /* y5 a2-b2 */ \
206 psrah($2, SHIFT_INV_COL, $2); \
207 psrah($18, SHIFT_INV_COL, $18); \
208 sq($2, 32, blk); \
209 sq($18, 80, blk); \
210 \
211 paddh($19, $23, $3); /* y3 a3+b3 */ \
212 psubh($19, $23, $19); /* y4 a3-b3 */ \
213 psrah($3, SHIFT_INV_COL, $3); \
214 psrah($19, SHIFT_INV_COL, $19); \
215 sq($3, 48, blk); \
216 sq($19, 64, blk);
217
218
219
220 #define DCT_8_INV_COL8_PMS() \
221 paddh($16, $20, $2); /* y0 a0+b0 */ \
222 psubh($16, $20, $20); /* y7 a0-b0 */ \
223 psrah($2, SHIFT_INV_COL, $16); \
224 psrah($20, SHIFT_INV_COL, $20); \
225 \
226 paddh($17, $21, $3); /* y1 a1+b1 */ \
227 psubh($17, $21, $21); /* y6 a1-b1 */ \
228 psrah($3, SHIFT_INV_COL, $17); \
229 psrah($21, SHIFT_INV_COL, $21); \
230 \
231 paddh($18, $22, $2); /* y2 a2+b2 */ \
232 psubh($18, $22, $22); /* y5 a2-b2 */ \
233 psrah($2, SHIFT_INV_COL, $18); \
234 psrah($22, SHIFT_INV_COL, $22); \
235 \
236 paddh($19, $23, $3); /* y3 a3+b3 */ \
237 psubh($19, $23, $23); /* y4 a3-b3 */ \
238 psrah($3, SHIFT_INV_COL, $19); \
239 psrah($23, SHIFT_INV_COL, $23);
240
241 #define PUT(rs) \
242 pminh(rs, $11, $2); \
243 pmaxh($2, $0, $2); \
244 ppacb($0, $2, $2); \
245 sd3(2, 0, 4); \
246 __asm__ __volatile__ ("add $4, $5, $4");
247
248 #define DCT_8_INV_COL8_PUT() \
249 PUT($16); \
250 PUT($17); \
251 PUT($18); \
252 PUT($19); \
253 PUT($23); \
254 PUT($22); \
255 PUT($21); \
256 PUT($20);
257
258 #define ADD(rs) \
259 ld3(4, 0, 2); \
260 pextlb($0, $2, $2); \
261 paddh($2, rs, $2); \
262 pminh($2, $11, $2); \
263 pmaxh($2, $0, $2); \
264 ppacb($0, $2, $2); \
265 sd3(2, 0, 4); \
266 __asm__ __volatile__ ("add $4, $5, $4");
267
268 /*fixme: schedule*/
269 #define DCT_8_INV_COL8_ADD() \
270 ADD($16); \
271 ADD($17); \
272 ADD($18); \
273 ADD($19); \
274 ADD($23); \
275 ADD($22); \
276 ADD($21); \
277 ADD($20);
278
279
280 void ff_mmi_idct(int16_t * block)
281 {
282 /* $4 = block */
283 __asm__ __volatile__("la $24, %0"::"m"(consttable[0]));
284 lq($24, ROUNDER_0, $8);
285 lq($24, ROUNDER_1, $7);
286 DCT_8_INV_ROW1($4, 0, TAB_i_04, $8, $8);
287 DCT_8_INV_ROW1($4, 16, TAB_i_17, $7, $9);
288 DCT_8_INV_ROW1($4, 32, TAB_i_26, $7, $10);
289 DCT_8_INV_ROW1($4, 48, TAB_i_35, $7, $11);
290 DCT_8_INV_ROW1($4, 64, TAB_i_04, $7, $12);
291 DCT_8_INV_ROW1($4, 80, TAB_i_35, $7, $13);
292 DCT_8_INV_ROW1($4, 96, TAB_i_26, $7, $14);
293 DCT_8_INV_ROW1($4, 112, TAB_i_17, $7, $15);
294 DCT_8_INV_COL8();
295 DCT_8_INV_COL8_STORE($4);
296
297 //let savedtemp regs be saved
298 __asm__ __volatile__(" ":::"$16", "$17", "$18", "$19", "$20", "$21", "$22", "$23");
299 }
300
301
302 void ff_mmi_idct_put(uint8_t *dest, int line_size, DCTELEM *block)
303 {
304 /* $4 = dest, $5 = line_size, $6 = block */
305 __asm__ __volatile__("la $24, %0"::"m"(consttable[0]));
306 lq($24, ROUNDER_0, $8);
307 lq($24, ROUNDER_1, $7);
308 DCT_8_INV_ROW1($6, 0, TAB_i_04, $8, $8);
309 DCT_8_INV_ROW1($6, 16, TAB_i_17, $7, $9);
310 DCT_8_INV_ROW1($6, 32, TAB_i_26, $7, $10);
311 DCT_8_INV_ROW1($6, 48, TAB_i_35, $7, $11);
312 DCT_8_INV_ROW1($6, 64, TAB_i_04, $7, $12);
313 DCT_8_INV_ROW1($6, 80, TAB_i_35, $7, $13);
314 DCT_8_INV_ROW1($6, 96, TAB_i_26, $7, $14);
315 DCT_8_INV_ROW1($6, 112, TAB_i_17, $7, $15);
316 DCT_8_INV_COL8();
317 lq($24, CLIPMAX, $11);
318 DCT_8_INV_COL8_PMS();
319 DCT_8_INV_COL8_PUT();
320
321 //let savedtemp regs be saved
322 __asm__ __volatile__(" ":::"$16", "$17", "$18", "$19", "$20", "$21", "$22", "$23");
323 }
324
325
326 void ff_mmi_idct_add(uint8_t *dest, int line_size, DCTELEM *block)
327 {
328 /* $4 = dest, $5 = line_size, $6 = block */
329 __asm__ __volatile__("la $24, %0"::"m"(consttable[0]));
330 lq($24, ROUNDER_0, $8);
331 lq($24, ROUNDER_1, $7);
332 DCT_8_INV_ROW1($6, 0, TAB_i_04, $8, $8);
333 DCT_8_INV_ROW1($6, 16, TAB_i_17, $7, $9);
334 DCT_8_INV_ROW1($6, 32, TAB_i_26, $7, $10);
335 DCT_8_INV_ROW1($6, 48, TAB_i_35, $7, $11);
336 DCT_8_INV_ROW1($6, 64, TAB_i_04, $7, $12);
337 DCT_8_INV_ROW1($6, 80, TAB_i_35, $7, $13);
338 DCT_8_INV_ROW1($6, 96, TAB_i_26, $7, $14);
339 DCT_8_INV_ROW1($6, 112, TAB_i_17, $7, $15);
340 DCT_8_INV_COL8();
341 lq($24, CLIPMAX, $11);
342 DCT_8_INV_COL8_PMS();
343 DCT_8_INV_COL8_ADD();
344
345 //let savedtemp regs be saved
346 __asm__ __volatile__(" ":::"$16", "$17", "$18", "$19", "$20", "$21", "$22", "$23");
347 }
348