1 /*
2 * Loongson SIMD optimized simple idct
3 *
4 * Copyright (c) 2015 Loongson Technology Corporation Limited
5 * Copyright (c) 2015 Zhou Xiaoyong <zhouxiaoyong@loongson.cn>
6 * Zhang Shuangshuang <zhangshuangshuang@ict.ac.cn>
7 *
8 * This file is part of FFmpeg.
9 *
10 * FFmpeg is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU Lesser General Public
12 * License as published by the Free Software Foundation; either
13 * version 2.1 of the License, or (at your option) any later version.
14 *
15 * FFmpeg is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * Lesser General Public License for more details.
19 *
20 * You should have received a copy of the GNU Lesser General Public
21 * License along with FFmpeg; if not, write to the Free Software
22 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23 */
24
25 #include "idctdsp_mips.h"
26 #include "constants.h"
27 #include "libavutil/mips/asmdefs.h"
28 #include "libavutil/mips/mmiutils.h"
29 #include "libavutil/mem_internal.h"
30
31 #define W1 22725 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
32 #define W2 21407 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
33 #define W3 19266 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
34 #define W4 16383 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
35 #define W5 12873 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
36 #define W6 8867 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
37 #define W7 4520 //cos(i*M_PI/16)*sqrt(2)*(1<<14) + 0.5
38
39 #define ROW_SHIFT 11
40 #define COL_SHIFT 20
41 #define DC_SHIFT 3
42
43 DECLARE_ALIGNED(16, const int16_t, W_arr)[46] = {
44 W4, W2, W4, W6,
45 W1, W3, W5, W7,
46 W4, W6, -W4, -W2,
47 W3, -W7, -W1, -W5,
48 W4, -W6, -W4, W2,
49 W5, -W1, W7, W3,
50 W4, -W2, W4, -W6,
51 W7, -W5, W3, -W1,
52 1024, 0, 1024, 0, //ff_p32_1024 = 0x0000040000000400ULL
53 0, -1, -1, -1, //mask = 0xffffffffffff0000ULL
54 32, 32, 32, 32 //ff_p16_32 = 0x0020002000200020ULL
55 };
56
ff_simple_idct_8_mmi(int16_t * block)57 void ff_simple_idct_8_mmi(int16_t *block)
58 {
59 DECLARE_VAR_ALL64;
60
61 BACKUP_REG
62 __asm__ volatile (
63
64 #define IDCT_ROW_COND_DC(src1, src2) \
65 "dmfc1 $11, "#src1" \n\t" \
66 "dmfc1 $12, "#src2" \n\t" \
67 "and $11, $11, $9 \n\t" \
68 "or $10, $11, $12 \n\t" \
69 "beqz $10, 1f \n\t" \
70 \
71 "punpcklhw $f30, "#src1", "#src2" \n\t" \
72 "punpckhhw $f31, "#src1", "#src2" \n\t" \
73 /* s6, s4, s2, s0 */ \
74 "punpcklhw "#src1", $f30, $f31 \n\t" \
75 /* s7, s5, s3, s1 */ \
76 "punpckhhw "#src2", $f30, $f31 \n\t" \
77 \
78 "pmaddhw $f30, "#src1", $f18 \n\t" \
79 "pmaddhw $f31, "#src2", $f19 \n\t" \
80 "paddw $f28, $f30, $f31 \n\t" \
81 "psubw $f29, $f30, $f31 \n\t" \
82 "punpcklwd $f30, $f28, $f29 \n\t" \
83 "punpckhwd $f31, $f28, $f29 \n\t" \
84 "paddw $f26, $f30, $f31 \n\t" \
85 "paddw $f26, $f26, $f16 \n\t" \
86 /* $f26: src[7], src[0] */ \
87 "psraw $f26, $f26, $f17 \n\t" \
88 \
89 "pmaddhw $f30, "#src1", $f20 \n\t" \
90 "pmaddhw $f31, "#src2", $f21 \n\t" \
91 "paddw $f28, $f30, $f31 \n\t" \
92 "psubw $f29, $f30, $f31 \n\t" \
93 "punpcklwd $f30, $f28, $f29 \n\t" \
94 "punpckhwd $f31, $f28, $f29 \n\t" \
95 "paddw $f27, $f30, $f31 \n\t" \
96 "paddw $f27, $f27, $f16 \n\t" \
97 /* $f27: src[6], src[1] */ \
98 "psraw $f27, $f27, $f17 \n\t" \
99 \
100 "pmaddhw $f30, "#src1", $f22 \n\t" \
101 "pmaddhw $f31, "#src2", $f23 \n\t" \
102 "paddw $f28, $f30, $f31 \n\t" \
103 "psubw $f29, $f30, $f31 \n\t" \
104 "punpcklwd $f30, $f28, $f29 \n\t" \
105 "punpckhwd $f31, $f28, $f29 \n\t" \
106 "paddw $f28, $f30, $f31 \n\t" \
107 "paddw $f28, $f28, $f16 \n\t" \
108 /* $f28: src[5], src[2] */ \
109 "psraw $f28, $f28, $f17 \n\t" \
110 \
111 "pmaddhw $f30, "#src1", $f24 \n\t" \
112 "pmaddhw $f31, "#src2", $f25 \n\t" \
113 "paddw "#src1", $f30, $f31 \n\t" \
114 "psubw "#src2", $f30, $f31 \n\t" \
115 "punpcklwd $f30, "#src1", "#src2" \n\t" \
116 "punpckhwd $f31, "#src1", "#src2" \n\t" \
117 "paddw $f29, $f30, $f31 \n\t" \
118 "paddw $f29, $f29, $f16 \n\t" \
119 /* $f29: src[4], src[3] */ \
120 "psraw $f29, $f29, $f17 \n\t" \
121 \
122 "punpcklhw "#src1", $f26, $f27 \n\t" \
123 "punpckhhw $f30, $f27, $f26 \n\t" \
124 "punpcklhw $f31, $f28, $f29 \n\t" \
125 "punpckhhw "#src2", $f29, $f28 \n\t" \
126 /* src[3], src[2], src[1], src[0] */ \
127 "punpcklwd "#src1", "#src1", $f31 \n\t" \
128 /* src[7], src[6], src[5], src[4] */ \
129 "punpcklwd "#src2", "#src2", $f30 \n\t" \
130 "j 2f \n\t" \
131 \
132 "1: \n\t" \
133 "li $10, 3 \n\t" \
134 "dmtc1 $10, $f30 \n\t" \
135 "psllh $f28, "#src1", $f30 \n\t" \
136 "dmtc1 $9, $f31 \n\t" \
137 "punpcklhw $f29, $f28, $f28 \n\t" \
138 "pand $f29, $f29, $f31 \n\t" \
139 "paddw $f28, $f28, $f29 \n\t" \
140 "punpcklwd "#src1", $f28, $f28 \n\t" \
141 "punpcklwd "#src2", $f28, $f28 \n\t" \
142 "2: \n\t" \
143
144 /* idctRowCondDC row0~8 */
145
146 /* load W */
147 MMI_LQC1($f19, $f18, %[w_arr], 0x00)
148 MMI_LQC1($f21, $f20, %[w_arr], 0x10)
149 MMI_LQC1($f23, $f22, %[w_arr], 0x20)
150 MMI_LQC1($f25, $f24, %[w_arr], 0x30)
151 MMI_LQC1($f17, $f16, %[w_arr], 0x40)
152 /* load source in block */
153 MMI_LQC1($f1, $f0, %[block], 0x00)
154 MMI_LQC1($f3, $f2, %[block], 0x10)
155 MMI_LQC1($f5, $f4, %[block], 0x20)
156 MMI_LQC1($f7, $f6, %[block], 0x30)
157 MMI_LQC1($f9, $f8, %[block], 0x40)
158 MMI_LQC1($f11, $f10, %[block], 0x50)
159 MMI_LQC1($f13, $f12, %[block], 0x60)
160 MMI_LQC1($f15, $f14, %[block], 0x70)
161
162 /* $9: mask ; $f17: ROW_SHIFT */
163 "dmfc1 $9, $f17 \n\t"
164 "li $10, 11 \n\t"
165 "mtc1 $10, $f17 \n\t"
166 IDCT_ROW_COND_DC($f0,$f1)
167 IDCT_ROW_COND_DC($f2,$f3)
168 IDCT_ROW_COND_DC($f4,$f5)
169 IDCT_ROW_COND_DC($f6,$f7)
170 IDCT_ROW_COND_DC($f8,$f9)
171 IDCT_ROW_COND_DC($f10,$f11)
172 IDCT_ROW_COND_DC($f12,$f13)
173 IDCT_ROW_COND_DC($f14,$f15)
174
175 #define IDCT_COL_CASE1(src, out1, out2) \
176 "pmaddhw $f26, "#src", $f18 \n\t" \
177 "pmaddhw $f27, "#src", $f20 \n\t" \
178 "pmaddhw $f28, "#src", $f22 \n\t" \
179 "pmaddhw $f29, "#src", $f24 \n\t" \
180 \
181 "punpcklwd $f30, $f26, $f26 \n\t" \
182 "punpckhwd $f31, $f26, $f26 \n\t" \
183 /* $f26: src[0], src[56] */ \
184 "paddw $f26, $f30, $f31 \n\t" \
185 "punpcklwd $f30, $f27, $f27 \n\t" \
186 "punpckhwd $f31, $f27, $f27 \n\t" \
187 /* $f27: src[8], src[48] */ \
188 "paddw $f27, $f30, $f31 \n\t" \
189 "punpcklwd $f30, $f28, $f28 \n\t" \
190 "punpckhwd $f31, $f28, $f28 \n\t" \
191 /* $f28: src[16], src[40] */ \
192 "paddw $f28, $f30, $f31 \n\t" \
193 "punpcklwd $f30, $f29, $f29 \n\t" \
194 "punpckhwd $f31, $f29, $f29 \n\t" \
195 /* $f29: src[24], src[32] */ \
196 "paddw $f29, $f30, $f31 \n\t" \
197 \
198 /* out1: src[24], src[16], src[8], src[0] */ \
199 /* out2: src[56], src[48], src[40], src[32] */ \
200 "punpcklhw $f30, $f26, $f27 \n\t" \
201 "punpcklhw $f31, $f28, $f29 \n\t" \
202 "punpckhwd "#out1", $f30, $f31 \n\t" \
203 "psrah "#out1", "#out1", $f16 \n\t" \
204 "punpcklhw $f30, $f27, $f26 \n\t" \
205 "punpcklhw $f31, $f29, $f28 \n\t" \
206 "punpckhwd "#out2", $f31, $f30 \n\t" \
207 "psrah "#out2", "#out2", $f16 \n\t"
208
209 #define IDCT_COL_CASE2(src1, src2, out1, out2) \
210 "pmaddhw $f28, "#src1", $f18 \n\t" \
211 "pmaddhw $f29, "#src2", $f19 \n\t" \
212 "paddw $f30, $f28, $f29 \n\t" \
213 "psubw $f31, $f28, $f29 \n\t" \
214 "punpcklwd $f28, $f30, $f31 \n\t" \
215 "punpckhwd $f29, $f30, $f31 \n\t" \
216 "pmaddhw $f30, "#src1", $f20 \n\t" \
217 "pmaddhw $f31, "#src2", $f21 \n\t" \
218 /* $f26: src[0], src[56] */ \
219 "paddw $f26, $f28, $f29 \n\t" \
220 "paddw $f28, $f30, $f31 \n\t" \
221 "psubw $f29, $f30, $f31 \n\t" \
222 "punpcklwd $f30, $f28, $f29 \n\t" \
223 "punpckhwd $f31, $f28, $f29 \n\t" \
224 "pmaddhw $f28, "#src1", $f22 \n\t" \
225 "pmaddhw $f29, "#src2", $f23 \n\t" \
226 /* $f27: src[8], src[48] */ \
227 "paddw $f27, $f30, $f31 \n\t" \
228 "paddw $f30, $f28, $f29 \n\t" \
229 "psubw $f31, $f28, $f29 \n\t" \
230 "punpcklwd $f28, $f30, $f31 \n\t" \
231 "punpckhwd $f29, $f30, $f31 \n\t" \
232 "pmaddhw $f30, "#src1", $f24 \n\t" \
233 "pmaddhw $f31, "#src2", $f25 \n\t" \
234 /* $f28: src[16], src[40] */ \
235 "paddw $f28, $f28, $f29 \n\t" \
236 "paddw "#out1", $f30, $f31 \n\t" \
237 "psubw "#out2", $f30, $f31 \n\t" \
238 "punpcklwd $f30, "#out1", "#out2" \n\t" \
239 "punpckhwd $f31, "#out1", "#out2" \n\t" \
240 /* $f29: src[24], src[32] */ \
241 "paddw $f29, $f30, $f31 \n\t" \
242 \
243 /* out1: src[24], src[16], src[8], src[0] */ \
244 /* out2: src[56], src[48], src[40], src[32] */ \
245 "punpcklhw "#out1", $f26, $f27 \n\t" \
246 "punpckhhw "#out2", $f27, $f26 \n\t" \
247 "punpcklhw $f30, $f28, $f29 \n\t" \
248 "punpckhhw $f31, $f29, $f28 \n\t" \
249 "punpckhwd "#out1", "#out1", $f30 \n\t" \
250 "punpckhwd "#out2", $f31, "#out2" \n\t" \
251 "psrah "#out1", "#out1", $f16 \n\t" \
252 "psrah "#out2", "#out2", $f16 \n\t"
253
254
255 /* idctSparseCol col0~3 */
256
257 /* $f17: ff_p16_32; $f16: COL_SHIFT-16 */
258 MMI_ULDC1($f17, %[w_arr], 0x50)
259 "li $10, 4 \n\t"
260 "dmtc1 $10, $f16 \n\t"
261 "paddh $f0, $f0, $f17 \n\t"
262 /* Transpose row[0,2,4,6] */
263 "punpcklhw $f26, $f0, $f4 \n\t"
264 "punpckhhw $f27, $f0, $f4 \n\t"
265 "punpcklhw $f28, $f8, $f12 \n\t"
266 "punpckhhw $f29, $f8, $f12 \n\t"
267 "punpcklwd $f0, $f26, $f28 \n\t"
268 "punpckhwd $f4, $f26, $f28 \n\t"
269 "punpcklwd $f8, $f27, $f29 \n\t"
270 "punpckhwd $f12, $f27, $f29 \n\t"
271
272 "por $f26, $f2, $f6 \n\t"
273 "por $f26, $f26, $f10 \n\t"
274 "por $f26, $f26, $f14 \n\t"
275 "dmfc1 $10, $f26 \n\t"
276 "bnez $10, 1f \n\t"
277 /* case1: In this case, row[1,3,5,7] are all zero */
278 /* col0: $f0: col[24,16,8,0]; $f2: col[56,48,40,32] */
279 IDCT_COL_CASE1($f0, $f0, $f2)
280 /* col1: $f4: col[25,17,9,1]; $f6: col[57,49,41,33] */
281 IDCT_COL_CASE1($f4, $f4, $f6)
282 /* col2: $f8: col[26,18,10,2]; $f10: col[58,50,42,34] */
283 IDCT_COL_CASE1($f8, $f8, $f10)
284 /* col3: $f12: col[27,19,11,3]; $f14: col[59,51,43,35] */
285 IDCT_COL_CASE1($f12, $f12, $f14)
286 "j 2f \n\t"
287
288 "1: \n\t"
289 /* case2: row[1,3,5,7] are not all zero */
290 /* Transpose */
291 "punpcklhw $f26, $f2, $f6 \n\t"
292 "punpckhhw $f27, $f2, $f6 \n\t"
293 "punpcklhw $f28, $f10, $f14 \n\t"
294 "punpckhhw $f29, $f10, $f14 \n\t"
295 "punpcklwd $f2, $f26, $f28 \n\t"
296 "punpckhwd $f6, $f26, $f28 \n\t"
297 "punpcklwd $f10, $f27, $f29 \n\t"
298 "punpckhwd $f14, $f27, $f29 \n\t"
299
300 /* col0: $f0: col[24,16,8,0]; $f2: col[56,48,40,32] */
301 IDCT_COL_CASE2($f0, $f2, $f0, $f2)
302 /* col1: $f4: col[25,17,9,1]; $f6: col[57,49,41,33] */
303 IDCT_COL_CASE2($f4, $f6, $f4, $f6)
304 /* col2: $f8: col[26,18,10,2]; $f10: col[58,50,42,34] */
305 IDCT_COL_CASE2($f8, $f10, $f8, $f10)
306 /* col3: $f12: col[27,19,11,3]; $f14: col[59,51,43,35] */
307 IDCT_COL_CASE2($f12, $f14, $f12, $f14)
308
309 "2: \n\t"
310 /* Transpose */
311 "punpcklhw $f26, $f0, $f4 \n\t"
312 "punpckhhw $f27, $f0, $f4 \n\t"
313 "punpcklhw $f28, $f8, $f12 \n\t"
314 "punpckhhw $f29, $f8, $f12 \n\t"
315 "punpcklwd $f0, $f26, $f28 \n\t"
316 "punpckhwd $f4, $f26, $f28 \n\t"
317 "punpcklwd $f8, $f27, $f29 \n\t"
318 "punpckhwd $f12, $f27, $f29 \n\t"
319 /* Transpose */
320 "punpcklhw $f26, $f2, $f6 \n\t"
321 "punpckhhw $f27, $f2, $f6 \n\t"
322 "punpcklhw $f28, $f10, $f14 \n\t"
323 "punpckhhw $f29, $f10, $f14 \n\t"
324 "punpcklwd $f2, $f26, $f28 \n\t"
325 "punpckhwd $f6, $f26, $f28 \n\t"
326 "punpcklwd $f10, $f27, $f29 \n\t"
327 "punpckhwd $f14, $f27, $f29 \n\t"
328
329 /* idctSparseCol col4~7 */
330
331 "paddh $f1, $f1, $f17 \n\t"
332 /* Transpose */
333 "punpcklhw $f26, $f1, $f5 \n\t"
334 "punpckhhw $f27, $f1, $f5 \n\t"
335 "punpcklhw $f28, $f9, $f13 \n\t"
336 "punpckhhw $f29, $f9, $f13 \n\t"
337 "punpcklwd $f1, $f26, $f28 \n\t"
338 "punpckhwd $f5, $f26, $f28 \n\t"
339 "punpcklwd $f9, $f27, $f29 \n\t"
340 "punpckhwd $f13, $f27, $f29 \n\t"
341
342 "por $f26, $f3, $f7 \n\t"
343 "por $f26, $f26, $f11 \n\t"
344 "por $f26, $f26, $f15 \n\t"
345 "dmfc1 $10, $f26 \n\t"
346 "bnez $10, 1f \n\t"
347 /* case1: In this case, row[1,3,5,7] are all zero */
348 /* col4: $f1: col[24,16,8,0]; $f3: col[56,48,40,32] */
349 IDCT_COL_CASE1($f1, $f1, $f3)
350 /* col5: $f5: col[25,17,9,1]; $f7: col[57,49,41,33] */
351 IDCT_COL_CASE1($f5, $f5, $f7)
352 /* col6: $f9: col[26,18,10,2]; $f11: col[58,50,42,34] */
353 IDCT_COL_CASE1($f9, $f9, $f11)
354 /* col7: $f13: col[27,19,11,3]; $f15: col[59,51,43,35] */
355 IDCT_COL_CASE1($f13, $f13, $f15)
356 "j 2f \n\t"
357
358 "1: \n\t"
359 /* case2: row[1,3,5,7] are not all zero */
360 /* Transpose */
361 "punpcklhw $f26, $f3, $f7 \n\t"
362 "punpckhhw $f27, $f3, $f7 \n\t"
363 "punpcklhw $f28, $f11, $f15 \n\t"
364 "punpckhhw $f29, $f11, $f15 \n\t"
365 "punpcklwd $f3, $f26, $f28 \n\t"
366 "punpckhwd $f7, $f26, $f28 \n\t"
367 "punpcklwd $f11, $f27, $f29 \n\t"
368 "punpckhwd $f15, $f27, $f29 \n\t"
369
370 /* col4: $f1: col[24,16,8,0]; $f3: col[56,48,40,32] */
371 IDCT_COL_CASE2($f1, $f3, $f1, $f3)
372 /* col5: $f5: col[25,17,9,1]; $f7: col[57,49,41,33] */
373 IDCT_COL_CASE2($f5, $f7, $f5, $f7)
374 /* col6: $f9: col[26,18,10,2]; $f11: col[58,50,42,34] */
375 IDCT_COL_CASE2($f9, $f11, $f9, $f11)
376 /* col7: $f13: col[27,19,11,3]; $f15: col[59,51,43,35] */
377 IDCT_COL_CASE2($f13, $f15, $f13, $f15)
378
379 "2: \n\t"
380 /* Transpose */
381 "punpcklhw $f26, $f1, $f5 \n\t"
382 "punpckhhw $f27, $f1, $f5 \n\t"
383 "punpcklhw $f28, $f9, $f13 \n\t"
384 "punpckhhw $f29, $f9, $f13 \n\t"
385 "punpcklwd $f1, $f26, $f28 \n\t"
386 "punpckhwd $f5, $f26, $f28 \n\t"
387 "punpcklwd $f9, $f27, $f29 \n\t"
388 "punpckhwd $f13, $f27, $f29 \n\t"
389 /* Transpose */
390 "punpcklhw $f26, $f3, $f7 \n\t"
391 "punpckhhw $f27, $f3, $f7 \n\t"
392 "punpcklhw $f28, $f11, $f15 \n\t"
393 "punpckhhw $f29, $f11, $f15 \n\t"
394 "punpcklwd $f3, $f26, $f28 \n\t"
395 "punpckhwd $f7, $f26, $f28 \n\t"
396 "punpcklwd $f11, $f27, $f29 \n\t"
397 "punpckhwd $f15, $f27, $f29 \n\t"
398 /* Store */
399 MMI_SQC1($f1, $f0, %[block], 0x00)
400 MMI_SQC1($f5, $f4, %[block], 0x10)
401 MMI_SQC1($f9, $f8, %[block], 0x20)
402 MMI_SQC1($f13, $f12, %[block], 0x30)
403 MMI_SQC1($f3, $f2, %[block], 0x40)
404 MMI_SQC1($f7, $f6, %[block], 0x50)
405 MMI_SQC1($f11, $f10, %[block], 0x60)
406 MMI_SQC1($f15, $f14, %[block], 0x70)
407
408 : RESTRICT_ASM_ALL64 [block]"+&r"(block)
409 : [w_arr]"r"(W_arr)
410 : "memory"
411 );
412
413 RECOVER_REG
414 }
415
ff_simple_idct_put_8_mmi(uint8_t * dest,ptrdiff_t line_size,int16_t * block)416 void ff_simple_idct_put_8_mmi(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
417 {
418 ff_simple_idct_8_mmi(block);
419 ff_put_pixels_clamped_mmi(block, dest, line_size);
420 }
ff_simple_idct_add_8_mmi(uint8_t * dest,ptrdiff_t line_size,int16_t * block)421 void ff_simple_idct_add_8_mmi(uint8_t *dest, ptrdiff_t line_size, int16_t *block)
422 {
423 ff_simple_idct_8_mmi(block);
424 ff_add_pixels_clamped_mmi(block, dest, line_size);
425 }
426