1 /*
2 * Copyright (c) 2013 The WebM project authors. All Rights Reserved.
3 *
4 * Use of this source code is governed by a BSD-style license
5 * that can be found in the LICENSE file in the root of the source
6 * tree. An additional intellectual property rights grant can be found
7 * in the file PATENTS. All contributing project authors may
8 * be found in the AUTHORS file in the root of the source tree.
9 */
10
11 #include <assert.h>
12 #include <stdio.h>
13
14 #include "./vpx_config.h"
15 #include "./vp9_rtcd.h"
16 #include "vp9/common/vp9_common.h"
17 #include "vp9/common/vp9_blockd.h"
18 #include "vp9/common/vp9_idct.h"
19 #include "vp9/common/mips/dspr2/vp9_common_dspr2.h"
20
21 #if HAVE_DSPR2
vp9_idct4_rows_dspr2(const int16_t * input,int16_t * output)22 static void vp9_idct4_rows_dspr2(const int16_t *input, int16_t *output) {
23 int16_t step_0, step_1, step_2, step_3;
24 int Temp0, Temp1, Temp2, Temp3;
25 const int const_2_power_13 = 8192;
26 int i;
27
28 for (i = 4; i--; ) {
29 __asm__ __volatile__ (
30 /*
31 temp_1 = (input[0] + input[2]) * cospi_16_64;
32 step_0 = dct_const_round_shift(temp_1);
33
34 temp_2 = (input[0] - input[2]) * cospi_16_64;
35 step_1 = dct_const_round_shift(temp_2);
36 */
37 "lh %[Temp0], 0(%[input]) \n\t"
38 "lh %[Temp1], 4(%[input]) \n\t"
39 "mtlo %[const_2_power_13], $ac0 \n\t"
40 "mthi $zero, $ac0 \n\t"
41 "mtlo %[const_2_power_13], $ac1 \n\t"
42 "mthi $zero, $ac1 \n\t"
43 "add %[Temp2], %[Temp0], %[Temp1] \n\t"
44 "sub %[Temp3], %[Temp0], %[Temp1] \n\t"
45 "madd $ac0, %[Temp2], %[cospi_16_64] \n\t"
46 "lh %[Temp0], 2(%[input]) \n\t"
47 "lh %[Temp1], 6(%[input]) \n\t"
48 "extp %[step_0], $ac0, 31 \n\t"
49 "mtlo %[const_2_power_13], $ac0 \n\t"
50 "mthi $zero, $ac0 \n\t"
51
52 "madd $ac1, %[Temp3], %[cospi_16_64] \n\t"
53 "extp %[step_1], $ac1, 31 \n\t"
54 "mtlo %[const_2_power_13], $ac1 \n\t"
55 "mthi $zero, $ac1 \n\t"
56
57 /*
58 temp1 = input[1] * cospi_24_64 - input[3] * cospi_8_64;
59 step_2 = dct_const_round_shift(temp1);
60 */
61 "madd $ac0, %[Temp0], %[cospi_24_64] \n\t"
62 "msub $ac0, %[Temp1], %[cospi_8_64] \n\t"
63 "extp %[step_2], $ac0, 31 \n\t"
64
65 /*
66 temp2 = input[1] * cospi_8_64 + input[3] * cospi_24_64;
67 step_3 = dct_const_round_shift(temp2);
68 */
69 "madd $ac1, %[Temp0], %[cospi_8_64] \n\t"
70 "madd $ac1, %[Temp1], %[cospi_24_64] \n\t"
71 "extp %[step_3], $ac1, 31 \n\t"
72
73 /*
74 output[0] = step_0 + step_3;
75 output[4] = step_1 + step_2;
76 output[8] = step_1 - step_2;
77 output[12] = step_0 - step_3;
78 */
79 "add %[Temp0], %[step_0], %[step_3] \n\t"
80 "sh %[Temp0], 0(%[output]) \n\t"
81
82 "add %[Temp1], %[step_1], %[step_2] \n\t"
83 "sh %[Temp1], 8(%[output]) \n\t"
84
85 "sub %[Temp2], %[step_1], %[step_2] \n\t"
86 "sh %[Temp2], 16(%[output]) \n\t"
87
88 "sub %[Temp3], %[step_0], %[step_3] \n\t"
89 "sh %[Temp3], 24(%[output]) \n\t"
90
91 : [Temp0] "=&r" (Temp0), [Temp1] "=&r" (Temp1),
92 [Temp2] "=&r" (Temp2), [Temp3] "=&r" (Temp3),
93 [step_0] "=&r" (step_0), [step_1] "=&r" (step_1),
94 [step_2] "=&r" (step_2), [step_3] "=&r" (step_3),
95 [output] "+r" (output)
96 : [const_2_power_13] "r" (const_2_power_13),
97 [cospi_8_64] "r" (cospi_8_64), [cospi_16_64] "r" (cospi_16_64),
98 [cospi_24_64] "r" (cospi_24_64),
99 [input] "r" (input)
100 );
101
102 input += 4;
103 output += 1;
104 }
105 }
106
vp9_idct4_columns_add_blk_dspr2(int16_t * input,uint8_t * dest,int dest_stride)107 static void vp9_idct4_columns_add_blk_dspr2(int16_t *input, uint8_t *dest,
108 int dest_stride) {
109 int16_t step_0, step_1, step_2, step_3;
110 int Temp0, Temp1, Temp2, Temp3;
111 const int const_2_power_13 = 8192;
112 int i;
113 uint8_t *dest_pix;
114 uint8_t *cm = vp9_ff_cropTbl;
115
116 /* prefetch vp9_ff_cropTbl */
117 vp9_prefetch_load(vp9_ff_cropTbl);
118 vp9_prefetch_load(vp9_ff_cropTbl + 32);
119 vp9_prefetch_load(vp9_ff_cropTbl + 64);
120 vp9_prefetch_load(vp9_ff_cropTbl + 96);
121 vp9_prefetch_load(vp9_ff_cropTbl + 128);
122 vp9_prefetch_load(vp9_ff_cropTbl + 160);
123 vp9_prefetch_load(vp9_ff_cropTbl + 192);
124 vp9_prefetch_load(vp9_ff_cropTbl + 224);
125
126 for (i = 0; i < 4; ++i) {
127 dest_pix = (dest + i);
128
129 __asm__ __volatile__ (
130 /*
131 temp_1 = (input[0] + input[2]) * cospi_16_64;
132 step_0 = dct_const_round_shift(temp_1);
133
134 temp_2 = (input[0] - input[2]) * cospi_16_64;
135 step_1 = dct_const_round_shift(temp_2);
136 */
137 "lh %[Temp0], 0(%[input]) \n\t"
138 "lh %[Temp1], 4(%[input]) \n\t"
139 "mtlo %[const_2_power_13], $ac0 \n\t"
140 "mthi $zero, $ac0 \n\t"
141 "mtlo %[const_2_power_13], $ac1 \n\t"
142 "mthi $zero, $ac1 \n\t"
143 "add %[Temp2], %[Temp0], %[Temp1] \n\t"
144 "sub %[Temp3], %[Temp0], %[Temp1] \n\t"
145 "madd $ac0, %[Temp2], %[cospi_16_64] \n\t"
146 "lh %[Temp0], 2(%[input]) \n\t"
147 "lh %[Temp1], 6(%[input]) \n\t"
148 "extp %[step_0], $ac0, 31 \n\t"
149 "mtlo %[const_2_power_13], $ac0 \n\t"
150 "mthi $zero, $ac0 \n\t"
151
152 "madd $ac1, %[Temp3], %[cospi_16_64] \n\t"
153 "extp %[step_1], $ac1, 31 \n\t"
154 "mtlo %[const_2_power_13], $ac1 \n\t"
155 "mthi $zero, $ac1 \n\t"
156
157 /*
158 temp1 = input[1] * cospi_24_64 - input[3] * cospi_8_64;
159 step_2 = dct_const_round_shift(temp1);
160 */
161 "madd $ac0, %[Temp0], %[cospi_24_64] \n\t"
162 "msub $ac0, %[Temp1], %[cospi_8_64] \n\t"
163 "extp %[step_2], $ac0, 31 \n\t"
164
165 /*
166 temp2 = input[1] * cospi_8_64 + input[3] * cospi_24_64;
167 step_3 = dct_const_round_shift(temp2);
168 */
169 "madd $ac1, %[Temp0], %[cospi_8_64] \n\t"
170 "madd $ac1, %[Temp1], %[cospi_24_64] \n\t"
171 "extp %[step_3], $ac1, 31 \n\t"
172
173 /*
174 output[0] = step_0 + step_3;
175 output[4] = step_1 + step_2;
176 output[8] = step_1 - step_2;
177 output[12] = step_0 - step_3;
178 */
179 "add %[Temp0], %[step_0], %[step_3] \n\t"
180 "addi %[Temp0], %[Temp0], 8 \n\t"
181 "sra %[Temp0], %[Temp0], 4 \n\t"
182 "lbu %[Temp1], 0(%[dest_pix]) \n\t"
183 "add %[Temp1], %[Temp1], %[Temp0] \n\t"
184 "add %[Temp0], %[step_1], %[step_2] \n\t"
185 "lbux %[Temp2], %[Temp1](%[cm]) \n\t"
186 "sb %[Temp2], 0(%[dest_pix]) \n\t"
187 "addu %[dest_pix], %[dest_pix], %[dest_stride] \n\t"
188
189 "addi %[Temp0], %[Temp0], 8 \n\t"
190 "sra %[Temp0], %[Temp0], 4 \n\t"
191 "lbu %[Temp1], 0(%[dest_pix]) \n\t"
192 "add %[Temp1], %[Temp1], %[Temp0] \n\t"
193 "sub %[Temp0], %[step_1], %[step_2] \n\t"
194 "lbux %[Temp2], %[Temp1](%[cm]) \n\t"
195 "sb %[Temp2], 0(%[dest_pix]) \n\t"
196 "addu %[dest_pix], %[dest_pix], %[dest_stride] \n\t"
197
198 "addi %[Temp0], %[Temp0], 8 \n\t"
199 "sra %[Temp0], %[Temp0], 4 \n\t"
200 "lbu %[Temp1], 0(%[dest_pix]) \n\t"
201 "add %[Temp1], %[Temp1], %[Temp0] \n\t"
202 "sub %[Temp0], %[step_0], %[step_3] \n\t"
203 "lbux %[Temp2], %[Temp1](%[cm]) \n\t"
204 "sb %[Temp2], 0(%[dest_pix]) \n\t"
205 "addu %[dest_pix], %[dest_pix], %[dest_stride] \n\t"
206
207 "addi %[Temp0], %[Temp0], 8 \n\t"
208 "sra %[Temp0], %[Temp0], 4 \n\t"
209 "lbu %[Temp1], 0(%[dest_pix]) \n\t"
210 "add %[Temp1], %[Temp1], %[Temp0] \n\t"
211 "lbux %[Temp2], %[Temp1](%[cm]) \n\t"
212 "sb %[Temp2], 0(%[dest_pix]) \n\t"
213
214 : [Temp0] "=&r" (Temp0), [Temp1] "=&r" (Temp1),
215 [Temp2] "=&r" (Temp2), [Temp3] "=&r" (Temp3),
216 [step_0] "=&r" (step_0), [step_1] "=&r" (step_1),
217 [step_2] "=&r" (step_2), [step_3] "=&r" (step_3),
218 [dest_pix] "+r" (dest_pix)
219 : [const_2_power_13] "r" (const_2_power_13),
220 [cospi_8_64] "r" (cospi_8_64), [cospi_16_64] "r" (cospi_16_64),
221 [cospi_24_64] "r" (cospi_24_64),
222 [input] "r" (input), [cm] "r" (cm), [dest_stride] "r" (dest_stride)
223 );
224
225 input += 4;
226 }
227 }
228
vp9_idct4x4_16_add_dspr2(const int16_t * input,uint8_t * dest,int dest_stride)229 void vp9_idct4x4_16_add_dspr2(const int16_t *input, uint8_t *dest,
230 int dest_stride) {
231 DECLARE_ALIGNED(32, int16_t, out[4 * 4]);
232 int16_t *outptr = out;
233 uint32_t pos = 45;
234
235 /* bit positon for extract from acc */
236 __asm__ __volatile__ (
237 "wrdsp %[pos], 1 \n\t"
238 :
239 : [pos] "r" (pos)
240 );
241
242 // Rows
243 vp9_idct4_rows_dspr2(input, outptr);
244
245 // Columns
246 vp9_idct4_columns_add_blk_dspr2(&out[0], dest, dest_stride);
247 }
248
vp9_idct4x4_1_add_dspr2(const int16_t * input,uint8_t * dest,int dest_stride)249 void vp9_idct4x4_1_add_dspr2(const int16_t *input, uint8_t *dest,
250 int dest_stride) {
251 int a1, absa1;
252 int r;
253 int32_t out;
254 int t2, vector_a1, vector_a;
255 uint32_t pos = 45;
256 int16_t input_dc = input[0];
257
258 /* bit positon for extract from acc */
259 __asm__ __volatile__ (
260 "wrdsp %[pos], 1 \n\t"
261
262 :
263 : [pos] "r" (pos)
264 );
265
266 out = DCT_CONST_ROUND_SHIFT_TWICE_COSPI_16_64(input_dc);
267 __asm__ __volatile__ (
268 "addi %[out], %[out], 8 \n\t"
269 "sra %[a1], %[out], 4 \n\t"
270
271 : [out] "+r" (out), [a1] "=r" (a1)
272 :
273 );
274
275 if (a1 < 0) {
276 /* use quad-byte
277 * input and output memory are four byte aligned */
278 __asm__ __volatile__ (
279 "abs %[absa1], %[a1] \n\t"
280 "replv.qb %[vector_a1], %[absa1] \n\t"
281
282 : [absa1] "=r" (absa1), [vector_a1] "=r" (vector_a1)
283 : [a1] "r" (a1)
284 );
285
286 for (r = 4; r--;) {
287 __asm__ __volatile__ (
288 "lw %[t2], 0(%[dest]) \n\t"
289 "subu_s.qb %[vector_a], %[t2], %[vector_a1] \n\t"
290 "sw %[vector_a], 0(%[dest]) \n\t"
291 "add %[dest], %[dest], %[dest_stride] \n\t"
292
293 : [t2] "=&r" (t2), [vector_a] "=&r" (vector_a),
294 [dest] "+&r" (dest)
295 : [dest_stride] "r" (dest_stride), [vector_a1] "r" (vector_a1)
296 );
297 }
298 } else {
299 /* use quad-byte
300 * input and output memory are four byte aligned */
301 __asm__ __volatile__ (
302 "replv.qb %[vector_a1], %[a1] \n\t"
303 : [vector_a1] "=r" (vector_a1)
304 : [a1] "r" (a1)
305 );
306
307 for (r = 4; r--;) {
308 __asm__ __volatile__ (
309 "lw %[t2], 0(%[dest]) \n\t"
310 "addu_s.qb %[vector_a], %[t2], %[vector_a1] \n\t"
311 "sw %[vector_a], 0(%[dest]) \n\t"
312 "add %[dest], %[dest], %[dest_stride] \n\t"
313
314 : [t2] "=&r" (t2), [vector_a] "=&r" (vector_a),
315 [dest] "+&r" (dest)
316 : [dest_stride] "r" (dest_stride), [vector_a1] "r" (vector_a1)
317 );
318 }
319 }
320 }
321
iadst4_dspr2(const int16_t * input,int16_t * output)322 static void iadst4_dspr2(const int16_t *input, int16_t *output) {
323 int s0, s1, s2, s3, s4, s5, s6, s7;
324 int x0, x1, x2, x3;
325
326 x0 = input[0];
327 x1 = input[1];
328 x2 = input[2];
329 x3 = input[3];
330
331 if (!(x0 | x1 | x2 | x3)) {
332 output[0] = output[1] = output[2] = output[3] = 0;
333 return;
334 }
335
336 s0 = sinpi_1_9 * x0;
337 s1 = sinpi_2_9 * x0;
338 s2 = sinpi_3_9 * x1;
339 s3 = sinpi_4_9 * x2;
340 s4 = sinpi_1_9 * x2;
341 s5 = sinpi_2_9 * x3;
342 s6 = sinpi_4_9 * x3;
343 s7 = x0 - x2 + x3;
344
345 x0 = s0 + s3 + s5;
346 x1 = s1 - s4 - s6;
347 x2 = sinpi_3_9 * s7;
348 x3 = s2;
349
350 s0 = x0 + x3;
351 s1 = x1 + x3;
352 s2 = x2;
353 s3 = x0 + x1 - x3;
354
355 // 1-D transform scaling factor is sqrt(2).
356 // The overall dynamic range is 14b (input) + 14b (multiplication scaling)
357 // + 1b (addition) = 29b.
358 // Hence the output bit depth is 15b.
359 output[0] = dct_const_round_shift(s0);
360 output[1] = dct_const_round_shift(s1);
361 output[2] = dct_const_round_shift(s2);
362 output[3] = dct_const_round_shift(s3);
363 }
364
vp9_iht4x4_16_add_dspr2(const int16_t * input,uint8_t * dest,int dest_stride,int tx_type)365 void vp9_iht4x4_16_add_dspr2(const int16_t *input, uint8_t *dest,
366 int dest_stride, int tx_type) {
367 int i, j;
368 DECLARE_ALIGNED(32, int16_t, out[4 * 4]);
369 int16_t *outptr = out;
370 int16_t temp_in[4 * 4], temp_out[4];
371 uint32_t pos = 45;
372
373 /* bit positon for extract from acc */
374 __asm__ __volatile__ (
375 "wrdsp %[pos], 1 \n\t"
376 :
377 : [pos] "r" (pos)
378 );
379
380 switch (tx_type) {
381 case DCT_DCT: // DCT in both horizontal and vertical
382 vp9_idct4_rows_dspr2(input, outptr);
383 vp9_idct4_columns_add_blk_dspr2(&out[0], dest, dest_stride);
384 break;
385 case ADST_DCT: // ADST in vertical, DCT in horizontal
386 vp9_idct4_rows_dspr2(input, outptr);
387
388 outptr = out;
389
390 for (i = 0; i < 4; ++i) {
391 iadst4_dspr2(outptr, temp_out);
392
393 for (j = 0; j < 4; ++j)
394 dest[j * dest_stride + i] =
395 clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 4)
396 + dest[j * dest_stride + i]);
397
398 outptr += 4;
399 }
400 break;
401 case DCT_ADST: // DCT in vertical, ADST in horizontal
402 for (i = 0; i < 4; ++i) {
403 iadst4_dspr2(input, outptr);
404 input += 4;
405 outptr += 4;
406 }
407
408 for (i = 0; i < 4; ++i) {
409 for (j = 0; j < 4; ++j) {
410 temp_in[i * 4 + j] = out[j * 4 + i];
411 }
412 }
413 vp9_idct4_columns_add_blk_dspr2(&temp_in[0], dest, dest_stride);
414 break;
415 case ADST_ADST: // ADST in both directions
416 for (i = 0; i < 4; ++i) {
417 iadst4_dspr2(input, outptr);
418 input += 4;
419 outptr += 4;
420 }
421
422 for (i = 0; i < 4; ++i) {
423 for (j = 0; j < 4; ++j)
424 temp_in[j] = out[j * 4 + i];
425 iadst4_dspr2(temp_in, temp_out);
426
427 for (j = 0; j < 4; ++j)
428 dest[j * dest_stride + i] =
429 clip_pixel(ROUND_POWER_OF_TWO(temp_out[j], 4)
430 + dest[j * dest_stride + i]);
431 }
432 break;
433 default:
434 printf("vp9_short_iht4x4_add_dspr2 : Invalid tx_type\n");
435 break;
436 }
437 }
438 #endif // #if HAVE_DSPR2
439