itrans4_dspr2.c revision 7ce0a1d1337c01056ba24006efab21f00e179e04 
1/*
2 *  Copyright (c) 2015 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 "./vpx_config.h"
12#include "./vpx_dsp_rtcd.h"
13#include "vpx_dsp/mips/inv_txfm_dspr2.h"
14#include "vpx_dsp/txfm_common.h"
15
16#if HAVE_DSPR2
17void vpx_idct4_rows_dspr2(const int16_t *input, int16_t *output) {
18  int16_t   step_0, step_1, step_2, step_3;
19  int       Temp0, Temp1, Temp2, Temp3;
20  const int const_2_power_13 = 8192;
21  int       i;
22
23  for (i = 4; i--; ) {
24    __asm__ __volatile__ (
25        /*
26          temp_1 = (input[0] + input[2]) * cospi_16_64;
27          step_0 = dct_const_round_shift(temp_1);
28
29          temp_2 = (input[0] - input[2]) * cospi_16_64;
30          step_1 = dct_const_round_shift(temp_2);
31        */
32        "lh       %[Temp0],             0(%[input])                     \n\t"
33        "lh       %[Temp1],             4(%[input])                     \n\t"
34        "mtlo     %[const_2_power_13],  $ac0                            \n\t"
35        "mthi     $zero,                $ac0                            \n\t"
36        "mtlo     %[const_2_power_13],  $ac1                            \n\t"
37        "mthi     $zero,                $ac1                            \n\t"
38        "add      %[Temp2],             %[Temp0],       %[Temp1]        \n\t"
39        "sub      %[Temp3],             %[Temp0],       %[Temp1]        \n\t"
40        "madd     $ac0,                 %[Temp2],       %[cospi_16_64]  \n\t"
41        "lh       %[Temp0],             2(%[input])                     \n\t"
42        "lh       %[Temp1],             6(%[input])                     \n\t"
43        "extp     %[step_0],            $ac0,           31              \n\t"
44        "mtlo     %[const_2_power_13],  $ac0                            \n\t"
45        "mthi     $zero,                $ac0                            \n\t"
46
47        "madd     $ac1,                 %[Temp3],       %[cospi_16_64]  \n\t"
48        "extp     %[step_1],            $ac1,           31              \n\t"
49        "mtlo     %[const_2_power_13],  $ac1                            \n\t"
50        "mthi     $zero,                $ac1                            \n\t"
51
52        /*
53          temp1 = input[1] * cospi_24_64 - input[3] * cospi_8_64;
54          step_2 = dct_const_round_shift(temp1);
55        */
56        "madd     $ac0,                 %[Temp0],       %[cospi_24_64]  \n\t"
57        "msub     $ac0,                 %[Temp1],       %[cospi_8_64]   \n\t"
58        "extp     %[step_2],            $ac0,           31              \n\t"
59
60        /*
61          temp2 = input[1] * cospi_8_64 + input[3] * cospi_24_64;
62          step_3 = dct_const_round_shift(temp2);
63        */
64        "madd     $ac1,                 %[Temp0],       %[cospi_8_64]   \n\t"
65        "madd     $ac1,                 %[Temp1],       %[cospi_24_64]  \n\t"
66        "extp     %[step_3],            $ac1,           31              \n\t"
67
68        /*
69          output[0]  = step_0 + step_3;
70          output[4]  = step_1 + step_2;
71          output[8]  = step_1 - step_2;
72          output[12] = step_0 - step_3;
73        */
74        "add      %[Temp0],             %[step_0],      %[step_3]       \n\t"
75        "sh       %[Temp0],             0(%[output])                    \n\t"
76
77        "add      %[Temp1],             %[step_1],      %[step_2]       \n\t"
78        "sh       %[Temp1],             8(%[output])                    \n\t"
79
80        "sub      %[Temp2],             %[step_1],      %[step_2]       \n\t"
81        "sh       %[Temp2],             16(%[output])                   \n\t"
82
83        "sub      %[Temp3],             %[step_0],      %[step_3]       \n\t"
84        "sh       %[Temp3],             24(%[output])                   \n\t"
85
86      : [Temp0] "=&r" (Temp0), [Temp1] "=&r" (Temp1),
87        [Temp2] "=&r" (Temp2), [Temp3] "=&r" (Temp3),
88        [step_0] "=&r" (step_0), [step_1] "=&r" (step_1),
89        [step_2] "=&r" (step_2), [step_3] "=&r" (step_3),
90        [output] "+r" (output)
91      : [const_2_power_13] "r" (const_2_power_13),
92        [cospi_8_64] "r" (cospi_8_64), [cospi_16_64] "r" (cospi_16_64),
93        [cospi_24_64] "r" (cospi_24_64),
94        [input] "r" (input)
95    );
96
97    input += 4;
98    output += 1;
99  }
100}
101
102void vpx_idct4_columns_add_blk_dspr2(int16_t *input, uint8_t *dest,
103                                     int dest_stride) {
104  int16_t   step_0, step_1, step_2, step_3;
105  int       Temp0, Temp1, Temp2, Temp3;
106  const int const_2_power_13 = 8192;
107  int       i;
108  uint8_t   *dest_pix;
109  uint8_t   *cm = vpx_ff_cropTbl;
110
111  /* prefetch vpx_ff_cropTbl */
112  prefetch_load(vpx_ff_cropTbl);
113  prefetch_load(vpx_ff_cropTbl +  32);
114  prefetch_load(vpx_ff_cropTbl +  64);
115  prefetch_load(vpx_ff_cropTbl +  96);
116  prefetch_load(vpx_ff_cropTbl + 128);
117  prefetch_load(vpx_ff_cropTbl + 160);
118  prefetch_load(vpx_ff_cropTbl + 192);
119  prefetch_load(vpx_ff_cropTbl + 224);
120
121  for (i = 0; i < 4; ++i) {
122      dest_pix = (dest + i);
123
124    __asm__ __volatile__ (
125        /*
126          temp_1 = (input[0] + input[2]) * cospi_16_64;
127          step_0 = dct_const_round_shift(temp_1);
128
129          temp_2 = (input[0] - input[2]) * cospi_16_64;
130          step_1 = dct_const_round_shift(temp_2);
131        */
132        "lh       %[Temp0],             0(%[input])                     \n\t"
133        "lh       %[Temp1],             4(%[input])                     \n\t"
134        "mtlo     %[const_2_power_13],  $ac0                            \n\t"
135        "mthi     $zero,                $ac0                            \n\t"
136        "mtlo     %[const_2_power_13],  $ac1                            \n\t"
137        "mthi     $zero,                $ac1                            \n\t"
138        "add      %[Temp2],             %[Temp0],       %[Temp1]        \n\t"
139        "sub      %[Temp3],             %[Temp0],       %[Temp1]        \n\t"
140        "madd     $ac0,                 %[Temp2],       %[cospi_16_64]  \n\t"
141        "lh       %[Temp0],             2(%[input])                     \n\t"
142        "lh       %[Temp1],             6(%[input])                     \n\t"
143        "extp     %[step_0],            $ac0,           31              \n\t"
144        "mtlo     %[const_2_power_13],  $ac0                            \n\t"
145        "mthi     $zero,                $ac0                            \n\t"
146
147        "madd     $ac1,                 %[Temp3],       %[cospi_16_64]  \n\t"
148        "extp     %[step_1],            $ac1,           31              \n\t"
149        "mtlo     %[const_2_power_13],  $ac1                            \n\t"
150        "mthi     $zero,                $ac1                            \n\t"
151
152        /*
153          temp1 = input[1] * cospi_24_64 - input[3] * cospi_8_64;
154          step_2 = dct_const_round_shift(temp1);
155        */
156        "madd     $ac0,                 %[Temp0],       %[cospi_24_64]  \n\t"
157        "msub     $ac0,                 %[Temp1],       %[cospi_8_64]   \n\t"
158        "extp     %[step_2],            $ac0,           31              \n\t"
159
160        /*
161          temp2 = input[1] * cospi_8_64 + input[3] * cospi_24_64;
162          step_3 = dct_const_round_shift(temp2);
163        */
164        "madd     $ac1,                 %[Temp0],       %[cospi_8_64]   \n\t"
165        "madd     $ac1,                 %[Temp1],       %[cospi_24_64]  \n\t"
166        "extp     %[step_3],            $ac1,           31              \n\t"
167
168        /*
169          output[0]  = step_0 + step_3;
170          output[4]  = step_1 + step_2;
171          output[8]  = step_1 - step_2;
172          output[12] = step_0 - step_3;
173        */
174        "add      %[Temp0],             %[step_0],      %[step_3]       \n\t"
175        "addi     %[Temp0],             %[Temp0],       8               \n\t"
176        "sra      %[Temp0],             %[Temp0],       4               \n\t"
177        "lbu      %[Temp1],             0(%[dest_pix])                  \n\t"
178        "add      %[Temp1],             %[Temp1],       %[Temp0]        \n\t"
179        "add      %[Temp0],             %[step_1],      %[step_2]       \n\t"
180        "lbux     %[Temp2],             %[Temp1](%[cm])                 \n\t"
181        "sb       %[Temp2],             0(%[dest_pix])                  \n\t"
182        "addu     %[dest_pix],          %[dest_pix],    %[dest_stride]  \n\t"
183
184        "addi     %[Temp0],             %[Temp0],       8               \n\t"
185        "sra      %[Temp0],             %[Temp0],       4               \n\t"
186        "lbu      %[Temp1],             0(%[dest_pix])                  \n\t"
187        "add      %[Temp1],             %[Temp1],       %[Temp0]        \n\t"
188        "sub      %[Temp0],             %[step_1],      %[step_2]       \n\t"
189        "lbux     %[Temp2],             %[Temp1](%[cm])                 \n\t"
190        "sb       %[Temp2],             0(%[dest_pix])                  \n\t"
191        "addu     %[dest_pix],          %[dest_pix],    %[dest_stride]  \n\t"
192
193        "addi     %[Temp0],             %[Temp0],       8               \n\t"
194        "sra      %[Temp0],             %[Temp0],       4               \n\t"
195        "lbu      %[Temp1],             0(%[dest_pix])                  \n\t"
196        "add      %[Temp1],             %[Temp1],       %[Temp0]        \n\t"
197        "sub      %[Temp0],             %[step_0],      %[step_3]       \n\t"
198        "lbux     %[Temp2],             %[Temp1](%[cm])                 \n\t"
199        "sb       %[Temp2],             0(%[dest_pix])                  \n\t"
200        "addu     %[dest_pix],          %[dest_pix],    %[dest_stride]  \n\t"
201
202        "addi     %[Temp0],             %[Temp0],       8               \n\t"
203        "sra      %[Temp0],             %[Temp0],       4               \n\t"
204        "lbu      %[Temp1],             0(%[dest_pix])                  \n\t"
205        "add      %[Temp1],             %[Temp1],       %[Temp0]        \n\t"
206        "lbux     %[Temp2],             %[Temp1](%[cm])                 \n\t"
207        "sb       %[Temp2],             0(%[dest_pix])                  \n\t"
208
209      : [Temp0] "=&r" (Temp0), [Temp1] "=&r" (Temp1),
210        [Temp2] "=&r" (Temp2), [Temp3] "=&r" (Temp3),
211        [step_0] "=&r" (step_0), [step_1] "=&r" (step_1),
212        [step_2] "=&r" (step_2), [step_3] "=&r" (step_3),
213        [dest_pix] "+r" (dest_pix)
214      : [const_2_power_13] "r" (const_2_power_13),
215        [cospi_8_64] "r" (cospi_8_64), [cospi_16_64] "r" (cospi_16_64),
216        [cospi_24_64] "r" (cospi_24_64),
217        [input] "r" (input), [cm] "r" (cm), [dest_stride] "r" (dest_stride)
218    );
219
220    input += 4;
221  }
222}
223
224void vpx_idct4x4_16_add_dspr2(const int16_t *input, uint8_t *dest,
225                              int dest_stride) {
226  DECLARE_ALIGNED(32, int16_t, out[4 * 4]);
227  int16_t *outptr = out;
228  uint32_t pos = 45;
229
230  /* bit positon for extract from acc */
231  __asm__ __volatile__ (
232    "wrdsp      %[pos],     1           \n\t"
233    :
234    : [pos] "r" (pos)
235  );
236
237  // Rows
238  vpx_idct4_rows_dspr2(input, outptr);
239
240  // Columns
241  vpx_idct4_columns_add_blk_dspr2(&out[0], dest, dest_stride);
242}
243
244void vpx_idct4x4_1_add_dspr2(const int16_t *input, uint8_t *dest,
245                             int dest_stride) {
246  int       a1, absa1;
247  int       r;
248  int32_t   out;
249  int       t2, vector_a1, vector_a;
250  uint32_t  pos = 45;
251  int16_t   input_dc = input[0];
252
253  /* bit positon for extract from acc */
254  __asm__ __volatile__ (
255    "wrdsp      %[pos],     1           \n\t"
256
257    :
258    : [pos] "r" (pos)
259  );
260
261  out = DCT_CONST_ROUND_SHIFT_TWICE_COSPI_16_64(input_dc);
262  __asm__ __volatile__ (
263      "addi     %[out],     %[out],    8       \n\t"
264      "sra      %[a1],      %[out],    4       \n\t"
265
266      : [out] "+r" (out), [a1] "=r" (a1)
267      :
268  );
269
270  if (a1 < 0) {
271    /* use quad-byte
272     * input and output memory are four byte aligned */
273    __asm__ __volatile__ (
274        "abs        %[absa1],     %[a1]         \n\t"
275        "replv.qb   %[vector_a1], %[absa1]      \n\t"
276
277        : [absa1] "=r" (absa1), [vector_a1] "=r" (vector_a1)
278        : [a1] "r" (a1)
279    );
280
281    for (r = 4; r--;) {
282      __asm__ __volatile__ (
283          "lw             %[t2],          0(%[dest])                      \n\t"
284          "subu_s.qb      %[vector_a],    %[t2],          %[vector_a1]    \n\t"
285          "sw             %[vector_a],    0(%[dest])                      \n\t"
286          "add            %[dest],        %[dest],        %[dest_stride]  \n\t"
287
288          : [t2] "=&r" (t2), [vector_a] "=&r" (vector_a),
289            [dest] "+&r" (dest)
290          : [dest_stride] "r" (dest_stride), [vector_a1] "r" (vector_a1)
291      );
292    }
293  } else {
294    /* use quad-byte
295     * input and output memory are four byte aligned */
296    __asm__ __volatile__ (
297        "replv.qb       %[vector_a1],   %[a1]     \n\t"
298        : [vector_a1] "=r" (vector_a1)
299        : [a1] "r" (a1)
300    );
301
302    for (r = 4; r--;) {
303      __asm__ __volatile__ (
304          "lw           %[t2],          0(%[dest])                        \n\t"
305          "addu_s.qb    %[vector_a],    %[t2],            %[vector_a1]    \n\t"
306          "sw           %[vector_a],    0(%[dest])                        \n\t"
307          "add          %[dest],        %[dest],          %[dest_stride]  \n\t"
308
309          : [t2] "=&r" (t2), [vector_a] "=&r" (vector_a),
310            [dest] "+&r" (dest)
311          : [dest_stride] "r" (dest_stride), [vector_a1] "r" (vector_a1)
312      );
313    }
314  }
315}
316
317void iadst4_dspr2(const int16_t *input, int16_t *output) {
318  int s0, s1, s2, s3, s4, s5, s6, s7;
319  int x0, x1, x2, x3;
320
321  x0 = input[0];
322  x1 = input[1];
323  x2 = input[2];
324  x3 = input[3];
325
326  if (!(x0 | x1 | x2 | x3)) {
327    output[0] = output[1] = output[2] = output[3] = 0;
328    return;
329  }
330
331  s0 = sinpi_1_9 * x0;
332  s1 = sinpi_2_9 * x0;
333  s2 = sinpi_3_9 * x1;
334  s3 = sinpi_4_9 * x2;
335  s4 = sinpi_1_9 * x2;
336  s5 = sinpi_2_9 * x3;
337  s6 = sinpi_4_9 * x3;
338  s7 = x0 - x2 + x3;
339
340  x0 = s0 + s3 + s5;
341  x1 = s1 - s4 - s6;
342  x2 = sinpi_3_9 * s7;
343  x3 = s2;
344
345  s0 = x0 + x3;
346  s1 = x1 + x3;
347  s2 = x2;
348  s3 = x0 + x1 - x3;
349
350  // 1-D transform scaling factor is sqrt(2).
351  // The overall dynamic range is 14b (input) + 14b (multiplication scaling)
352  // + 1b (addition) = 29b.
353  // Hence the output bit depth is 15b.
354  output[0] = dct_const_round_shift(s0);
355  output[1] = dct_const_round_shift(s1);
356  output[2] = dct_const_round_shift(s2);
357  output[3] = dct_const_round_shift(s3);
358}
359#endif  // #if HAVE_DSPR2
360