Line data Source code
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 <assert.h>
12 : #include "./vpx_dsp_rtcd.h"
13 : #include "vpx_dsp/fwd_txfm.h"
14 :
15 0 : void vpx_fdct4x4_c(const int16_t *input, tran_low_t *output, int stride) {
16 : // The 2D transform is done with two passes which are actually pretty
17 : // similar. In the first one, we transform the columns and transpose
18 : // the results. In the second one, we transform the rows. To achieve that,
19 : // as the first pass results are transposed, we transpose the columns (that
20 : // is the transposed rows) and transpose the results (so that it goes back
21 : // in normal/row positions).
22 : int pass;
23 : // We need an intermediate buffer between passes.
24 : tran_low_t intermediate[4 * 4];
25 0 : const tran_low_t *in_low = NULL;
26 0 : tran_low_t *out = intermediate;
27 : // Do the two transform/transpose passes
28 0 : for (pass = 0; pass < 2; ++pass) {
29 : tran_high_t in_high[4]; // canbe16
30 : tran_high_t step[4]; // canbe16
31 : tran_high_t temp1, temp2; // needs32
32 : int i;
33 0 : for (i = 0; i < 4; ++i) {
34 : // Load inputs.
35 0 : if (pass == 0) {
36 0 : in_high[0] = input[0 * stride] * 16;
37 0 : in_high[1] = input[1 * stride] * 16;
38 0 : in_high[2] = input[2 * stride] * 16;
39 0 : in_high[3] = input[3 * stride] * 16;
40 0 : if (i == 0 && in_high[0]) {
41 0 : ++in_high[0];
42 : }
43 : } else {
44 0 : assert(in_low != NULL);
45 0 : in_high[0] = in_low[0 * 4];
46 0 : in_high[1] = in_low[1 * 4];
47 0 : in_high[2] = in_low[2 * 4];
48 0 : in_high[3] = in_low[3 * 4];
49 0 : ++in_low;
50 : }
51 : // Transform.
52 0 : step[0] = in_high[0] + in_high[3];
53 0 : step[1] = in_high[1] + in_high[2];
54 0 : step[2] = in_high[1] - in_high[2];
55 0 : step[3] = in_high[0] - in_high[3];
56 0 : temp1 = (step[0] + step[1]) * cospi_16_64;
57 0 : temp2 = (step[0] - step[1]) * cospi_16_64;
58 0 : out[0] = (tran_low_t)fdct_round_shift(temp1);
59 0 : out[2] = (tran_low_t)fdct_round_shift(temp2);
60 0 : temp1 = step[2] * cospi_24_64 + step[3] * cospi_8_64;
61 0 : temp2 = -step[2] * cospi_8_64 + step[3] * cospi_24_64;
62 0 : out[1] = (tran_low_t)fdct_round_shift(temp1);
63 0 : out[3] = (tran_low_t)fdct_round_shift(temp2);
64 : // Do next column (which is a transposed row in second/horizontal pass)
65 0 : ++input;
66 0 : out += 4;
67 : }
68 : // Setup in/out for next pass.
69 0 : in_low = intermediate;
70 0 : out = output;
71 : }
72 :
73 : {
74 : int i, j;
75 0 : for (i = 0; i < 4; ++i) {
76 0 : for (j = 0; j < 4; ++j) output[j + i * 4] = (output[j + i * 4] + 1) >> 2;
77 : }
78 : }
79 0 : }
80 :
81 0 : void vpx_fdct4x4_1_c(const int16_t *input, tran_low_t *output, int stride) {
82 : int r, c;
83 0 : tran_low_t sum = 0;
84 0 : for (r = 0; r < 4; ++r)
85 0 : for (c = 0; c < 4; ++c) sum += input[r * stride + c];
86 :
87 0 : output[0] = sum << 1;
88 0 : }
89 :
90 0 : void vpx_fdct8x8_c(const int16_t *input, tran_low_t *final_output, int stride) {
91 : int i, j;
92 : tran_low_t intermediate[64];
93 : int pass;
94 0 : tran_low_t *output = intermediate;
95 0 : const tran_low_t *in = NULL;
96 :
97 : // Transform columns
98 0 : for (pass = 0; pass < 2; ++pass) {
99 : tran_high_t s0, s1, s2, s3, s4, s5, s6, s7; // canbe16
100 : tran_high_t t0, t1, t2, t3; // needs32
101 : tran_high_t x0, x1, x2, x3; // canbe16
102 :
103 0 : for (i = 0; i < 8; i++) {
104 : // stage 1
105 0 : if (pass == 0) {
106 0 : s0 = (input[0 * stride] + input[7 * stride]) * 4;
107 0 : s1 = (input[1 * stride] + input[6 * stride]) * 4;
108 0 : s2 = (input[2 * stride] + input[5 * stride]) * 4;
109 0 : s3 = (input[3 * stride] + input[4 * stride]) * 4;
110 0 : s4 = (input[3 * stride] - input[4 * stride]) * 4;
111 0 : s5 = (input[2 * stride] - input[5 * stride]) * 4;
112 0 : s6 = (input[1 * stride] - input[6 * stride]) * 4;
113 0 : s7 = (input[0 * stride] - input[7 * stride]) * 4;
114 0 : ++input;
115 : } else {
116 0 : s0 = in[0 * 8] + in[7 * 8];
117 0 : s1 = in[1 * 8] + in[6 * 8];
118 0 : s2 = in[2 * 8] + in[5 * 8];
119 0 : s3 = in[3 * 8] + in[4 * 8];
120 0 : s4 = in[3 * 8] - in[4 * 8];
121 0 : s5 = in[2 * 8] - in[5 * 8];
122 0 : s6 = in[1 * 8] - in[6 * 8];
123 0 : s7 = in[0 * 8] - in[7 * 8];
124 0 : ++in;
125 : }
126 :
127 : // fdct4(step, step);
128 0 : x0 = s0 + s3;
129 0 : x1 = s1 + s2;
130 0 : x2 = s1 - s2;
131 0 : x3 = s0 - s3;
132 0 : t0 = (x0 + x1) * cospi_16_64;
133 0 : t1 = (x0 - x1) * cospi_16_64;
134 0 : t2 = x2 * cospi_24_64 + x3 * cospi_8_64;
135 0 : t3 = -x2 * cospi_8_64 + x3 * cospi_24_64;
136 0 : output[0] = (tran_low_t)fdct_round_shift(t0);
137 0 : output[2] = (tran_low_t)fdct_round_shift(t2);
138 0 : output[4] = (tran_low_t)fdct_round_shift(t1);
139 0 : output[6] = (tran_low_t)fdct_round_shift(t3);
140 :
141 : // Stage 2
142 0 : t0 = (s6 - s5) * cospi_16_64;
143 0 : t1 = (s6 + s5) * cospi_16_64;
144 0 : t2 = fdct_round_shift(t0);
145 0 : t3 = fdct_round_shift(t1);
146 :
147 : // Stage 3
148 0 : x0 = s4 + t2;
149 0 : x1 = s4 - t2;
150 0 : x2 = s7 - t3;
151 0 : x3 = s7 + t3;
152 :
153 : // Stage 4
154 0 : t0 = x0 * cospi_28_64 + x3 * cospi_4_64;
155 0 : t1 = x1 * cospi_12_64 + x2 * cospi_20_64;
156 0 : t2 = x2 * cospi_12_64 + x1 * -cospi_20_64;
157 0 : t3 = x3 * cospi_28_64 + x0 * -cospi_4_64;
158 0 : output[1] = (tran_low_t)fdct_round_shift(t0);
159 0 : output[3] = (tran_low_t)fdct_round_shift(t2);
160 0 : output[5] = (tran_low_t)fdct_round_shift(t1);
161 0 : output[7] = (tran_low_t)fdct_round_shift(t3);
162 0 : output += 8;
163 : }
164 0 : in = intermediate;
165 0 : output = final_output;
166 : }
167 :
168 : // Rows
169 0 : for (i = 0; i < 8; ++i) {
170 0 : for (j = 0; j < 8; ++j) final_output[j + i * 8] /= 2;
171 : }
172 0 : }
173 :
174 0 : void vpx_fdct8x8_1_c(const int16_t *input, tran_low_t *output, int stride) {
175 : int r, c;
176 0 : tran_low_t sum = 0;
177 0 : for (r = 0; r < 8; ++r)
178 0 : for (c = 0; c < 8; ++c) sum += input[r * stride + c];
179 :
180 0 : output[0] = sum;
181 0 : }
182 :
183 0 : void vpx_fdct16x16_c(const int16_t *input, tran_low_t *output, int stride) {
184 : // The 2D transform is done with two passes which are actually pretty
185 : // similar. In the first one, we transform the columns and transpose
186 : // the results. In the second one, we transform the rows. To achieve that,
187 : // as the first pass results are transposed, we transpose the columns (that
188 : // is the transposed rows) and transpose the results (so that it goes back
189 : // in normal/row positions).
190 : int pass;
191 : // We need an intermediate buffer between passes.
192 : tran_low_t intermediate[256];
193 0 : const tran_low_t *in_low = NULL;
194 0 : tran_low_t *out = intermediate;
195 : // Do the two transform/transpose passes
196 0 : for (pass = 0; pass < 2; ++pass) {
197 : tran_high_t step1[8]; // canbe16
198 : tran_high_t step2[8]; // canbe16
199 : tran_high_t step3[8]; // canbe16
200 : tran_high_t in_high[8]; // canbe16
201 : tran_high_t temp1, temp2; // needs32
202 : int i;
203 0 : for (i = 0; i < 16; i++) {
204 0 : if (0 == pass) {
205 : // Calculate input for the first 8 results.
206 0 : in_high[0] = (input[0 * stride] + input[15 * stride]) * 4;
207 0 : in_high[1] = (input[1 * stride] + input[14 * stride]) * 4;
208 0 : in_high[2] = (input[2 * stride] + input[13 * stride]) * 4;
209 0 : in_high[3] = (input[3 * stride] + input[12 * stride]) * 4;
210 0 : in_high[4] = (input[4 * stride] + input[11 * stride]) * 4;
211 0 : in_high[5] = (input[5 * stride] + input[10 * stride]) * 4;
212 0 : in_high[6] = (input[6 * stride] + input[9 * stride]) * 4;
213 0 : in_high[7] = (input[7 * stride] + input[8 * stride]) * 4;
214 : // Calculate input for the next 8 results.
215 0 : step1[0] = (input[7 * stride] - input[8 * stride]) * 4;
216 0 : step1[1] = (input[6 * stride] - input[9 * stride]) * 4;
217 0 : step1[2] = (input[5 * stride] - input[10 * stride]) * 4;
218 0 : step1[3] = (input[4 * stride] - input[11 * stride]) * 4;
219 0 : step1[4] = (input[3 * stride] - input[12 * stride]) * 4;
220 0 : step1[5] = (input[2 * stride] - input[13 * stride]) * 4;
221 0 : step1[6] = (input[1 * stride] - input[14 * stride]) * 4;
222 0 : step1[7] = (input[0 * stride] - input[15 * stride]) * 4;
223 : } else {
224 : // Calculate input for the first 8 results.
225 0 : assert(in_low != NULL);
226 0 : in_high[0] = ((in_low[0 * 16] + 1) >> 2) + ((in_low[15 * 16] + 1) >> 2);
227 0 : in_high[1] = ((in_low[1 * 16] + 1) >> 2) + ((in_low[14 * 16] + 1) >> 2);
228 0 : in_high[2] = ((in_low[2 * 16] + 1) >> 2) + ((in_low[13 * 16] + 1) >> 2);
229 0 : in_high[3] = ((in_low[3 * 16] + 1) >> 2) + ((in_low[12 * 16] + 1) >> 2);
230 0 : in_high[4] = ((in_low[4 * 16] + 1) >> 2) + ((in_low[11 * 16] + 1) >> 2);
231 0 : in_high[5] = ((in_low[5 * 16] + 1) >> 2) + ((in_low[10 * 16] + 1) >> 2);
232 0 : in_high[6] = ((in_low[6 * 16] + 1) >> 2) + ((in_low[9 * 16] + 1) >> 2);
233 0 : in_high[7] = ((in_low[7 * 16] + 1) >> 2) + ((in_low[8 * 16] + 1) >> 2);
234 : // Calculate input for the next 8 results.
235 0 : step1[0] = ((in_low[7 * 16] + 1) >> 2) - ((in_low[8 * 16] + 1) >> 2);
236 0 : step1[1] = ((in_low[6 * 16] + 1) >> 2) - ((in_low[9 * 16] + 1) >> 2);
237 0 : step1[2] = ((in_low[5 * 16] + 1) >> 2) - ((in_low[10 * 16] + 1) >> 2);
238 0 : step1[3] = ((in_low[4 * 16] + 1) >> 2) - ((in_low[11 * 16] + 1) >> 2);
239 0 : step1[4] = ((in_low[3 * 16] + 1) >> 2) - ((in_low[12 * 16] + 1) >> 2);
240 0 : step1[5] = ((in_low[2 * 16] + 1) >> 2) - ((in_low[13 * 16] + 1) >> 2);
241 0 : step1[6] = ((in_low[1 * 16] + 1) >> 2) - ((in_low[14 * 16] + 1) >> 2);
242 0 : step1[7] = ((in_low[0 * 16] + 1) >> 2) - ((in_low[15 * 16] + 1) >> 2);
243 0 : in_low++;
244 : }
245 : // Work on the first eight values; fdct8(input, even_results);
246 : {
247 : tran_high_t s0, s1, s2, s3, s4, s5, s6, s7; // canbe16
248 : tran_high_t t0, t1, t2, t3; // needs32
249 : tran_high_t x0, x1, x2, x3; // canbe16
250 :
251 : // stage 1
252 0 : s0 = in_high[0] + in_high[7];
253 0 : s1 = in_high[1] + in_high[6];
254 0 : s2 = in_high[2] + in_high[5];
255 0 : s3 = in_high[3] + in_high[4];
256 0 : s4 = in_high[3] - in_high[4];
257 0 : s5 = in_high[2] - in_high[5];
258 0 : s6 = in_high[1] - in_high[6];
259 0 : s7 = in_high[0] - in_high[7];
260 :
261 : // fdct4(step, step);
262 0 : x0 = s0 + s3;
263 0 : x1 = s1 + s2;
264 0 : x2 = s1 - s2;
265 0 : x3 = s0 - s3;
266 0 : t0 = (x0 + x1) * cospi_16_64;
267 0 : t1 = (x0 - x1) * cospi_16_64;
268 0 : t2 = x3 * cospi_8_64 + x2 * cospi_24_64;
269 0 : t3 = x3 * cospi_24_64 - x2 * cospi_8_64;
270 0 : out[0] = (tran_low_t)fdct_round_shift(t0);
271 0 : out[4] = (tran_low_t)fdct_round_shift(t2);
272 0 : out[8] = (tran_low_t)fdct_round_shift(t1);
273 0 : out[12] = (tran_low_t)fdct_round_shift(t3);
274 :
275 : // Stage 2
276 0 : t0 = (s6 - s5) * cospi_16_64;
277 0 : t1 = (s6 + s5) * cospi_16_64;
278 0 : t2 = fdct_round_shift(t0);
279 0 : t3 = fdct_round_shift(t1);
280 :
281 : // Stage 3
282 0 : x0 = s4 + t2;
283 0 : x1 = s4 - t2;
284 0 : x2 = s7 - t3;
285 0 : x3 = s7 + t3;
286 :
287 : // Stage 4
288 0 : t0 = x0 * cospi_28_64 + x3 * cospi_4_64;
289 0 : t1 = x1 * cospi_12_64 + x2 * cospi_20_64;
290 0 : t2 = x2 * cospi_12_64 + x1 * -cospi_20_64;
291 0 : t3 = x3 * cospi_28_64 + x0 * -cospi_4_64;
292 0 : out[2] = (tran_low_t)fdct_round_shift(t0);
293 0 : out[6] = (tran_low_t)fdct_round_shift(t2);
294 0 : out[10] = (tran_low_t)fdct_round_shift(t1);
295 0 : out[14] = (tran_low_t)fdct_round_shift(t3);
296 : }
297 : // Work on the next eight values; step1 -> odd_results
298 : {
299 : // step 2
300 0 : temp1 = (step1[5] - step1[2]) * cospi_16_64;
301 0 : temp2 = (step1[4] - step1[3]) * cospi_16_64;
302 0 : step2[2] = fdct_round_shift(temp1);
303 0 : step2[3] = fdct_round_shift(temp2);
304 0 : temp1 = (step1[4] + step1[3]) * cospi_16_64;
305 0 : temp2 = (step1[5] + step1[2]) * cospi_16_64;
306 0 : step2[4] = fdct_round_shift(temp1);
307 0 : step2[5] = fdct_round_shift(temp2);
308 : // step 3
309 0 : step3[0] = step1[0] + step2[3];
310 0 : step3[1] = step1[1] + step2[2];
311 0 : step3[2] = step1[1] - step2[2];
312 0 : step3[3] = step1[0] - step2[3];
313 0 : step3[4] = step1[7] - step2[4];
314 0 : step3[5] = step1[6] - step2[5];
315 0 : step3[6] = step1[6] + step2[5];
316 0 : step3[7] = step1[7] + step2[4];
317 : // step 4
318 0 : temp1 = step3[1] * -cospi_8_64 + step3[6] * cospi_24_64;
319 0 : temp2 = step3[2] * cospi_24_64 + step3[5] * cospi_8_64;
320 0 : step2[1] = fdct_round_shift(temp1);
321 0 : step2[2] = fdct_round_shift(temp2);
322 0 : temp1 = step3[2] * cospi_8_64 - step3[5] * cospi_24_64;
323 0 : temp2 = step3[1] * cospi_24_64 + step3[6] * cospi_8_64;
324 0 : step2[5] = fdct_round_shift(temp1);
325 0 : step2[6] = fdct_round_shift(temp2);
326 : // step 5
327 0 : step1[0] = step3[0] + step2[1];
328 0 : step1[1] = step3[0] - step2[1];
329 0 : step1[2] = step3[3] + step2[2];
330 0 : step1[3] = step3[3] - step2[2];
331 0 : step1[4] = step3[4] - step2[5];
332 0 : step1[5] = step3[4] + step2[5];
333 0 : step1[6] = step3[7] - step2[6];
334 0 : step1[7] = step3[7] + step2[6];
335 : // step 6
336 0 : temp1 = step1[0] * cospi_30_64 + step1[7] * cospi_2_64;
337 0 : temp2 = step1[1] * cospi_14_64 + step1[6] * cospi_18_64;
338 0 : out[1] = (tran_low_t)fdct_round_shift(temp1);
339 0 : out[9] = (tran_low_t)fdct_round_shift(temp2);
340 0 : temp1 = step1[2] * cospi_22_64 + step1[5] * cospi_10_64;
341 0 : temp2 = step1[3] * cospi_6_64 + step1[4] * cospi_26_64;
342 0 : out[5] = (tran_low_t)fdct_round_shift(temp1);
343 0 : out[13] = (tran_low_t)fdct_round_shift(temp2);
344 0 : temp1 = step1[3] * -cospi_26_64 + step1[4] * cospi_6_64;
345 0 : temp2 = step1[2] * -cospi_10_64 + step1[5] * cospi_22_64;
346 0 : out[3] = (tran_low_t)fdct_round_shift(temp1);
347 0 : out[11] = (tran_low_t)fdct_round_shift(temp2);
348 0 : temp1 = step1[1] * -cospi_18_64 + step1[6] * cospi_14_64;
349 0 : temp2 = step1[0] * -cospi_2_64 + step1[7] * cospi_30_64;
350 0 : out[7] = (tran_low_t)fdct_round_shift(temp1);
351 0 : out[15] = (tran_low_t)fdct_round_shift(temp2);
352 : }
353 : // Do next column (which is a transposed row in second/horizontal pass)
354 0 : input++;
355 0 : out += 16;
356 : }
357 : // Setup in/out for next pass.
358 0 : in_low = intermediate;
359 0 : out = output;
360 : }
361 0 : }
362 :
363 0 : void vpx_fdct16x16_1_c(const int16_t *input, tran_low_t *output, int stride) {
364 : int r, c;
365 0 : int sum = 0;
366 0 : for (r = 0; r < 16; ++r)
367 0 : for (c = 0; c < 16; ++c) sum += input[r * stride + c];
368 :
369 0 : output[0] = (tran_low_t)(sum >> 1);
370 0 : }
371 :
372 0 : static INLINE tran_high_t dct_32_round(tran_high_t input) {
373 0 : tran_high_t rv = ROUND_POWER_OF_TWO(input, DCT_CONST_BITS);
374 : // TODO(debargha, peter.derivaz): Find new bounds for this assert,
375 : // and make the bounds consts.
376 : // assert(-131072 <= rv && rv <= 131071);
377 0 : return rv;
378 : }
379 :
380 0 : static INLINE tran_high_t half_round_shift(tran_high_t input) {
381 0 : tran_high_t rv = (input + 1 + (input < 0)) >> 2;
382 0 : return rv;
383 : }
384 :
385 0 : void vpx_fdct32(const tran_high_t *input, tran_high_t *output, int round) {
386 : tran_high_t step[32];
387 : // Stage 1
388 0 : step[0] = input[0] + input[(32 - 1)];
389 0 : step[1] = input[1] + input[(32 - 2)];
390 0 : step[2] = input[2] + input[(32 - 3)];
391 0 : step[3] = input[3] + input[(32 - 4)];
392 0 : step[4] = input[4] + input[(32 - 5)];
393 0 : step[5] = input[5] + input[(32 - 6)];
394 0 : step[6] = input[6] + input[(32 - 7)];
395 0 : step[7] = input[7] + input[(32 - 8)];
396 0 : step[8] = input[8] + input[(32 - 9)];
397 0 : step[9] = input[9] + input[(32 - 10)];
398 0 : step[10] = input[10] + input[(32 - 11)];
399 0 : step[11] = input[11] + input[(32 - 12)];
400 0 : step[12] = input[12] + input[(32 - 13)];
401 0 : step[13] = input[13] + input[(32 - 14)];
402 0 : step[14] = input[14] + input[(32 - 15)];
403 0 : step[15] = input[15] + input[(32 - 16)];
404 0 : step[16] = -input[16] + input[(32 - 17)];
405 0 : step[17] = -input[17] + input[(32 - 18)];
406 0 : step[18] = -input[18] + input[(32 - 19)];
407 0 : step[19] = -input[19] + input[(32 - 20)];
408 0 : step[20] = -input[20] + input[(32 - 21)];
409 0 : step[21] = -input[21] + input[(32 - 22)];
410 0 : step[22] = -input[22] + input[(32 - 23)];
411 0 : step[23] = -input[23] + input[(32 - 24)];
412 0 : step[24] = -input[24] + input[(32 - 25)];
413 0 : step[25] = -input[25] + input[(32 - 26)];
414 0 : step[26] = -input[26] + input[(32 - 27)];
415 0 : step[27] = -input[27] + input[(32 - 28)];
416 0 : step[28] = -input[28] + input[(32 - 29)];
417 0 : step[29] = -input[29] + input[(32 - 30)];
418 0 : step[30] = -input[30] + input[(32 - 31)];
419 0 : step[31] = -input[31] + input[(32 - 32)];
420 :
421 : // Stage 2
422 0 : output[0] = step[0] + step[16 - 1];
423 0 : output[1] = step[1] + step[16 - 2];
424 0 : output[2] = step[2] + step[16 - 3];
425 0 : output[3] = step[3] + step[16 - 4];
426 0 : output[4] = step[4] + step[16 - 5];
427 0 : output[5] = step[5] + step[16 - 6];
428 0 : output[6] = step[6] + step[16 - 7];
429 0 : output[7] = step[7] + step[16 - 8];
430 0 : output[8] = -step[8] + step[16 - 9];
431 0 : output[9] = -step[9] + step[16 - 10];
432 0 : output[10] = -step[10] + step[16 - 11];
433 0 : output[11] = -step[11] + step[16 - 12];
434 0 : output[12] = -step[12] + step[16 - 13];
435 0 : output[13] = -step[13] + step[16 - 14];
436 0 : output[14] = -step[14] + step[16 - 15];
437 0 : output[15] = -step[15] + step[16 - 16];
438 :
439 0 : output[16] = step[16];
440 0 : output[17] = step[17];
441 0 : output[18] = step[18];
442 0 : output[19] = step[19];
443 :
444 0 : output[20] = dct_32_round((-step[20] + step[27]) * cospi_16_64);
445 0 : output[21] = dct_32_round((-step[21] + step[26]) * cospi_16_64);
446 0 : output[22] = dct_32_round((-step[22] + step[25]) * cospi_16_64);
447 0 : output[23] = dct_32_round((-step[23] + step[24]) * cospi_16_64);
448 :
449 0 : output[24] = dct_32_round((step[24] + step[23]) * cospi_16_64);
450 0 : output[25] = dct_32_round((step[25] + step[22]) * cospi_16_64);
451 0 : output[26] = dct_32_round((step[26] + step[21]) * cospi_16_64);
452 0 : output[27] = dct_32_round((step[27] + step[20]) * cospi_16_64);
453 :
454 0 : output[28] = step[28];
455 0 : output[29] = step[29];
456 0 : output[30] = step[30];
457 0 : output[31] = step[31];
458 :
459 : // dump the magnitude by 4, hence the intermediate values are within
460 : // the range of 16 bits.
461 0 : if (round) {
462 0 : output[0] = half_round_shift(output[0]);
463 0 : output[1] = half_round_shift(output[1]);
464 0 : output[2] = half_round_shift(output[2]);
465 0 : output[3] = half_round_shift(output[3]);
466 0 : output[4] = half_round_shift(output[4]);
467 0 : output[5] = half_round_shift(output[5]);
468 0 : output[6] = half_round_shift(output[6]);
469 0 : output[7] = half_round_shift(output[7]);
470 0 : output[8] = half_round_shift(output[8]);
471 0 : output[9] = half_round_shift(output[9]);
472 0 : output[10] = half_round_shift(output[10]);
473 0 : output[11] = half_round_shift(output[11]);
474 0 : output[12] = half_round_shift(output[12]);
475 0 : output[13] = half_round_shift(output[13]);
476 0 : output[14] = half_round_shift(output[14]);
477 0 : output[15] = half_round_shift(output[15]);
478 :
479 0 : output[16] = half_round_shift(output[16]);
480 0 : output[17] = half_round_shift(output[17]);
481 0 : output[18] = half_round_shift(output[18]);
482 0 : output[19] = half_round_shift(output[19]);
483 0 : output[20] = half_round_shift(output[20]);
484 0 : output[21] = half_round_shift(output[21]);
485 0 : output[22] = half_round_shift(output[22]);
486 0 : output[23] = half_round_shift(output[23]);
487 0 : output[24] = half_round_shift(output[24]);
488 0 : output[25] = half_round_shift(output[25]);
489 0 : output[26] = half_round_shift(output[26]);
490 0 : output[27] = half_round_shift(output[27]);
491 0 : output[28] = half_round_shift(output[28]);
492 0 : output[29] = half_round_shift(output[29]);
493 0 : output[30] = half_round_shift(output[30]);
494 0 : output[31] = half_round_shift(output[31]);
495 : }
496 :
497 : // Stage 3
498 0 : step[0] = output[0] + output[(8 - 1)];
499 0 : step[1] = output[1] + output[(8 - 2)];
500 0 : step[2] = output[2] + output[(8 - 3)];
501 0 : step[3] = output[3] + output[(8 - 4)];
502 0 : step[4] = -output[4] + output[(8 - 5)];
503 0 : step[5] = -output[5] + output[(8 - 6)];
504 0 : step[6] = -output[6] + output[(8 - 7)];
505 0 : step[7] = -output[7] + output[(8 - 8)];
506 0 : step[8] = output[8];
507 0 : step[9] = output[9];
508 0 : step[10] = dct_32_round((-output[10] + output[13]) * cospi_16_64);
509 0 : step[11] = dct_32_round((-output[11] + output[12]) * cospi_16_64);
510 0 : step[12] = dct_32_round((output[12] + output[11]) * cospi_16_64);
511 0 : step[13] = dct_32_round((output[13] + output[10]) * cospi_16_64);
512 0 : step[14] = output[14];
513 0 : step[15] = output[15];
514 :
515 0 : step[16] = output[16] + output[23];
516 0 : step[17] = output[17] + output[22];
517 0 : step[18] = output[18] + output[21];
518 0 : step[19] = output[19] + output[20];
519 0 : step[20] = -output[20] + output[19];
520 0 : step[21] = -output[21] + output[18];
521 0 : step[22] = -output[22] + output[17];
522 0 : step[23] = -output[23] + output[16];
523 0 : step[24] = -output[24] + output[31];
524 0 : step[25] = -output[25] + output[30];
525 0 : step[26] = -output[26] + output[29];
526 0 : step[27] = -output[27] + output[28];
527 0 : step[28] = output[28] + output[27];
528 0 : step[29] = output[29] + output[26];
529 0 : step[30] = output[30] + output[25];
530 0 : step[31] = output[31] + output[24];
531 :
532 : // Stage 4
533 0 : output[0] = step[0] + step[3];
534 0 : output[1] = step[1] + step[2];
535 0 : output[2] = -step[2] + step[1];
536 0 : output[3] = -step[3] + step[0];
537 0 : output[4] = step[4];
538 0 : output[5] = dct_32_round((-step[5] + step[6]) * cospi_16_64);
539 0 : output[6] = dct_32_round((step[6] + step[5]) * cospi_16_64);
540 0 : output[7] = step[7];
541 0 : output[8] = step[8] + step[11];
542 0 : output[9] = step[9] + step[10];
543 0 : output[10] = -step[10] + step[9];
544 0 : output[11] = -step[11] + step[8];
545 0 : output[12] = -step[12] + step[15];
546 0 : output[13] = -step[13] + step[14];
547 0 : output[14] = step[14] + step[13];
548 0 : output[15] = step[15] + step[12];
549 :
550 0 : output[16] = step[16];
551 0 : output[17] = step[17];
552 0 : output[18] = dct_32_round(step[18] * -cospi_8_64 + step[29] * cospi_24_64);
553 0 : output[19] = dct_32_round(step[19] * -cospi_8_64 + step[28] * cospi_24_64);
554 0 : output[20] = dct_32_round(step[20] * -cospi_24_64 + step[27] * -cospi_8_64);
555 0 : output[21] = dct_32_round(step[21] * -cospi_24_64 + step[26] * -cospi_8_64);
556 0 : output[22] = step[22];
557 0 : output[23] = step[23];
558 0 : output[24] = step[24];
559 0 : output[25] = step[25];
560 0 : output[26] = dct_32_round(step[26] * cospi_24_64 + step[21] * -cospi_8_64);
561 0 : output[27] = dct_32_round(step[27] * cospi_24_64 + step[20] * -cospi_8_64);
562 0 : output[28] = dct_32_round(step[28] * cospi_8_64 + step[19] * cospi_24_64);
563 0 : output[29] = dct_32_round(step[29] * cospi_8_64 + step[18] * cospi_24_64);
564 0 : output[30] = step[30];
565 0 : output[31] = step[31];
566 :
567 : // Stage 5
568 0 : step[0] = dct_32_round((output[0] + output[1]) * cospi_16_64);
569 0 : step[1] = dct_32_round((-output[1] + output[0]) * cospi_16_64);
570 0 : step[2] = dct_32_round(output[2] * cospi_24_64 + output[3] * cospi_8_64);
571 0 : step[3] = dct_32_round(output[3] * cospi_24_64 - output[2] * cospi_8_64);
572 0 : step[4] = output[4] + output[5];
573 0 : step[5] = -output[5] + output[4];
574 0 : step[6] = -output[6] + output[7];
575 0 : step[7] = output[7] + output[6];
576 0 : step[8] = output[8];
577 0 : step[9] = dct_32_round(output[9] * -cospi_8_64 + output[14] * cospi_24_64);
578 0 : step[10] = dct_32_round(output[10] * -cospi_24_64 + output[13] * -cospi_8_64);
579 0 : step[11] = output[11];
580 0 : step[12] = output[12];
581 0 : step[13] = dct_32_round(output[13] * cospi_24_64 + output[10] * -cospi_8_64);
582 0 : step[14] = dct_32_round(output[14] * cospi_8_64 + output[9] * cospi_24_64);
583 0 : step[15] = output[15];
584 :
585 0 : step[16] = output[16] + output[19];
586 0 : step[17] = output[17] + output[18];
587 0 : step[18] = -output[18] + output[17];
588 0 : step[19] = -output[19] + output[16];
589 0 : step[20] = -output[20] + output[23];
590 0 : step[21] = -output[21] + output[22];
591 0 : step[22] = output[22] + output[21];
592 0 : step[23] = output[23] + output[20];
593 0 : step[24] = output[24] + output[27];
594 0 : step[25] = output[25] + output[26];
595 0 : step[26] = -output[26] + output[25];
596 0 : step[27] = -output[27] + output[24];
597 0 : step[28] = -output[28] + output[31];
598 0 : step[29] = -output[29] + output[30];
599 0 : step[30] = output[30] + output[29];
600 0 : step[31] = output[31] + output[28];
601 :
602 : // Stage 6
603 0 : output[0] = step[0];
604 0 : output[1] = step[1];
605 0 : output[2] = step[2];
606 0 : output[3] = step[3];
607 0 : output[4] = dct_32_round(step[4] * cospi_28_64 + step[7] * cospi_4_64);
608 0 : output[5] = dct_32_round(step[5] * cospi_12_64 + step[6] * cospi_20_64);
609 0 : output[6] = dct_32_round(step[6] * cospi_12_64 + step[5] * -cospi_20_64);
610 0 : output[7] = dct_32_round(step[7] * cospi_28_64 + step[4] * -cospi_4_64);
611 0 : output[8] = step[8] + step[9];
612 0 : output[9] = -step[9] + step[8];
613 0 : output[10] = -step[10] + step[11];
614 0 : output[11] = step[11] + step[10];
615 0 : output[12] = step[12] + step[13];
616 0 : output[13] = -step[13] + step[12];
617 0 : output[14] = -step[14] + step[15];
618 0 : output[15] = step[15] + step[14];
619 :
620 0 : output[16] = step[16];
621 0 : output[17] = dct_32_round(step[17] * -cospi_4_64 + step[30] * cospi_28_64);
622 0 : output[18] = dct_32_round(step[18] * -cospi_28_64 + step[29] * -cospi_4_64);
623 0 : output[19] = step[19];
624 0 : output[20] = step[20];
625 0 : output[21] = dct_32_round(step[21] * -cospi_20_64 + step[26] * cospi_12_64);
626 0 : output[22] = dct_32_round(step[22] * -cospi_12_64 + step[25] * -cospi_20_64);
627 0 : output[23] = step[23];
628 0 : output[24] = step[24];
629 0 : output[25] = dct_32_round(step[25] * cospi_12_64 + step[22] * -cospi_20_64);
630 0 : output[26] = dct_32_round(step[26] * cospi_20_64 + step[21] * cospi_12_64);
631 0 : output[27] = step[27];
632 0 : output[28] = step[28];
633 0 : output[29] = dct_32_round(step[29] * cospi_28_64 + step[18] * -cospi_4_64);
634 0 : output[30] = dct_32_round(step[30] * cospi_4_64 + step[17] * cospi_28_64);
635 0 : output[31] = step[31];
636 :
637 : // Stage 7
638 0 : step[0] = output[0];
639 0 : step[1] = output[1];
640 0 : step[2] = output[2];
641 0 : step[3] = output[3];
642 0 : step[4] = output[4];
643 0 : step[5] = output[5];
644 0 : step[6] = output[6];
645 0 : step[7] = output[7];
646 0 : step[8] = dct_32_round(output[8] * cospi_30_64 + output[15] * cospi_2_64);
647 0 : step[9] = dct_32_round(output[9] * cospi_14_64 + output[14] * cospi_18_64);
648 0 : step[10] = dct_32_round(output[10] * cospi_22_64 + output[13] * cospi_10_64);
649 0 : step[11] = dct_32_round(output[11] * cospi_6_64 + output[12] * cospi_26_64);
650 0 : step[12] = dct_32_round(output[12] * cospi_6_64 + output[11] * -cospi_26_64);
651 0 : step[13] = dct_32_round(output[13] * cospi_22_64 + output[10] * -cospi_10_64);
652 0 : step[14] = dct_32_round(output[14] * cospi_14_64 + output[9] * -cospi_18_64);
653 0 : step[15] = dct_32_round(output[15] * cospi_30_64 + output[8] * -cospi_2_64);
654 :
655 0 : step[16] = output[16] + output[17];
656 0 : step[17] = -output[17] + output[16];
657 0 : step[18] = -output[18] + output[19];
658 0 : step[19] = output[19] + output[18];
659 0 : step[20] = output[20] + output[21];
660 0 : step[21] = -output[21] + output[20];
661 0 : step[22] = -output[22] + output[23];
662 0 : step[23] = output[23] + output[22];
663 0 : step[24] = output[24] + output[25];
664 0 : step[25] = -output[25] + output[24];
665 0 : step[26] = -output[26] + output[27];
666 0 : step[27] = output[27] + output[26];
667 0 : step[28] = output[28] + output[29];
668 0 : step[29] = -output[29] + output[28];
669 0 : step[30] = -output[30] + output[31];
670 0 : step[31] = output[31] + output[30];
671 :
672 : // Final stage --- outputs indices are bit-reversed.
673 0 : output[0] = step[0];
674 0 : output[16] = step[1];
675 0 : output[8] = step[2];
676 0 : output[24] = step[3];
677 0 : output[4] = step[4];
678 0 : output[20] = step[5];
679 0 : output[12] = step[6];
680 0 : output[28] = step[7];
681 0 : output[2] = step[8];
682 0 : output[18] = step[9];
683 0 : output[10] = step[10];
684 0 : output[26] = step[11];
685 0 : output[6] = step[12];
686 0 : output[22] = step[13];
687 0 : output[14] = step[14];
688 0 : output[30] = step[15];
689 :
690 0 : output[1] = dct_32_round(step[16] * cospi_31_64 + step[31] * cospi_1_64);
691 0 : output[17] = dct_32_round(step[17] * cospi_15_64 + step[30] * cospi_17_64);
692 0 : output[9] = dct_32_round(step[18] * cospi_23_64 + step[29] * cospi_9_64);
693 0 : output[25] = dct_32_round(step[19] * cospi_7_64 + step[28] * cospi_25_64);
694 0 : output[5] = dct_32_round(step[20] * cospi_27_64 + step[27] * cospi_5_64);
695 0 : output[21] = dct_32_round(step[21] * cospi_11_64 + step[26] * cospi_21_64);
696 0 : output[13] = dct_32_round(step[22] * cospi_19_64 + step[25] * cospi_13_64);
697 0 : output[29] = dct_32_round(step[23] * cospi_3_64 + step[24] * cospi_29_64);
698 0 : output[3] = dct_32_round(step[24] * cospi_3_64 + step[23] * -cospi_29_64);
699 0 : output[19] = dct_32_round(step[25] * cospi_19_64 + step[22] * -cospi_13_64);
700 0 : output[11] = dct_32_round(step[26] * cospi_11_64 + step[21] * -cospi_21_64);
701 0 : output[27] = dct_32_round(step[27] * cospi_27_64 + step[20] * -cospi_5_64);
702 0 : output[7] = dct_32_round(step[28] * cospi_7_64 + step[19] * -cospi_25_64);
703 0 : output[23] = dct_32_round(step[29] * cospi_23_64 + step[18] * -cospi_9_64);
704 0 : output[15] = dct_32_round(step[30] * cospi_15_64 + step[17] * -cospi_17_64);
705 0 : output[31] = dct_32_round(step[31] * cospi_31_64 + step[16] * -cospi_1_64);
706 0 : }
707 :
708 0 : void vpx_fdct32x32_c(const int16_t *input, tran_low_t *out, int stride) {
709 : int i, j;
710 : tran_high_t output[32 * 32];
711 :
712 : // Columns
713 0 : for (i = 0; i < 32; ++i) {
714 : tran_high_t temp_in[32], temp_out[32];
715 0 : for (j = 0; j < 32; ++j) temp_in[j] = input[j * stride + i] * 4;
716 0 : vpx_fdct32(temp_in, temp_out, 0);
717 0 : for (j = 0; j < 32; ++j)
718 0 : output[j * 32 + i] = (temp_out[j] + 1 + (temp_out[j] > 0)) >> 2;
719 : }
720 :
721 : // Rows
722 0 : for (i = 0; i < 32; ++i) {
723 : tran_high_t temp_in[32], temp_out[32];
724 0 : for (j = 0; j < 32; ++j) temp_in[j] = output[j + i * 32];
725 0 : vpx_fdct32(temp_in, temp_out, 0);
726 0 : for (j = 0; j < 32; ++j)
727 0 : out[j + i * 32] =
728 0 : (tran_low_t)((temp_out[j] + 1 + (temp_out[j] < 0)) >> 2);
729 : }
730 0 : }
731 :
732 : // Note that although we use dct_32_round in dct32 computation flow,
733 : // this 2d fdct32x32 for rate-distortion optimization loop is operating
734 : // within 16 bits precision.
735 0 : void vpx_fdct32x32_rd_c(const int16_t *input, tran_low_t *out, int stride) {
736 : int i, j;
737 : tran_high_t output[32 * 32];
738 :
739 : // Columns
740 0 : for (i = 0; i < 32; ++i) {
741 : tran_high_t temp_in[32], temp_out[32];
742 0 : for (j = 0; j < 32; ++j) temp_in[j] = input[j * stride + i] * 4;
743 0 : vpx_fdct32(temp_in, temp_out, 0);
744 0 : for (j = 0; j < 32; ++j)
745 : // TODO(cd): see quality impact of only doing
746 : // output[j * 32 + i] = (temp_out[j] + 1) >> 2;
747 : // PS: also change code in vpx_dsp/x86/vpx_dct_sse2.c
748 0 : output[j * 32 + i] = (temp_out[j] + 1 + (temp_out[j] > 0)) >> 2;
749 : }
750 :
751 : // Rows
752 0 : for (i = 0; i < 32; ++i) {
753 : tran_high_t temp_in[32], temp_out[32];
754 0 : for (j = 0; j < 32; ++j) temp_in[j] = output[j + i * 32];
755 0 : vpx_fdct32(temp_in, temp_out, 1);
756 0 : for (j = 0; j < 32; ++j) out[j + i * 32] = (tran_low_t)temp_out[j];
757 : }
758 0 : }
759 :
760 0 : void vpx_fdct32x32_1_c(const int16_t *input, tran_low_t *output, int stride) {
761 : int r, c;
762 0 : int sum = 0;
763 0 : for (r = 0; r < 32; ++r)
764 0 : for (c = 0; c < 32; ++c) sum += input[r * stride + c];
765 :
766 0 : output[0] = (tran_low_t)(sum >> 3);
767 0 : }
768 :
769 : #if CONFIG_VP9_HIGHBITDEPTH
770 : void vpx_highbd_fdct4x4_c(const int16_t *input, tran_low_t *output,
771 : int stride) {
772 : vpx_fdct4x4_c(input, output, stride);
773 : }
774 :
775 : void vpx_highbd_fdct8x8_c(const int16_t *input, tran_low_t *final_output,
776 : int stride) {
777 : vpx_fdct8x8_c(input, final_output, stride);
778 : }
779 :
780 : void vpx_highbd_fdct8x8_1_c(const int16_t *input, tran_low_t *final_output,
781 : int stride) {
782 : vpx_fdct8x8_1_c(input, final_output, stride);
783 : }
784 :
785 : void vpx_highbd_fdct16x16_c(const int16_t *input, tran_low_t *output,
786 : int stride) {
787 : vpx_fdct16x16_c(input, output, stride);
788 : }
789 :
790 : void vpx_highbd_fdct16x16_1_c(const int16_t *input, tran_low_t *output,
791 : int stride) {
792 : vpx_fdct16x16_1_c(input, output, stride);
793 : }
794 :
795 : void vpx_highbd_fdct32x32_c(const int16_t *input, tran_low_t *out, int stride) {
796 : vpx_fdct32x32_c(input, out, stride);
797 : }
798 :
799 : void vpx_highbd_fdct32x32_rd_c(const int16_t *input, tran_low_t *out,
800 : int stride) {
801 : vpx_fdct32x32_rd_c(input, out, stride);
802 : }
803 :
804 : void vpx_highbd_fdct32x32_1_c(const int16_t *input, tran_low_t *out,
805 : int stride) {
806 : vpx_fdct32x32_1_c(input, out, stride);
807 : }
808 : #endif // CONFIG_VP9_HIGHBITDEPTH
|