2 * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
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.
11 #include "vpx_dsp/fwd_txfm.h"
13 void vp9_fdct4x4_c(const int16_t *input, tran_low_t *output, int stride) {
14 // The 2D transform is done with two passes which are actually pretty
15 // similar. In the first one, we transform the columns and transpose
16 // the results. In the second one, we transform the rows. To achieve that,
17 // as the first pass results are transposed, we transpose the columns (that
18 // is the transposed rows) and transpose the results (so that it goes back
19 // in normal/row positions).
21 // We need an intermediate buffer between passes.
22 tran_low_t intermediate[4 * 4];
23 const int16_t *in_pass0 = input;
24 const tran_low_t *in = NULL;
25 tran_low_t *out = intermediate;
26 // Do the two transform/transpose passes
27 for (pass = 0; pass < 2; ++pass) {
28 tran_high_t input[4]; // canbe16
29 tran_high_t step[4]; // canbe16
30 tran_high_t temp1, temp2; // needs32
32 for (i = 0; i < 4; ++i) {
35 input[0] = in_pass0[0 * stride] * 16;
36 input[1] = in_pass0[1 * stride] * 16;
37 input[2] = in_pass0[2 * stride] * 16;
38 input[3] = in_pass0[3 * stride] * 16;
39 if (i == 0 && input[0]) {
49 step[0] = input[0] + input[3];
50 step[1] = input[1] + input[2];
51 step[2] = input[1] - input[2];
52 step[3] = input[0] - input[3];
53 temp1 = (step[0] + step[1]) * cospi_16_64;
54 temp2 = (step[0] - step[1]) * cospi_16_64;
55 out[0] = (tran_low_t)fdct_round_shift(temp1);
56 out[2] = (tran_low_t)fdct_round_shift(temp2);
57 temp1 = step[2] * cospi_24_64 + step[3] * cospi_8_64;
58 temp2 = -step[2] * cospi_8_64 + step[3] * cospi_24_64;
59 out[1] = (tran_low_t)fdct_round_shift(temp1);
60 out[3] = (tran_low_t)fdct_round_shift(temp2);
61 // Do next column (which is a transposed row in second/horizontal pass)
66 // Setup in/out for next pass.
73 for (i = 0; i < 4; ++i) {
74 for (j = 0; j < 4; ++j)
75 output[j + i * 4] = (output[j + i * 4] + 1) >> 2;
80 void vp9_fdct8x8_c(const int16_t *input, tran_low_t *final_output, int stride) {
82 tran_low_t intermediate[64];
84 tran_low_t *output = intermediate;
85 const tran_low_t *in = NULL;
88 for (pass = 0; pass < 2; ++pass) {
89 tran_high_t s0, s1, s2, s3, s4, s5, s6, s7; // canbe16
90 tran_high_t t0, t1, t2, t3; // needs32
91 tran_high_t x0, x1, x2, x3; // canbe16
94 for (i = 0; i < 8; i++) {
97 s0 = (input[0 * stride] + input[7 * stride]) * 4;
98 s1 = (input[1 * stride] + input[6 * stride]) * 4;
99 s2 = (input[2 * stride] + input[5 * stride]) * 4;
100 s3 = (input[3 * stride] + input[4 * stride]) * 4;
101 s4 = (input[3 * stride] - input[4 * stride]) * 4;
102 s5 = (input[2 * stride] - input[5 * stride]) * 4;
103 s6 = (input[1 * stride] - input[6 * stride]) * 4;
104 s7 = (input[0 * stride] - input[7 * stride]) * 4;
107 s0 = in[0 * 8] + in[7 * 8];
108 s1 = in[1 * 8] + in[6 * 8];
109 s2 = in[2 * 8] + in[5 * 8];
110 s3 = in[3 * 8] + in[4 * 8];
111 s4 = in[3 * 8] - in[4 * 8];
112 s5 = in[2 * 8] - in[5 * 8];
113 s6 = in[1 * 8] - in[6 * 8];
114 s7 = in[0 * 8] - in[7 * 8];
118 // fdct4(step, step);
123 t0 = (x0 + x1) * cospi_16_64;
124 t1 = (x0 - x1) * cospi_16_64;
125 t2 = x2 * cospi_24_64 + x3 * cospi_8_64;
126 t3 = -x2 * cospi_8_64 + x3 * cospi_24_64;
127 output[0] = (tran_low_t)fdct_round_shift(t0);
128 output[2] = (tran_low_t)fdct_round_shift(t2);
129 output[4] = (tran_low_t)fdct_round_shift(t1);
130 output[6] = (tran_low_t)fdct_round_shift(t3);
133 t0 = (s6 - s5) * cospi_16_64;
134 t1 = (s6 + s5) * cospi_16_64;
135 t2 = fdct_round_shift(t0);
136 t3 = fdct_round_shift(t1);
145 t0 = x0 * cospi_28_64 + x3 * cospi_4_64;
146 t1 = x1 * cospi_12_64 + x2 * cospi_20_64;
147 t2 = x2 * cospi_12_64 + x1 * -cospi_20_64;
148 t3 = x3 * cospi_28_64 + x0 * -cospi_4_64;
149 output[1] = (tran_low_t)fdct_round_shift(t0);
150 output[3] = (tran_low_t)fdct_round_shift(t2);
151 output[5] = (tran_low_t)fdct_round_shift(t1);
152 output[7] = (tran_low_t)fdct_round_shift(t3);
156 output = final_output;
160 for (i = 0; i < 8; ++i) {
161 for (j = 0; j < 8; ++j)
162 final_output[j + i * 8] /= 2;
166 void vp9_fdct16x16_c(const int16_t *input, tran_low_t *output, int stride) {
167 // The 2D transform is done with two passes which are actually pretty
168 // similar. In the first one, we transform the columns and transpose
169 // the results. In the second one, we transform the rows. To achieve that,
170 // as the first pass results are transposed, we transpose the columns (that
171 // is the transposed rows) and transpose the results (so that it goes back
172 // in normal/row positions).
174 // We need an intermediate buffer between passes.
175 tran_low_t intermediate[256];
176 const int16_t *in_pass0 = input;
177 const tran_low_t *in = NULL;
178 tran_low_t *out = intermediate;
179 // Do the two transform/transpose passes
180 for (pass = 0; pass < 2; ++pass) {
181 tran_high_t step1[8]; // canbe16
182 tran_high_t step2[8]; // canbe16
183 tran_high_t step3[8]; // canbe16
184 tran_high_t input[8]; // canbe16
185 tran_high_t temp1, temp2; // needs32
187 for (i = 0; i < 16; i++) {
189 // Calculate input for the first 8 results.
190 input[0] = (in_pass0[0 * stride] + in_pass0[15 * stride]) * 4;
191 input[1] = (in_pass0[1 * stride] + in_pass0[14 * stride]) * 4;
192 input[2] = (in_pass0[2 * stride] + in_pass0[13 * stride]) * 4;
193 input[3] = (in_pass0[3 * stride] + in_pass0[12 * stride]) * 4;
194 input[4] = (in_pass0[4 * stride] + in_pass0[11 * stride]) * 4;
195 input[5] = (in_pass0[5 * stride] + in_pass0[10 * stride]) * 4;
196 input[6] = (in_pass0[6 * stride] + in_pass0[ 9 * stride]) * 4;
197 input[7] = (in_pass0[7 * stride] + in_pass0[ 8 * stride]) * 4;
198 // Calculate input for the next 8 results.
199 step1[0] = (in_pass0[7 * stride] - in_pass0[ 8 * stride]) * 4;
200 step1[1] = (in_pass0[6 * stride] - in_pass0[ 9 * stride]) * 4;
201 step1[2] = (in_pass0[5 * stride] - in_pass0[10 * stride]) * 4;
202 step1[3] = (in_pass0[4 * stride] - in_pass0[11 * stride]) * 4;
203 step1[4] = (in_pass0[3 * stride] - in_pass0[12 * stride]) * 4;
204 step1[5] = (in_pass0[2 * stride] - in_pass0[13 * stride]) * 4;
205 step1[6] = (in_pass0[1 * stride] - in_pass0[14 * stride]) * 4;
206 step1[7] = (in_pass0[0 * stride] - in_pass0[15 * stride]) * 4;
208 // Calculate input for the first 8 results.
209 input[0] = ((in[0 * 16] + 1) >> 2) + ((in[15 * 16] + 1) >> 2);
210 input[1] = ((in[1 * 16] + 1) >> 2) + ((in[14 * 16] + 1) >> 2);
211 input[2] = ((in[2 * 16] + 1) >> 2) + ((in[13 * 16] + 1) >> 2);
212 input[3] = ((in[3 * 16] + 1) >> 2) + ((in[12 * 16] + 1) >> 2);
213 input[4] = ((in[4 * 16] + 1) >> 2) + ((in[11 * 16] + 1) >> 2);
214 input[5] = ((in[5 * 16] + 1) >> 2) + ((in[10 * 16] + 1) >> 2);
215 input[6] = ((in[6 * 16] + 1) >> 2) + ((in[ 9 * 16] + 1) >> 2);
216 input[7] = ((in[7 * 16] + 1) >> 2) + ((in[ 8 * 16] + 1) >> 2);
217 // Calculate input for the next 8 results.
218 step1[0] = ((in[7 * 16] + 1) >> 2) - ((in[ 8 * 16] + 1) >> 2);
219 step1[1] = ((in[6 * 16] + 1) >> 2) - ((in[ 9 * 16] + 1) >> 2);
220 step1[2] = ((in[5 * 16] + 1) >> 2) - ((in[10 * 16] + 1) >> 2);
221 step1[3] = ((in[4 * 16] + 1) >> 2) - ((in[11 * 16] + 1) >> 2);
222 step1[4] = ((in[3 * 16] + 1) >> 2) - ((in[12 * 16] + 1) >> 2);
223 step1[5] = ((in[2 * 16] + 1) >> 2) - ((in[13 * 16] + 1) >> 2);
224 step1[6] = ((in[1 * 16] + 1) >> 2) - ((in[14 * 16] + 1) >> 2);
225 step1[7] = ((in[0 * 16] + 1) >> 2) - ((in[15 * 16] + 1) >> 2);
227 // Work on the first eight values; fdct8(input, even_results);
229 tran_high_t s0, s1, s2, s3, s4, s5, s6, s7; // canbe16
230 tran_high_t t0, t1, t2, t3; // needs32
231 tran_high_t x0, x1, x2, x3; // canbe16
234 s0 = input[0] + input[7];
235 s1 = input[1] + input[6];
236 s2 = input[2] + input[5];
237 s3 = input[3] + input[4];
238 s4 = input[3] - input[4];
239 s5 = input[2] - input[5];
240 s6 = input[1] - input[6];
241 s7 = input[0] - input[7];
243 // fdct4(step, step);
248 t0 = (x0 + x1) * cospi_16_64;
249 t1 = (x0 - x1) * cospi_16_64;
250 t2 = x3 * cospi_8_64 + x2 * cospi_24_64;
251 t3 = x3 * cospi_24_64 - x2 * cospi_8_64;
252 out[0] = (tran_low_t)fdct_round_shift(t0);
253 out[4] = (tran_low_t)fdct_round_shift(t2);
254 out[8] = (tran_low_t)fdct_round_shift(t1);
255 out[12] = (tran_low_t)fdct_round_shift(t3);
258 t0 = (s6 - s5) * cospi_16_64;
259 t1 = (s6 + s5) * cospi_16_64;
260 t2 = fdct_round_shift(t0);
261 t3 = fdct_round_shift(t1);
270 t0 = x0 * cospi_28_64 + x3 * cospi_4_64;
271 t1 = x1 * cospi_12_64 + x2 * cospi_20_64;
272 t2 = x2 * cospi_12_64 + x1 * -cospi_20_64;
273 t3 = x3 * cospi_28_64 + x0 * -cospi_4_64;
274 out[2] = (tran_low_t)fdct_round_shift(t0);
275 out[6] = (tran_low_t)fdct_round_shift(t2);
276 out[10] = (tran_low_t)fdct_round_shift(t1);
277 out[14] = (tran_low_t)fdct_round_shift(t3);
279 // Work on the next eight values; step1 -> odd_results
282 temp1 = (step1[5] - step1[2]) * cospi_16_64;
283 temp2 = (step1[4] - step1[3]) * cospi_16_64;
284 step2[2] = fdct_round_shift(temp1);
285 step2[3] = fdct_round_shift(temp2);
286 temp1 = (step1[4] + step1[3]) * cospi_16_64;
287 temp2 = (step1[5] + step1[2]) * cospi_16_64;
288 step2[4] = fdct_round_shift(temp1);
289 step2[5] = fdct_round_shift(temp2);
291 step3[0] = step1[0] + step2[3];
292 step3[1] = step1[1] + step2[2];
293 step3[2] = step1[1] - step2[2];
294 step3[3] = step1[0] - step2[3];
295 step3[4] = step1[7] - step2[4];
296 step3[5] = step1[6] - step2[5];
297 step3[6] = step1[6] + step2[5];
298 step3[7] = step1[7] + step2[4];
300 temp1 = step3[1] * -cospi_8_64 + step3[6] * cospi_24_64;
301 temp2 = step3[2] * cospi_24_64 + step3[5] * cospi_8_64;
302 step2[1] = fdct_round_shift(temp1);
303 step2[2] = fdct_round_shift(temp2);
304 temp1 = step3[2] * cospi_8_64 - step3[5] * cospi_24_64;
305 temp2 = step3[1] * cospi_24_64 + step3[6] * cospi_8_64;
306 step2[5] = fdct_round_shift(temp1);
307 step2[6] = fdct_round_shift(temp2);
309 step1[0] = step3[0] + step2[1];
310 step1[1] = step3[0] - step2[1];
311 step1[2] = step3[3] + step2[2];
312 step1[3] = step3[3] - step2[2];
313 step1[4] = step3[4] - step2[5];
314 step1[5] = step3[4] + step2[5];
315 step1[6] = step3[7] - step2[6];
316 step1[7] = step3[7] + step2[6];
318 temp1 = step1[0] * cospi_30_64 + step1[7] * cospi_2_64;
319 temp2 = step1[1] * cospi_14_64 + step1[6] * cospi_18_64;
320 out[1] = (tran_low_t)fdct_round_shift(temp1);
321 out[9] = (tran_low_t)fdct_round_shift(temp2);
322 temp1 = step1[2] * cospi_22_64 + step1[5] * cospi_10_64;
323 temp2 = step1[3] * cospi_6_64 + step1[4] * cospi_26_64;
324 out[5] = (tran_low_t)fdct_round_shift(temp1);
325 out[13] = (tran_low_t)fdct_round_shift(temp2);
326 temp1 = step1[3] * -cospi_26_64 + step1[4] * cospi_6_64;
327 temp2 = step1[2] * -cospi_10_64 + step1[5] * cospi_22_64;
328 out[3] = (tran_low_t)fdct_round_shift(temp1);
329 out[11] = (tran_low_t)fdct_round_shift(temp2);
330 temp1 = step1[1] * -cospi_18_64 + step1[6] * cospi_14_64;
331 temp2 = step1[0] * -cospi_2_64 + step1[7] * cospi_30_64;
332 out[7] = (tran_low_t)fdct_round_shift(temp1);
333 out[15] = (tran_low_t)fdct_round_shift(temp2);
335 // Do next column (which is a transposed row in second/horizontal pass)
340 // Setup in/out for next pass.
346 #if CONFIG_VP9_HIGHBITDEPTH
347 void vp9_highbd_fdct4x4_c(const int16_t *input, tran_low_t *output,
349 vp9_fdct4x4_c(input, output, stride);
352 void vp9_highbd_fdct8x8_c(const int16_t *input, tran_low_t *final_output,
354 vp9_fdct8x8_c(input, final_output, stride);
357 void vp9_highbd_fdct16x16_c(const int16_t *input, tran_low_t *output,
359 vp9_fdct16x16_c(input, output, stride);
361 #endif // CONFIG_VP9_HIGHBITDEPTH