1 /*
2 * Copyright (c) 2016, Alliance for Open Media. All rights reserved
3 *
4 * This source code is subject to the terms of the BSD 2 Clause License and
5 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
6 * was not distributed with this source code in the LICENSE file, you can
7 * obtain it at www.aomedia.org/license/software. If the Alliance for Open
8 * Media Patent License 1.0 was not distributed with this source code in the
9 * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
10 */
11
12 #include "config/aom_dsp_rtcd.h"
13 #include "config/av1_rtcd.h"
14
15 #include "av1/common/enums.h"
16 #include "av1/common/av1_txfm.h"
17 #include "av1/common/av1_inv_txfm1d.h"
18 #include "av1/common/av1_inv_txfm1d_cfg.h"
19
av1_highbd_iwht4x4_16_add_c(const tran_low_t * input,uint8_t * dest8,int stride,int bd)20 void av1_highbd_iwht4x4_16_add_c(const tran_low_t *input, uint8_t *dest8,
21 int stride, int bd) {
22 /* 4-point reversible, orthonormal inverse Walsh-Hadamard in 3.5 adds,
23 0.5 shifts per pixel. */
24 int i;
25 tran_low_t output[16];
26 tran_low_t a1, b1, c1, d1, e1;
27 const tran_low_t *ip = input;
28 tran_low_t *op = output;
29 uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
30
31 for (i = 0; i < 4; i++) {
32 a1 = ip[0] >> UNIT_QUANT_SHIFT;
33 c1 = ip[1] >> UNIT_QUANT_SHIFT;
34 d1 = ip[2] >> UNIT_QUANT_SHIFT;
35 b1 = ip[3] >> UNIT_QUANT_SHIFT;
36 a1 += c1;
37 d1 -= b1;
38 e1 = (a1 - d1) >> 1;
39 b1 = e1 - b1;
40 c1 = e1 - c1;
41 a1 -= b1;
42 d1 += c1;
43
44 op[0] = a1;
45 op[1] = b1;
46 op[2] = c1;
47 op[3] = d1;
48 ip += 4;
49 op += 4;
50 }
51
52 ip = output;
53 for (i = 0; i < 4; i++) {
54 a1 = ip[4 * 0];
55 c1 = ip[4 * 1];
56 d1 = ip[4 * 2];
57 b1 = ip[4 * 3];
58 a1 += c1;
59 d1 -= b1;
60 e1 = (a1 - d1) >> 1;
61 b1 = e1 - b1;
62 c1 = e1 - c1;
63 a1 -= b1;
64 d1 += c1;
65
66 range_check_value(a1, bd + 1);
67 range_check_value(b1, bd + 1);
68 range_check_value(c1, bd + 1);
69 range_check_value(d1, bd + 1);
70
71 dest[stride * 0] = highbd_clip_pixel_add(dest[stride * 0], a1, bd);
72 dest[stride * 1] = highbd_clip_pixel_add(dest[stride * 1], b1, bd);
73 dest[stride * 2] = highbd_clip_pixel_add(dest[stride * 2], c1, bd);
74 dest[stride * 3] = highbd_clip_pixel_add(dest[stride * 3], d1, bd);
75
76 ip++;
77 dest++;
78 }
79 }
80
av1_highbd_iwht4x4_1_add_c(const tran_low_t * in,uint8_t * dest8,int dest_stride,int bd)81 void av1_highbd_iwht4x4_1_add_c(const tran_low_t *in, uint8_t *dest8,
82 int dest_stride, int bd) {
83 int i;
84 tran_low_t a1, e1;
85 tran_low_t tmp[4];
86 const tran_low_t *ip = in;
87 tran_low_t *op = tmp;
88 uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);
89 (void)bd;
90
91 a1 = ip[0] >> UNIT_QUANT_SHIFT;
92 e1 = a1 >> 1;
93 a1 -= e1;
94 op[0] = a1;
95 op[1] = op[2] = op[3] = e1;
96
97 ip = tmp;
98 for (i = 0; i < 4; i++) {
99 e1 = ip[0] >> 1;
100 a1 = ip[0] - e1;
101 dest[dest_stride * 0] =
102 highbd_clip_pixel_add(dest[dest_stride * 0], a1, bd);
103 dest[dest_stride * 1] =
104 highbd_clip_pixel_add(dest[dest_stride * 1], e1, bd);
105 dest[dest_stride * 2] =
106 highbd_clip_pixel_add(dest[dest_stride * 2], e1, bd);
107 dest[dest_stride * 3] =
108 highbd_clip_pixel_add(dest[dest_stride * 3], e1, bd);
109 ip++;
110 dest++;
111 }
112 }
113
inv_txfm_type_to_func(TXFM_TYPE txfm_type)114 static INLINE TxfmFunc inv_txfm_type_to_func(TXFM_TYPE txfm_type) {
115 switch (txfm_type) {
116 case TXFM_TYPE_DCT4: return av1_idct4;
117 case TXFM_TYPE_DCT8: return av1_idct8;
118 case TXFM_TYPE_DCT16: return av1_idct16;
119 case TXFM_TYPE_DCT32: return av1_idct32;
120 case TXFM_TYPE_DCT64: return av1_idct64;
121 case TXFM_TYPE_ADST4: return av1_iadst4;
122 case TXFM_TYPE_ADST8: return av1_iadst8;
123 case TXFM_TYPE_ADST16: return av1_iadst16;
124 case TXFM_TYPE_IDENTITY4: return av1_iidentity4_c;
125 case TXFM_TYPE_IDENTITY8: return av1_iidentity8_c;
126 case TXFM_TYPE_IDENTITY16: return av1_iidentity16_c;
127 case TXFM_TYPE_IDENTITY32: return av1_iidentity32_c;
128 default: assert(0); return NULL;
129 }
130 }
131
132 static const int8_t inv_shift_4x4[2] = { 0, -4 };
133 static const int8_t inv_shift_8x8[2] = { -1, -4 };
134 static const int8_t inv_shift_16x16[2] = { -2, -4 };
135 static const int8_t inv_shift_32x32[2] = { -2, -4 };
136 static const int8_t inv_shift_64x64[2] = { -2, -4 };
137 static const int8_t inv_shift_4x8[2] = { 0, -4 };
138 static const int8_t inv_shift_8x4[2] = { 0, -4 };
139 static const int8_t inv_shift_8x16[2] = { -1, -4 };
140 static const int8_t inv_shift_16x8[2] = { -1, -4 };
141 static const int8_t inv_shift_16x32[2] = { -1, -4 };
142 static const int8_t inv_shift_32x16[2] = { -1, -4 };
143 static const int8_t inv_shift_32x64[2] = { -1, -4 };
144 static const int8_t inv_shift_64x32[2] = { -1, -4 };
145 static const int8_t inv_shift_4x16[2] = { -1, -4 };
146 static const int8_t inv_shift_16x4[2] = { -1, -4 };
147 static const int8_t inv_shift_8x32[2] = { -2, -4 };
148 static const int8_t inv_shift_32x8[2] = { -2, -4 };
149 static const int8_t inv_shift_16x64[2] = { -2, -4 };
150 static const int8_t inv_shift_64x16[2] = { -2, -4 };
151
152 const int8_t *av1_inv_txfm_shift_ls[TX_SIZES_ALL] = {
153 inv_shift_4x4, inv_shift_8x8, inv_shift_16x16, inv_shift_32x32,
154 inv_shift_64x64, inv_shift_4x8, inv_shift_8x4, inv_shift_8x16,
155 inv_shift_16x8, inv_shift_16x32, inv_shift_32x16, inv_shift_32x64,
156 inv_shift_64x32, inv_shift_4x16, inv_shift_16x4, inv_shift_8x32,
157 inv_shift_32x8, inv_shift_16x64, inv_shift_64x16,
158 };
159
160 /* clang-format off */
161 const int8_t av1_inv_cos_bit_col[MAX_TXWH_IDX] // txw_idx
162 [MAX_TXWH_IDX] = { // txh_idx
163 { INV_COS_BIT, INV_COS_BIT, INV_COS_BIT, 0, 0 },
164 { INV_COS_BIT, INV_COS_BIT, INV_COS_BIT, INV_COS_BIT, 0 },
165 { INV_COS_BIT, INV_COS_BIT, INV_COS_BIT, INV_COS_BIT, INV_COS_BIT },
166 { 0, INV_COS_BIT, INV_COS_BIT, INV_COS_BIT, INV_COS_BIT },
167 { 0, 0, INV_COS_BIT, INV_COS_BIT, INV_COS_BIT }
168 };
169
170 const int8_t av1_inv_cos_bit_row[MAX_TXWH_IDX] // txw_idx
171 [MAX_TXWH_IDX] = { // txh_idx
172 { INV_COS_BIT, INV_COS_BIT, INV_COS_BIT, 0, 0 },
173 { INV_COS_BIT, INV_COS_BIT, INV_COS_BIT, INV_COS_BIT, 0 },
174 { INV_COS_BIT, INV_COS_BIT, INV_COS_BIT, INV_COS_BIT, INV_COS_BIT },
175 { 0, INV_COS_BIT, INV_COS_BIT, INV_COS_BIT, INV_COS_BIT },
176 { 0, 0, INV_COS_BIT, INV_COS_BIT, INV_COS_BIT }
177 };
178 /* clang-format on */
179
180 static const int8_t iadst4_range[7] = { 0, 1, 0, 0, 0, 0, 0 };
181
av1_get_inv_txfm_cfg(TX_TYPE tx_type,TX_SIZE tx_size,TXFM_2D_FLIP_CFG * cfg)182 void av1_get_inv_txfm_cfg(TX_TYPE tx_type, TX_SIZE tx_size,
183 TXFM_2D_FLIP_CFG *cfg) {
184 assert(cfg != NULL);
185 cfg->tx_size = tx_size;
186 av1_zero(cfg->stage_range_col);
187 av1_zero(cfg->stage_range_row);
188 set_flip_cfg(tx_type, cfg);
189 const TX_TYPE_1D tx_type_1d_col = vtx_tab[tx_type];
190 const TX_TYPE_1D tx_type_1d_row = htx_tab[tx_type];
191 cfg->shift = av1_inv_txfm_shift_ls[tx_size];
192 const int txw_idx = get_txw_idx(tx_size);
193 const int txh_idx = get_txh_idx(tx_size);
194 cfg->cos_bit_col = av1_inv_cos_bit_col[txw_idx][txh_idx];
195 cfg->cos_bit_row = av1_inv_cos_bit_row[txw_idx][txh_idx];
196 cfg->txfm_type_col = av1_txfm_type_ls[txh_idx][tx_type_1d_col];
197 if (cfg->txfm_type_col == TXFM_TYPE_ADST4) {
198 memcpy(cfg->stage_range_col, iadst4_range, sizeof(iadst4_range));
199 }
200 cfg->txfm_type_row = av1_txfm_type_ls[txw_idx][tx_type_1d_row];
201 if (cfg->txfm_type_row == TXFM_TYPE_ADST4) {
202 memcpy(cfg->stage_range_row, iadst4_range, sizeof(iadst4_range));
203 }
204 cfg->stage_num_col = av1_txfm_stage_num_list[cfg->txfm_type_col];
205 cfg->stage_num_row = av1_txfm_stage_num_list[cfg->txfm_type_row];
206 }
207
av1_gen_inv_stage_range(int8_t * stage_range_col,int8_t * stage_range_row,const TXFM_2D_FLIP_CFG * cfg,TX_SIZE tx_size,int bd)208 void av1_gen_inv_stage_range(int8_t *stage_range_col, int8_t *stage_range_row,
209 const TXFM_2D_FLIP_CFG *cfg, TX_SIZE tx_size,
210 int bd) {
211 const int fwd_shift = inv_start_range[tx_size];
212 const int8_t *shift = cfg->shift;
213 int8_t opt_range_row, opt_range_col;
214 if (bd == 8) {
215 opt_range_row = 16;
216 opt_range_col = 16;
217 } else if (bd == 10) {
218 opt_range_row = 18;
219 opt_range_col = 16;
220 } else {
221 assert(bd == 12);
222 opt_range_row = 20;
223 opt_range_col = 18;
224 }
225 // i < MAX_TXFM_STAGE_NUM will mute above array bounds warning
226 for (int i = 0; i < cfg->stage_num_row && i < MAX_TXFM_STAGE_NUM; ++i) {
227 int real_range_row = cfg->stage_range_row[i] + fwd_shift + bd + 1;
228 (void)real_range_row;
229 if (cfg->txfm_type_row == TXFM_TYPE_ADST4 && i == 1) {
230 // the adst4 may use 1 extra bit on top of opt_range_row at stage 1
231 // so opt_range_row >= real_range_row will not hold
232 stage_range_row[i] = opt_range_row;
233 } else {
234 assert(opt_range_row >= real_range_row);
235 stage_range_row[i] = opt_range_row;
236 }
237 }
238 // i < MAX_TXFM_STAGE_NUM will mute above array bounds warning
239 for (int i = 0; i < cfg->stage_num_col && i < MAX_TXFM_STAGE_NUM; ++i) {
240 int real_range_col =
241 cfg->stage_range_col[i] + fwd_shift + shift[0] + bd + 1;
242 (void)real_range_col;
243 if (cfg->txfm_type_col == TXFM_TYPE_ADST4 && i == 1) {
244 // the adst4 may use 1 extra bit on top of opt_range_col at stage 1
245 // so opt_range_col >= real_range_col will not hold
246 stage_range_col[i] = opt_range_col;
247 } else {
248 assert(opt_range_col >= real_range_col);
249 stage_range_col[i] = opt_range_col;
250 }
251 }
252 }
253
inv_txfm2d_add_c(const int32_t * input,uint16_t * output,int stride,TXFM_2D_FLIP_CFG * cfg,int32_t * txfm_buf,TX_SIZE tx_size,int bd)254 static INLINE void inv_txfm2d_add_c(const int32_t *input, uint16_t *output,
255 int stride, TXFM_2D_FLIP_CFG *cfg,
256 int32_t *txfm_buf, TX_SIZE tx_size,
257 int bd) {
258 // Note when assigning txfm_size_col, we use the txfm_size from the
259 // row configuration and vice versa. This is intentionally done to
260 // accurately perform rectangular transforms. When the transform is
261 // rectangular, the number of columns will be the same as the
262 // txfm_size stored in the row cfg struct. It will make no difference
263 // for square transforms.
264 const int txfm_size_col = tx_size_wide[cfg->tx_size];
265 const int txfm_size_row = tx_size_high[cfg->tx_size];
266 // Take the shift from the larger dimension in the rectangular case.
267 const int8_t *shift = cfg->shift;
268 const int rect_type = get_rect_tx_log_ratio(txfm_size_col, txfm_size_row);
269 int8_t stage_range_row[MAX_TXFM_STAGE_NUM];
270 int8_t stage_range_col[MAX_TXFM_STAGE_NUM];
271 assert(cfg->stage_num_row <= MAX_TXFM_STAGE_NUM);
272 assert(cfg->stage_num_col <= MAX_TXFM_STAGE_NUM);
273 av1_gen_inv_stage_range(stage_range_col, stage_range_row, cfg, tx_size, bd);
274
275 const int8_t cos_bit_col = cfg->cos_bit_col;
276 const int8_t cos_bit_row = cfg->cos_bit_row;
277 const TxfmFunc txfm_func_col = inv_txfm_type_to_func(cfg->txfm_type_col);
278 const TxfmFunc txfm_func_row = inv_txfm_type_to_func(cfg->txfm_type_row);
279
280 // txfm_buf's length is txfm_size_row * txfm_size_col + 2 *
281 // AOMMAX(txfm_size_row, txfm_size_col)
282 // it is used for intermediate data buffering
283 const int buf_offset = AOMMAX(txfm_size_row, txfm_size_col);
284 int32_t *temp_in = txfm_buf;
285 int32_t *temp_out = temp_in + buf_offset;
286 int32_t *buf = temp_out + buf_offset;
287 int32_t *buf_ptr = buf;
288 int c, r;
289
290 // Rows
291 for (r = 0; r < txfm_size_row; ++r) {
292 if (abs(rect_type) == 1) {
293 for (c = 0; c < txfm_size_col; ++c) {
294 temp_in[c] = round_shift((int64_t)input[c] * NewInvSqrt2, NewSqrt2Bits);
295 }
296 clamp_buf(temp_in, txfm_size_col, bd + 8);
297 txfm_func_row(temp_in, buf_ptr, cos_bit_row, stage_range_row);
298 } else {
299 for (c = 0; c < txfm_size_col; ++c) {
300 temp_in[c] = input[c];
301 }
302 clamp_buf(temp_in, txfm_size_col, bd + 8);
303 txfm_func_row(temp_in, buf_ptr, cos_bit_row, stage_range_row);
304 }
305 av1_round_shift_array(buf_ptr, txfm_size_col, -shift[0]);
306 input += txfm_size_col;
307 buf_ptr += txfm_size_col;
308 }
309
310 // Columns
311 for (c = 0; c < txfm_size_col; ++c) {
312 if (cfg->lr_flip == 0) {
313 for (r = 0; r < txfm_size_row; ++r)
314 temp_in[r] = buf[r * txfm_size_col + c];
315 } else {
316 // flip left right
317 for (r = 0; r < txfm_size_row; ++r)
318 temp_in[r] = buf[r * txfm_size_col + (txfm_size_col - c - 1)];
319 }
320 clamp_buf(temp_in, txfm_size_row, AOMMAX(bd + 6, 16));
321 txfm_func_col(temp_in, temp_out, cos_bit_col, stage_range_col);
322 av1_round_shift_array(temp_out, txfm_size_row, -shift[1]);
323 if (cfg->ud_flip == 0) {
324 for (r = 0; r < txfm_size_row; ++r) {
325 output[r * stride + c] =
326 highbd_clip_pixel_add(output[r * stride + c], temp_out[r], bd);
327 }
328 } else {
329 // flip upside down
330 for (r = 0; r < txfm_size_row; ++r) {
331 output[r * stride + c] = highbd_clip_pixel_add(
332 output[r * stride + c], temp_out[txfm_size_row - r - 1], bd);
333 }
334 }
335 }
336 }
337
inv_txfm2d_add_facade(const int32_t * input,uint16_t * output,int stride,int32_t * txfm_buf,TX_TYPE tx_type,TX_SIZE tx_size,int bd)338 static INLINE void inv_txfm2d_add_facade(const int32_t *input, uint16_t *output,
339 int stride, int32_t *txfm_buf,
340 TX_TYPE tx_type, TX_SIZE tx_size,
341 int bd) {
342 TXFM_2D_FLIP_CFG cfg;
343 av1_get_inv_txfm_cfg(tx_type, tx_size, &cfg);
344 // Forward shift sum uses larger square size, to be consistent with what
345 // av1_gen_inv_stage_range() does for inverse shifts.
346 inv_txfm2d_add_c(input, output, stride, &cfg, txfm_buf, tx_size, bd);
347 }
348
av1_inv_txfm2d_add_4x8_c(const int32_t * input,uint16_t * output,int stride,TX_TYPE tx_type,int bd)349 void av1_inv_txfm2d_add_4x8_c(const int32_t *input, uint16_t *output,
350 int stride, TX_TYPE tx_type, int bd) {
351 DECLARE_ALIGNED(32, int, txfm_buf[4 * 8 + 8 + 8]);
352 inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_4X8, bd);
353 }
354
av1_inv_txfm2d_add_8x4_c(const int32_t * input,uint16_t * output,int stride,TX_TYPE tx_type,int bd)355 void av1_inv_txfm2d_add_8x4_c(const int32_t *input, uint16_t *output,
356 int stride, TX_TYPE tx_type, int bd) {
357 DECLARE_ALIGNED(32, int, txfm_buf[8 * 4 + 8 + 8]);
358 inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_8X4, bd);
359 }
360
av1_inv_txfm2d_add_8x16_c(const int32_t * input,uint16_t * output,int stride,TX_TYPE tx_type,int bd)361 void av1_inv_txfm2d_add_8x16_c(const int32_t *input, uint16_t *output,
362 int stride, TX_TYPE tx_type, int bd) {
363 DECLARE_ALIGNED(32, int, txfm_buf[8 * 16 + 16 + 16]);
364 inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_8X16, bd);
365 }
366
av1_inv_txfm2d_add_16x8_c(const int32_t * input,uint16_t * output,int stride,TX_TYPE tx_type,int bd)367 void av1_inv_txfm2d_add_16x8_c(const int32_t *input, uint16_t *output,
368 int stride, TX_TYPE tx_type, int bd) {
369 DECLARE_ALIGNED(32, int, txfm_buf[16 * 8 + 16 + 16]);
370 inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_16X8, bd);
371 }
372
av1_inv_txfm2d_add_16x32_c(const int32_t * input,uint16_t * output,int stride,TX_TYPE tx_type,int bd)373 void av1_inv_txfm2d_add_16x32_c(const int32_t *input, uint16_t *output,
374 int stride, TX_TYPE tx_type, int bd) {
375 DECLARE_ALIGNED(32, int, txfm_buf[16 * 32 + 32 + 32]);
376 inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_16X32, bd);
377 }
378
av1_inv_txfm2d_add_32x16_c(const int32_t * input,uint16_t * output,int stride,TX_TYPE tx_type,int bd)379 void av1_inv_txfm2d_add_32x16_c(const int32_t *input, uint16_t *output,
380 int stride, TX_TYPE tx_type, int bd) {
381 DECLARE_ALIGNED(32, int, txfm_buf[32 * 16 + 32 + 32]);
382 inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_32X16, bd);
383 }
384
av1_inv_txfm2d_add_4x4_c(const int32_t * input,uint16_t * output,int stride,TX_TYPE tx_type,int bd)385 void av1_inv_txfm2d_add_4x4_c(const int32_t *input, uint16_t *output,
386 int stride, TX_TYPE tx_type, int bd) {
387 DECLARE_ALIGNED(32, int, txfm_buf[4 * 4 + 4 + 4]);
388 inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_4X4, bd);
389 }
390
av1_inv_txfm2d_add_8x8_c(const int32_t * input,uint16_t * output,int stride,TX_TYPE tx_type,int bd)391 void av1_inv_txfm2d_add_8x8_c(const int32_t *input, uint16_t *output,
392 int stride, TX_TYPE tx_type, int bd) {
393 DECLARE_ALIGNED(32, int, txfm_buf[8 * 8 + 8 + 8]);
394 inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_8X8, bd);
395 }
396
av1_inv_txfm2d_add_16x16_c(const int32_t * input,uint16_t * output,int stride,TX_TYPE tx_type,int bd)397 void av1_inv_txfm2d_add_16x16_c(const int32_t *input, uint16_t *output,
398 int stride, TX_TYPE tx_type, int bd) {
399 DECLARE_ALIGNED(32, int, txfm_buf[16 * 16 + 16 + 16]);
400 inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_16X16, bd);
401 }
402
av1_inv_txfm2d_add_32x32_c(const int32_t * input,uint16_t * output,int stride,TX_TYPE tx_type,int bd)403 void av1_inv_txfm2d_add_32x32_c(const int32_t *input, uint16_t *output,
404 int stride, TX_TYPE tx_type, int bd) {
405 DECLARE_ALIGNED(32, int, txfm_buf[32 * 32 + 32 + 32]);
406 inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_32X32, bd);
407 }
408
av1_inv_txfm2d_add_64x64_c(const int32_t * input,uint16_t * output,int stride,TX_TYPE tx_type,int bd)409 void av1_inv_txfm2d_add_64x64_c(const int32_t *input, uint16_t *output,
410 int stride, TX_TYPE tx_type, int bd) {
411 // TODO(urvang): Can the same array be reused, instead of using a new array?
412 // Remap 32x32 input into a modified 64x64 by:
413 // - Copying over these values in top-left 32x32 locations.
414 // - Setting the rest of the locations to 0.
415 int32_t mod_input[64 * 64];
416 for (int row = 0; row < 32; ++row) {
417 memcpy(mod_input + row * 64, input + row * 32, 32 * sizeof(*mod_input));
418 memset(mod_input + row * 64 + 32, 0, 32 * sizeof(*mod_input));
419 }
420 memset(mod_input + 32 * 64, 0, 32 * 64 * sizeof(*mod_input));
421 DECLARE_ALIGNED(32, int, txfm_buf[64 * 64 + 64 + 64]);
422 inv_txfm2d_add_facade(mod_input, output, stride, txfm_buf, tx_type, TX_64X64,
423 bd);
424 }
425
av1_inv_txfm2d_add_64x32_c(const int32_t * input,uint16_t * output,int stride,TX_TYPE tx_type,int bd)426 void av1_inv_txfm2d_add_64x32_c(const int32_t *input, uint16_t *output,
427 int stride, TX_TYPE tx_type, int bd) {
428 // Remap 32x32 input into a modified 64x32 by:
429 // - Copying over these values in top-left 32x32 locations.
430 // - Setting the rest of the locations to 0.
431 int32_t mod_input[64 * 32];
432 for (int row = 0; row < 32; ++row) {
433 memcpy(mod_input + row * 64, input + row * 32, 32 * sizeof(*mod_input));
434 memset(mod_input + row * 64 + 32, 0, 32 * sizeof(*mod_input));
435 }
436 DECLARE_ALIGNED(32, int, txfm_buf[64 * 32 + 64 + 64]);
437 inv_txfm2d_add_facade(mod_input, output, stride, txfm_buf, tx_type, TX_64X32,
438 bd);
439 }
440
av1_inv_txfm2d_add_32x64_c(const int32_t * input,uint16_t * output,int stride,TX_TYPE tx_type,int bd)441 void av1_inv_txfm2d_add_32x64_c(const int32_t *input, uint16_t *output,
442 int stride, TX_TYPE tx_type, int bd) {
443 // Remap 32x32 input into a modified 32x64 input by:
444 // - Copying over these values in top-left 32x32 locations.
445 // - Setting the rest of the locations to 0.
446 int32_t mod_input[32 * 64];
447 memcpy(mod_input, input, 32 * 32 * sizeof(*mod_input));
448 memset(mod_input + 32 * 32, 0, 32 * 32 * sizeof(*mod_input));
449 DECLARE_ALIGNED(32, int, txfm_buf[64 * 32 + 64 + 64]);
450 inv_txfm2d_add_facade(mod_input, output, stride, txfm_buf, tx_type, TX_32X64,
451 bd);
452 }
453
av1_inv_txfm2d_add_16x64_c(const int32_t * input,uint16_t * output,int stride,TX_TYPE tx_type,int bd)454 void av1_inv_txfm2d_add_16x64_c(const int32_t *input, uint16_t *output,
455 int stride, TX_TYPE tx_type, int bd) {
456 // Remap 16x32 input into a modified 16x64 input by:
457 // - Copying over these values in top-left 16x32 locations.
458 // - Setting the rest of the locations to 0.
459 int32_t mod_input[16 * 64];
460 memcpy(mod_input, input, 16 * 32 * sizeof(*mod_input));
461 memset(mod_input + 16 * 32, 0, 16 * 32 * sizeof(*mod_input));
462 DECLARE_ALIGNED(32, int, txfm_buf[16 * 64 + 64 + 64]);
463 inv_txfm2d_add_facade(mod_input, output, stride, txfm_buf, tx_type, TX_16X64,
464 bd);
465 }
466
av1_inv_txfm2d_add_64x16_c(const int32_t * input,uint16_t * output,int stride,TX_TYPE tx_type,int bd)467 void av1_inv_txfm2d_add_64x16_c(const int32_t *input, uint16_t *output,
468 int stride, TX_TYPE tx_type, int bd) {
469 // Remap 32x16 input into a modified 64x16 by:
470 // - Copying over these values in top-left 32x16 locations.
471 // - Setting the rest of the locations to 0.
472 int32_t mod_input[64 * 16];
473 for (int row = 0; row < 16; ++row) {
474 memcpy(mod_input + row * 64, input + row * 32, 32 * sizeof(*mod_input));
475 memset(mod_input + row * 64 + 32, 0, 32 * sizeof(*mod_input));
476 }
477 DECLARE_ALIGNED(32, int, txfm_buf[16 * 64 + 64 + 64]);
478 inv_txfm2d_add_facade(mod_input, output, stride, txfm_buf, tx_type, TX_64X16,
479 bd);
480 }
481
av1_inv_txfm2d_add_4x16_c(const int32_t * input,uint16_t * output,int stride,TX_TYPE tx_type,int bd)482 void av1_inv_txfm2d_add_4x16_c(const int32_t *input, uint16_t *output,
483 int stride, TX_TYPE tx_type, int bd) {
484 DECLARE_ALIGNED(32, int, txfm_buf[4 * 16 + 16 + 16]);
485 inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_4X16, bd);
486 }
487
av1_inv_txfm2d_add_16x4_c(const int32_t * input,uint16_t * output,int stride,TX_TYPE tx_type,int bd)488 void av1_inv_txfm2d_add_16x4_c(const int32_t *input, uint16_t *output,
489 int stride, TX_TYPE tx_type, int bd) {
490 DECLARE_ALIGNED(32, int, txfm_buf[4 * 16 + 16 + 16]);
491 inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_16X4, bd);
492 }
493
av1_inv_txfm2d_add_8x32_c(const int32_t * input,uint16_t * output,int stride,TX_TYPE tx_type,int bd)494 void av1_inv_txfm2d_add_8x32_c(const int32_t *input, uint16_t *output,
495 int stride, TX_TYPE tx_type, int bd) {
496 DECLARE_ALIGNED(32, int, txfm_buf[8 * 32 + 32 + 32]);
497 inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_8X32, bd);
498 }
499
av1_inv_txfm2d_add_32x8_c(const int32_t * input,uint16_t * output,int stride,TX_TYPE tx_type,int bd)500 void av1_inv_txfm2d_add_32x8_c(const int32_t *input, uint16_t *output,
501 int stride, TX_TYPE tx_type, int bd) {
502 DECLARE_ALIGNED(32, int, txfm_buf[8 * 32 + 32 + 32]);
503 inv_txfm2d_add_facade(input, output, stride, txfm_buf, tx_type, TX_32X8, bd);
504 }
505