2 * Copyright (c) 2016 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.
13 #include "./vpx_config.h"
14 #include "./vpx_dsp_rtcd.h"
15 #include "vpx_dsp/arm/idct_neon.h"
16 #include "vpx_dsp/arm/transpose_neon.h"
17 #include "vpx_dsp/txfm_common.h"
19 static INLINE void load_8x8_s16(const tran_low_t *input, int16x8_t *const in0,
20 int16x8_t *const in1, int16x8_t *const in2,
21 int16x8_t *const in3, int16x8_t *const in4,
22 int16x8_t *const in5, int16x8_t *const in6,
23 int16x8_t *const in7) {
24 *in0 = load_tran_low_to_s16q(input);
26 *in1 = load_tran_low_to_s16q(input);
28 *in2 = load_tran_low_to_s16q(input);
30 *in3 = load_tran_low_to_s16q(input);
32 *in4 = load_tran_low_to_s16q(input);
34 *in5 = load_tran_low_to_s16q(input);
36 *in6 = load_tran_low_to_s16q(input);
38 *in7 = load_tran_low_to_s16q(input);
41 static INLINE void load_4x8_s16(const tran_low_t *input, int16x4_t *const in0,
42 int16x4_t *const in1, int16x4_t *const in2,
43 int16x4_t *const in3, int16x4_t *const in4,
44 int16x4_t *const in5, int16x4_t *const in6,
45 int16x4_t *const in7) {
46 *in0 = load_tran_low_to_s16d(input);
48 *in1 = load_tran_low_to_s16d(input);
50 *in2 = load_tran_low_to_s16d(input);
52 *in3 = load_tran_low_to_s16d(input);
54 *in4 = load_tran_low_to_s16d(input);
56 *in5 = load_tran_low_to_s16d(input);
58 *in6 = load_tran_low_to_s16d(input);
60 *in7 = load_tran_low_to_s16d(input);
63 // Only for the first pass of the _135_ variant. Since it only uses values from
64 // the top left 16x16 it can safely assume all the remaining values are 0 and
65 // skip an awful lot of calculations. In fact, only the first 12 columns make
66 // the cut. None of the elements in the 13th, 14th, 15th or 16th columns are
67 // used so it skips any calls to input[12|13|14|15] too.
68 // In C this does a single row of 32 for each call. Here it transposes the top
69 // left 12x8 to allow using SIMD.
71 // vp9/common/vp9_scan.c:vp9_default_iscan_32x32 arranges the first 135 non-zero
72 // coefficients as follows:
73 // 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
74 // 0 0 2 5 10 17 25 38 47 62 83 101 121
75 // 1 1 4 8 15 22 30 45 58 74 92 112 133
76 // 2 3 7 12 18 28 36 52 64 82 102 118
77 // 3 6 11 16 23 31 43 60 73 90 109 126
78 // 4 9 14 19 29 37 50 65 78 98 116 134
79 // 5 13 20 26 35 44 54 72 85 105 123
80 // 6 21 27 33 42 53 63 80 94 113 132
81 // 7 24 32 39 48 57 71 88 104 120
82 // 8 34 40 46 56 68 81 96 111 130
83 // 9 41 49 55 67 77 91 107 124
84 // 10 51 59 66 76 89 99 119 131
85 // 11 61 69 75 87 100 114 129
86 // 12 70 79 86 97 108 122
87 // 13 84 93 103 110 125
90 void vpx_idct32_12_neon(const tran_low_t *const input, int16_t *output) {
92 int16x8_t in[12], s1[32], s2[32], s3[32], s4[32], s5[32], s6[32], s7[32];
94 load_8x8_s16(input, &in[0], &in[1], &in[2], &in[3], &in[4], &in[5], &in[6],
96 transpose_s16_8x8(&in[0], &in[1], &in[2], &in[3], &in[4], &in[5], &in[6],
99 load_4x8_s16(input + 8, &tmp[0], &tmp[1], &tmp[2], &tmp[3], &tmp[4], &tmp[5],
101 transpose_s16_4x8(tmp[0], tmp[1], tmp[2], tmp[3], tmp[4], tmp[5], tmp[6],
102 tmp[7], &in[8], &in[9], &in[10], &in[11]);
105 s1[16] = multiply_shift_and_narrow_s16(in[1], cospi_31_64);
106 s1[31] = multiply_shift_and_narrow_s16(in[1], cospi_1_64);
108 s1[18] = multiply_shift_and_narrow_s16(in[9], cospi_23_64);
109 s1[29] = multiply_shift_and_narrow_s16(in[9], cospi_9_64);
111 s1[19] = multiply_shift_and_narrow_s16(in[7], -cospi_25_64);
112 s1[28] = multiply_shift_and_narrow_s16(in[7], cospi_7_64);
114 s1[20] = multiply_shift_and_narrow_s16(in[5], cospi_27_64);
115 s1[27] = multiply_shift_and_narrow_s16(in[5], cospi_5_64);
117 s1[21] = multiply_shift_and_narrow_s16(in[11], -cospi_21_64);
118 s1[26] = multiply_shift_and_narrow_s16(in[11], cospi_11_64);
120 s1[23] = multiply_shift_and_narrow_s16(in[3], -cospi_29_64);
121 s1[24] = multiply_shift_and_narrow_s16(in[3], cospi_3_64);
124 s2[8] = multiply_shift_and_narrow_s16(in[2], cospi_30_64);
125 s2[15] = multiply_shift_and_narrow_s16(in[2], cospi_2_64);
127 s2[10] = multiply_shift_and_narrow_s16(in[10], cospi_22_64);
128 s2[13] = multiply_shift_and_narrow_s16(in[10], cospi_10_64);
130 s2[11] = multiply_shift_and_narrow_s16(in[6], -cospi_26_64);
131 s2[12] = multiply_shift_and_narrow_s16(in[6], cospi_6_64);
133 s2[18] = vsubq_s16(s1[19], s1[18]);
134 s2[19] = vaddq_s16(s1[18], s1[19]);
135 s2[20] = vaddq_s16(s1[20], s1[21]);
136 s2[21] = vsubq_s16(s1[20], s1[21]);
137 s2[26] = vsubq_s16(s1[27], s1[26]);
138 s2[27] = vaddq_s16(s1[26], s1[27]);
139 s2[28] = vaddq_s16(s1[28], s1[29]);
140 s2[29] = vsubq_s16(s1[28], s1[29]);
143 s3[4] = multiply_shift_and_narrow_s16(in[4], cospi_28_64);
144 s3[7] = multiply_shift_and_narrow_s16(in[4], cospi_4_64);
146 s3[10] = vsubq_s16(s2[11], s2[10]);
147 s3[11] = vaddq_s16(s2[10], s2[11]);
148 s3[12] = vaddq_s16(s2[12], s2[13]);
149 s3[13] = vsubq_s16(s2[12], s2[13]);
151 s3[17] = multiply_accumulate_shift_and_narrow_s16(s1[16], -cospi_4_64, s1[31],
153 s3[30] = multiply_accumulate_shift_and_narrow_s16(s1[16], cospi_28_64, s1[31],
156 s3[18] = multiply_accumulate_shift_and_narrow_s16(s2[18], -cospi_28_64,
157 s2[29], -cospi_4_64);
158 s3[29] = multiply_accumulate_shift_and_narrow_s16(s2[18], -cospi_4_64, s2[29],
161 s3[21] = multiply_accumulate_shift_and_narrow_s16(s2[21], -cospi_20_64,
162 s2[26], cospi_12_64);
163 s3[26] = multiply_accumulate_shift_and_narrow_s16(s2[21], cospi_12_64, s2[26],
166 s3[22] = multiply_accumulate_shift_and_narrow_s16(s1[23], -cospi_12_64,
167 s1[24], -cospi_20_64);
168 s3[25] = multiply_accumulate_shift_and_narrow_s16(s1[23], -cospi_20_64,
169 s1[24], cospi_12_64);
172 s4[0] = multiply_shift_and_narrow_s16(in[0], cospi_16_64);
173 s4[2] = multiply_shift_and_narrow_s16(in[8], cospi_24_64);
174 s4[3] = multiply_shift_and_narrow_s16(in[8], cospi_8_64);
176 s4[9] = multiply_accumulate_shift_and_narrow_s16(s2[8], -cospi_8_64, s2[15],
178 s4[14] = multiply_accumulate_shift_and_narrow_s16(s2[8], cospi_24_64, s2[15],
181 s4[10] = multiply_accumulate_shift_and_narrow_s16(s3[10], -cospi_24_64,
182 s3[13], -cospi_8_64);
183 s4[13] = multiply_accumulate_shift_and_narrow_s16(s3[10], -cospi_8_64, s3[13],
186 s4[16] = vaddq_s16(s1[16], s2[19]);
187 s4[17] = vaddq_s16(s3[17], s3[18]);
188 s4[18] = vsubq_s16(s3[17], s3[18]);
189 s4[19] = vsubq_s16(s1[16], s2[19]);
190 s4[20] = vsubq_s16(s1[23], s2[20]);
191 s4[21] = vsubq_s16(s3[22], s3[21]);
192 s4[22] = vaddq_s16(s3[21], s3[22]);
193 s4[23] = vaddq_s16(s2[20], s1[23]);
194 s4[24] = vaddq_s16(s1[24], s2[27]);
195 s4[25] = vaddq_s16(s3[25], s3[26]);
196 s4[26] = vsubq_s16(s3[25], s3[26]);
197 s4[27] = vsubq_s16(s1[24], s2[27]);
198 s4[28] = vsubq_s16(s1[31], s2[28]);
199 s4[29] = vsubq_s16(s3[30], s3[29]);
200 s4[30] = vaddq_s16(s3[29], s3[30]);
201 s4[31] = vaddq_s16(s2[28], s1[31]);
204 s5[0] = vaddq_s16(s4[0], s4[3]);
205 s5[1] = vaddq_s16(s4[0], s4[2]);
206 s5[2] = vsubq_s16(s4[0], s4[2]);
207 s5[3] = vsubq_s16(s4[0], s4[3]);
209 s5[5] = sub_multiply_shift_and_narrow_s16(s3[7], s3[4], cospi_16_64);
210 s5[6] = add_multiply_shift_and_narrow_s16(s3[4], s3[7], cospi_16_64);
212 s5[8] = vaddq_s16(s2[8], s3[11]);
213 s5[9] = vaddq_s16(s4[9], s4[10]);
214 s5[10] = vsubq_s16(s4[9], s4[10]);
215 s5[11] = vsubq_s16(s2[8], s3[11]);
216 s5[12] = vsubq_s16(s2[15], s3[12]);
217 s5[13] = vsubq_s16(s4[14], s4[13]);
218 s5[14] = vaddq_s16(s4[13], s4[14]);
219 s5[15] = vaddq_s16(s2[15], s3[12]);
221 s5[18] = multiply_accumulate_shift_and_narrow_s16(s4[18], -cospi_8_64, s4[29],
223 s5[29] = multiply_accumulate_shift_and_narrow_s16(s4[18], cospi_24_64, s4[29],
226 s5[19] = multiply_accumulate_shift_and_narrow_s16(s4[19], -cospi_8_64, s4[28],
228 s5[28] = multiply_accumulate_shift_and_narrow_s16(s4[19], cospi_24_64, s4[28],
231 s5[20] = multiply_accumulate_shift_and_narrow_s16(s4[20], -cospi_24_64,
232 s4[27], -cospi_8_64);
233 s5[27] = multiply_accumulate_shift_and_narrow_s16(s4[20], -cospi_8_64, s4[27],
236 s5[21] = multiply_accumulate_shift_and_narrow_s16(s4[21], -cospi_24_64,
237 s4[26], -cospi_8_64);
238 s5[26] = multiply_accumulate_shift_and_narrow_s16(s4[21], -cospi_8_64, s4[26],
242 s6[0] = vaddq_s16(s5[0], s3[7]);
243 s6[1] = vaddq_s16(s5[1], s5[6]);
244 s6[2] = vaddq_s16(s5[2], s5[5]);
245 s6[3] = vaddq_s16(s5[3], s3[4]);
246 s6[4] = vsubq_s16(s5[3], s3[4]);
247 s6[5] = vsubq_s16(s5[2], s5[5]);
248 s6[6] = vsubq_s16(s5[1], s5[6]);
249 s6[7] = vsubq_s16(s5[0], s3[7]);
251 s6[10] = sub_multiply_shift_and_narrow_s16(s5[13], s5[10], cospi_16_64);
252 s6[13] = add_multiply_shift_and_narrow_s16(s5[10], s5[13], cospi_16_64);
254 s6[11] = sub_multiply_shift_and_narrow_s16(s5[12], s5[11], cospi_16_64);
255 s6[12] = add_multiply_shift_and_narrow_s16(s5[11], s5[12], cospi_16_64);
257 s6[16] = vaddq_s16(s4[16], s4[23]);
258 s6[17] = vaddq_s16(s4[17], s4[22]);
259 s6[18] = vaddq_s16(s5[18], s5[21]);
260 s6[19] = vaddq_s16(s5[19], s5[20]);
261 s6[20] = vsubq_s16(s5[19], s5[20]);
262 s6[21] = vsubq_s16(s5[18], s5[21]);
263 s6[22] = vsubq_s16(s4[17], s4[22]);
264 s6[23] = vsubq_s16(s4[16], s4[23]);
266 s6[24] = vsubq_s16(s4[31], s4[24]);
267 s6[25] = vsubq_s16(s4[30], s4[25]);
268 s6[26] = vsubq_s16(s5[29], s5[26]);
269 s6[27] = vsubq_s16(s5[28], s5[27]);
270 s6[28] = vaddq_s16(s5[27], s5[28]);
271 s6[29] = vaddq_s16(s5[26], s5[29]);
272 s6[30] = vaddq_s16(s4[25], s4[30]);
273 s6[31] = vaddq_s16(s4[24], s4[31]);
276 s7[0] = vaddq_s16(s6[0], s5[15]);
277 s7[1] = vaddq_s16(s6[1], s5[14]);
278 s7[2] = vaddq_s16(s6[2], s6[13]);
279 s7[3] = vaddq_s16(s6[3], s6[12]);
280 s7[4] = vaddq_s16(s6[4], s6[11]);
281 s7[5] = vaddq_s16(s6[5], s6[10]);
282 s7[6] = vaddq_s16(s6[6], s5[9]);
283 s7[7] = vaddq_s16(s6[7], s5[8]);
284 s7[8] = vsubq_s16(s6[7], s5[8]);
285 s7[9] = vsubq_s16(s6[6], s5[9]);
286 s7[10] = vsubq_s16(s6[5], s6[10]);
287 s7[11] = vsubq_s16(s6[4], s6[11]);
288 s7[12] = vsubq_s16(s6[3], s6[12]);
289 s7[13] = vsubq_s16(s6[2], s6[13]);
290 s7[14] = vsubq_s16(s6[1], s5[14]);
291 s7[15] = vsubq_s16(s6[0], s5[15]);
293 s7[20] = sub_multiply_shift_and_narrow_s16(s6[27], s6[20], cospi_16_64);
294 s7[27] = add_multiply_shift_and_narrow_s16(s6[20], s6[27], cospi_16_64);
296 s7[21] = sub_multiply_shift_and_narrow_s16(s6[26], s6[21], cospi_16_64);
297 s7[26] = add_multiply_shift_and_narrow_s16(s6[21], s6[26], cospi_16_64);
299 s7[22] = sub_multiply_shift_and_narrow_s16(s6[25], s6[22], cospi_16_64);
300 s7[25] = add_multiply_shift_and_narrow_s16(s6[22], s6[25], cospi_16_64);
302 s7[23] = sub_multiply_shift_and_narrow_s16(s6[24], s6[23], cospi_16_64);
303 s7[24] = add_multiply_shift_and_narrow_s16(s6[23], s6[24], cospi_16_64);
306 vst1q_s16(output, vaddq_s16(s7[0], s6[31]));
308 vst1q_s16(output, vaddq_s16(s7[1], s6[30]));
310 vst1q_s16(output, vaddq_s16(s7[2], s6[29]));
312 vst1q_s16(output, vaddq_s16(s7[3], s6[28]));
314 vst1q_s16(output, vaddq_s16(s7[4], s7[27]));
316 vst1q_s16(output, vaddq_s16(s7[5], s7[26]));
318 vst1q_s16(output, vaddq_s16(s7[6], s7[25]));
320 vst1q_s16(output, vaddq_s16(s7[7], s7[24]));
323 vst1q_s16(output, vaddq_s16(s7[8], s7[23]));
325 vst1q_s16(output, vaddq_s16(s7[9], s7[22]));
327 vst1q_s16(output, vaddq_s16(s7[10], s7[21]));
329 vst1q_s16(output, vaddq_s16(s7[11], s7[20]));
331 vst1q_s16(output, vaddq_s16(s7[12], s6[19]));
333 vst1q_s16(output, vaddq_s16(s7[13], s6[18]));
335 vst1q_s16(output, vaddq_s16(s7[14], s6[17]));
337 vst1q_s16(output, vaddq_s16(s7[15], s6[16]));
340 vst1q_s16(output, vsubq_s16(s7[15], s6[16]));
342 vst1q_s16(output, vsubq_s16(s7[14], s6[17]));
344 vst1q_s16(output, vsubq_s16(s7[13], s6[18]));
346 vst1q_s16(output, vsubq_s16(s7[12], s6[19]));
348 vst1q_s16(output, vsubq_s16(s7[11], s7[20]));
350 vst1q_s16(output, vsubq_s16(s7[10], s7[21]));
352 vst1q_s16(output, vsubq_s16(s7[9], s7[22]));
354 vst1q_s16(output, vsubq_s16(s7[8], s7[23]));
357 vst1q_s16(output, vsubq_s16(s7[7], s7[24]));
359 vst1q_s16(output, vsubq_s16(s7[6], s7[25]));
361 vst1q_s16(output, vsubq_s16(s7[5], s7[26]));
363 vst1q_s16(output, vsubq_s16(s7[4], s7[27]));
365 vst1q_s16(output, vsubq_s16(s7[3], s6[28]));
367 vst1q_s16(output, vsubq_s16(s7[2], s6[29]));
369 vst1q_s16(output, vsubq_s16(s7[1], s6[30]));
371 vst1q_s16(output, vsubq_s16(s7[0], s6[31]));
374 void vpx_idct32_16_neon(const int16_t *const input, void *const output,
375 const int stride, const int highbd_flag) {
376 int16x8_t in[16], s1[32], s2[32], s3[32], s4[32], s5[32], s6[32], s7[32],
379 load_and_transpose_s16_8x8(input, 16, &in[0], &in[1], &in[2], &in[3], &in[4],
380 &in[5], &in[6], &in[7]);
382 load_and_transpose_s16_8x8(input + 8, 16, &in[8], &in[9], &in[10], &in[11],
383 &in[12], &in[13], &in[14], &in[15]);
386 s1[16] = multiply_shift_and_narrow_s16(in[1], cospi_31_64);
387 s1[31] = multiply_shift_and_narrow_s16(in[1], cospi_1_64);
389 s1[17] = multiply_shift_and_narrow_s16(in[15], -cospi_17_64);
390 s1[30] = multiply_shift_and_narrow_s16(in[15], cospi_15_64);
392 s1[18] = multiply_shift_and_narrow_s16(in[9], cospi_23_64);
393 s1[29] = multiply_shift_and_narrow_s16(in[9], cospi_9_64);
395 s1[19] = multiply_shift_and_narrow_s16(in[7], -cospi_25_64);
396 s1[28] = multiply_shift_and_narrow_s16(in[7], cospi_7_64);
398 s1[20] = multiply_shift_and_narrow_s16(in[5], cospi_27_64);
399 s1[27] = multiply_shift_and_narrow_s16(in[5], cospi_5_64);
401 s1[21] = multiply_shift_and_narrow_s16(in[11], -cospi_21_64);
402 s1[26] = multiply_shift_and_narrow_s16(in[11], cospi_11_64);
404 s1[22] = multiply_shift_and_narrow_s16(in[13], cospi_19_64);
405 s1[25] = multiply_shift_and_narrow_s16(in[13], cospi_13_64);
407 s1[23] = multiply_shift_and_narrow_s16(in[3], -cospi_29_64);
408 s1[24] = multiply_shift_and_narrow_s16(in[3], cospi_3_64);
411 s2[8] = multiply_shift_and_narrow_s16(in[2], cospi_30_64);
412 s2[15] = multiply_shift_and_narrow_s16(in[2], cospi_2_64);
414 s2[9] = multiply_shift_and_narrow_s16(in[14], -cospi_18_64);
415 s2[14] = multiply_shift_and_narrow_s16(in[14], cospi_14_64);
417 s2[10] = multiply_shift_and_narrow_s16(in[10], cospi_22_64);
418 s2[13] = multiply_shift_and_narrow_s16(in[10], cospi_10_64);
420 s2[11] = multiply_shift_and_narrow_s16(in[6], -cospi_26_64);
421 s2[12] = multiply_shift_and_narrow_s16(in[6], cospi_6_64);
423 s2[16] = vaddq_s16(s1[16], s1[17]);
424 s2[17] = vsubq_s16(s1[16], s1[17]);
425 s2[18] = vsubq_s16(s1[19], s1[18]);
426 s2[19] = vaddq_s16(s1[18], s1[19]);
427 s2[20] = vaddq_s16(s1[20], s1[21]);
428 s2[21] = vsubq_s16(s1[20], s1[21]);
429 s2[22] = vsubq_s16(s1[23], s1[22]);
430 s2[23] = vaddq_s16(s1[22], s1[23]);
431 s2[24] = vaddq_s16(s1[24], s1[25]);
432 s2[25] = vsubq_s16(s1[24], s1[25]);
433 s2[26] = vsubq_s16(s1[27], s1[26]);
434 s2[27] = vaddq_s16(s1[26], s1[27]);
435 s2[28] = vaddq_s16(s1[28], s1[29]);
436 s2[29] = vsubq_s16(s1[28], s1[29]);
437 s2[30] = vsubq_s16(s1[31], s1[30]);
438 s2[31] = vaddq_s16(s1[30], s1[31]);
441 s3[4] = multiply_shift_and_narrow_s16(in[4], cospi_28_64);
442 s3[7] = multiply_shift_and_narrow_s16(in[4], cospi_4_64);
444 s3[5] = multiply_shift_and_narrow_s16(in[12], -cospi_20_64);
445 s3[6] = multiply_shift_and_narrow_s16(in[12], cospi_12_64);
447 s3[8] = vaddq_s16(s2[8], s2[9]);
448 s3[9] = vsubq_s16(s2[8], s2[9]);
449 s3[10] = vsubq_s16(s2[11], s2[10]);
450 s3[11] = vaddq_s16(s2[10], s2[11]);
451 s3[12] = vaddq_s16(s2[12], s2[13]);
452 s3[13] = vsubq_s16(s2[12], s2[13]);
453 s3[14] = vsubq_s16(s2[15], s2[14]);
454 s3[15] = vaddq_s16(s2[14], s2[15]);
456 s3[17] = multiply_accumulate_shift_and_narrow_s16(s2[17], -cospi_4_64, s2[30],
458 s3[30] = multiply_accumulate_shift_and_narrow_s16(s2[17], cospi_28_64, s2[30],
461 s3[18] = multiply_accumulate_shift_and_narrow_s16(s2[18], -cospi_28_64,
462 s2[29], -cospi_4_64);
463 s3[29] = multiply_accumulate_shift_and_narrow_s16(s2[18], -cospi_4_64, s2[29],
466 s3[21] = multiply_accumulate_shift_and_narrow_s16(s2[21], -cospi_20_64,
467 s2[26], cospi_12_64);
468 s3[26] = multiply_accumulate_shift_and_narrow_s16(s2[21], cospi_12_64, s2[26],
471 s3[22] = multiply_accumulate_shift_and_narrow_s16(s2[22], -cospi_12_64,
472 s2[25], -cospi_20_64);
473 s3[25] = multiply_accumulate_shift_and_narrow_s16(s2[22], -cospi_20_64,
474 s2[25], cospi_12_64);
477 s4[0] = multiply_shift_and_narrow_s16(in[0], cospi_16_64);
478 s4[2] = multiply_shift_and_narrow_s16(in[8], cospi_24_64);
479 s4[3] = multiply_shift_and_narrow_s16(in[8], cospi_8_64);
481 s4[4] = vaddq_s16(s3[4], s3[5]);
482 s4[5] = vsubq_s16(s3[4], s3[5]);
483 s4[6] = vsubq_s16(s3[7], s3[6]);
484 s4[7] = vaddq_s16(s3[6], s3[7]);
486 s4[9] = multiply_accumulate_shift_and_narrow_s16(s3[9], -cospi_8_64, s3[14],
488 s4[14] = multiply_accumulate_shift_and_narrow_s16(s3[9], cospi_24_64, s3[14],
491 s4[10] = multiply_accumulate_shift_and_narrow_s16(s3[10], -cospi_24_64,
492 s3[13], -cospi_8_64);
493 s4[13] = multiply_accumulate_shift_and_narrow_s16(s3[10], -cospi_8_64, s3[13],
496 s4[16] = vaddq_s16(s2[16], s2[19]);
497 s4[17] = vaddq_s16(s3[17], s3[18]);
498 s4[18] = vsubq_s16(s3[17], s3[18]);
499 s4[19] = vsubq_s16(s2[16], s2[19]);
500 s4[20] = vsubq_s16(s2[23], s2[20]);
501 s4[21] = vsubq_s16(s3[22], s3[21]);
502 s4[22] = vaddq_s16(s3[21], s3[22]);
503 s4[23] = vaddq_s16(s2[20], s2[23]);
504 s4[24] = vaddq_s16(s2[24], s2[27]);
505 s4[25] = vaddq_s16(s3[25], s3[26]);
506 s4[26] = vsubq_s16(s3[25], s3[26]);
507 s4[27] = vsubq_s16(s2[24], s2[27]);
508 s4[28] = vsubq_s16(s2[31], s2[28]);
509 s4[29] = vsubq_s16(s3[30], s3[29]);
510 s4[30] = vaddq_s16(s3[29], s3[30]);
511 s4[31] = vaddq_s16(s2[28], s2[31]);
514 s5[0] = vaddq_s16(s4[0], s4[3]);
515 s5[1] = vaddq_s16(s4[0], s4[2]);
516 s5[2] = vsubq_s16(s4[0], s4[2]);
517 s5[3] = vsubq_s16(s4[0], s4[3]);
519 s5[5] = sub_multiply_shift_and_narrow_s16(s4[6], s4[5], cospi_16_64);
520 s5[6] = add_multiply_shift_and_narrow_s16(s4[5], s4[6], cospi_16_64);
522 s5[8] = vaddq_s16(s3[8], s3[11]);
523 s5[9] = vaddq_s16(s4[9], s4[10]);
524 s5[10] = vsubq_s16(s4[9], s4[10]);
525 s5[11] = vsubq_s16(s3[8], s3[11]);
526 s5[12] = vsubq_s16(s3[15], s3[12]);
527 s5[13] = vsubq_s16(s4[14], s4[13]);
528 s5[14] = vaddq_s16(s4[13], s4[14]);
529 s5[15] = vaddq_s16(s3[15], s3[12]);
531 s5[18] = multiply_accumulate_shift_and_narrow_s16(s4[18], -cospi_8_64, s4[29],
533 s5[29] = multiply_accumulate_shift_and_narrow_s16(s4[18], cospi_24_64, s4[29],
536 s5[19] = multiply_accumulate_shift_and_narrow_s16(s4[19], -cospi_8_64, s4[28],
538 s5[28] = multiply_accumulate_shift_and_narrow_s16(s4[19], cospi_24_64, s4[28],
541 s5[20] = multiply_accumulate_shift_and_narrow_s16(s4[20], -cospi_24_64,
542 s4[27], -cospi_8_64);
543 s5[27] = multiply_accumulate_shift_and_narrow_s16(s4[20], -cospi_8_64, s4[27],
546 s5[21] = multiply_accumulate_shift_and_narrow_s16(s4[21], -cospi_24_64,
547 s4[26], -cospi_8_64);
548 s5[26] = multiply_accumulate_shift_and_narrow_s16(s4[21], -cospi_8_64, s4[26],
552 s6[0] = vaddq_s16(s5[0], s4[7]);
553 s6[1] = vaddq_s16(s5[1], s5[6]);
554 s6[2] = vaddq_s16(s5[2], s5[5]);
555 s6[3] = vaddq_s16(s5[3], s4[4]);
556 s6[4] = vsubq_s16(s5[3], s4[4]);
557 s6[5] = vsubq_s16(s5[2], s5[5]);
558 s6[6] = vsubq_s16(s5[1], s5[6]);
559 s6[7] = vsubq_s16(s5[0], s4[7]);
561 s6[10] = sub_multiply_shift_and_narrow_s16(s5[13], s5[10], cospi_16_64);
562 s6[13] = add_multiply_shift_and_narrow_s16(s5[10], s5[13], cospi_16_64);
564 s6[11] = sub_multiply_shift_and_narrow_s16(s5[12], s5[11], cospi_16_64);
565 s6[12] = add_multiply_shift_and_narrow_s16(s5[11], s5[12], cospi_16_64);
567 s6[16] = vaddq_s16(s4[16], s4[23]);
568 s6[17] = vaddq_s16(s4[17], s4[22]);
569 s6[18] = vaddq_s16(s5[18], s5[21]);
570 s6[19] = vaddq_s16(s5[19], s5[20]);
571 s6[20] = vsubq_s16(s5[19], s5[20]);
572 s6[21] = vsubq_s16(s5[18], s5[21]);
573 s6[22] = vsubq_s16(s4[17], s4[22]);
574 s6[23] = vsubq_s16(s4[16], s4[23]);
575 s6[24] = vsubq_s16(s4[31], s4[24]);
576 s6[25] = vsubq_s16(s4[30], s4[25]);
577 s6[26] = vsubq_s16(s5[29], s5[26]);
578 s6[27] = vsubq_s16(s5[28], s5[27]);
579 s6[28] = vaddq_s16(s5[27], s5[28]);
580 s6[29] = vaddq_s16(s5[26], s5[29]);
581 s6[30] = vaddq_s16(s4[25], s4[30]);
582 s6[31] = vaddq_s16(s4[24], s4[31]);
585 s7[0] = vaddq_s16(s6[0], s5[15]);
586 s7[1] = vaddq_s16(s6[1], s5[14]);
587 s7[2] = vaddq_s16(s6[2], s6[13]);
588 s7[3] = vaddq_s16(s6[3], s6[12]);
589 s7[4] = vaddq_s16(s6[4], s6[11]);
590 s7[5] = vaddq_s16(s6[5], s6[10]);
591 s7[6] = vaddq_s16(s6[6], s5[9]);
592 s7[7] = vaddq_s16(s6[7], s5[8]);
593 s7[8] = vsubq_s16(s6[7], s5[8]);
594 s7[9] = vsubq_s16(s6[6], s5[9]);
595 s7[10] = vsubq_s16(s6[5], s6[10]);
596 s7[11] = vsubq_s16(s6[4], s6[11]);
597 s7[12] = vsubq_s16(s6[3], s6[12]);
598 s7[13] = vsubq_s16(s6[2], s6[13]);
599 s7[14] = vsubq_s16(s6[1], s5[14]);
600 s7[15] = vsubq_s16(s6[0], s5[15]);
602 s7[20] = sub_multiply_shift_and_narrow_s16(s6[27], s6[20], cospi_16_64);
603 s7[27] = add_multiply_shift_and_narrow_s16(s6[20], s6[27], cospi_16_64);
605 s7[21] = sub_multiply_shift_and_narrow_s16(s6[26], s6[21], cospi_16_64);
606 s7[26] = add_multiply_shift_and_narrow_s16(s6[21], s6[26], cospi_16_64);
608 s7[22] = sub_multiply_shift_and_narrow_s16(s6[25], s6[22], cospi_16_64);
609 s7[25] = add_multiply_shift_and_narrow_s16(s6[22], s6[25], cospi_16_64);
611 s7[23] = sub_multiply_shift_and_narrow_s16(s6[24], s6[23], cospi_16_64);
612 s7[24] = add_multiply_shift_and_narrow_s16(s6[23], s6[24], cospi_16_64);
615 out[0] = final_add(s7[0], s6[31]);
616 out[1] = final_add(s7[1], s6[30]);
617 out[2] = final_add(s7[2], s6[29]);
618 out[3] = final_add(s7[3], s6[28]);
619 out[4] = final_add(s7[4], s7[27]);
620 out[5] = final_add(s7[5], s7[26]);
621 out[6] = final_add(s7[6], s7[25]);
622 out[7] = final_add(s7[7], s7[24]);
623 out[8] = final_add(s7[8], s7[23]);
624 out[9] = final_add(s7[9], s7[22]);
625 out[10] = final_add(s7[10], s7[21]);
626 out[11] = final_add(s7[11], s7[20]);
627 out[12] = final_add(s7[12], s6[19]);
628 out[13] = final_add(s7[13], s6[18]);
629 out[14] = final_add(s7[14], s6[17]);
630 out[15] = final_add(s7[15], s6[16]);
631 out[16] = final_sub(s7[15], s6[16]);
632 out[17] = final_sub(s7[14], s6[17]);
633 out[18] = final_sub(s7[13], s6[18]);
634 out[19] = final_sub(s7[12], s6[19]);
635 out[20] = final_sub(s7[11], s7[20]);
636 out[21] = final_sub(s7[10], s7[21]);
637 out[22] = final_sub(s7[9], s7[22]);
638 out[23] = final_sub(s7[8], s7[23]);
639 out[24] = final_sub(s7[7], s7[24]);
640 out[25] = final_sub(s7[6], s7[25]);
641 out[26] = final_sub(s7[5], s7[26]);
642 out[27] = final_sub(s7[4], s7[27]);
643 out[28] = final_sub(s7[3], s6[28]);
644 out[29] = final_sub(s7[2], s6[29]);
645 out[30] = final_sub(s7[1], s6[30]);
646 out[31] = final_sub(s7[0], s6[31]);
649 highbd_add_and_store_bd8(out, output, stride);
651 uint8_t *const outputT = (uint8_t *)output;
652 add_and_store_u8_s16(out[0], out[1], out[2], out[3], out[4], out[5], out[6],
653 out[7], outputT, stride);
654 add_and_store_u8_s16(out[8], out[9], out[10], out[11], out[12], out[13],
655 out[14], out[15], outputT + (8 * stride), stride);
656 add_and_store_u8_s16(out[16], out[17], out[18], out[19], out[20], out[21],
657 out[22], out[23], outputT + (16 * stride), stride);
658 add_and_store_u8_s16(out[24], out[25], out[26], out[27], out[28], out[29],
659 out[30], out[31], outputT + (24 * stride), stride);
663 void vpx_idct32x32_135_add_neon(const tran_low_t *input, uint8_t *dest,
666 int16_t temp[32 * 16];
669 vpx_idct32_12_neon(input, temp);
670 vpx_idct32_12_neon(input + 32 * 8, temp + 8);
672 for (i = 0; i < 32; i += 8) {
673 vpx_idct32_16_neon(t, dest, stride, 0);