2 * Copyright (c) 2014 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.
18 #include "./vpx_config.h"
19 #if CONFIG_VP9_HIGHBITDEPTH
20 #include "vpx_dsp/vpx_dsp_common.h"
21 #endif // CONFIG_VP9_HIGHBITDEPTH
22 #include "vpx_ports/mem.h"
23 #include "vp9/common/vp9_common.h"
24 #include "vp9/encoder/vp9_resize.h"
30 #define SUBPEL_MASK ((1 << SUBPEL_BITS) - 1)
31 #define INTERP_PRECISION_BITS 32
33 typedef int16_t interp_kernel[INTERP_TAPS];
35 // Filters for interpolation (0.5-band) - note this also filters integer pels.
36 static const interp_kernel filteredinterp_filters500[(1 << SUBPEL_BITS)] = {
37 { -3, 0, 35, 64, 35, 0, -3, 0 }, { -3, -1, 34, 64, 36, 1, -3, 0 },
38 { -3, -1, 32, 64, 38, 1, -3, 0 }, { -2, -2, 31, 63, 39, 2, -3, 0 },
39 { -2, -2, 29, 63, 41, 2, -3, 0 }, { -2, -2, 28, 63, 42, 3, -4, 0 },
40 { -2, -3, 27, 63, 43, 4, -4, 0 }, { -2, -3, 25, 62, 45, 5, -4, 0 },
41 { -2, -3, 24, 62, 46, 5, -4, 0 }, { -2, -3, 23, 61, 47, 6, -4, 0 },
42 { -2, -3, 21, 60, 49, 7, -4, 0 }, { -1, -4, 20, 60, 50, 8, -4, -1 },
43 { -1, -4, 19, 59, 51, 9, -4, -1 }, { -1, -4, 17, 58, 52, 10, -4, 0 },
44 { -1, -4, 16, 57, 53, 12, -4, -1 }, { -1, -4, 15, 56, 54, 13, -4, -1 },
45 { -1, -4, 14, 55, 55, 14, -4, -1 }, { -1, -4, 13, 54, 56, 15, -4, -1 },
46 { -1, -4, 12, 53, 57, 16, -4, -1 }, { 0, -4, 10, 52, 58, 17, -4, -1 },
47 { -1, -4, 9, 51, 59, 19, -4, -1 }, { -1, -4, 8, 50, 60, 20, -4, -1 },
48 { 0, -4, 7, 49, 60, 21, -3, -2 }, { 0, -4, 6, 47, 61, 23, -3, -2 },
49 { 0, -4, 5, 46, 62, 24, -3, -2 }, { 0, -4, 5, 45, 62, 25, -3, -2 },
50 { 0, -4, 4, 43, 63, 27, -3, -2 }, { 0, -4, 3, 42, 63, 28, -2, -2 },
51 { 0, -3, 2, 41, 63, 29, -2, -2 }, { 0, -3, 2, 39, 63, 31, -2, -2 },
52 { 0, -3, 1, 38, 64, 32, -1, -3 }, { 0, -3, 1, 36, 64, 34, -1, -3 }
55 // Filters for interpolation (0.625-band) - note this also filters integer pels.
56 static const interp_kernel filteredinterp_filters625[(1 << SUBPEL_BITS)] = {
57 { -1, -8, 33, 80, 33, -8, -1, 0 }, { -1, -8, 30, 80, 35, -8, -1, 1 },
58 { -1, -8, 28, 80, 37, -7, -2, 1 }, { 0, -8, 26, 79, 39, -7, -2, 1 },
59 { 0, -8, 24, 79, 41, -7, -2, 1 }, { 0, -8, 22, 78, 43, -6, -2, 1 },
60 { 0, -8, 20, 78, 45, -5, -3, 1 }, { 0, -8, 18, 77, 48, -5, -3, 1 },
61 { 0, -8, 16, 76, 50, -4, -3, 1 }, { 0, -8, 15, 75, 52, -3, -4, 1 },
62 { 0, -7, 13, 74, 54, -3, -4, 1 }, { 0, -7, 11, 73, 56, -2, -4, 1 },
63 { 0, -7, 10, 71, 58, -1, -4, 1 }, { 1, -7, 8, 70, 60, 0, -5, 1 },
64 { 1, -6, 6, 68, 62, 1, -5, 1 }, { 1, -6, 5, 67, 63, 2, -5, 1 },
65 { 1, -6, 4, 65, 65, 4, -6, 1 }, { 1, -5, 2, 63, 67, 5, -6, 1 },
66 { 1, -5, 1, 62, 68, 6, -6, 1 }, { 1, -5, 0, 60, 70, 8, -7, 1 },
67 { 1, -4, -1, 58, 71, 10, -7, 0 }, { 1, -4, -2, 56, 73, 11, -7, 0 },
68 { 1, -4, -3, 54, 74, 13, -7, 0 }, { 1, -4, -3, 52, 75, 15, -8, 0 },
69 { 1, -3, -4, 50, 76, 16, -8, 0 }, { 1, -3, -5, 48, 77, 18, -8, 0 },
70 { 1, -3, -5, 45, 78, 20, -8, 0 }, { 1, -2, -6, 43, 78, 22, -8, 0 },
71 { 1, -2, -7, 41, 79, 24, -8, 0 }, { 1, -2, -7, 39, 79, 26, -8, 0 },
72 { 1, -2, -7, 37, 80, 28, -8, -1 }, { 1, -1, -8, 35, 80, 30, -8, -1 },
75 // Filters for interpolation (0.75-band) - note this also filters integer pels.
76 static const interp_kernel filteredinterp_filters750[(1 << SUBPEL_BITS)] = {
77 { 2, -11, 25, 96, 25, -11, 2, 0 }, { 2, -11, 22, 96, 28, -11, 2, 0 },
78 { 2, -10, 19, 95, 31, -11, 2, 0 }, { 2, -10, 17, 95, 34, -12, 2, 0 },
79 { 2, -9, 14, 94, 37, -12, 2, 0 }, { 2, -8, 12, 93, 40, -12, 1, 0 },
80 { 2, -8, 9, 92, 43, -12, 1, 1 }, { 2, -7, 7, 91, 46, -12, 1, 0 },
81 { 2, -7, 5, 90, 49, -12, 1, 0 }, { 2, -6, 3, 88, 52, -12, 0, 1 },
82 { 2, -5, 1, 86, 55, -12, 0, 1 }, { 2, -5, -1, 84, 58, -11, 0, 1 },
83 { 2, -4, -2, 82, 61, -11, -1, 1 }, { 2, -4, -4, 80, 64, -10, -1, 1 },
84 { 1, -3, -5, 77, 67, -9, -1, 1 }, { 1, -3, -6, 75, 70, -8, -2, 1 },
85 { 1, -2, -7, 72, 72, -7, -2, 1 }, { 1, -2, -8, 70, 75, -6, -3, 1 },
86 { 1, -1, -9, 67, 77, -5, -3, 1 }, { 1, -1, -10, 64, 80, -4, -4, 2 },
87 { 1, -1, -11, 61, 82, -2, -4, 2 }, { 1, 0, -11, 58, 84, -1, -5, 2 },
88 { 1, 0, -12, 55, 86, 1, -5, 2 }, { 1, 0, -12, 52, 88, 3, -6, 2 },
89 { 0, 1, -12, 49, 90, 5, -7, 2 }, { 0, 1, -12, 46, 91, 7, -7, 2 },
90 { 1, 1, -12, 43, 92, 9, -8, 2 }, { 0, 1, -12, 40, 93, 12, -8, 2 },
91 { 0, 2, -12, 37, 94, 14, -9, 2 }, { 0, 2, -12, 34, 95, 17, -10, 2 },
92 { 0, 2, -11, 31, 95, 19, -10, 2 }, { 0, 2, -11, 28, 96, 22, -11, 2 }
95 // Filters for interpolation (0.875-band) - note this also filters integer pels.
96 static const interp_kernel filteredinterp_filters875[(1 << SUBPEL_BITS)] = {
97 { 3, -8, 13, 112, 13, -8, 3, 0 }, { 3, -7, 10, 112, 17, -9, 3, -1 },
98 { 2, -6, 7, 111, 21, -9, 3, -1 }, { 2, -5, 4, 111, 24, -10, 3, -1 },
99 { 2, -4, 1, 110, 28, -11, 3, -1 }, { 1, -3, -1, 108, 32, -12, 4, -1 },
100 { 1, -2, -3, 106, 36, -13, 4, -1 }, { 1, -1, -6, 105, 40, -14, 4, -1 },
101 { 1, -1, -7, 102, 44, -14, 4, -1 }, { 1, 0, -9, 100, 48, -15, 4, -1 },
102 { 1, 1, -11, 97, 53, -16, 4, -1 }, { 0, 1, -12, 95, 57, -16, 4, -1 },
103 { 0, 2, -13, 91, 61, -16, 4, -1 }, { 0, 2, -14, 88, 65, -16, 4, -1 },
104 { 0, 3, -15, 84, 69, -17, 4, 0 }, { 0, 3, -16, 81, 73, -16, 3, 0 },
105 { 0, 3, -16, 77, 77, -16, 3, 0 }, { 0, 3, -16, 73, 81, -16, 3, 0 },
106 { 0, 4, -17, 69, 84, -15, 3, 0 }, { -1, 4, -16, 65, 88, -14, 2, 0 },
107 { -1, 4, -16, 61, 91, -13, 2, 0 }, { -1, 4, -16, 57, 95, -12, 1, 0 },
108 { -1, 4, -16, 53, 97, -11, 1, 1 }, { -1, 4, -15, 48, 100, -9, 0, 1 },
109 { -1, 4, -14, 44, 102, -7, -1, 1 }, { -1, 4, -14, 40, 105, -6, -1, 1 },
110 { -1, 4, -13, 36, 106, -3, -2, 1 }, { -1, 4, -12, 32, 108, -1, -3, 1 },
111 { -1, 3, -11, 28, 110, 1, -4, 2 }, { -1, 3, -10, 24, 111, 4, -5, 2 },
112 { -1, 3, -9, 21, 111, 7, -6, 2 }, { -1, 3, -9, 17, 112, 10, -7, 3 }
115 // Filters for interpolation (full-band) - no filtering for integer pixels
116 static const interp_kernel filteredinterp_filters1000[(1 << SUBPEL_BITS)] = {
117 { 0, 0, 0, 128, 0, 0, 0, 0 }, { 0, 1, -3, 128, 3, -1, 0, 0 },
118 { -1, 2, -6, 127, 7, -2, 1, 0 }, { -1, 3, -9, 126, 12, -4, 1, 0 },
119 { -1, 4, -12, 125, 16, -5, 1, 0 }, { -1, 4, -14, 123, 20, -6, 2, 0 },
120 { -1, 5, -15, 120, 25, -8, 2, 0 }, { -1, 5, -17, 118, 30, -9, 3, -1 },
121 { -1, 6, -18, 114, 35, -10, 3, -1 }, { -1, 6, -19, 111, 41, -12, 3, -1 },
122 { -1, 6, -20, 107, 46, -13, 4, -1 }, { -1, 6, -21, 103, 52, -14, 4, -1 },
123 { -1, 6, -21, 99, 57, -16, 5, -1 }, { -1, 6, -21, 94, 63, -17, 5, -1 },
124 { -1, 6, -20, 89, 68, -18, 5, -1 }, { -1, 6, -20, 84, 73, -19, 6, -1 },
125 { -1, 6, -20, 79, 79, -20, 6, -1 }, { -1, 6, -19, 73, 84, -20, 6, -1 },
126 { -1, 5, -18, 68, 89, -20, 6, -1 }, { -1, 5, -17, 63, 94, -21, 6, -1 },
127 { -1, 5, -16, 57, 99, -21, 6, -1 }, { -1, 4, -14, 52, 103, -21, 6, -1 },
128 { -1, 4, -13, 46, 107, -20, 6, -1 }, { -1, 3, -12, 41, 111, -19, 6, -1 },
129 { -1, 3, -10, 35, 114, -18, 6, -1 }, { -1, 3, -9, 30, 118, -17, 5, -1 },
130 { 0, 2, -8, 25, 120, -15, 5, -1 }, { 0, 2, -6, 20, 123, -14, 4, -1 },
131 { 0, 1, -5, 16, 125, -12, 4, -1 }, { 0, 1, -4, 12, 126, -9, 3, -1 },
132 { 0, 1, -2, 7, 127, -6, 2, -1 }, { 0, 0, -1, 3, 128, -3, 1, 0 }
135 // Filters for factor of 2 downsampling.
136 static const int16_t vp9_down2_symeven_half_filter[] = { 56, 12, -3, -1 };
137 static const int16_t vp9_down2_symodd_half_filter[] = { 64, 35, 0, -3 };
139 static const interp_kernel *choose_interp_filter(int inlength, int outlength) {
140 int outlength16 = outlength * 16;
141 if (outlength16 >= inlength * 16)
142 return filteredinterp_filters1000;
143 else if (outlength16 >= inlength * 13)
144 return filteredinterp_filters875;
145 else if (outlength16 >= inlength * 11)
146 return filteredinterp_filters750;
147 else if (outlength16 >= inlength * 9)
148 return filteredinterp_filters625;
150 return filteredinterp_filters500;
153 static void interpolate(const uint8_t *const input, int inlength,
154 uint8_t *output, int outlength) {
155 const int64_t delta =
156 (((uint64_t)inlength << 32) + outlength / 2) / outlength;
157 const int64_t offset =
159 ? (((int64_t)(inlength - outlength) << 31) + outlength / 2) /
161 : -(((int64_t)(outlength - inlength) << 31) + outlength / 2) /
163 uint8_t *optr = output;
164 int x, x1, x2, sum, k, int_pel, sub_pel;
167 const interp_kernel *interp_filters =
168 choose_interp_filter(inlength, outlength);
172 while ((y >> INTERP_PRECISION_BITS) < (INTERP_TAPS / 2 - 1)) {
178 y = delta * x + offset;
179 while ((y >> INTERP_PRECISION_BITS) + (int64_t)(INTERP_TAPS / 2) >=
186 for (x = 0, y = offset; x < outlength; ++x, y += delta) {
187 const int16_t *filter;
188 int_pel = y >> INTERP_PRECISION_BITS;
189 sub_pel = (y >> (INTERP_PRECISION_BITS - SUBPEL_BITS)) & SUBPEL_MASK;
190 filter = interp_filters[sub_pel];
192 for (k = 0; k < INTERP_TAPS; ++k) {
193 const int pk = int_pel - INTERP_TAPS / 2 + 1 + k;
195 input[(pk < 0 ? 0 : (pk >= inlength ? inlength - 1 : pk))];
197 *optr++ = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
201 for (x = 0, y = offset; x < x1; ++x, y += delta) {
202 const int16_t *filter;
203 int_pel = y >> INTERP_PRECISION_BITS;
204 sub_pel = (y >> (INTERP_PRECISION_BITS - SUBPEL_BITS)) & SUBPEL_MASK;
205 filter = interp_filters[sub_pel];
207 for (k = 0; k < INTERP_TAPS; ++k)
208 sum += filter[k] * input[(int_pel - INTERP_TAPS / 2 + 1 + k < 0
210 : int_pel - INTERP_TAPS / 2 + 1 + k)];
211 *optr++ = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
214 for (; x <= x2; ++x, y += delta) {
215 const int16_t *filter;
216 int_pel = y >> INTERP_PRECISION_BITS;
217 sub_pel = (y >> (INTERP_PRECISION_BITS - SUBPEL_BITS)) & SUBPEL_MASK;
218 filter = interp_filters[sub_pel];
220 for (k = 0; k < INTERP_TAPS; ++k)
221 sum += filter[k] * input[int_pel - INTERP_TAPS / 2 + 1 + k];
222 *optr++ = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
225 for (; x < outlength; ++x, y += delta) {
226 const int16_t *filter;
227 int_pel = y >> INTERP_PRECISION_BITS;
228 sub_pel = (y >> (INTERP_PRECISION_BITS - SUBPEL_BITS)) & SUBPEL_MASK;
229 filter = interp_filters[sub_pel];
231 for (k = 0; k < INTERP_TAPS; ++k)
232 sum += filter[k] * input[(int_pel - INTERP_TAPS / 2 + 1 + k >= inlength
234 : int_pel - INTERP_TAPS / 2 + 1 + k)];
235 *optr++ = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
240 static void down2_symeven(const uint8_t *const input, int length,
242 // Actual filter len = 2 * filter_len_half.
243 const int16_t *filter = vp9_down2_symeven_half_filter;
244 const int filter_len_half = sizeof(vp9_down2_symeven_half_filter) / 2;
246 uint8_t *optr = output;
247 int l1 = filter_len_half;
248 int l2 = (length - filter_len_half);
252 // Short input length.
253 for (i = 0; i < length; i += 2) {
254 int sum = (1 << (FILTER_BITS - 1));
255 for (j = 0; j < filter_len_half; ++j) {
256 sum += (input[(i - j < 0 ? 0 : i - j)] +
257 input[(i + 1 + j >= length ? length - 1 : i + 1 + j)]) *
261 *optr++ = clip_pixel(sum);
265 for (i = 0; i < l1; i += 2) {
266 int sum = (1 << (FILTER_BITS - 1));
267 for (j = 0; j < filter_len_half; ++j) {
268 sum += (input[(i - j < 0 ? 0 : i - j)] + input[i + 1 + j]) * filter[j];
271 *optr++ = clip_pixel(sum);
274 for (; i < l2; i += 2) {
275 int sum = (1 << (FILTER_BITS - 1));
276 for (j = 0; j < filter_len_half; ++j) {
277 sum += (input[i - j] + input[i + 1 + j]) * filter[j];
280 *optr++ = clip_pixel(sum);
283 for (; i < length; i += 2) {
284 int sum = (1 << (FILTER_BITS - 1));
285 for (j = 0; j < filter_len_half; ++j) {
286 sum += (input[i - j] +
287 input[(i + 1 + j >= length ? length - 1 : i + 1 + j)]) *
291 *optr++ = clip_pixel(sum);
296 static void down2_symodd(const uint8_t *const input, int length,
298 // Actual filter len = 2 * filter_len_half - 1.
299 const int16_t *filter = vp9_down2_symodd_half_filter;
300 const int filter_len_half = sizeof(vp9_down2_symodd_half_filter) / 2;
302 uint8_t *optr = output;
303 int l1 = filter_len_half - 1;
304 int l2 = (length - filter_len_half + 1);
308 // Short input length.
309 for (i = 0; i < length; i += 2) {
310 int sum = (1 << (FILTER_BITS - 1)) + input[i] * filter[0];
311 for (j = 1; j < filter_len_half; ++j) {
312 sum += (input[(i - j < 0 ? 0 : i - j)] +
313 input[(i + j >= length ? length - 1 : i + j)]) *
317 *optr++ = clip_pixel(sum);
321 for (i = 0; i < l1; i += 2) {
322 int sum = (1 << (FILTER_BITS - 1)) + input[i] * filter[0];
323 for (j = 1; j < filter_len_half; ++j) {
324 sum += (input[(i - j < 0 ? 0 : i - j)] + input[i + j]) * filter[j];
327 *optr++ = clip_pixel(sum);
330 for (; i < l2; i += 2) {
331 int sum = (1 << (FILTER_BITS - 1)) + input[i] * filter[0];
332 for (j = 1; j < filter_len_half; ++j) {
333 sum += (input[i - j] + input[i + j]) * filter[j];
336 *optr++ = clip_pixel(sum);
339 for (; i < length; i += 2) {
340 int sum = (1 << (FILTER_BITS - 1)) + input[i] * filter[0];
341 for (j = 1; j < filter_len_half; ++j) {
342 sum += (input[i - j] + input[(i + j >= length ? length - 1 : i + j)]) *
346 *optr++ = clip_pixel(sum);
351 static int get_down2_length(int length, int steps) {
353 for (s = 0; s < steps; ++s) length = (length + 1) >> 1;
357 static int get_down2_steps(int in_length, int out_length) {
360 while ((proj_in_length = get_down2_length(in_length, 1)) >= out_length) {
362 in_length = proj_in_length;
367 static void resize_multistep(const uint8_t *const input, int length,
368 uint8_t *output, int olength, uint8_t *otmp) {
370 if (length == olength) {
371 memcpy(output, input, sizeof(output[0]) * length);
374 steps = get_down2_steps(length, olength);
380 int filteredlength = length;
382 assert(otmp != NULL);
383 otmp2 = otmp + get_down2_length(length, 1);
384 for (s = 0; s < steps; ++s) {
385 const int proj_filteredlength = get_down2_length(filteredlength, 1);
386 const uint8_t *const in = (s == 0 ? input : out);
387 if (s == steps - 1 && proj_filteredlength == olength)
390 out = (s & 1 ? otmp2 : otmp);
391 if (filteredlength & 1)
392 down2_symodd(in, filteredlength, out);
394 down2_symeven(in, filteredlength, out);
395 filteredlength = proj_filteredlength;
397 if (filteredlength != olength) {
398 interpolate(out, filteredlength, output, olength);
401 interpolate(input, length, output, olength);
405 static void fill_col_to_arr(uint8_t *img, int stride, int len, uint8_t *arr) {
409 for (i = 0; i < len; ++i, iptr += stride) {
414 static void fill_arr_to_col(uint8_t *img, int stride, int len, uint8_t *arr) {
418 for (i = 0; i < len; ++i, iptr += stride) {
423 void vp9_resize_plane(const uint8_t *const input, int height, int width,
424 int in_stride, uint8_t *output, int height2, int width2,
427 uint8_t *intbuf = (uint8_t *)malloc(sizeof(uint8_t) * width2 * height);
429 (uint8_t *)malloc(sizeof(uint8_t) * (width < height ? height : width));
430 uint8_t *arrbuf = (uint8_t *)malloc(sizeof(uint8_t) * height);
431 uint8_t *arrbuf2 = (uint8_t *)malloc(sizeof(uint8_t) * height2);
432 if (intbuf == NULL || tmpbuf == NULL || arrbuf == NULL || arrbuf2 == NULL)
438 for (i = 0; i < height; ++i)
439 resize_multistep(input + in_stride * i, width, intbuf + width2 * i, width2,
441 for (i = 0; i < width2; ++i) {
442 fill_col_to_arr(intbuf + i, width2, height, arrbuf);
443 resize_multistep(arrbuf, height, arrbuf2, height2, tmpbuf);
444 fill_arr_to_col(output + i, out_stride, height2, arrbuf2);
454 #if CONFIG_VP9_HIGHBITDEPTH
455 static void highbd_interpolate(const uint16_t *const input, int inlength,
456 uint16_t *output, int outlength, int bd) {
457 const int64_t delta =
458 (((uint64_t)inlength << 32) + outlength / 2) / outlength;
459 const int64_t offset =
461 ? (((int64_t)(inlength - outlength) << 31) + outlength / 2) /
463 : -(((int64_t)(outlength - inlength) << 31) + outlength / 2) /
465 uint16_t *optr = output;
466 int x, x1, x2, sum, k, int_pel, sub_pel;
469 const interp_kernel *interp_filters =
470 choose_interp_filter(inlength, outlength);
474 while ((y >> INTERP_PRECISION_BITS) < (INTERP_TAPS / 2 - 1)) {
480 y = delta * x + offset;
481 while ((y >> INTERP_PRECISION_BITS) + (int64_t)(INTERP_TAPS / 2) >=
488 for (x = 0, y = offset; x < outlength; ++x, y += delta) {
489 const int16_t *filter;
490 int_pel = y >> INTERP_PRECISION_BITS;
491 sub_pel = (y >> (INTERP_PRECISION_BITS - SUBPEL_BITS)) & SUBPEL_MASK;
492 filter = interp_filters[sub_pel];
494 for (k = 0; k < INTERP_TAPS; ++k) {
495 const int pk = int_pel - INTERP_TAPS / 2 + 1 + k;
497 input[(pk < 0 ? 0 : (pk >= inlength ? inlength - 1 : pk))];
499 *optr++ = clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
503 for (x = 0, y = offset; x < x1; ++x, y += delta) {
504 const int16_t *filter;
505 int_pel = y >> INTERP_PRECISION_BITS;
506 sub_pel = (y >> (INTERP_PRECISION_BITS - SUBPEL_BITS)) & SUBPEL_MASK;
507 filter = interp_filters[sub_pel];
509 for (k = 0; k < INTERP_TAPS; ++k)
510 sum += filter[k] * input[(int_pel - INTERP_TAPS / 2 + 1 + k < 0
512 : int_pel - INTERP_TAPS / 2 + 1 + k)];
513 *optr++ = clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
516 for (; x <= x2; ++x, y += delta) {
517 const int16_t *filter;
518 int_pel = y >> INTERP_PRECISION_BITS;
519 sub_pel = (y >> (INTERP_PRECISION_BITS - SUBPEL_BITS)) & SUBPEL_MASK;
520 filter = interp_filters[sub_pel];
522 for (k = 0; k < INTERP_TAPS; ++k)
523 sum += filter[k] * input[int_pel - INTERP_TAPS / 2 + 1 + k];
524 *optr++ = clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
527 for (; x < outlength; ++x, y += delta) {
528 const int16_t *filter;
529 int_pel = y >> INTERP_PRECISION_BITS;
530 sub_pel = (y >> (INTERP_PRECISION_BITS - SUBPEL_BITS)) & SUBPEL_MASK;
531 filter = interp_filters[sub_pel];
533 for (k = 0; k < INTERP_TAPS; ++k)
534 sum += filter[k] * input[(int_pel - INTERP_TAPS / 2 + 1 + k >= inlength
536 : int_pel - INTERP_TAPS / 2 + 1 + k)];
537 *optr++ = clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
542 static void highbd_down2_symeven(const uint16_t *const input, int length,
543 uint16_t *output, int bd) {
544 // Actual filter len = 2 * filter_len_half.
545 static const int16_t *filter = vp9_down2_symeven_half_filter;
546 const int filter_len_half = sizeof(vp9_down2_symeven_half_filter) / 2;
548 uint16_t *optr = output;
549 int l1 = filter_len_half;
550 int l2 = (length - filter_len_half);
554 // Short input length.
555 for (i = 0; i < length; i += 2) {
556 int sum = (1 << (FILTER_BITS - 1));
557 for (j = 0; j < filter_len_half; ++j) {
558 sum += (input[(i - j < 0 ? 0 : i - j)] +
559 input[(i + 1 + j >= length ? length - 1 : i + 1 + j)]) *
563 *optr++ = clip_pixel_highbd(sum, bd);
567 for (i = 0; i < l1; i += 2) {
568 int sum = (1 << (FILTER_BITS - 1));
569 for (j = 0; j < filter_len_half; ++j) {
570 sum += (input[(i - j < 0 ? 0 : i - j)] + input[i + 1 + j]) * filter[j];
573 *optr++ = clip_pixel_highbd(sum, bd);
576 for (; i < l2; i += 2) {
577 int sum = (1 << (FILTER_BITS - 1));
578 for (j = 0; j < filter_len_half; ++j) {
579 sum += (input[i - j] + input[i + 1 + j]) * filter[j];
582 *optr++ = clip_pixel_highbd(sum, bd);
585 for (; i < length; i += 2) {
586 int sum = (1 << (FILTER_BITS - 1));
587 for (j = 0; j < filter_len_half; ++j) {
588 sum += (input[i - j] +
589 input[(i + 1 + j >= length ? length - 1 : i + 1 + j)]) *
593 *optr++ = clip_pixel_highbd(sum, bd);
598 static void highbd_down2_symodd(const uint16_t *const input, int length,
599 uint16_t *output, int bd) {
600 // Actual filter len = 2 * filter_len_half - 1.
601 static const int16_t *filter = vp9_down2_symodd_half_filter;
602 const int filter_len_half = sizeof(vp9_down2_symodd_half_filter) / 2;
604 uint16_t *optr = output;
605 int l1 = filter_len_half - 1;
606 int l2 = (length - filter_len_half + 1);
610 // Short input length.
611 for (i = 0; i < length; i += 2) {
612 int sum = (1 << (FILTER_BITS - 1)) + input[i] * filter[0];
613 for (j = 1; j < filter_len_half; ++j) {
614 sum += (input[(i - j < 0 ? 0 : i - j)] +
615 input[(i + j >= length ? length - 1 : i + j)]) *
619 *optr++ = clip_pixel_highbd(sum, bd);
623 for (i = 0; i < l1; i += 2) {
624 int sum = (1 << (FILTER_BITS - 1)) + input[i] * filter[0];
625 for (j = 1; j < filter_len_half; ++j) {
626 sum += (input[(i - j < 0 ? 0 : i - j)] + input[i + j]) * filter[j];
629 *optr++ = clip_pixel_highbd(sum, bd);
632 for (; i < l2; i += 2) {
633 int sum = (1 << (FILTER_BITS - 1)) + input[i] * filter[0];
634 for (j = 1; j < filter_len_half; ++j) {
635 sum += (input[i - j] + input[i + j]) * filter[j];
638 *optr++ = clip_pixel_highbd(sum, bd);
641 for (; i < length; i += 2) {
642 int sum = (1 << (FILTER_BITS - 1)) + input[i] * filter[0];
643 for (j = 1; j < filter_len_half; ++j) {
644 sum += (input[i - j] + input[(i + j >= length ? length - 1 : i + j)]) *
648 *optr++ = clip_pixel_highbd(sum, bd);
653 static void highbd_resize_multistep(const uint16_t *const input, int length,
654 uint16_t *output, int olength,
655 uint16_t *otmp, int bd) {
657 if (length == olength) {
658 memcpy(output, input, sizeof(output[0]) * length);
661 steps = get_down2_steps(length, olength);
665 uint16_t *out = NULL;
667 int filteredlength = length;
669 assert(otmp != NULL);
670 otmp2 = otmp + get_down2_length(length, 1);
671 for (s = 0; s < steps; ++s) {
672 const int proj_filteredlength = get_down2_length(filteredlength, 1);
673 const uint16_t *const in = (s == 0 ? input : out);
674 if (s == steps - 1 && proj_filteredlength == olength)
677 out = (s & 1 ? otmp2 : otmp);
678 if (filteredlength & 1)
679 highbd_down2_symodd(in, filteredlength, out, bd);
681 highbd_down2_symeven(in, filteredlength, out, bd);
682 filteredlength = proj_filteredlength;
684 if (filteredlength != olength) {
685 highbd_interpolate(out, filteredlength, output, olength, bd);
688 highbd_interpolate(input, length, output, olength, bd);
692 static void highbd_fill_col_to_arr(uint16_t *img, int stride, int len,
695 uint16_t *iptr = img;
696 uint16_t *aptr = arr;
697 for (i = 0; i < len; ++i, iptr += stride) {
702 static void highbd_fill_arr_to_col(uint16_t *img, int stride, int len,
705 uint16_t *iptr = img;
706 uint16_t *aptr = arr;
707 for (i = 0; i < len; ++i, iptr += stride) {
712 void vp9_highbd_resize_plane(const uint8_t *const input, int height, int width,
713 int in_stride, uint8_t *output, int height2,
714 int width2, int out_stride, int bd) {
716 uint16_t *intbuf = (uint16_t *)malloc(sizeof(uint16_t) * width2 * height);
718 (uint16_t *)malloc(sizeof(uint16_t) * (width < height ? height : width));
719 uint16_t *arrbuf = (uint16_t *)malloc(sizeof(uint16_t) * height);
720 uint16_t *arrbuf2 = (uint16_t *)malloc(sizeof(uint16_t) * height2);
721 if (intbuf == NULL || tmpbuf == NULL || arrbuf == NULL || arrbuf2 == NULL)
723 for (i = 0; i < height; ++i) {
724 highbd_resize_multistep(CONVERT_TO_SHORTPTR(input + in_stride * i), width,
725 intbuf + width2 * i, width2, tmpbuf, bd);
727 for (i = 0; i < width2; ++i) {
728 highbd_fill_col_to_arr(intbuf + i, width2, height, arrbuf);
729 highbd_resize_multistep(arrbuf, height, arrbuf2, height2, tmpbuf, bd);
730 highbd_fill_arr_to_col(CONVERT_TO_SHORTPTR(output + i), out_stride, height2,
740 #endif // CONFIG_VP9_HIGHBITDEPTH
742 void vp9_resize_frame420(const uint8_t *const y, int y_stride,
743 const uint8_t *const u, const uint8_t *const v,
744 int uv_stride, int height, int width, uint8_t *oy,
745 int oy_stride, uint8_t *ou, uint8_t *ov,
746 int ouv_stride, int oheight, int owidth) {
747 vp9_resize_plane(y, height, width, y_stride, oy, oheight, owidth, oy_stride);
748 vp9_resize_plane(u, height / 2, width / 2, uv_stride, ou, oheight / 2,
749 owidth / 2, ouv_stride);
750 vp9_resize_plane(v, height / 2, width / 2, uv_stride, ov, oheight / 2,
751 owidth / 2, ouv_stride);
754 void vp9_resize_frame422(const uint8_t *const y, int y_stride,
755 const uint8_t *const u, const uint8_t *const v,
756 int uv_stride, int height, int width, uint8_t *oy,
757 int oy_stride, uint8_t *ou, uint8_t *ov,
758 int ouv_stride, int oheight, int owidth) {
759 vp9_resize_plane(y, height, width, y_stride, oy, oheight, owidth, oy_stride);
760 vp9_resize_plane(u, height, width / 2, uv_stride, ou, oheight, owidth / 2,
762 vp9_resize_plane(v, height, width / 2, uv_stride, ov, oheight, owidth / 2,
766 void vp9_resize_frame444(const uint8_t *const y, int y_stride,
767 const uint8_t *const u, const uint8_t *const v,
768 int uv_stride, int height, int width, uint8_t *oy,
769 int oy_stride, uint8_t *ou, uint8_t *ov,
770 int ouv_stride, int oheight, int owidth) {
771 vp9_resize_plane(y, height, width, y_stride, oy, oheight, owidth, oy_stride);
772 vp9_resize_plane(u, height, width, uv_stride, ou, oheight, owidth,
774 vp9_resize_plane(v, height, width, uv_stride, ov, oheight, owidth,
778 #if CONFIG_VP9_HIGHBITDEPTH
779 void vp9_highbd_resize_frame420(const uint8_t *const y, int y_stride,
780 const uint8_t *const u, const uint8_t *const v,
781 int uv_stride, int height, int width,
782 uint8_t *oy, int oy_stride, uint8_t *ou,
783 uint8_t *ov, int ouv_stride, int oheight,
784 int owidth, int bd) {
785 vp9_highbd_resize_plane(y, height, width, y_stride, oy, oheight, owidth,
787 vp9_highbd_resize_plane(u, height / 2, width / 2, uv_stride, ou, oheight / 2,
788 owidth / 2, ouv_stride, bd);
789 vp9_highbd_resize_plane(v, height / 2, width / 2, uv_stride, ov, oheight / 2,
790 owidth / 2, ouv_stride, bd);
793 void vp9_highbd_resize_frame422(const uint8_t *const y, int y_stride,
794 const uint8_t *const u, const uint8_t *const v,
795 int uv_stride, int height, int width,
796 uint8_t *oy, int oy_stride, uint8_t *ou,
797 uint8_t *ov, int ouv_stride, int oheight,
798 int owidth, int bd) {
799 vp9_highbd_resize_plane(y, height, width, y_stride, oy, oheight, owidth,
801 vp9_highbd_resize_plane(u, height, width / 2, uv_stride, ou, oheight,
802 owidth / 2, ouv_stride, bd);
803 vp9_highbd_resize_plane(v, height, width / 2, uv_stride, ov, oheight,
804 owidth / 2, ouv_stride, bd);
807 void vp9_highbd_resize_frame444(const uint8_t *const y, int y_stride,
808 const uint8_t *const u, const uint8_t *const v,
809 int uv_stride, int height, int width,
810 uint8_t *oy, int oy_stride, uint8_t *ou,
811 uint8_t *ov, int ouv_stride, int oheight,
812 int owidth, int bd) {
813 vp9_highbd_resize_plane(y, height, width, y_stride, oy, oheight, owidth,
815 vp9_highbd_resize_plane(u, height, width, uv_stride, ou, oheight, owidth,
817 vp9_highbd_resize_plane(v, height, width, uv_stride, ov, oheight, owidth,
820 #endif // CONFIG_VP9_HIGHBITDEPTH