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 <arm_neon.h>
13
14 #include "config/aom_dsp_rtcd.h"
15 #include "config/aom_config.h"
16
17 #include "aom_ports/mem.h"
18 #include "aom/aom_integer.h"
19
20 #include "aom_dsp/aom_filter.h"
21 #include "aom_dsp/variance.h"
22
23 // Load 2 sets of 4 bytes when alignment is not guaranteed.
load_unaligned_u8(const uint8_t * buf,int stride)24 static INLINE uint8x8_t load_unaligned_u8(const uint8_t *buf, int stride) {
25 uint32_t a;
26 uint32x2_t a_u32 = vdup_n_u32(0);
27 if (stride == 4) return vld1_u8(buf);
28 memcpy(&a, buf, 4);
29 buf += stride;
30 a_u32 = vld1_lane_u32(&a, a_u32, 0);
31 memcpy(&a, buf, 4);
32 a_u32 = vld1_lane_u32(&a, a_u32, 1);
33 return vreinterpret_u8_u32(a_u32);
34 }
35
36 // Process a block exactly 4 wide and a multiple of 2 high.
var_filter_block2d_bil_w4(const uint8_t * src_ptr,uint8_t * output_ptr,unsigned int src_pixels_per_line,int pixel_step,unsigned int output_height,const uint8_t * filter)37 static void var_filter_block2d_bil_w4(const uint8_t *src_ptr,
38 uint8_t *output_ptr,
39 unsigned int src_pixels_per_line,
40 int pixel_step,
41 unsigned int output_height,
42 const uint8_t *filter) {
43 const uint8x8_t f0 = vdup_n_u8(filter[0]);
44 const uint8x8_t f1 = vdup_n_u8(filter[1]);
45 unsigned int i;
46 for (i = 0; i < output_height; i += 2) {
47 const uint8x8_t src_0 = load_unaligned_u8(src_ptr, src_pixels_per_line);
48 const uint8x8_t src_1 =
49 load_unaligned_u8(src_ptr + pixel_step, src_pixels_per_line);
50 const uint16x8_t a = vmull_u8(src_0, f0);
51 const uint16x8_t b = vmlal_u8(a, src_1, f1);
52 const uint8x8_t out = vrshrn_n_u16(b, FILTER_BITS);
53 vst1_u8(output_ptr, out);
54 src_ptr += 2 * src_pixels_per_line;
55 output_ptr += 8;
56 }
57 }
58
var_filter_block2d_bil_w8(const uint8_t * src_ptr,uint8_t * output_ptr,unsigned int src_pixels_per_line,int pixel_step,unsigned int output_height,unsigned int output_width,const uint8_t * filter)59 static void var_filter_block2d_bil_w8(const uint8_t *src_ptr,
60 uint8_t *output_ptr,
61 unsigned int src_pixels_per_line,
62 int pixel_step,
63 unsigned int output_height,
64 unsigned int output_width,
65 const uint8_t *filter) {
66 const uint8x8_t f0 = vdup_n_u8(filter[0]);
67 const uint8x8_t f1 = vdup_n_u8(filter[1]);
68 unsigned int i;
69 for (i = 0; i < output_height; ++i) {
70 const uint8x8_t src_0 = vld1_u8(&src_ptr[0]);
71 const uint8x8_t src_1 = vld1_u8(&src_ptr[pixel_step]);
72 const uint16x8_t a = vmull_u8(src_0, f0);
73 const uint16x8_t b = vmlal_u8(a, src_1, f1);
74 const uint8x8_t out = vrshrn_n_u16(b, FILTER_BITS);
75 vst1_u8(output_ptr, out);
76 // Next row...
77 src_ptr += src_pixels_per_line;
78 output_ptr += output_width;
79 }
80 }
81
82 // Process a block which is a mutiple of 16 wide and any height.
var_filter_block2d_bil_w16(const uint8_t * src_ptr,uint8_t * output_ptr,unsigned int src_pixels_per_line,int pixel_step,unsigned int output_height,unsigned int output_width,const uint8_t * filter)83 static void var_filter_block2d_bil_w16(const uint8_t *src_ptr,
84 uint8_t *output_ptr,
85 unsigned int src_pixels_per_line,
86 int pixel_step,
87 unsigned int output_height,
88 unsigned int output_width,
89 const uint8_t *filter) {
90 const uint8x8_t f0 = vdup_n_u8(filter[0]);
91 const uint8x8_t f1 = vdup_n_u8(filter[1]);
92 unsigned int i, j;
93 for (i = 0; i < output_height; ++i) {
94 for (j = 0; j < output_width; j += 16) {
95 const uint8x16_t src_0 = vld1q_u8(&src_ptr[j]);
96 const uint8x16_t src_1 = vld1q_u8(&src_ptr[j + pixel_step]);
97 const uint16x8_t a = vmull_u8(vget_low_u8(src_0), f0);
98 const uint16x8_t b = vmlal_u8(a, vget_low_u8(src_1), f1);
99 const uint8x8_t out_lo = vrshrn_n_u16(b, FILTER_BITS);
100 const uint16x8_t c = vmull_u8(vget_high_u8(src_0), f0);
101 const uint16x8_t d = vmlal_u8(c, vget_high_u8(src_1), f1);
102 const uint8x8_t out_hi = vrshrn_n_u16(d, FILTER_BITS);
103 vst1q_u8(output_ptr + j, vcombine_u8(out_lo, out_hi));
104 }
105 src_ptr += src_pixels_per_line;
106 output_ptr += output_width;
107 }
108 }
109
aom_sub_pixel_variance8x8_neon(const uint8_t * src,int src_stride,int xoffset,int yoffset,const uint8_t * dst,int dst_stride,unsigned int * sse)110 unsigned int aom_sub_pixel_variance8x8_neon(const uint8_t *src, int src_stride,
111 int xoffset, int yoffset,
112 const uint8_t *dst, int dst_stride,
113 unsigned int *sse) {
114 DECLARE_ALIGNED(16, uint8_t, temp2[8 * 8]);
115 DECLARE_ALIGNED(16, uint8_t, fdata3[9 * 8]);
116
117 var_filter_block2d_bil_w8(src, fdata3, src_stride, 1, 9, 8,
118 bilinear_filters_2t[xoffset]);
119 var_filter_block2d_bil_w8(fdata3, temp2, 8, 8, 8, 8,
120 bilinear_filters_2t[yoffset]);
121 return aom_variance8x8_neon(temp2, 8, dst, dst_stride, sse);
122 }
123
aom_sub_pixel_variance16x16_neon(const uint8_t * src,int src_stride,int xoffset,int yoffset,const uint8_t * dst,int dst_stride,unsigned int * sse)124 unsigned int aom_sub_pixel_variance16x16_neon(const uint8_t *src,
125 int src_stride, int xoffset,
126 int yoffset, const uint8_t *dst,
127 int dst_stride,
128 unsigned int *sse) {
129 DECLARE_ALIGNED(16, uint8_t, temp2[16 * 16]);
130 DECLARE_ALIGNED(16, uint8_t, fdata3[17 * 16]);
131
132 var_filter_block2d_bil_w16(src, fdata3, src_stride, 1, 17, 16,
133 bilinear_filters_2t[xoffset]);
134 var_filter_block2d_bil_w16(fdata3, temp2, 16, 16, 16, 16,
135 bilinear_filters_2t[yoffset]);
136 return aom_variance16x16_neon(temp2, 16, dst, dst_stride, sse);
137 }
138
aom_sub_pixel_variance32x32_neon(const uint8_t * src,int src_stride,int xoffset,int yoffset,const uint8_t * dst,int dst_stride,unsigned int * sse)139 unsigned int aom_sub_pixel_variance32x32_neon(const uint8_t *src,
140 int src_stride, int xoffset,
141 int yoffset, const uint8_t *dst,
142 int dst_stride,
143 unsigned int *sse) {
144 DECLARE_ALIGNED(16, uint8_t, temp2[32 * 32]);
145 DECLARE_ALIGNED(16, uint8_t, fdata3[33 * 32]);
146
147 var_filter_block2d_bil_w16(src, fdata3, src_stride, 1, 33, 32,
148 bilinear_filters_2t[xoffset]);
149 var_filter_block2d_bil_w16(fdata3, temp2, 32, 32, 32, 32,
150 bilinear_filters_2t[yoffset]);
151 return aom_variance32x32_neon(temp2, 32, dst, dst_stride, sse);
152 }
153
aom_sub_pixel_variance64x64_neon(const uint8_t * src,int src_stride,int xoffset,int yoffset,const uint8_t * dst,int dst_stride,unsigned int * sse)154 unsigned int aom_sub_pixel_variance64x64_neon(const uint8_t *src,
155 int src_stride, int xoffset,
156 int yoffset, const uint8_t *dst,
157 int dst_stride,
158 unsigned int *sse) {
159 DECLARE_ALIGNED(16, uint8_t, temp2[64 * 64]);
160 DECLARE_ALIGNED(16, uint8_t, fdata3[65 * 64]);
161
162 var_filter_block2d_bil_w16(src, fdata3, src_stride, 1, 65, 64,
163 bilinear_filters_2t[xoffset]);
164 var_filter_block2d_bil_w16(fdata3, temp2, 64, 64, 64, 64,
165 bilinear_filters_2t[yoffset]);
166 return aom_variance64x64_neon(temp2, 64, dst, dst_stride, sse);
167 }
168
aom_sub_pixel_variance4x4_neon(const uint8_t * a,int a_stride,int xoffset,int yoffset,const uint8_t * b,int b_stride,uint32_t * sse)169 unsigned int aom_sub_pixel_variance4x4_neon(const uint8_t *a, int a_stride,
170 int xoffset, int yoffset,
171 const uint8_t *b, int b_stride,
172 uint32_t *sse) {
173 uint8_t temp0[4 * (4 + 2)];
174 uint8_t temp1[4 * 4];
175
176 var_filter_block2d_bil_w4(a, temp0, a_stride, 1, (4 + 2),
177 bilinear_filters_2t[xoffset]);
178 var_filter_block2d_bil_w4(temp0, temp1, 4, 4, 4,
179 bilinear_filters_2t[yoffset]);
180
181 return aom_variance4x4(temp1, 4, b, b_stride, sse);
182 }
183
aom_sub_pixel_variance4x8_neon(const uint8_t * a,int a_stride,int xoffset,int yoffset,const uint8_t * b,int b_stride,uint32_t * sse)184 unsigned int aom_sub_pixel_variance4x8_neon(const uint8_t *a, int a_stride,
185 int xoffset, int yoffset,
186 const uint8_t *b, int b_stride,
187 uint32_t *sse) {
188 uint8_t temp0[4 * (8 + 2)];
189 uint8_t temp1[4 * 8];
190
191 var_filter_block2d_bil_w4(a, temp0, a_stride, 1, (8 + 2),
192 bilinear_filters_2t[xoffset]);
193 var_filter_block2d_bil_w4(temp0, temp1, 4, 4, 8,
194 bilinear_filters_2t[yoffset]);
195
196 return aom_variance4x8(temp1, 4, b, b_stride, sse);
197 }
198
aom_sub_pixel_variance8x4_neon(const uint8_t * a,int a_stride,int xoffset,int yoffset,const uint8_t * b,int b_stride,uint32_t * sse)199 unsigned int aom_sub_pixel_variance8x4_neon(const uint8_t *a, int a_stride,
200 int xoffset, int yoffset,
201 const uint8_t *b, int b_stride,
202 uint32_t *sse) {
203 uint8_t temp0[8 * (4 + 1)];
204 uint8_t temp1[8 * 4];
205
206 var_filter_block2d_bil_w8(a, temp0, a_stride, 1, (4 + 1), 8,
207 bilinear_filters_2t[xoffset]);
208 var_filter_block2d_bil_w8(temp0, temp1, 8, 8, 4, 8,
209 bilinear_filters_2t[yoffset]);
210
211 return aom_variance8x4(temp1, 8, b, b_stride, sse);
212 }
213
aom_sub_pixel_variance8x16_neon(const uint8_t * a,int a_stride,int xoffset,int yoffset,const uint8_t * b,int b_stride,uint32_t * sse)214 unsigned int aom_sub_pixel_variance8x16_neon(const uint8_t *a, int a_stride,
215 int xoffset, int yoffset,
216 const uint8_t *b, int b_stride,
217 uint32_t *sse) {
218 uint8_t temp0[8 * (16 + 1)];
219 uint8_t temp1[8 * 16];
220
221 var_filter_block2d_bil_w8(a, temp0, a_stride, 1, (16 + 1), 8,
222 bilinear_filters_2t[xoffset]);
223 var_filter_block2d_bil_w8(temp0, temp1, 8, 8, 16, 8,
224 bilinear_filters_2t[yoffset]);
225
226 return aom_variance8x16(temp1, 8, b, b_stride, sse);
227 }
228
aom_sub_pixel_variance16x8_neon(const uint8_t * a,int a_stride,int xoffset,int yoffset,const uint8_t * b,int b_stride,uint32_t * sse)229 unsigned int aom_sub_pixel_variance16x8_neon(const uint8_t *a, int a_stride,
230 int xoffset, int yoffset,
231 const uint8_t *b, int b_stride,
232 uint32_t *sse) {
233 uint8_t temp0[16 * (8 + 1)];
234 uint8_t temp1[16 * 8];
235
236 var_filter_block2d_bil_w16(a, temp0, a_stride, 1, (8 + 1), 16,
237 bilinear_filters_2t[xoffset]);
238 var_filter_block2d_bil_w16(temp0, temp1, 16, 16, 8, 16,
239 bilinear_filters_2t[yoffset]);
240
241 return aom_variance16x8(temp1, 16, b, b_stride, sse);
242 }
243
aom_sub_pixel_variance16x32_neon(const uint8_t * a,int a_stride,int xoffset,int yoffset,const uint8_t * b,int b_stride,uint32_t * sse)244 unsigned int aom_sub_pixel_variance16x32_neon(const uint8_t *a, int a_stride,
245 int xoffset, int yoffset,
246 const uint8_t *b, int b_stride,
247 uint32_t *sse) {
248 uint8_t temp0[16 * (32 + 1)];
249 uint8_t temp1[16 * 32];
250
251 var_filter_block2d_bil_w16(a, temp0, a_stride, 1, (32 + 1), 16,
252 bilinear_filters_2t[xoffset]);
253 var_filter_block2d_bil_w16(temp0, temp1, 16, 16, 32, 16,
254 bilinear_filters_2t[yoffset]);
255
256 return aom_variance16x32(temp1, 16, b, b_stride, sse);
257 }
258
aom_sub_pixel_variance32x16_neon(const uint8_t * a,int a_stride,int xoffset,int yoffset,const uint8_t * b,int b_stride,uint32_t * sse)259 unsigned int aom_sub_pixel_variance32x16_neon(const uint8_t *a, int a_stride,
260 int xoffset, int yoffset,
261 const uint8_t *b, int b_stride,
262 uint32_t *sse) {
263 uint8_t temp0[32 * (16 + 1)];
264 uint8_t temp1[32 * 16];
265
266 var_filter_block2d_bil_w16(a, temp0, a_stride, 1, (16 + 1), 32,
267 bilinear_filters_2t[xoffset]);
268 var_filter_block2d_bil_w16(temp0, temp1, 32, 32, 16, 32,
269 bilinear_filters_2t[yoffset]);
270
271 return aom_variance32x16(temp1, 32, b, b_stride, sse);
272 }
273
aom_sub_pixel_variance32x64_neon(const uint8_t * a,int a_stride,int xoffset,int yoffset,const uint8_t * b,int b_stride,uint32_t * sse)274 unsigned int aom_sub_pixel_variance32x64_neon(const uint8_t *a, int a_stride,
275 int xoffset, int yoffset,
276 const uint8_t *b, int b_stride,
277 uint32_t *sse) {
278 uint8_t temp0[32 * (64 + 1)];
279 uint8_t temp1[32 * 64];
280
281 var_filter_block2d_bil_w16(a, temp0, a_stride, 1, (64 + 1), 32,
282 bilinear_filters_2t[xoffset]);
283 var_filter_block2d_bil_w16(temp0, temp1, 32, 32, 64, 32,
284 bilinear_filters_2t[yoffset]);
285
286 return aom_variance32x64(temp1, 32, b, b_stride, sse);
287 }
288
aom_sub_pixel_variance64x32_neon(const uint8_t * a,int a_stride,int xoffset,int yoffset,const uint8_t * b,int b_stride,uint32_t * sse)289 unsigned int aom_sub_pixel_variance64x32_neon(const uint8_t *a, int a_stride,
290 int xoffset, int yoffset,
291 const uint8_t *b, int b_stride,
292 uint32_t *sse) {
293 uint8_t temp0[64 * (32 + 1)];
294 uint8_t temp1[64 * 32];
295
296 var_filter_block2d_bil_w16(a, temp0, a_stride, 1, (32 + 1), 64,
297 bilinear_filters_2t[xoffset]);
298 var_filter_block2d_bil_w16(temp0, temp1, 64, 64, 32, 64,
299 bilinear_filters_2t[yoffset]);
300
301 return aom_variance64x32(temp1, 64, b, b_stride, sse);
302 }
303
aom_sub_pixel_variance64x128_neon(const uint8_t * a,int a_stride,int xoffset,int yoffset,const uint8_t * b,int b_stride,uint32_t * sse)304 unsigned int aom_sub_pixel_variance64x128_neon(const uint8_t *a, int a_stride,
305 int xoffset, int yoffset,
306 const uint8_t *b, int b_stride,
307 uint32_t *sse) {
308 uint8_t temp0[64 * (128 + 1)];
309 uint8_t temp1[64 * 128];
310
311 var_filter_block2d_bil_w16(a, temp0, a_stride, 1, (128 + 1), 64,
312 bilinear_filters_2t[xoffset]);
313 var_filter_block2d_bil_w16(temp0, temp1, 64, 64, 128, 64,
314 bilinear_filters_2t[yoffset]);
315
316 return aom_variance64x128(temp1, 64, b, b_stride, sse);
317 }
318
aom_sub_pixel_variance128x64_neon(const uint8_t * a,int a_stride,int xoffset,int yoffset,const uint8_t * b,int b_stride,uint32_t * sse)319 unsigned int aom_sub_pixel_variance128x64_neon(const uint8_t *a, int a_stride,
320 int xoffset, int yoffset,
321 const uint8_t *b, int b_stride,
322 uint32_t *sse) {
323 uint8_t temp0[128 * (64 + 1)];
324 uint8_t temp1[128 * 64];
325
326 var_filter_block2d_bil_w16(a, temp0, a_stride, 1, (64 + 1), 128,
327 bilinear_filters_2t[xoffset]);
328 var_filter_block2d_bil_w16(temp0, temp1, 128, 128, 64, 128,
329 bilinear_filters_2t[yoffset]);
330
331 return aom_variance128x64(temp1, 128, b, b_stride, sse);
332 }
333
aom_sub_pixel_variance128x128_neon(const uint8_t * a,int a_stride,int xoffset,int yoffset,const uint8_t * b,int b_stride,uint32_t * sse)334 unsigned int aom_sub_pixel_variance128x128_neon(const uint8_t *a, int a_stride,
335 int xoffset, int yoffset,
336 const uint8_t *b, int b_stride,
337 uint32_t *sse) {
338 uint8_t temp0[128 * (128 + 1)];
339 uint8_t temp1[128 * 128];
340
341 var_filter_block2d_bil_w16(a, temp0, a_stride, 1, (128 + 1), 128,
342 bilinear_filters_2t[xoffset]);
343 var_filter_block2d_bil_w16(temp0, temp1, 128, 128, 128, 128,
344 bilinear_filters_2t[yoffset]);
345
346 return aom_variance128x128(temp1, 128, b, b_stride, sse);
347 }
348
349 // Realtime mode doesn't use 4x rectangular blocks.
350 #if !CONFIG_REALTIME_ONLY
aom_sub_pixel_variance4x16_neon(const uint8_t * a,int a_stride,int xoffset,int yoffset,const uint8_t * b,int b_stride,uint32_t * sse)351 unsigned int aom_sub_pixel_variance4x16_neon(const uint8_t *a, int a_stride,
352 int xoffset, int yoffset,
353 const uint8_t *b, int b_stride,
354 uint32_t *sse) {
355 uint8_t temp0[4 * (16 + 2)];
356 uint8_t temp1[4 * 16];
357
358 var_filter_block2d_bil_w4(a, temp0, a_stride, 1, (16 + 2),
359 bilinear_filters_2t[xoffset]);
360 var_filter_block2d_bil_w4(temp0, temp1, 4, 4, 16,
361 bilinear_filters_2t[yoffset]);
362
363 return aom_variance4x16(temp1, 4, b, b_stride, sse);
364 }
365
aom_sub_pixel_variance8x32_neon(const uint8_t * a,int a_stride,int xoffset,int yoffset,const uint8_t * b,int b_stride,uint32_t * sse)366 unsigned int aom_sub_pixel_variance8x32_neon(const uint8_t *a, int a_stride,
367 int xoffset, int yoffset,
368 const uint8_t *b, int b_stride,
369 uint32_t *sse) {
370 uint8_t temp0[8 * (32 + 1)];
371 uint8_t temp1[8 * 32];
372
373 var_filter_block2d_bil_w8(a, temp0, a_stride, 1, (32 + 1), 8,
374 bilinear_filters_2t[xoffset]);
375 var_filter_block2d_bil_w8(temp0, temp1, 8, 8, 32, 8,
376 bilinear_filters_2t[yoffset]);
377
378 return aom_variance8x32(temp1, 8, b, b_stride, sse);
379 }
380
aom_sub_pixel_variance16x4_neon(const uint8_t * a,int a_stride,int xoffset,int yoffset,const uint8_t * b,int b_stride,uint32_t * sse)381 unsigned int aom_sub_pixel_variance16x4_neon(const uint8_t *a, int a_stride,
382 int xoffset, int yoffset,
383 const uint8_t *b, int b_stride,
384 uint32_t *sse) {
385 uint8_t temp0[16 * (4 + 1)];
386 uint8_t temp1[16 * 4];
387
388 var_filter_block2d_bil_w16(a, temp0, a_stride, 1, (4 + 1), 16,
389 bilinear_filters_2t[xoffset]);
390 var_filter_block2d_bil_w16(temp0, temp1, 16, 16, 4, 16,
391 bilinear_filters_2t[yoffset]);
392
393 return aom_variance16x4(temp1, 16, b, b_stride, sse);
394 }
395
aom_sub_pixel_variance64x16_neon(const uint8_t * a,int a_stride,int xoffset,int yoffset,const uint8_t * b,int b_stride,uint32_t * sse)396 unsigned int aom_sub_pixel_variance64x16_neon(const uint8_t *a, int a_stride,
397 int xoffset, int yoffset,
398 const uint8_t *b, int b_stride,
399 uint32_t *sse) {
400 uint8_t temp0[64 * (16 + 1)];
401 uint8_t temp1[64 * 16];
402
403 var_filter_block2d_bil_w16(a, temp0, a_stride, 1, (16 + 1), 64,
404 bilinear_filters_2t[xoffset]);
405 var_filter_block2d_bil_w16(temp0, temp1, 64, 64, 16, 64,
406 bilinear_filters_2t[yoffset]);
407
408 return aom_variance64x16(temp1, 64, b, b_stride, sse);
409 }
410
aom_sub_pixel_variance16x64_neon(const uint8_t * a,int a_stride,int xoffset,int yoffset,const uint8_t * b,int b_stride,uint32_t * sse)411 unsigned int aom_sub_pixel_variance16x64_neon(const uint8_t *a, int a_stride,
412 int xoffset, int yoffset,
413 const uint8_t *b, int b_stride,
414 uint32_t *sse) {
415 uint8_t temp0[16 * (64 + 1)];
416 uint8_t temp1[16 * 64];
417
418 var_filter_block2d_bil_w16(a, temp0, a_stride, 1, (64 + 1), 16,
419 bilinear_filters_2t[xoffset]);
420 var_filter_block2d_bil_w16(temp0, temp1, 16, 16, 64, 16,
421 bilinear_filters_2t[yoffset]);
422
423 return aom_variance16x64(temp1, 16, b, b_stride, sse);
424 }
425
aom_sub_pixel_variance32x8_neon(const uint8_t * a,int a_stride,int xoffset,int yoffset,const uint8_t * b,int b_stride,uint32_t * sse)426 unsigned int aom_sub_pixel_variance32x8_neon(const uint8_t *a, int a_stride,
427 int xoffset, int yoffset,
428 const uint8_t *b, int b_stride,
429 uint32_t *sse) {
430 uint8_t temp0[32 * (8 + 1)];
431 uint8_t temp1[32 * 8];
432
433 var_filter_block2d_bil_w16(a, temp0, a_stride, 1, (8 + 1), 32,
434 bilinear_filters_2t[xoffset]);
435 var_filter_block2d_bil_w16(temp0, temp1, 32, 32, 8, 32,
436 bilinear_filters_2t[yoffset]);
437
438 return aom_variance32x8(temp1, 32, b, b_stride, sse);
439 }
440 #endif // !CONFIG_REALTIME_ONLY
441