1 /*
2 * Copyright (c) 2019, 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 <emmintrin.h>
13
14 #include "aom_dsp/x86/synonyms.h"
15 #include "av1/common/warped_motion.h"
16 #include "config/av1_rtcd.h"
17
av1_calc_frame_error_sse2(const uint8_t * const ref,int ref_stride,const uint8_t * const dst,int p_width,int p_height,int dst_stride)18 int64_t av1_calc_frame_error_sse2(const uint8_t *const ref, int ref_stride,
19 const uint8_t *const dst, int p_width,
20 int p_height, int dst_stride) {
21 int64_t sum_error = 0;
22 int i, j;
23 __m128i row_error, col_error;
24 __m128i zero = _mm_set1_epi16(0);
25 __m128i dup_255 = _mm_set1_epi16(255);
26 col_error = zero;
27 for (i = 0; i < (p_height); i++) {
28 row_error = zero;
29 for (j = 0; j < (p_width / 16); j++) {
30 __m128i ref_8 =
31 _mm_load_si128((__m128i *)(ref + (j * 16) + (i * ref_stride)));
32 __m128i dst_8 =
33 _mm_load_si128((__m128i *)(dst + (j * 16) + (i * dst_stride)));
34 __m128i ref_16_lo = _mm_unpacklo_epi8(ref_8, zero);
35 __m128i ref_16_hi = _mm_unpackhi_epi8(ref_8, zero);
36 __m128i dst_16_lo = _mm_unpacklo_epi8(dst_8, zero);
37 __m128i dst_16_hi = _mm_unpackhi_epi8(dst_8, zero);
38
39 __m128i diff_1 =
40 _mm_add_epi16(_mm_sub_epi16(dst_16_lo, ref_16_lo), dup_255);
41 __m128i diff_2 =
42 _mm_add_epi16(_mm_sub_epi16(dst_16_hi, ref_16_hi), dup_255);
43
44 __m128i error_1_lo =
45 _mm_set_epi32(error_measure_lut[_mm_extract_epi16(diff_1, 3)],
46 error_measure_lut[_mm_extract_epi16(diff_1, 2)],
47 error_measure_lut[_mm_extract_epi16(diff_1, 1)],
48 error_measure_lut[_mm_extract_epi16(diff_1, 0)]);
49 __m128i error_1_hi =
50 _mm_set_epi32(error_measure_lut[_mm_extract_epi16(diff_1, 7)],
51 error_measure_lut[_mm_extract_epi16(diff_1, 6)],
52 error_measure_lut[_mm_extract_epi16(diff_1, 5)],
53 error_measure_lut[_mm_extract_epi16(diff_1, 4)]);
54 __m128i error_2_lo =
55 _mm_set_epi32(error_measure_lut[_mm_extract_epi16(diff_2, 3)],
56 error_measure_lut[_mm_extract_epi16(diff_2, 2)],
57 error_measure_lut[_mm_extract_epi16(diff_2, 1)],
58 error_measure_lut[_mm_extract_epi16(diff_2, 0)]);
59 __m128i error_2_hi =
60 _mm_set_epi32(error_measure_lut[_mm_extract_epi16(diff_2, 7)],
61 error_measure_lut[_mm_extract_epi16(diff_2, 6)],
62 error_measure_lut[_mm_extract_epi16(diff_2, 5)],
63 error_measure_lut[_mm_extract_epi16(diff_2, 4)]);
64
65 __m128i error_1 = _mm_add_epi32(error_1_lo, error_1_hi);
66 __m128i error_2 = _mm_add_epi32(error_2_lo, error_2_hi);
67 __m128i error_1_2 = _mm_add_epi32(error_1, error_2);
68
69 row_error = _mm_add_epi32(row_error, error_1_2);
70 }
71 __m128i col_error_lo = _mm_unpacklo_epi32(row_error, zero);
72 __m128i col_error_hi = _mm_unpackhi_epi32(row_error, zero);
73 __m128i col_error_temp = _mm_add_epi64(col_error_lo, col_error_hi);
74 col_error = _mm_add_epi64(col_error, col_error_temp);
75 // Error summation for remaining width, which is not multiple of 16
76 if (p_width & 0xf) {
77 for (int l = j * 16; l < p_width; ++l) {
78 sum_error += (int64_t)error_measure(dst[l + i * dst_stride] -
79 ref[l + i * ref_stride]);
80 }
81 }
82 }
83 int64_t sum_error_d_0, sum_error_d_1;
84 xx_storel_64(&sum_error_d_0, col_error);
85 xx_storel_64(&sum_error_d_1, _mm_srli_si128(col_error, 8));
86 sum_error = (sum_error + sum_error_d_0 + sum_error_d_1);
87 return sum_error;
88 }
89