1 /*
2 * Copyright (c) 2021, 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 #ifndef AOM_AOM_DSP_X86_CONVOLVE_SSSE3_H_
13 #define AOM_AOM_DSP_X86_CONVOLVE_SSSE3_H_
14
15 #include <tmmintrin.h> // SSSE3
16
shuffle_filter_ssse3(const int16_t * const filter,__m128i * const f)17 static INLINE void shuffle_filter_ssse3(const int16_t *const filter,
18 __m128i *const f) {
19 const __m128i f_values = _mm_load_si128((const __m128i *)filter);
20 // pack and duplicate the filter values
21 f[0] = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0200u));
22 f[1] = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0604u));
23 f[2] = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0a08u));
24 f[3] = _mm_shuffle_epi8(f_values, _mm_set1_epi16(0x0e0cu));
25 }
26
convolve8_8_ssse3(const __m128i * const s,const __m128i * const f)27 static INLINE __m128i convolve8_8_ssse3(const __m128i *const s,
28 const __m128i *const f) {
29 // multiply 2 adjacent elements with the filter and add the result
30 const __m128i k_64 = _mm_set1_epi16(1 << 6);
31 const __m128i x0 = _mm_maddubs_epi16(s[0], f[0]);
32 const __m128i x1 = _mm_maddubs_epi16(s[1], f[1]);
33 const __m128i x2 = _mm_maddubs_epi16(s[2], f[2]);
34 const __m128i x3 = _mm_maddubs_epi16(s[3], f[3]);
35 __m128i sum1, sum2;
36
37 // sum the results together, saturating only on the final step
38 // adding x0 with x2 and x1 with x3 is the only order that prevents
39 // outranges for all filters
40 sum1 = _mm_add_epi16(x0, x2);
41 sum2 = _mm_add_epi16(x1, x3);
42 // add the rounding offset early to avoid another saturated add
43 sum1 = _mm_add_epi16(sum1, k_64);
44 sum1 = _mm_adds_epi16(sum1, sum2);
45 // shift by 7 bit each 16 bit
46 sum1 = _mm_srai_epi16(sum1, 7);
47 return sum1;
48 }
49
50 #endif // AOM_AOM_DSP_X86_CONVOLVE_SSSE3_H_
51