1 /*
2 * Copyright (c) 2018, 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_SSE2_H_
13 #define AOM_AOM_DSP_X86_CONVOLVE_SSE2_H_
14
15 #include "config/aom_scale_rtcd.h"
16
17 // Note:
18 // This header file should be put below any x86 intrinsics head file
prepare_coeffs(const InterpFilterParams * const filter_params,const int subpel_q4,__m128i * const coeffs)19 static INLINE void prepare_coeffs(const InterpFilterParams *const filter_params,
20 const int subpel_q4,
21 __m128i *const coeffs /* [4] */) {
22 const int16_t *filter = av1_get_interp_filter_subpel_kernel(
23 filter_params, subpel_q4 & SUBPEL_MASK);
24 const __m128i coeff = _mm_loadu_si128((__m128i *)filter);
25
26 // coeffs 0 1 0 1 0 1 0 1
27 coeffs[0] = _mm_shuffle_epi32(coeff, 0x00);
28 // coeffs 2 3 2 3 2 3 2 3
29 coeffs[1] = _mm_shuffle_epi32(coeff, 0x55);
30 // coeffs 4 5 4 5 4 5 4 5
31 coeffs[2] = _mm_shuffle_epi32(coeff, 0xaa);
32 // coeffs 6 7 6 7 6 7 6 7
33 coeffs[3] = _mm_shuffle_epi32(coeff, 0xff);
34 }
35
convolve(const __m128i * const s,const __m128i * const coeffs)36 static INLINE __m128i convolve(const __m128i *const s,
37 const __m128i *const coeffs) {
38 const __m128i res_0 = _mm_madd_epi16(s[0], coeffs[0]);
39 const __m128i res_1 = _mm_madd_epi16(s[1], coeffs[1]);
40 const __m128i res_2 = _mm_madd_epi16(s[2], coeffs[2]);
41 const __m128i res_3 = _mm_madd_epi16(s[3], coeffs[3]);
42
43 const __m128i res =
44 _mm_add_epi32(_mm_add_epi32(res_0, res_1), _mm_add_epi32(res_2, res_3));
45
46 return res;
47 }
48
convolve_lo_x(const __m128i * const s,const __m128i * const coeffs)49 static INLINE __m128i convolve_lo_x(const __m128i *const s,
50 const __m128i *const coeffs) {
51 __m128i ss[4];
52 ss[0] = _mm_unpacklo_epi8(s[0], _mm_setzero_si128());
53 ss[1] = _mm_unpacklo_epi8(s[1], _mm_setzero_si128());
54 ss[2] = _mm_unpacklo_epi8(s[2], _mm_setzero_si128());
55 ss[3] = _mm_unpacklo_epi8(s[3], _mm_setzero_si128());
56 return convolve(ss, coeffs);
57 }
58
convolve_lo_y(const __m128i * const s,const __m128i * const coeffs)59 static INLINE __m128i convolve_lo_y(const __m128i *const s,
60 const __m128i *const coeffs) {
61 __m128i ss[4];
62 ss[0] = _mm_unpacklo_epi8(s[0], _mm_setzero_si128());
63 ss[1] = _mm_unpacklo_epi8(s[2], _mm_setzero_si128());
64 ss[2] = _mm_unpacklo_epi8(s[4], _mm_setzero_si128());
65 ss[3] = _mm_unpacklo_epi8(s[6], _mm_setzero_si128());
66 return convolve(ss, coeffs);
67 }
68
convolve_hi_y(const __m128i * const s,const __m128i * const coeffs)69 static INLINE __m128i convolve_hi_y(const __m128i *const s,
70 const __m128i *const coeffs) {
71 __m128i ss[4];
72 ss[0] = _mm_unpackhi_epi8(s[0], _mm_setzero_si128());
73 ss[1] = _mm_unpackhi_epi8(s[2], _mm_setzero_si128());
74 ss[2] = _mm_unpackhi_epi8(s[4], _mm_setzero_si128());
75 ss[3] = _mm_unpackhi_epi8(s[6], _mm_setzero_si128());
76 return convolve(ss, coeffs);
77 }
78
comp_avg(const __m128i * const data_ref_0,const __m128i * const res_unsigned,const __m128i * const wt,const int use_dist_wtd_avg)79 static INLINE __m128i comp_avg(const __m128i *const data_ref_0,
80 const __m128i *const res_unsigned,
81 const __m128i *const wt,
82 const int use_dist_wtd_avg) {
83 __m128i res;
84 if (use_dist_wtd_avg) {
85 const __m128i data_lo = _mm_unpacklo_epi16(*data_ref_0, *res_unsigned);
86 const __m128i data_hi = _mm_unpackhi_epi16(*data_ref_0, *res_unsigned);
87
88 const __m128i wt_res_lo = _mm_madd_epi16(data_lo, *wt);
89 const __m128i wt_res_hi = _mm_madd_epi16(data_hi, *wt);
90
91 const __m128i res_lo = _mm_srai_epi32(wt_res_lo, DIST_PRECISION_BITS);
92 const __m128i res_hi = _mm_srai_epi32(wt_res_hi, DIST_PRECISION_BITS);
93
94 res = _mm_packs_epi32(res_lo, res_hi);
95 } else {
96 const __m128i wt_res = _mm_add_epi16(*data_ref_0, *res_unsigned);
97 res = _mm_srai_epi16(wt_res, 1);
98 }
99 return res;
100 }
101
convolve_rounding(const __m128i * const res_unsigned,const __m128i * const offset_const,const __m128i * const round_const,const int round_shift)102 static INLINE __m128i convolve_rounding(const __m128i *const res_unsigned,
103 const __m128i *const offset_const,
104 const __m128i *const round_const,
105 const int round_shift) {
106 const __m128i res_signed = _mm_sub_epi16(*res_unsigned, *offset_const);
107 const __m128i res_round =
108 _mm_srai_epi16(_mm_add_epi16(res_signed, *round_const), round_shift);
109 return res_round;
110 }
111
highbd_convolve_rounding_sse2(const __m128i * const res_unsigned,const __m128i * const offset_const,const __m128i * const round_const,const int round_shift)112 static INLINE __m128i highbd_convolve_rounding_sse2(
113 const __m128i *const res_unsigned, const __m128i *const offset_const,
114 const __m128i *const round_const, const int round_shift) {
115 const __m128i res_signed = _mm_sub_epi32(*res_unsigned, *offset_const);
116 const __m128i res_round =
117 _mm_srai_epi32(_mm_add_epi32(res_signed, *round_const), round_shift);
118
119 return res_round;
120 }
121
122 #endif // AOM_AOM_DSP_X86_CONVOLVE_SSE2_H_
123