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 #include <tmmintrin.h>
13 
14 #include "config/av1_rtcd.h"
15 
16 #include "aom/aom_integer.h"
17 #include "aom_dsp/blend.h"
18 #include "aom_dsp/x86/synonyms.h"
19 #include "av1/common/blockd.h"
20 
av1_build_compound_diffwtd_mask_highbd_ssse3(uint8_t * mask,DIFFWTD_MASK_TYPE mask_type,const uint8_t * src0,int src0_stride,const uint8_t * src1,int src1_stride,int h,int w,int bd)21 void av1_build_compound_diffwtd_mask_highbd_ssse3(
22     uint8_t *mask, DIFFWTD_MASK_TYPE mask_type, const uint8_t *src0,
23     int src0_stride, const uint8_t *src1, int src1_stride, int h, int w,
24     int bd) {
25   if (w < 8) {
26     av1_build_compound_diffwtd_mask_highbd_c(mask, mask_type, src0, src0_stride,
27                                              src1, src1_stride, h, w, bd);
28   } else {
29     assert(bd >= 8);
30     assert((w % 8) == 0);
31     assert(mask_type == DIFFWTD_38 || mask_type == DIFFWTD_38_INV);
32     const __m128i x0 = _mm_setzero_si128();
33     const __m128i xAOM_BLEND_A64_MAX_ALPHA =
34         _mm_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);
35     const int mask_base = 38;
36     const __m128i xmask_base = _mm_set1_epi16(mask_base);
37     const uint16_t *ssrc0 = CONVERT_TO_SHORTPTR(src0);
38     const uint16_t *ssrc1 = CONVERT_TO_SHORTPTR(src1);
39     if (bd == 8) {
40       if (mask_type == DIFFWTD_38_INV) {
41         for (int i = 0; i < h; ++i) {
42           for (int j = 0; j < w; j += 8) {
43             __m128i s0 = _mm_loadu_si128((const __m128i *)&ssrc0[j]);
44             __m128i s1 = _mm_loadu_si128((const __m128i *)&ssrc1[j]);
45             __m128i diff = _mm_srai_epi16(_mm_abs_epi16(_mm_sub_epi16(s0, s1)),
46                                           DIFF_FACTOR_LOG2);
47             __m128i m = _mm_min_epi16(
48                 _mm_max_epi16(x0, _mm_add_epi16(diff, xmask_base)),
49                 xAOM_BLEND_A64_MAX_ALPHA);
50             m = _mm_sub_epi16(xAOM_BLEND_A64_MAX_ALPHA, m);
51             m = _mm_packus_epi16(m, m);
52             _mm_storel_epi64((__m128i *)&mask[j], m);
53           }
54           ssrc0 += src0_stride;
55           ssrc1 += src1_stride;
56           mask += w;
57         }
58       } else {
59         for (int i = 0; i < h; ++i) {
60           for (int j = 0; j < w; j += 8) {
61             __m128i s0 = _mm_loadu_si128((const __m128i *)&ssrc0[j]);
62             __m128i s1 = _mm_loadu_si128((const __m128i *)&ssrc1[j]);
63             __m128i diff = _mm_srai_epi16(_mm_abs_epi16(_mm_sub_epi16(s0, s1)),
64                                           DIFF_FACTOR_LOG2);
65             __m128i m = _mm_min_epi16(
66                 _mm_max_epi16(x0, _mm_add_epi16(diff, xmask_base)),
67                 xAOM_BLEND_A64_MAX_ALPHA);
68             m = _mm_packus_epi16(m, m);
69             _mm_storel_epi64((__m128i *)&mask[j], m);
70           }
71           ssrc0 += src0_stride;
72           ssrc1 += src1_stride;
73           mask += w;
74         }
75       }
76     } else {
77       const __m128i xshift = xx_set1_64_from_32i(bd - 8 + DIFF_FACTOR_LOG2);
78       if (mask_type == DIFFWTD_38_INV) {
79         for (int i = 0; i < h; ++i) {
80           for (int j = 0; j < w; j += 8) {
81             __m128i s0 = _mm_loadu_si128((const __m128i *)&ssrc0[j]);
82             __m128i s1 = _mm_loadu_si128((const __m128i *)&ssrc1[j]);
83             __m128i diff =
84                 _mm_sra_epi16(_mm_abs_epi16(_mm_sub_epi16(s0, s1)), xshift);
85             __m128i m = _mm_min_epi16(
86                 _mm_max_epi16(x0, _mm_add_epi16(diff, xmask_base)),
87                 xAOM_BLEND_A64_MAX_ALPHA);
88             m = _mm_sub_epi16(xAOM_BLEND_A64_MAX_ALPHA, m);
89             m = _mm_packus_epi16(m, m);
90             _mm_storel_epi64((__m128i *)&mask[j], m);
91           }
92           ssrc0 += src0_stride;
93           ssrc1 += src1_stride;
94           mask += w;
95         }
96       } else {
97         for (int i = 0; i < h; ++i) {
98           for (int j = 0; j < w; j += 8) {
99             __m128i s0 = _mm_loadu_si128((const __m128i *)&ssrc0[j]);
100             __m128i s1 = _mm_loadu_si128((const __m128i *)&ssrc1[j]);
101             __m128i diff =
102                 _mm_sra_epi16(_mm_abs_epi16(_mm_sub_epi16(s0, s1)), xshift);
103             __m128i m = _mm_min_epi16(
104                 _mm_max_epi16(x0, _mm_add_epi16(diff, xmask_base)),
105                 xAOM_BLEND_A64_MAX_ALPHA);
106             m = _mm_packus_epi16(m, m);
107             _mm_storel_epi64((__m128i *)&mask[j], m);
108           }
109           ssrc0 += src0_stride;
110           ssrc1 += src1_stride;
111           mask += w;
112         }
113       }
114     }
115   }
116 }
117