1/*
2 * MIPS SIMD optimized H.264 deblocking code
3 *
4 * Copyright (c) 2020 Loongson Technology Corporation Limited
5 *                    Gu Xiwei <guxiwei-hf@loongson.cn>
6 *
7 * This file is part of FFmpeg.
8 *
9 * FFmpeg is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public
11 * License as published by the Free Software Foundation; either
12 * version 2.1 of the License, or (at your option) any later version.
13 *
14 * FFmpeg is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
17 * Lesser General Public License for more details.
18 *
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with FFmpeg; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
22 */
23
24#include "libavcodec/bit_depth_template.c"
25#include "h264dsp_mips.h"
26#include "libavutil/mips/generic_macros_msa.h"
27#include "libavcodec/mips/h264dsp_mips.h"
28
29#define h264_loop_filter_strength_iteration_msa(edges, step, mask_mv, dir, \
30                                                d_idx, mask_dir)           \
31do {                                                                       \
32    int b_idx = 0; \
33    int step_x4 = step << 2; \
34    int d_idx_12 = d_idx + 12; \
35    int d_idx_52 = d_idx + 52; \
36    int d_idx_x4 = d_idx << 2; \
37    int d_idx_x4_48 = d_idx_x4 + 48; \
38    int dir_x32  = dir * 32; \
39    uint8_t *ref_t = (uint8_t*)ref; \
40    uint8_t *mv_t  = (uint8_t*)mv; \
41    uint8_t *nnz_t = (uint8_t*)nnz; \
42    uint8_t *bS_t  = (uint8_t*)bS; \
43    mask_mv <<= 3; \
44    for (; b_idx < edges; b_idx += step) { \
45        out &= mask_dir; \
46        if (!(mask_mv & b_idx)) { \
47            if (bidir) { \
48                ref_2 = LD_SB(ref_t + d_idx_12); \
49                ref_3 = LD_SB(ref_t + d_idx_52); \
50                ref_0 = LD_SB(ref_t + 12); \
51                ref_1 = LD_SB(ref_t + 52); \
52                ref_2 = (v16i8)__msa_ilvr_w((v4i32)ref_3, (v4i32)ref_2); \
53                ref_0 = (v16i8)__msa_ilvr_w((v4i32)ref_0, (v4i32)ref_0); \
54                ref_1 = (v16i8)__msa_ilvr_w((v4i32)ref_1, (v4i32)ref_1); \
55                ref_3 = (v16i8)__msa_shf_h((v8i16)ref_2, 0x4e); \
56                ref_0 -= ref_2; \
57                ref_1 -= ref_3; \
58                ref_0 = (v16i8)__msa_or_v((v16u8)ref_0, (v16u8)ref_1); \
59\
60                tmp_2 = LD_SH(mv_t + d_idx_x4_48);   \
61                tmp_3 = LD_SH(mv_t + 48); \
62                tmp_4 = LD_SH(mv_t + 208); \
63                tmp_5 = tmp_2 - tmp_3; \
64                tmp_6 = tmp_2 - tmp_4; \
65                SAT_SH2_SH(tmp_5, tmp_6, 7); \
66                tmp_0 = __msa_pckev_b((v16i8)tmp_6, (v16i8)tmp_5); \
67                tmp_0 += cnst_1; \
68                tmp_0 = (v16i8)__msa_subs_u_b((v16u8)tmp_0, (v16u8)cnst_0);\
69                tmp_0 = (v16i8)__msa_sat_s_h((v8i16)tmp_0, 7); \
70                tmp_0 = __msa_pckev_b(tmp_0, tmp_0); \
71                out   = (v16i8)__msa_or_v((v16u8)ref_0, (v16u8)tmp_0); \
72\
73                tmp_2 = LD_SH(mv_t + 208 + d_idx_x4); \
74                tmp_5 = tmp_2 - tmp_3; \
75                tmp_6 = tmp_2 - tmp_4; \
76                SAT_SH2_SH(tmp_5, tmp_6, 7); \
77                tmp_1 = __msa_pckev_b((v16i8)tmp_6, (v16i8)tmp_5); \
78                tmp_1 += cnst_1; \
79                tmp_1 = (v16i8)__msa_subs_u_b((v16u8)tmp_1, (v16u8)cnst_0); \
80                tmp_1 = (v16i8)__msa_sat_s_h((v8i16)tmp_1, 7); \
81                tmp_1 = __msa_pckev_b(tmp_1, tmp_1); \
82\
83                tmp_1 = (v16i8)__msa_shf_h((v8i16)tmp_1, 0x4e); \
84                out   = (v16i8)__msa_or_v((v16u8)out, (v16u8)tmp_1); \
85                tmp_0 = (v16i8)__msa_shf_h((v8i16)out, 0x4e); \
86                out   = (v16i8)__msa_min_u_b((v16u8)out, (v16u8)tmp_0); \
87            } else { \
88                ref_0 = LD_SB(ref_t + d_idx_12); \
89                ref_3 = LD_SB(ref_t + 12); \
90                tmp_2 = LD_SH(mv_t + d_idx_x4_48); \
91                tmp_3 = LD_SH(mv_t + 48); \
92                tmp_4 = tmp_3 - tmp_2; \
93                tmp_1 = (v16i8)__msa_sat_s_h(tmp_4, 7); \
94                tmp_1 = __msa_pckev_b(tmp_1, tmp_1); \
95                tmp_1 += cnst_1; \
96                out   = (v16i8)__msa_subs_u_b((v16u8)tmp_1, (v16u8)cnst_0); \
97                out   = (v16i8)__msa_sat_s_h((v8i16)out, 7); \
98                out   = __msa_pckev_b(out, out); \
99                ref_0 = ref_3 - ref_0; \
100                out   = (v16i8)__msa_or_v((v16u8)out, (v16u8)ref_0); \
101            } \
102        } \
103        tmp_0 = LD_SB(nnz_t + 12); \
104        tmp_1 = LD_SB(nnz_t + d_idx_12); \
105        tmp_0 = (v16i8)__msa_or_v((v16u8)tmp_0, (v16u8)tmp_1); \
106        tmp_0 = (v16i8)__msa_min_u_b((v16u8)tmp_0, (v16u8)cnst_2); \
107        out   = (v16i8)__msa_min_u_b((v16u8)out, (v16u8)cnst_2); \
108        tmp_0 = (v16i8)((v8i16)tmp_0 << 1); \
109        tmp_0 = (v16i8)__msa_max_u_b((v16u8)out, (v16u8)tmp_0); \
110        tmp_0 = __msa_ilvr_b(zero, tmp_0); \
111        ST_D1(tmp_0, 0, bS_t + dir_x32); \
112        ref_t += step; \
113        mv_t  += step_x4; \
114        nnz_t += step; \
115        bS_t  += step; \
116    } \
117} while(0)
118
119void ff_h264_loop_filter_strength_msa(int16_t bS[2][4][4], uint8_t nnz[40],
120                                      int8_t ref[2][40], int16_t mv[2][40][2],
121                                      int bidir, int edges, int step,
122                                      int mask_mv0, int mask_mv1, int field)
123{
124    v16i8 out;
125    v16i8 ref_0, ref_1, ref_2, ref_3;
126    v16i8 tmp_0, tmp_1;
127    v8i16 tmp_2, tmp_3, tmp_4, tmp_5, tmp_6;
128    v16i8 cnst_0, cnst_1, cnst_2;
129    v16i8 zero = { 0 };
130    v16i8 one  = __msa_fill_b(0xff);
131    if (field) {
132        cnst_0 = (v16i8)__msa_fill_h(0x206);
133        cnst_1 = (v16i8)__msa_fill_h(0x103);
134        cnst_2 = (v16i8)__msa_fill_h(0x101);
135    } else {
136        cnst_0 = __msa_fill_b(0x6);
137        cnst_1 = __msa_fill_b(0x3);
138        cnst_2 = __msa_fill_b(0x1);
139    }
140    step  <<= 3;
141    edges <<= 3;
142
143    h264_loop_filter_strength_iteration_msa(edges, step, mask_mv1, 1, -8, zero);
144    h264_loop_filter_strength_iteration_msa(32, 8, mask_mv0, 0, -1, one);
145
146    LD_SB2((int8_t*)bS, 16, tmp_0, tmp_1);
147    tmp_2 = (v8i16)__msa_ilvl_d((v2i64)tmp_0, (v2i64)tmp_0);
148    tmp_3 = (v8i16)__msa_ilvl_d((v2i64)tmp_1, (v2i64)tmp_1);
149    TRANSPOSE4x4_SH_SH(tmp_0, tmp_2, tmp_1, tmp_3, tmp_2, tmp_3, tmp_4, tmp_5);
150    tmp_0 = (v16i8)__msa_ilvr_d((v2i64)tmp_3, (v2i64)tmp_2);
151    tmp_1 = (v16i8)__msa_ilvr_d((v2i64)tmp_5, (v2i64)tmp_4);
152    ST_SB2(tmp_0, tmp_1, (int8_t*)bS, 16);
153}
154