1 /*
2 * Copyright (c) 2015 The WebM project authors. All Rights Reserved.
3 *
4 * Use of this source code is governed by a BSD-style license
5 * that can be found in the LICENSE file in the root of the source
6 * tree. An additional intellectual property rights grant can be found
7 * in the file PATENTS. All contributing project authors may
8 * be found in the AUTHORS file in the root of the source tree.
9 */
10
11 #include "vpx_dsp/mips/loopfilter_msa.h"
12
vpx_lpf_horizontal_4_msa(uint8_t * src,int32_t pitch,const uint8_t * b_limit_ptr,const uint8_t * limit_ptr,const uint8_t * thresh_ptr)13 void vpx_lpf_horizontal_4_msa(uint8_t *src, int32_t pitch,
14 const uint8_t *b_limit_ptr,
15 const uint8_t *limit_ptr,
16 const uint8_t *thresh_ptr) {
17 uint64_t p1_d, p0_d, q0_d, q1_d;
18 v16u8 mask, hev, flat, thresh, b_limit, limit;
19 v16u8 p3, p2, p1, p0, q3, q2, q1, q0, p1_out, p0_out, q0_out, q1_out;
20
21 /* load vector elements */
22 LD_UB8((src - 4 * pitch), pitch, p3, p2, p1, p0, q0, q1, q2, q3);
23
24 thresh = (v16u8)__msa_fill_b(*thresh_ptr);
25 b_limit = (v16u8)__msa_fill_b(*b_limit_ptr);
26 limit = (v16u8)__msa_fill_b(*limit_ptr);
27
28 LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh, hev,
29 mask, flat);
30 VP9_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev, p1_out, p0_out, q0_out, q1_out);
31
32 p1_d = __msa_copy_u_d((v2i64)p1_out, 0);
33 p0_d = __msa_copy_u_d((v2i64)p0_out, 0);
34 q0_d = __msa_copy_u_d((v2i64)q0_out, 0);
35 q1_d = __msa_copy_u_d((v2i64)q1_out, 0);
36 SD4(p1_d, p0_d, q0_d, q1_d, (src - 2 * pitch), pitch);
37 }
38
vpx_lpf_horizontal_4_dual_msa(uint8_t * src,int32_t pitch,const uint8_t * b_limit0_ptr,const uint8_t * limit0_ptr,const uint8_t * thresh0_ptr,const uint8_t * b_limit1_ptr,const uint8_t * limit1_ptr,const uint8_t * thresh1_ptr)39 void vpx_lpf_horizontal_4_dual_msa(uint8_t *src, int32_t pitch,
40 const uint8_t *b_limit0_ptr,
41 const uint8_t *limit0_ptr,
42 const uint8_t *thresh0_ptr,
43 const uint8_t *b_limit1_ptr,
44 const uint8_t *limit1_ptr,
45 const uint8_t *thresh1_ptr) {
46 v16u8 mask, hev, flat, thresh0, b_limit0, limit0, thresh1, b_limit1, limit1;
47 v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
48
49 /* load vector elements */
50 LD_UB8((src - 4 * pitch), pitch, p3, p2, p1, p0, q0, q1, q2, q3);
51
52 thresh0 = (v16u8)__msa_fill_b(*thresh0_ptr);
53 thresh1 = (v16u8)__msa_fill_b(*thresh1_ptr);
54 thresh0 = (v16u8)__msa_ilvr_d((v2i64)thresh1, (v2i64)thresh0);
55
56 b_limit0 = (v16u8)__msa_fill_b(*b_limit0_ptr);
57 b_limit1 = (v16u8)__msa_fill_b(*b_limit1_ptr);
58 b_limit0 = (v16u8)__msa_ilvr_d((v2i64)b_limit1, (v2i64)b_limit0);
59
60 limit0 = (v16u8)__msa_fill_b(*limit0_ptr);
61 limit1 = (v16u8)__msa_fill_b(*limit1_ptr);
62 limit0 = (v16u8)__msa_ilvr_d((v2i64)limit1, (v2i64)limit0);
63
64 LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit0, b_limit0, thresh0, hev,
65 mask, flat);
66 VP9_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev, p1, p0, q0, q1);
67
68 ST_UB4(p1, p0, q0, q1, (src - 2 * pitch), pitch);
69 }
70
vpx_lpf_vertical_4_msa(uint8_t * src,int32_t pitch,const uint8_t * b_limit_ptr,const uint8_t * limit_ptr,const uint8_t * thresh_ptr)71 void vpx_lpf_vertical_4_msa(uint8_t *src, int32_t pitch,
72 const uint8_t *b_limit_ptr,
73 const uint8_t *limit_ptr,
74 const uint8_t *thresh_ptr) {
75 v16u8 mask, hev, flat, limit, thresh, b_limit;
76 v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
77 v8i16 vec0, vec1, vec2, vec3;
78
79 LD_UB8((src - 4), pitch, p3, p2, p1, p0, q0, q1, q2, q3);
80
81 thresh = (v16u8)__msa_fill_b(*thresh_ptr);
82 b_limit = (v16u8)__msa_fill_b(*b_limit_ptr);
83 limit = (v16u8)__msa_fill_b(*limit_ptr);
84
85 TRANSPOSE8x8_UB_UB(p3, p2, p1, p0, q0, q1, q2, q3, p3, p2, p1, p0, q0, q1, q2,
86 q3);
87 LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit, b_limit, thresh, hev,
88 mask, flat);
89 VP9_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev, p1, p0, q0, q1);
90 ILVR_B2_SH(p0, p1, q1, q0, vec0, vec1);
91 ILVRL_H2_SH(vec1, vec0, vec2, vec3);
92
93 src -= 2;
94 ST4x4_UB(vec2, vec2, 0, 1, 2, 3, src, pitch);
95 src += 4 * pitch;
96 ST4x4_UB(vec3, vec3, 0, 1, 2, 3, src, pitch);
97 }
98
vpx_lpf_vertical_4_dual_msa(uint8_t * src,int32_t pitch,const uint8_t * b_limit0_ptr,const uint8_t * limit0_ptr,const uint8_t * thresh0_ptr,const uint8_t * b_limit1_ptr,const uint8_t * limit1_ptr,const uint8_t * thresh1_ptr)99 void vpx_lpf_vertical_4_dual_msa(uint8_t *src, int32_t pitch,
100 const uint8_t *b_limit0_ptr,
101 const uint8_t *limit0_ptr,
102 const uint8_t *thresh0_ptr,
103 const uint8_t *b_limit1_ptr,
104 const uint8_t *limit1_ptr,
105 const uint8_t *thresh1_ptr) {
106 v16u8 mask, hev, flat;
107 v16u8 thresh0, b_limit0, limit0, thresh1, b_limit1, limit1;
108 v16u8 p3, p2, p1, p0, q3, q2, q1, q0;
109 v16u8 row0, row1, row2, row3, row4, row5, row6, row7;
110 v16u8 row8, row9, row10, row11, row12, row13, row14, row15;
111 v8i16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5;
112
113 LD_UB8(src - 4, pitch, row0, row1, row2, row3, row4, row5, row6, row7);
114 LD_UB8(src - 4 + (8 * pitch), pitch, row8, row9, row10, row11, row12, row13,
115 row14, row15);
116
117 TRANSPOSE16x8_UB_UB(row0, row1, row2, row3, row4, row5, row6, row7, row8,
118 row9, row10, row11, row12, row13, row14, row15, p3, p2,
119 p1, p0, q0, q1, q2, q3);
120
121 thresh0 = (v16u8)__msa_fill_b(*thresh0_ptr);
122 thresh1 = (v16u8)__msa_fill_b(*thresh1_ptr);
123 thresh0 = (v16u8)__msa_ilvr_d((v2i64)thresh1, (v2i64)thresh0);
124
125 b_limit0 = (v16u8)__msa_fill_b(*b_limit0_ptr);
126 b_limit1 = (v16u8)__msa_fill_b(*b_limit1_ptr);
127 b_limit0 = (v16u8)__msa_ilvr_d((v2i64)b_limit1, (v2i64)b_limit0);
128
129 limit0 = (v16u8)__msa_fill_b(*limit0_ptr);
130 limit1 = (v16u8)__msa_fill_b(*limit1_ptr);
131 limit0 = (v16u8)__msa_ilvr_d((v2i64)limit1, (v2i64)limit0);
132
133 LPF_MASK_HEV(p3, p2, p1, p0, q0, q1, q2, q3, limit0, b_limit0, thresh0, hev,
134 mask, flat);
135 VP9_LPF_FILTER4_4W(p1, p0, q0, q1, mask, hev, p1, p0, q0, q1);
136 ILVR_B2_SH(p0, p1, q1, q0, tmp0, tmp1);
137 ILVRL_H2_SH(tmp1, tmp0, tmp2, tmp3);
138 ILVL_B2_SH(p0, p1, q1, q0, tmp0, tmp1);
139 ILVRL_H2_SH(tmp1, tmp0, tmp4, tmp5);
140
141 src -= 2;
142
143 ST4x8_UB(tmp2, tmp3, src, pitch);
144 src += (8 * pitch);
145 ST4x8_UB(tmp4, tmp5, src, pitch);
146 }
147