1/*
2 * Copyright (C) 2009 David Conrad
3 *
4 * This file is part of Libav.
5 *
6 * Libav is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
10 *
11 * Libav is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with Libav; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 */
20
21#include "libavcodec/dsputil.h"
22#include "util_altivec.h"
23#include "types_altivec.h"
24#include "dsputil_altivec.h"
25
26static const vec_s16 constants =
27    {0, 64277, 60547, 54491, 46341, 36410, 25080, 12785};
28static const vec_u8 interleave_high =
29    {0, 1, 16, 17, 4, 5, 20, 21, 8, 9, 24, 25, 12, 13, 28, 29};
30
31#define IDCT_START \
32    vec_s16 A, B, C, D, Ad, Bd, Cd, Dd, E, F, G, H;\
33    vec_s16 Ed, Gd, Add, Bdd, Fd, Hd;\
34    vec_s16 eight = vec_splat_s16(8);\
35    vec_u16 four = vec_splat_u16(4);\
36\
37    vec_s16 C1 = vec_splat(constants, 1);\
38    vec_s16 C2 = vec_splat(constants, 2);\
39    vec_s16 C3 = vec_splat(constants, 3);\
40    vec_s16 C4 = vec_splat(constants, 4);\
41    vec_s16 C5 = vec_splat(constants, 5);\
42    vec_s16 C6 = vec_splat(constants, 6);\
43    vec_s16 C7 = vec_splat(constants, 7);\
44\
45    vec_s16 b0 = vec_ld(0x00, block);\
46    vec_s16 b1 = vec_ld(0x10, block);\
47    vec_s16 b2 = vec_ld(0x20, block);\
48    vec_s16 b3 = vec_ld(0x30, block);\
49    vec_s16 b4 = vec_ld(0x40, block);\
50    vec_s16 b5 = vec_ld(0x50, block);\
51    vec_s16 b6 = vec_ld(0x60, block);\
52    vec_s16 b7 = vec_ld(0x70, block);
53
54// these functions do (a*C)>>16
55// things are tricky because a is signed, but C unsigned.
56// M15 is used if C fits in 15 bit unsigned (C6,C7)
57// M16 is used if C requires 16 bits unsigned
58static inline vec_s16 M15(vec_s16 a, vec_s16 C)
59{
60    return (vec_s16)vec_perm(vec_mule(a,C), vec_mulo(a,C), interleave_high);
61}
62static inline vec_s16 M16(vec_s16 a, vec_s16 C)
63{
64    return vec_add(a, M15(a, C));
65}
66
67#define IDCT_1D(ADD, SHIFT)\
68    A = vec_add(M16(b1, C1), M15(b7, C7));\
69    B = vec_sub(M15(b1, C7), M16(b7, C1));\
70    C = vec_add(M16(b3, C3), M16(b5, C5));\
71    D = vec_sub(M16(b5, C3), M16(b3, C5));\
72\
73    Ad = M16(vec_sub(A, C), C4);\
74    Bd = M16(vec_sub(B, D), C4);\
75\
76    Cd = vec_add(A, C);\
77    Dd = vec_add(B, D);\
78\
79    E = ADD(M16(vec_add(b0, b4), C4));\
80    F = ADD(M16(vec_sub(b0, b4), C4));\
81\
82    G = vec_add(M16(b2, C2), M15(b6, C6));\
83    H = vec_sub(M15(b2, C6), M16(b6, C2));\
84\
85    Ed = vec_sub(E, G);\
86    Gd = vec_add(E, G);\
87\
88    Add = vec_add(F, Ad);\
89    Bdd = vec_sub(Bd, H);\
90\
91    Fd = vec_sub(F, Ad);\
92    Hd = vec_add(Bd, H);\
93\
94    b0 = SHIFT(vec_add(Gd, Cd));\
95    b7 = SHIFT(vec_sub(Gd, Cd));\
96\
97    b1 = SHIFT(vec_add(Add, Hd));\
98    b2 = SHIFT(vec_sub(Add, Hd));\
99\
100    b3 = SHIFT(vec_add(Ed, Dd));\
101    b4 = SHIFT(vec_sub(Ed, Dd));\
102\
103    b5 = SHIFT(vec_add(Fd, Bdd));\
104    b6 = SHIFT(vec_sub(Fd, Bdd));
105
106#define NOP(a) a
107#define ADD8(a) vec_add(a, eight)
108#define SHIFT4(a) vec_sra(a, four)
109
110void ff_vp3_idct_altivec(DCTELEM block[64])
111{
112    IDCT_START
113
114    IDCT_1D(NOP, NOP)
115    TRANSPOSE8(b0, b1, b2, b3, b4, b5, b6, b7);
116    IDCT_1D(ADD8, SHIFT4)
117
118    vec_st(b0, 0x00, block);
119    vec_st(b1, 0x10, block);
120    vec_st(b2, 0x20, block);
121    vec_st(b3, 0x30, block);
122    vec_st(b4, 0x40, block);
123    vec_st(b5, 0x50, block);
124    vec_st(b6, 0x60, block);
125    vec_st(b7, 0x70, block);
126}
127
128void ff_vp3_idct_put_altivec(uint8_t *dst, int stride, DCTELEM block[64])
129{
130    vec_u8 t;
131    IDCT_START
132
133    // pixels are signed; so add 128*16 in addition to the normal 8
134    vec_s16 v2048 = vec_sl(vec_splat_s16(1), vec_splat_u16(11));
135    eight = vec_add(eight, v2048);
136
137    IDCT_1D(NOP, NOP)
138    TRANSPOSE8(b0, b1, b2, b3, b4, b5, b6, b7);
139    IDCT_1D(ADD8, SHIFT4)
140
141#define PUT(a)\
142    t = vec_packsu(a, a);\
143    vec_ste((vec_u32)t, 0, (unsigned int *)dst);\
144    vec_ste((vec_u32)t, 4, (unsigned int *)dst);
145
146    PUT(b0)     dst += stride;
147    PUT(b1)     dst += stride;
148    PUT(b2)     dst += stride;
149    PUT(b3)     dst += stride;
150    PUT(b4)     dst += stride;
151    PUT(b5)     dst += stride;
152    PUT(b6)     dst += stride;
153    PUT(b7)
154}
155
156void ff_vp3_idct_add_altivec(uint8_t *dst, int stride, DCTELEM block[64])
157{
158    LOAD_ZERO;
159    vec_u8 t, vdst;
160    vec_s16 vdst_16;
161    vec_u8 vdst_mask = vec_mergeh(vec_splat_u8(-1), vec_lvsl(0, dst));
162
163    IDCT_START
164
165    IDCT_1D(NOP, NOP)
166    TRANSPOSE8(b0, b1, b2, b3, b4, b5, b6, b7);
167    IDCT_1D(ADD8, SHIFT4)
168
169#define ADD(a)\
170    vdst = vec_ld(0, dst);\
171    vdst_16 = (vec_s16)vec_perm(vdst, zero_u8v, vdst_mask);\
172    vdst_16 = vec_adds(a, vdst_16);\
173    t = vec_packsu(vdst_16, vdst_16);\
174    vec_ste((vec_u32)t, 0, (unsigned int *)dst);\
175    vec_ste((vec_u32)t, 4, (unsigned int *)dst);
176
177    ADD(b0)     dst += stride;
178    ADD(b1)     dst += stride;
179    ADD(b2)     dst += stride;
180    ADD(b3)     dst += stride;
181    ADD(b4)     dst += stride;
182    ADD(b5)     dst += stride;
183    ADD(b6)     dst += stride;
184    ADD(b7)
185}
186