1/*
2 * x86 optimized discrete wavelet transform
3 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
4 * Copyright (c) 2010 David Conrad
5 *
6 * This file is part of FFmpeg.
7 *
8 * FFmpeg is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
12 *
13 * FFmpeg is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
16 * Lesser General Public License for more details.
17 *
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with FFmpeg; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 */
22
23#include "libavutil/x86/asm.h"
24#include "libavutil/x86/cpu.h"
25#include "libavcodec/dirac_dwt.h"
26
27#define COMPOSE_VERTICAL(ext, align) \
28void ff_vertical_compose53iL0##ext(int16_t *b0, int16_t *b1, int16_t *b2, int width); \
29void ff_vertical_compose_dirac53iH0##ext(int16_t *b0, int16_t *b1, int16_t *b2, int width); \
30void ff_vertical_compose_dd137iL0##ext(int16_t *b0, int16_t *b1, int16_t *b2, int16_t *b3, int16_t *b4, int width); \
31void ff_vertical_compose_dd97iH0##ext(int16_t *b0, int16_t *b1, int16_t *b2, int16_t *b3, int16_t *b4, int width); \
32void ff_vertical_compose_haar##ext(int16_t *b0, int16_t *b1, int width); \
33void ff_horizontal_compose_haar0i##ext(int16_t *b, int16_t *tmp, int w);\
34void ff_horizontal_compose_haar1i##ext(int16_t *b, int16_t *tmp, int w);\
35\
36static void vertical_compose53iL0##ext(uint8_t *_b0, uint8_t *_b1, uint8_t *_b2, int width) \
37{ \
38    int i, width_align = width&~(align-1); \
39    int16_t *b0 = (int16_t *)_b0; \
40    int16_t *b1 = (int16_t *)_b1; \
41    int16_t *b2 = (int16_t *)_b2; \
42\
43    for(i=width_align; i<width; i++) \
44        b1[i] = COMPOSE_53iL0(b0[i], b1[i], b2[i]); \
45\
46    ff_vertical_compose53iL0##ext(b0, b1, b2, width_align); \
47} \
48\
49static void vertical_compose_dirac53iH0##ext(uint8_t *_b0, uint8_t *_b1, uint8_t *_b2, int width) \
50{ \
51    int i, width_align = width&~(align-1); \
52    int16_t *b0 = (int16_t *)_b0; \
53    int16_t *b1 = (int16_t *)_b1; \
54    int16_t *b2 = (int16_t *)_b2; \
55\
56    for(i=width_align; i<width; i++) \
57        b1[i] = COMPOSE_DIRAC53iH0(b0[i], b1[i], b2[i]); \
58\
59    ff_vertical_compose_dirac53iH0##ext(b0, b1, b2, width_align); \
60} \
61\
62static void vertical_compose_dd137iL0##ext(uint8_t *_b0, uint8_t *_b1, uint8_t *_b2, \
63                                           uint8_t *_b3, uint8_t *_b4, int width) \
64{ \
65    int i, width_align = width&~(align-1); \
66    int16_t *b0 = (int16_t *)_b0; \
67    int16_t *b1 = (int16_t *)_b1; \
68    int16_t *b2 = (int16_t *)_b2; \
69    int16_t *b3 = (int16_t *)_b3; \
70    int16_t *b4 = (int16_t *)_b4; \
71\
72    for(i=width_align; i<width; i++) \
73        b2[i] = COMPOSE_DD137iL0(b0[i], b1[i], b2[i], b3[i], b4[i]); \
74\
75    ff_vertical_compose_dd137iL0##ext(b0, b1, b2, b3, b4, width_align); \
76} \
77\
78static void vertical_compose_dd97iH0##ext(uint8_t *_b0, uint8_t *_b1, uint8_t *_b2, \
79                                          uint8_t *_b3, uint8_t *_b4, int width) \
80{ \
81    int i, width_align = width&~(align-1); \
82    int16_t *b0 = (int16_t *)_b0; \
83    int16_t *b1 = (int16_t *)_b1; \
84    int16_t *b2 = (int16_t *)_b2; \
85    int16_t *b3 = (int16_t *)_b3; \
86    int16_t *b4 = (int16_t *)_b4; \
87\
88    for(i=width_align; i<width; i++) \
89        b2[i] = COMPOSE_DD97iH0(b0[i], b1[i], b2[i], b3[i], b4[i]); \
90\
91    ff_vertical_compose_dd97iH0##ext(b0, b1, b2, b3, b4, width_align); \
92} \
93static void vertical_compose_haar##ext(uint8_t *_b0, uint8_t *_b1, int width) \
94{ \
95    int i, width_align = width&~(align-1); \
96    int16_t *b0 = (int16_t *)_b0; \
97    int16_t *b1 = (int16_t *)_b1; \
98\
99    for(i=width_align; i<width; i++) { \
100        b0[i] = COMPOSE_HAARiL0(b0[i], b1[i]); \
101        b1[i] = COMPOSE_HAARiH0(b1[i], b0[i]); \
102    } \
103\
104    ff_vertical_compose_haar##ext(b0, b1, width_align); \
105} \
106static void horizontal_compose_haar0i##ext(uint8_t *_b, uint8_t *_tmp, int w)\
107{\
108    int w2= w>>1;\
109    int x= w2 - (w2&(align-1));\
110    int16_t *b = (int16_t *)_b; \
111    int16_t *tmp = (int16_t *)_tmp; \
112\
113    ff_horizontal_compose_haar0i##ext(b, tmp, w);\
114\
115    for (; x < w2; x++) {\
116        b[2*x  ] = tmp[x];\
117        b[2*x+1] = COMPOSE_HAARiH0(b[x+w2], tmp[x]);\
118    }\
119}\
120static void horizontal_compose_haar1i##ext(uint8_t *_b, uint8_t *_tmp, int w)\
121{\
122    int w2= w>>1;\
123    int x= w2 - (w2&(align-1));\
124    int16_t *b = (int16_t *)_b; \
125    int16_t *tmp = (int16_t *)_tmp; \
126\
127    ff_horizontal_compose_haar1i##ext(b, tmp, w);\
128\
129    for (; x < w2; x++) {\
130        b[2*x  ] = (tmp[x] + 1)>>1;\
131        b[2*x+1] = (COMPOSE_HAARiH0(b[x+w2], tmp[x]) + 1)>>1;\
132    }\
133}\
134\
135
136#if HAVE_X86ASM
137COMPOSE_VERTICAL(_sse2, 8)
138
139
140void ff_horizontal_compose_dd97i_ssse3(int16_t *_b, int16_t *_tmp, int w);
141
142static void horizontal_compose_dd97i_ssse3(uint8_t *_b, uint8_t *_tmp, int w)
143{
144    int w2= w>>1;
145    int x= w2 - (w2&7);
146    int16_t *b = (int16_t *)_b;
147    int16_t *tmp = (int16_t *)_tmp;
148
149    ff_horizontal_compose_dd97i_ssse3(b, tmp, w);
150
151    for (; x < w2; x++) {
152        b[2*x  ] = (tmp[x] + 1)>>1;
153        b[2*x+1] = (COMPOSE_DD97iH0(tmp[x-1], tmp[x], b[x+w2], tmp[x+1], tmp[x+2]) + 1)>>1;
154    }
155}
156#endif
157
158void ff_spatial_idwt_init_x86(DWTContext *d, enum dwt_type type)
159{
160#if HAVE_X86ASM
161  int mm_flags = av_get_cpu_flags();
162
163    if (!(mm_flags & AV_CPU_FLAG_SSE2))
164        return;
165
166    switch (type) {
167    case DWT_DIRAC_DD9_7:
168        d->vertical_compose_l0 = (void*)vertical_compose53iL0_sse2;
169        d->vertical_compose_h0 = (void*)vertical_compose_dd97iH0_sse2;
170        break;
171    case DWT_DIRAC_LEGALL5_3:
172        d->vertical_compose_l0 = (void*)vertical_compose53iL0_sse2;
173        d->vertical_compose_h0 = (void*)vertical_compose_dirac53iH0_sse2;
174        break;
175    case DWT_DIRAC_DD13_7:
176        d->vertical_compose_l0 = (void*)vertical_compose_dd137iL0_sse2;
177        d->vertical_compose_h0 = (void*)vertical_compose_dd97iH0_sse2;
178        break;
179    case DWT_DIRAC_HAAR0:
180        d->vertical_compose   = (void*)vertical_compose_haar_sse2;
181        d->horizontal_compose = horizontal_compose_haar0i_sse2;
182        break;
183    case DWT_DIRAC_HAAR1:
184        d->vertical_compose   = (void*)vertical_compose_haar_sse2;
185        d->horizontal_compose = horizontal_compose_haar1i_sse2;
186        break;
187    }
188
189    if (!(mm_flags & AV_CPU_FLAG_SSSE3))
190        return;
191
192    switch (type) {
193    case DWT_DIRAC_DD9_7:
194        d->horizontal_compose = horizontal_compose_dd97i_ssse3;
195        break;
196    }
197#endif // HAVE_X86ASM
198}
199