2 * MMX optimized discrete wavelet transform
3 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
4 * Copyright (c) 2010 David Conrad
6 * This file is part of FFmpeg.
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.
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.
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
23 #include "libavutil/x86/asm.h"
24 #include "dsputil_mmx.h"
25 #include "dirac_dwt.h"
27 #define COMPOSE_VERTICAL(ext, align) \
28 void ff_vertical_compose53iL0##ext(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width); \
29 void ff_vertical_compose_dirac53iH0##ext(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width); \
30 void ff_vertical_compose_dd137iL0##ext(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, int width); \
31 void ff_vertical_compose_dd97iH0##ext(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, int width); \
32 void ff_vertical_compose_haar##ext(IDWTELEM *b0, IDWTELEM *b1, int width); \
33 void ff_horizontal_compose_haar0i##ext(IDWTELEM *b, IDWTELEM *tmp, int w);\
34 void ff_horizontal_compose_haar1i##ext(IDWTELEM *b, IDWTELEM *tmp, int w);\
36 static void vertical_compose53iL0##ext(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width) \
38 int i, width_align = width&~(align-1); \
40 for(i=width_align; i<width; i++) \
41 b1[i] = COMPOSE_53iL0(b0[i], b1[i], b2[i]); \
43 ff_vertical_compose53iL0##ext(b0, b1, b2, width_align); \
46 static void vertical_compose_dirac53iH0##ext(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width) \
48 int i, width_align = width&~(align-1); \
50 for(i=width_align; i<width; i++) \
51 b1[i] = COMPOSE_DIRAC53iH0(b0[i], b1[i], b2[i]); \
53 ff_vertical_compose_dirac53iH0##ext(b0, b1, b2, width_align); \
56 static void vertical_compose_dd137iL0##ext(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, \
57 IDWTELEM *b3, IDWTELEM *b4, int width) \
59 int i, width_align = width&~(align-1); \
61 for(i=width_align; i<width; i++) \
62 b2[i] = COMPOSE_DD137iL0(b0[i], b1[i], b2[i], b3[i], b4[i]); \
64 ff_vertical_compose_dd137iL0##ext(b0, b1, b2, b3, b4, width_align); \
67 static void vertical_compose_dd97iH0##ext(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, \
68 IDWTELEM *b3, IDWTELEM *b4, int width) \
70 int i, width_align = width&~(align-1); \
72 for(i=width_align; i<width; i++) \
73 b2[i] = COMPOSE_DD97iH0(b0[i], b1[i], b2[i], b3[i], b4[i]); \
75 ff_vertical_compose_dd97iH0##ext(b0, b1, b2, b3, b4, width_align); \
77 static void vertical_compose_haar##ext(IDWTELEM *b0, IDWTELEM *b1, int width) \
79 int i, width_align = width&~(align-1); \
81 for(i=width_align; i<width; i++) { \
82 b0[i] = COMPOSE_HAARiL0(b0[i], b1[i]); \
83 b1[i] = COMPOSE_HAARiH0(b1[i], b0[i]); \
86 ff_vertical_compose_haar##ext(b0, b1, width_align); \
88 static void horizontal_compose_haar0i##ext(IDWTELEM *b, IDWTELEM *tmp, int w)\
91 int x= w2 - (w2&(align-1));\
92 ff_horizontal_compose_haar0i##ext(b, tmp, w);\
94 for (; x < w2; x++) {\
96 b[2*x+1] = COMPOSE_HAARiH0(b[x+w2], tmp[x]);\
99 static void horizontal_compose_haar1i##ext(IDWTELEM *b, IDWTELEM *tmp, int w)\
102 int x= w2 - (w2&(align-1));\
103 ff_horizontal_compose_haar1i##ext(b, tmp, w);\
105 for (; x < w2; x++) {\
106 b[2*x ] = (tmp[x] + 1)>>1;\
107 b[2*x+1] = (COMPOSE_HAARiH0(b[x+w2], tmp[x]) + 1)>>1;\
114 COMPOSE_VERTICAL(_mmx, 4)
116 COMPOSE_VERTICAL(_sse2, 8)
119 void ff_horizontal_compose_dd97i_ssse3(IDWTELEM *b, IDWTELEM *tmp, int w);
121 static void horizontal_compose_dd97i_ssse3(IDWTELEM *b, IDWTELEM *tmp, int w)
125 ff_horizontal_compose_dd97i_ssse3(b, tmp, w);
127 for (; x < w2; x++) {
128 b[2*x ] = (tmp[x] + 1)>>1;
129 b[2*x+1] = (COMPOSE_DD97iH0(tmp[x-1], tmp[x], b[x+w2], tmp[x+1], tmp[x+2]) + 1)>>1;
134 void ff_spatial_idwt_init_mmx(DWTContext *d, enum dwt_type type)
137 int mm_flags = av_get_cpu_flags();
140 if (!(mm_flags & AV_CPU_FLAG_MMX))
144 case DWT_DIRAC_DD9_7:
145 d->vertical_compose_l0 = (void*)vertical_compose53iL0_mmx;
146 d->vertical_compose_h0 = (void*)vertical_compose_dd97iH0_mmx;
148 case DWT_DIRAC_LEGALL5_3:
149 d->vertical_compose_l0 = (void*)vertical_compose53iL0_mmx;
150 d->vertical_compose_h0 = (void*)vertical_compose_dirac53iH0_mmx;
152 case DWT_DIRAC_DD13_7:
153 d->vertical_compose_l0 = (void*)vertical_compose_dd137iL0_mmx;
154 d->vertical_compose_h0 = (void*)vertical_compose_dd97iH0_mmx;
156 case DWT_DIRAC_HAAR0:
157 d->vertical_compose = (void*)vertical_compose_haar_mmx;
158 d->horizontal_compose = horizontal_compose_haar0i_mmx;
160 case DWT_DIRAC_HAAR1:
161 d->vertical_compose = (void*)vertical_compose_haar_mmx;
162 d->horizontal_compose = horizontal_compose_haar1i_mmx;
167 if (!(mm_flags & AV_CPU_FLAG_SSE2))
171 case DWT_DIRAC_DD9_7:
172 d->vertical_compose_l0 = (void*)vertical_compose53iL0_sse2;
173 d->vertical_compose_h0 = (void*)vertical_compose_dd97iH0_sse2;
175 case DWT_DIRAC_LEGALL5_3:
176 d->vertical_compose_l0 = (void*)vertical_compose53iL0_sse2;
177 d->vertical_compose_h0 = (void*)vertical_compose_dirac53iH0_sse2;
179 case DWT_DIRAC_DD13_7:
180 d->vertical_compose_l0 = (void*)vertical_compose_dd137iL0_sse2;
181 d->vertical_compose_h0 = (void*)vertical_compose_dd97iH0_sse2;
183 case DWT_DIRAC_HAAR0:
184 d->vertical_compose = (void*)vertical_compose_haar_sse2;
185 d->horizontal_compose = horizontal_compose_haar0i_sse2;
187 case DWT_DIRAC_HAAR1:
188 d->vertical_compose = (void*)vertical_compose_haar_sse2;
189 d->horizontal_compose = horizontal_compose_haar1i_sse2;
193 if (!(mm_flags & AV_CPU_FLAG_SSSE3))
197 case DWT_DIRAC_DD9_7:
198 d->horizontal_compose = horizontal_compose_dd97i_ssse3;