| 1 | /* |
| 2 | * MMX 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 "dirac_dwt.h" |
| 26 | |
| 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);\ |
| 35 | \ |
| 36 | static void vertical_compose53iL0##ext(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width) \ |
| 37 | { \ |
| 38 | int i, width_align = width&~(align-1); \ |
| 39 | \ |
| 40 | for(i=width_align; i<width; i++) \ |
| 41 | b1[i] = COMPOSE_53iL0(b0[i], b1[i], b2[i]); \ |
| 42 | \ |
| 43 | ff_vertical_compose53iL0##ext(b0, b1, b2, width_align); \ |
| 44 | } \ |
| 45 | \ |
| 46 | static void vertical_compose_dirac53iH0##ext(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, int width) \ |
| 47 | { \ |
| 48 | int i, width_align = width&~(align-1); \ |
| 49 | \ |
| 50 | for(i=width_align; i<width; i++) \ |
| 51 | b1[i] = COMPOSE_DIRAC53iH0(b0[i], b1[i], b2[i]); \ |
| 52 | \ |
| 53 | ff_vertical_compose_dirac53iH0##ext(b0, b1, b2, width_align); \ |
| 54 | } \ |
| 55 | \ |
| 56 | static void vertical_compose_dd137iL0##ext(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, \ |
| 57 | IDWTELEM *b3, IDWTELEM *b4, int width) \ |
| 58 | { \ |
| 59 | int i, width_align = width&~(align-1); \ |
| 60 | \ |
| 61 | for(i=width_align; i<width; i++) \ |
| 62 | b2[i] = COMPOSE_DD137iL0(b0[i], b1[i], b2[i], b3[i], b4[i]); \ |
| 63 | \ |
| 64 | ff_vertical_compose_dd137iL0##ext(b0, b1, b2, b3, b4, width_align); \ |
| 65 | } \ |
| 66 | \ |
| 67 | static void vertical_compose_dd97iH0##ext(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, \ |
| 68 | IDWTELEM *b3, IDWTELEM *b4, int width) \ |
| 69 | { \ |
| 70 | int i, width_align = width&~(align-1); \ |
| 71 | \ |
| 72 | for(i=width_align; i<width; i++) \ |
| 73 | b2[i] = COMPOSE_DD97iH0(b0[i], b1[i], b2[i], b3[i], b4[i]); \ |
| 74 | \ |
| 75 | ff_vertical_compose_dd97iH0##ext(b0, b1, b2, b3, b4, width_align); \ |
| 76 | } \ |
| 77 | static void vertical_compose_haar##ext(IDWTELEM *b0, IDWTELEM *b1, int width) \ |
| 78 | { \ |
| 79 | int i, width_align = width&~(align-1); \ |
| 80 | \ |
| 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]); \ |
| 84 | } \ |
| 85 | \ |
| 86 | ff_vertical_compose_haar##ext(b0, b1, width_align); \ |
| 87 | } \ |
| 88 | static void horizontal_compose_haar0i##ext(IDWTELEM *b, IDWTELEM *tmp, int w)\ |
| 89 | {\ |
| 90 | int w2= w>>1;\ |
| 91 | int x= w2 - (w2&(align-1));\ |
| 92 | ff_horizontal_compose_haar0i##ext(b, tmp, w);\ |
| 93 | \ |
| 94 | for (; x < w2; x++) {\ |
| 95 | b[2*x ] = tmp[x];\ |
| 96 | b[2*x+1] = COMPOSE_HAARiH0(b[x+w2], tmp[x]);\ |
| 97 | }\ |
| 98 | }\ |
| 99 | static void horizontal_compose_haar1i##ext(IDWTELEM *b, IDWTELEM *tmp, int w)\ |
| 100 | {\ |
| 101 | int w2= w>>1;\ |
| 102 | int x= w2 - (w2&(align-1));\ |
| 103 | ff_horizontal_compose_haar1i##ext(b, tmp, w);\ |
| 104 | \ |
| 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;\ |
| 108 | }\ |
| 109 | }\ |
| 110 | \ |
| 111 | |
| 112 | #if HAVE_YASM |
| 113 | #if !ARCH_X86_64 |
| 114 | COMPOSE_VERTICAL(_mmx, 4) |
| 115 | #endif |
| 116 | COMPOSE_VERTICAL(_sse2, 8) |
| 117 | |
| 118 | |
| 119 | void ff_horizontal_compose_dd97i_ssse3(IDWTELEM *b, IDWTELEM *tmp, int w); |
| 120 | |
| 121 | static void horizontal_compose_dd97i_ssse3(IDWTELEM *b, IDWTELEM *tmp, int w) |
| 122 | { |
| 123 | int w2= w>>1; |
| 124 | int x= w2 - (w2&7); |
| 125 | ff_horizontal_compose_dd97i_ssse3(b, tmp, w); |
| 126 | |
| 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; |
| 130 | } |
| 131 | } |
| 132 | #endif |
| 133 | |
| 134 | void ff_spatial_idwt_init_mmx(DWTContext *d, enum dwt_type type) |
| 135 | { |
| 136 | #if HAVE_YASM |
| 137 | int mm_flags = av_get_cpu_flags(); |
| 138 | |
| 139 | #if !ARCH_X86_64 |
| 140 | if (!(mm_flags & AV_CPU_FLAG_MMX)) |
| 141 | return; |
| 142 | |
| 143 | switch (type) { |
| 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; |
| 147 | break; |
| 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; |
| 151 | break; |
| 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; |
| 155 | break; |
| 156 | case DWT_DIRAC_HAAR0: |
| 157 | d->vertical_compose = (void*)vertical_compose_haar_mmx; |
| 158 | d->horizontal_compose = horizontal_compose_haar0i_mmx; |
| 159 | break; |
| 160 | case DWT_DIRAC_HAAR1: |
| 161 | d->vertical_compose = (void*)vertical_compose_haar_mmx; |
| 162 | d->horizontal_compose = horizontal_compose_haar1i_mmx; |
| 163 | break; |
| 164 | } |
| 165 | #endif |
| 166 | |
| 167 | if (!(mm_flags & AV_CPU_FLAG_SSE2)) |
| 168 | return; |
| 169 | |
| 170 | switch (type) { |
| 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; |
| 174 | break; |
| 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; |
| 178 | break; |
| 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; |
| 182 | break; |
| 183 | case DWT_DIRAC_HAAR0: |
| 184 | d->vertical_compose = (void*)vertical_compose_haar_sse2; |
| 185 | d->horizontal_compose = horizontal_compose_haar0i_sse2; |
| 186 | break; |
| 187 | case DWT_DIRAC_HAAR1: |
| 188 | d->vertical_compose = (void*)vertical_compose_haar_sse2; |
| 189 | d->horizontal_compose = horizontal_compose_haar1i_sse2; |
| 190 | break; |
| 191 | } |
| 192 | |
| 193 | if (!(mm_flags & AV_CPU_FLAG_SSSE3)) |
| 194 | return; |
| 195 | |
| 196 | switch (type) { |
| 197 | case DWT_DIRAC_DD9_7: |
| 198 | d->horizontal_compose = horizontal_compose_dd97i_ssse3; |
| 199 | break; |
| 200 | } |
| 201 | #endif // HAVE_YASM |
| 202 | } |