| 1 | ;***************************************************************************** |
| 2 | ;* x86-optimized functions for yadif filter |
| 3 | ;* |
| 4 | ;* Copyright (C) 2006 Michael Niedermayer <michaelni@gmx.at> |
| 5 | ;* Copyright (c) 2013 Daniel Kang <daniel.d.kang@gmail.com> |
| 6 | ;* |
| 7 | ;* This file is part of FFmpeg. |
| 8 | ;* |
| 9 | ;* FFmpeg is free software; you can redistribute it and/or |
| 10 | ;* modify it under the terms of the GNU Lesser General Public |
| 11 | ;* License as published by the Free Software Foundation; either |
| 12 | ;* version 2.1 of the License, or (at your option) any later version. |
| 13 | ;* |
| 14 | ;* FFmpeg is distributed in the hope that it will be useful, |
| 15 | ;* but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 16 | ;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 17 | ;* Lesser General Public License for more details. |
| 18 | ;* |
| 19 | ;* You should have received a copy of the GNU Lesser General Public |
| 20 | ;* License along with FFmpeg; if not, write to the Free Software |
| 21 | ;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
| 22 | ;****************************************************************************** |
| 23 | |
| 24 | %include "libavutil/x86/x86util.asm" |
| 25 | |
| 26 | SECTION_RODATA |
| 27 | |
| 28 | pb_1: times 16 db 1 |
| 29 | pw_1: times 8 dw 1 |
| 30 | |
| 31 | SECTION .text |
| 32 | |
| 33 | %macro CHECK 2 |
| 34 | movu m2, [curq+t1+%1] |
| 35 | movu m3, [curq+t0+%2] |
| 36 | mova m4, m2 |
| 37 | mova m5, m2 |
| 38 | pxor m4, m3 |
| 39 | pavgb m5, m3 |
| 40 | pand m4, [pb_1] |
| 41 | psubusb m5, m4 |
| 42 | RSHIFT m5, 1 |
| 43 | punpcklbw m5, m7 |
| 44 | mova m4, m2 |
| 45 | psubusb m2, m3 |
| 46 | psubusb m3, m4 |
| 47 | pmaxub m2, m3 |
| 48 | mova m3, m2 |
| 49 | mova m4, m2 |
| 50 | RSHIFT m3, 1 |
| 51 | RSHIFT m4, 2 |
| 52 | punpcklbw m2, m7 |
| 53 | punpcklbw m3, m7 |
| 54 | punpcklbw m4, m7 |
| 55 | paddw m2, m3 |
| 56 | paddw m2, m4 |
| 57 | %endmacro |
| 58 | |
| 59 | %macro CHECK1 0 |
| 60 | mova m3, m0 |
| 61 | pcmpgtw m3, m2 |
| 62 | pminsw m0, m2 |
| 63 | mova m6, m3 |
| 64 | pand m5, m3 |
| 65 | pandn m3, m1 |
| 66 | por m3, m5 |
| 67 | mova m1, m3 |
| 68 | %endmacro |
| 69 | |
| 70 | %macro CHECK2 0 |
| 71 | paddw m6, [pw_1] |
| 72 | psllw m6, 14 |
| 73 | paddsw m2, m6 |
| 74 | mova m3, m0 |
| 75 | pcmpgtw m3, m2 |
| 76 | pminsw m0, m2 |
| 77 | pand m5, m3 |
| 78 | pandn m3, m1 |
| 79 | por m3, m5 |
| 80 | mova m1, m3 |
| 81 | %endmacro |
| 82 | |
| 83 | %macro LOAD 2 |
| 84 | movh %1, %2 |
| 85 | punpcklbw %1, m7 |
| 86 | %endmacro |
| 87 | |
| 88 | %macro FILTER 3 |
| 89 | .loop%1: |
| 90 | pxor m7, m7 |
| 91 | LOAD m0, [curq+t1] |
| 92 | LOAD m1, [curq+t0] |
| 93 | LOAD m2, [%2] |
| 94 | LOAD m3, [%3] |
| 95 | mova m4, m3 |
| 96 | paddw m3, m2 |
| 97 | psraw m3, 1 |
| 98 | mova [rsp+ 0], m0 |
| 99 | mova [rsp+16], m3 |
| 100 | mova [rsp+32], m1 |
| 101 | psubw m2, m4 |
| 102 | ABS1 m2, m4 |
| 103 | LOAD m3, [prevq+t1] |
| 104 | LOAD m4, [prevq+t0] |
| 105 | psubw m3, m0 |
| 106 | psubw m4, m1 |
| 107 | ABS1 m3, m5 |
| 108 | ABS1 m4, m5 |
| 109 | paddw m3, m4 |
| 110 | psrlw m2, 1 |
| 111 | psrlw m3, 1 |
| 112 | pmaxsw m2, m3 |
| 113 | LOAD m3, [nextq+t1] |
| 114 | LOAD m4, [nextq+t0] |
| 115 | psubw m3, m0 |
| 116 | psubw m4, m1 |
| 117 | ABS1 m3, m5 |
| 118 | ABS1 m4, m5 |
| 119 | paddw m3, m4 |
| 120 | psrlw m3, 1 |
| 121 | pmaxsw m2, m3 |
| 122 | mova [rsp+48], m2 |
| 123 | |
| 124 | paddw m1, m0 |
| 125 | paddw m0, m0 |
| 126 | psubw m0, m1 |
| 127 | psrlw m1, 1 |
| 128 | ABS1 m0, m2 |
| 129 | |
| 130 | movu m2, [curq+t1-1] |
| 131 | movu m3, [curq+t0-1] |
| 132 | mova m4, m2 |
| 133 | psubusb m2, m3 |
| 134 | psubusb m3, m4 |
| 135 | pmaxub m2, m3 |
| 136 | %if mmsize == 16 |
| 137 | mova m3, m2 |
| 138 | psrldq m3, 2 |
| 139 | %else |
| 140 | pshufw m3, m2, q0021 |
| 141 | %endif |
| 142 | punpcklbw m2, m7 |
| 143 | punpcklbw m3, m7 |
| 144 | paddw m0, m2 |
| 145 | paddw m0, m3 |
| 146 | psubw m0, [pw_1] |
| 147 | |
| 148 | CHECK -2, 0 |
| 149 | CHECK1 |
| 150 | CHECK -3, 1 |
| 151 | CHECK2 |
| 152 | CHECK 0, -2 |
| 153 | CHECK1 |
| 154 | CHECK 1, -3 |
| 155 | CHECK2 |
| 156 | |
| 157 | mova m6, [rsp+48] |
| 158 | cmp DWORD r8m, 2 |
| 159 | jge .end%1 |
| 160 | LOAD m2, [%2+t1*2] |
| 161 | LOAD m4, [%3+t1*2] |
| 162 | LOAD m3, [%2+t0*2] |
| 163 | LOAD m5, [%3+t0*2] |
| 164 | paddw m2, m4 |
| 165 | paddw m3, m5 |
| 166 | psrlw m2, 1 |
| 167 | psrlw m3, 1 |
| 168 | mova m4, [rsp+ 0] |
| 169 | mova m5, [rsp+16] |
| 170 | mova m7, [rsp+32] |
| 171 | psubw m2, m4 |
| 172 | psubw m3, m7 |
| 173 | mova m0, m5 |
| 174 | psubw m5, m4 |
| 175 | psubw m0, m7 |
| 176 | mova m4, m2 |
| 177 | pminsw m2, m3 |
| 178 | pmaxsw m3, m4 |
| 179 | pmaxsw m2, m5 |
| 180 | pminsw m3, m5 |
| 181 | pmaxsw m2, m0 |
| 182 | pminsw m3, m0 |
| 183 | pxor m4, m4 |
| 184 | pmaxsw m6, m3 |
| 185 | psubw m4, m2 |
| 186 | pmaxsw m6, m4 |
| 187 | |
| 188 | .end%1: |
| 189 | mova m2, [rsp+16] |
| 190 | mova m3, m2 |
| 191 | psubw m2, m6 |
| 192 | paddw m3, m6 |
| 193 | pmaxsw m1, m2 |
| 194 | pminsw m1, m3 |
| 195 | packuswb m1, m1 |
| 196 | |
| 197 | movh [dstq], m1 |
| 198 | add dstq, mmsize/2 |
| 199 | add prevq, mmsize/2 |
| 200 | add curq, mmsize/2 |
| 201 | add nextq, mmsize/2 |
| 202 | sub DWORD r4m, mmsize/2 |
| 203 | jg .loop%1 |
| 204 | %endmacro |
| 205 | |
| 206 | %macro YADIF 0 |
| 207 | %if ARCH_X86_32 |
| 208 | cglobal yadif_filter_line, 4, 6, 8, 80, dst, prev, cur, next, w, prefs, \ |
| 209 | mrefs, parity, mode |
| 210 | %else |
| 211 | cglobal yadif_filter_line, 4, 7, 8, 80, dst, prev, cur, next, w, prefs, \ |
| 212 | mrefs, parity, mode |
| 213 | %endif |
| 214 | %if ARCH_X86_32 |
| 215 | mov r4, r5mp |
| 216 | mov r5, r6mp |
| 217 | DECLARE_REG_TMP 4,5 |
| 218 | %else |
| 219 | movsxd r5, DWORD r5m |
| 220 | movsxd r6, DWORD r6m |
| 221 | DECLARE_REG_TMP 5,6 |
| 222 | %endif |
| 223 | |
| 224 | cmp DWORD paritym, 0 |
| 225 | je .parity0 |
| 226 | FILTER 1, prevq, curq |
| 227 | jmp .ret |
| 228 | |
| 229 | .parity0: |
| 230 | FILTER 0, curq, nextq |
| 231 | |
| 232 | .ret: |
| 233 | RET |
| 234 | %endmacro |
| 235 | |
| 236 | INIT_XMM ssse3 |
| 237 | YADIF |
| 238 | INIT_XMM sse2 |
| 239 | YADIF |
| 240 | %if ARCH_X86_32 |
| 241 | INIT_MMX mmxext |
| 242 | YADIF |
| 243 | %endif |