Parent Directory
| 
Revision Log
Revision 1793 - (view) (download)
| 1 : | edgomez | 1586 | ;/**************************************************************************** |
| 2 : | ; * | ||
| 3 : | ; * XVID MPEG-4 VIDEO CODEC | ||
| 4 : | ; * - Interlacing Field test - | ||
| 5 : | ; * | ||
| 6 : | ; * Copyright(C) 2002 Daniel Smith <danielsmith@astroboymail.com> | ||
| 7 : | ; * 2004 Andre Werthmann <wertmann@aei.mpg.de> | ||
| 8 : | ; * | ||
| 9 : | ; * This program is free software ; you can redistribute it and/or modify | ||
| 10 : | ; * it under the terms of the GNU General Public License as published by | ||
| 11 : | ; * the Free Software Foundation ; either version 2 of the License, or | ||
| 12 : | ; * (at your option) any later version. | ||
| 13 : | ; * | ||
| 14 : | ; * This program 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 | ||
| 17 : | ; * GNU General Public License for more details. | ||
| 18 : | ; * | ||
| 19 : | ; * You should have received a copy of the GNU General Public License | ||
| 20 : | ; * along with this program ; if not, write to the Free Software | ||
| 21 : | ; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | ||
| 22 : | ; * | ||
| 23 : | Isibaar | 1793 | ; * $Id: interlacing_mmx.asm,v 1.5 2008-11-11 20:46:24 Isibaar Exp $ |
| 24 : | edgomez | 1586 | ; * |
| 25 : | ; ***************************************************************************/ | ||
| 26 : | |||
| 27 : | BITS 64 | ||
| 28 : | |||
| 29 : | %macro cglobal 1 | ||
| 30 : | %ifdef PREFIX | ||
| 31 : | %ifdef MARK_FUNCS | ||
| 32 : | global _%1:function %1.endfunc-%1 | ||
| 33 : | %define %1 _%1:function %1.endfunc-%1 | ||
| 34 : | Isibaar | 1793 | %define ENDFUNC .endfunc |
| 35 : | edgomez | 1586 | %else |
| 36 : | global _%1 | ||
| 37 : | %define %1 _%1 | ||
| 38 : | Isibaar | 1793 | %define ENDFUNC |
| 39 : | edgomez | 1586 | %endif |
| 40 : | %else | ||
| 41 : | %ifdef MARK_FUNCS | ||
| 42 : | global %1:function %1.endfunc-%1 | ||
| 43 : | Isibaar | 1793 | %define ENDFUNC .endfunc |
| 44 : | edgomez | 1586 | %else |
| 45 : | global %1 | ||
| 46 : | Isibaar | 1793 | %define ENDFUNC |
| 47 : | edgomez | 1586 | %endif |
| 48 : | %endif | ||
| 49 : | %endmacro | ||
| 50 : | |||
| 51 : | ;============================================================================= | ||
| 52 : | ; Read only data | ||
| 53 : | ;============================================================================= | ||
| 54 : | |||
| 55 : | %ifdef FORMAT_COFF | ||
| 56 : | SECTION .rodata | ||
| 57 : | %else | ||
| 58 : | SECTION .rodata align=16 | ||
| 59 : | %endif | ||
| 60 : | |||
| 61 : | ; advances to next block on right | ||
| 62 : | ALIGN 16 | ||
| 63 : | nexts: | ||
| 64 : | dd 0, 0, 8, 120, 8 | ||
| 65 : | |||
| 66 : | ; multiply word sums into dwords | ||
| 67 : | ALIGN 16 | ||
| 68 : | ones: | ||
| 69 : | times 4 dw 1 | ||
| 70 : | |||
| 71 : | ;============================================================================= | ||
| 72 : | ; Code | ||
| 73 : | ;============================================================================= | ||
| 74 : | |||
| 75 : | SECTION .text align=16 | ||
| 76 : | |||
| 77 : | cglobal MBFieldTest_x86_64 | ||
| 78 : | |||
| 79 : | ; neater | ||
| 80 : | %define line0 rsi | ||
| 81 : | %define line1 rsi+16 | ||
| 82 : | %define line2 rsi+32 | ||
| 83 : | %define line3 rsi+48 | ||
| 84 : | %define line4 rsi+64 | ||
| 85 : | %define line5 rsi+80 | ||
| 86 : | %define line6 rsi+96 | ||
| 87 : | %define line7 rsi+112 | ||
| 88 : | %define line8 rdi | ||
| 89 : | %define line9 rdi+16 | ||
| 90 : | %define line10 rdi+32 | ||
| 91 : | %define line11 rdi+48 | ||
| 92 : | %define line12 rdi+64 | ||
| 93 : | %define line13 rdi+80 | ||
| 94 : | %define line14 rdi+96 | ||
| 95 : | %define line15 rdi+112 | ||
| 96 : | |||
| 97 : | ; keep from losing track which reg holds which line - these never overlap | ||
| 98 : | %define m00 mm0 | ||
| 99 : | %define m01 mm1 | ||
| 100 : | %define m02 mm2 | ||
| 101 : | %define m03 mm0 | ||
| 102 : | %define m04 mm1 | ||
| 103 : | %define m05 mm2 | ||
| 104 : | %define m06 mm0 | ||
| 105 : | %define m07 mm1 | ||
| 106 : | %define m08 mm2 | ||
| 107 : | %define m09 mm0 | ||
| 108 : | %define m10 mm1 | ||
| 109 : | %define m11 mm2 | ||
| 110 : | %define m12 mm0 | ||
| 111 : | %define m13 mm1 | ||
| 112 : | %define m14 mm2 | ||
| 113 : | %define m15 mm0 | ||
| 114 : | |||
| 115 : | ; gets diff between three lines low(%2),mid(%3),hi(%4): frame = mid-low, field = hi-low | ||
| 116 : | %macro ABS8 4 | ||
| 117 : | movq %4, [%1] ; m02 = hi | ||
| 118 : | movq mm3, %2 ; mm3 = low copy | ||
| 119 : | |||
| 120 : | pxor mm4, mm4 ; mm4 = 0 | ||
| 121 : | pxor mm5, mm5 ; mm5 = 0 | ||
| 122 : | |||
| 123 : | psubw %2, %3 ; diff(med,low) for frame | ||
| 124 : | psubw mm3, %4 ; diff(hi,low) for field | ||
| 125 : | |||
| 126 : | pcmpgtw mm4, %2 ; if (diff<0), mm4 will be all 1's, else all 0's | ||
| 127 : | pcmpgtw mm5, mm3 | ||
| 128 : | pxor %2, mm4 ; this will get abs(), but off by 1 if (diff<0) | ||
| 129 : | pxor mm3, mm5 | ||
| 130 : | psubw %2, mm4 ; correct abs being off by 1 when (diff<0) | ||
| 131 : | psubw mm3, mm5 | ||
| 132 : | |||
| 133 : | paddw mm6, %2 ; add to totals | ||
| 134 : | paddw mm7, mm3 | ||
| 135 : | %endmacro | ||
| 136 : | |||
| 137 : | ;----------------------------------------------------------------------------- | ||
| 138 : | ; | ||
| 139 : | ; uint32_t MBFieldTest_mmx(int16_t * const data); | ||
| 140 : | ; | ||
| 141 : | ;----------------------------------------------------------------------------- | ||
| 142 : | |||
| 143 : | ALIGN 16 | ||
| 144 : | MBFieldTest_x86_64: | ||
| 145 : | mov rsi, rdi ; rsi = top left block | ||
| 146 : | ; rdi is already data[] | ||
| 147 : | add rdi, 256 ; rdi = bottom left block | ||
| 148 : | |||
| 149 : | pxor mm6, mm6 ; frame total | ||
| 150 : | pxor mm7, mm7 ; field total | ||
| 151 : | |||
| 152 : | mov rax, 4 ; we do left 8 bytes of data[0*64], then right 8 bytes | ||
| 153 : | ; then left 8 bytes of data[1*64], then last 8 bytes | ||
| 154 : | .loop: | ||
| 155 : | movq m00, [line0] ; line0 | ||
| 156 : | movq m01, [line1] ; line1 | ||
| 157 : | |||
| 158 : | ABS8 line2, m00, m01, m02 ; frame += (line2-line1), field += (line2-line0) | ||
| 159 : | ABS8 line3, m01, m02, m03 | ||
| 160 : | ABS8 line4, m02, m03, m04 | ||
| 161 : | ABS8 line5, m03, m04, m05 | ||
| 162 : | ABS8 line6, m04, m05, m06 | ||
| 163 : | ABS8 line7, m05, m06, m07 | ||
| 164 : | ABS8 line8, m06, m07, m08 | ||
| 165 : | |||
| 166 : | movq m09, [line9] ; line9-line7, no frame comp for line9-line8! | ||
| 167 : | pxor mm4, mm4 | ||
| 168 : | psubw m07, m09 | ||
| 169 : | pcmpgtw mm4, mm1 | ||
| 170 : | pxor m07, mm4 | ||
| 171 : | psubw m07, mm4 | ||
| 172 : | paddw mm7, m07 ; add to field total | ||
| 173 : | |||
| 174 : | ABS8 line10, m08, m09, m10 ; frame += (line10-line9), field += (line10-line8) | ||
| 175 : | ABS8 line11, m09, m10, m11 | ||
| 176 : | ABS8 line12, m10, m11, m12 | ||
| 177 : | ABS8 line13, m11, m12, m13 | ||
| 178 : | ABS8 line14, m12, m13, m14 | ||
| 179 : | ABS8 line15, m13, m14, m15 | ||
| 180 : | |||
| 181 : | pxor mm4, mm4 ; line15-line14, we're done with field comps! | ||
| 182 : | psubw m14, m15 | ||
| 183 : | pcmpgtw mm4, m14 | ||
| 184 : | pxor m14, mm4 | ||
| 185 : | psubw m14, mm4 | ||
| 186 : | paddw mm6, m14 ; add to frame total | ||
| 187 : | |||
| 188 : | lea r9, [nexts wrt rip] ; pic code | ||
| 189 : | edgomez | 1606 | mov ecx, [r9+rax*4] ; move rsi/rdi 8 pixels to the right |
| 190 : | edgomez | 1586 | add rsi, rcx |
| 191 : | add rdi, rcx | ||
| 192 : | |||
| 193 : | dec rax | ||
| 194 : | jnz near .loop | ||
| 195 : | |||
| 196 : | .decide: | ||
| 197 : | movq mm0, [ones wrt rip] ; add packed words into single dwords | ||
| 198 : | pmaddwd mm6, mm0 | ||
| 199 : | pmaddwd mm7, mm0 | ||
| 200 : | |||
| 201 : | movq mm0, mm6 ; ecx will be frame total, edx field | ||
| 202 : | movq mm1, mm7 | ||
| 203 : | psrlq mm0, 32 | ||
| 204 : | psrlq mm1, 32 | ||
| 205 : | paddd mm0, mm6 | ||
| 206 : | paddd mm1, mm7 | ||
| 207 : | movd rcx, mm0 | ||
| 208 : | movd rdx, mm1 | ||
| 209 : | |||
| 210 : | add rdx, 350 ; add bias against field decision | ||
| 211 : | cmp rcx, rdx | ||
| 212 : | jb .end ; if frame<field, don't use field dct | ||
| 213 : | inc rax ; if frame>=field, use field dct (return 1) | ||
| 214 : | |||
| 215 : | .end: | ||
| 216 : | ret | ||
| 217 : | Isibaar | 1793 | ENDFUNC |
| 218 : | edgomez | 1586 | |
| 219 : | Isibaar | 1790 | |
| 220 : | %ifidn __OUTPUT_FORMAT__,elf | ||
| 221 : | section ".note.GNU-stack" noalloc noexec nowrite progbits | ||
| 222 : | %endif | ||
| 223 : |
| No admin address has been configured | ViewVC Help |
| Powered by ViewVC 1.0.4 |