Parent Directory
| 
Revision Log
Revision 1793 - (view) (download)
| 1 : | edgomez | 1586 | ;/**************************************************************************** |
| 2 : | ; * | ||
| 3 : | ; * XVID MPEG-4 VIDEO CODEC | ||
| 4 : | ; * - K7 optimized SAD operators - | ||
| 5 : | ; * | ||
| 6 : | ; * Copyright(C) 2001 Peter Ross <pross@xvid.org> | ||
| 7 : | ; * 2002 Pascal Massimino <skal@planet-d.net> | ||
| 8 : | ; * 2004 Andre Werthmann <wertmann@aei.mpg.de> | ||
| 9 : | ; * | ||
| 10 : | ; * This program is free software; you can redistribute it and/or modify it | ||
| 11 : | ; * under the terms of the GNU General Public License as published by | ||
| 12 : | ; * the Free Software Foundation; either version 2 of the License, or | ||
| 13 : | ; * (at your option) any later version. | ||
| 14 : | ; * | ||
| 15 : | ; * This program is distributed in the hope that it will be useful, | ||
| 16 : | ; * but WITHOUT ANY WARRANTY; without even the implied warranty of | ||
| 17 : | ; * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | ||
| 18 : | ; * GNU General Public License for more details. | ||
| 19 : | ; * | ||
| 20 : | ; * You should have received a copy of the GNU General Public License | ||
| 21 : | ; * along with this program; if not, write to the Free Software | ||
| 22 : | ; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | ||
| 23 : | ; * | ||
| 24 : | Isibaar | 1793 | ; * $Id: sad_mmx.asm,v 1.3 2008-11-11 20:46:24 Isibaar Exp $ |
| 25 : | edgomez | 1586 | ; * |
| 26 : | ; ***************************************************************************/ | ||
| 27 : | |||
| 28 : | BITS 64 | ||
| 29 : | |||
| 30 : | %macro cglobal 1 | ||
| 31 : | %ifdef PREFIX | ||
| 32 : | %ifdef MARK_FUNCS | ||
| 33 : | global _%1:function %1.endfunc-%1 | ||
| 34 : | %define %1 _%1:function %1.endfunc-%1 | ||
| 35 : | Isibaar | 1793 | %define ENDFUNC .endfunc |
| 36 : | edgomez | 1586 | %else |
| 37 : | global _%1 | ||
| 38 : | %define %1 _%1 | ||
| 39 : | Isibaar | 1793 | %define ENDFUNC |
| 40 : | edgomez | 1586 | %endif |
| 41 : | %else | ||
| 42 : | %ifdef MARK_FUNCS | ||
| 43 : | global %1:function %1.endfunc-%1 | ||
| 44 : | Isibaar | 1793 | %define ENDFUNC .endfunc |
| 45 : | edgomez | 1586 | %else |
| 46 : | global %1 | ||
| 47 : | Isibaar | 1793 | %define ENDFUNC |
| 48 : | edgomez | 1586 | %endif |
| 49 : | %endif | ||
| 50 : | %endmacro | ||
| 51 : | |||
| 52 : | ;============================================================================= | ||
| 53 : | ; Read only data | ||
| 54 : | ;============================================================================= | ||
| 55 : | |||
| 56 : | %ifdef FORMAT_COFF | ||
| 57 : | SECTION .rodata | ||
| 58 : | %else | ||
| 59 : | SECTION .rodata align=16 | ||
| 60 : | %endif | ||
| 61 : | |||
| 62 : | ;============================================================================= | ||
| 63 : | ; Code | ||
| 64 : | ;============================================================================= | ||
| 65 : | |||
| 66 : | SECTION .text align=16 | ||
| 67 : | |||
| 68 : | cglobal sse8_16bit_x86_64 | ||
| 69 : | cglobal sse8_8bit_x86_64 | ||
| 70 : | |||
| 71 : | ;----------------------------------------------------------------------------- | ||
| 72 : | ; | ||
| 73 : | ; uint32_t sse8_16bit_x86_64x(const int16_t *b1, | ||
| 74 : | ; const int16_t *b2, | ||
| 75 : | ; const uint32_t stride); | ||
| 76 : | ; | ||
| 77 : | ;----------------------------------------------------------------------------- | ||
| 78 : | |||
| 79 : | %macro ROW_SSE_16Bit_MMX 2 | ||
| 80 : | movq mm0, [%1] | ||
| 81 : | movq mm1, [%1+8] | ||
| 82 : | psubw mm0, [%2] | ||
| 83 : | psubw mm1, [%2+8] | ||
| 84 : | pmaddwd mm0, mm0 | ||
| 85 : | pmaddwd mm1, mm1 | ||
| 86 : | paddd mm2, mm0 | ||
| 87 : | paddd mm2, mm1 | ||
| 88 : | %endmacro | ||
| 89 : | |||
| 90 : | sse8_16bit_x86_64: | ||
| 91 : | |||
| 92 : | ; rdx is stride | ||
| 93 : | ; rsi is b2 | ||
| 94 : | ; rdi is b1 | ||
| 95 : | |||
| 96 : | ;; Reset the sse accumulator | ||
| 97 : | pxor mm2, mm2 | ||
| 98 : | |||
| 99 : | ;; Let's go | ||
| 100 : | %rep 8 | ||
| 101 : | ROW_SSE_16Bit_MMX rsi, rdi | ||
| 102 : | lea rsi, [rsi+rdx] | ||
| 103 : | lea rdi, [rdi+rdx] | ||
| 104 : | %endrep | ||
| 105 : | |||
| 106 : | ;; Finish adding each dword of the accumulator | ||
| 107 : | movq mm3, mm2 | ||
| 108 : | psrlq mm2, 32 | ||
| 109 : | paddd mm2, mm3 | ||
| 110 : | movd eax, mm2 | ||
| 111 : | |||
| 112 : | ;; All done | ||
| 113 : | ret | ||
| 114 : | Isibaar | 1793 | ENDFUNC |
| 115 : | edgomez | 1586 | |
| 116 : | ;----------------------------------------------------------------------------- | ||
| 117 : | ; | ||
| 118 : | ; uint32_t sse8_8bit_x86_64(const int8_t *b1, | ||
| 119 : | ; const int8_t *b2, | ||
| 120 : | ; const uint32_t stride); | ||
| 121 : | ; | ||
| 122 : | ;----------------------------------------------------------------------------- | ||
| 123 : | |||
| 124 : | %macro ROW_SSE_8bit_MMX 2 | ||
| 125 : | movq mm0, [%1] ; load a row | ||
| 126 : | movq mm2, [%2] ; load a row | ||
| 127 : | |||
| 128 : | movq mm1, mm0 ; copy row | ||
| 129 : | movq mm3, mm2 ; copy row | ||
| 130 : | |||
| 131 : | punpcklbw mm0, mm7 ; turn the 4low elements into 16bit | ||
| 132 : | punpckhbw mm1, mm7 ; turn the 4high elements into 16bit | ||
| 133 : | |||
| 134 : | punpcklbw mm2, mm7 ; turn the 4low elements into 16bit | ||
| 135 : | punpckhbw mm3, mm7 ; turn the 4high elements into 16bit | ||
| 136 : | |||
| 137 : | psubw mm0, mm2 ; low part of src-dst | ||
| 138 : | psubw mm1, mm3 ; high part of src-dst | ||
| 139 : | |||
| 140 : | pmaddwd mm0, mm0 ; compute the square sum | ||
| 141 : | pmaddwd mm1, mm1 ; compute the square sum | ||
| 142 : | |||
| 143 : | paddd mm6, mm0 ; add to the accumulator | ||
| 144 : | paddd mm6, mm1 ; add to the accumulator | ||
| 145 : | %endmacro | ||
| 146 : | |||
| 147 : | sse8_8bit_x86_64: | ||
| 148 : | ;; Reset the sse accumulator | ||
| 149 : | pxor mm6, mm6 | ||
| 150 : | |||
| 151 : | ;; Used to interleave 8bit data with 0x00 values | ||
| 152 : | pxor mm7, mm7 | ||
| 153 : | |||
| 154 : | ;; Let's go | ||
| 155 : | %rep 8 | ||
| 156 : | ROW_SSE_8bit_MMX rsi, rdi | ||
| 157 : | lea rsi, [rsi+rdx] | ||
| 158 : | lea rdi, [rdi+rdx] | ||
| 159 : | %endrep | ||
| 160 : | |||
| 161 : | ;; Finish adding each dword of the accumulator | ||
| 162 : | movq mm7, mm6 | ||
| 163 : | psrlq mm6, 32 | ||
| 164 : | paddd mm6, mm7 | ||
| 165 : | movd eax, mm6 | ||
| 166 : | |||
| 167 : | ;; All done | ||
| 168 : | ret | ||
| 169 : | Isibaar | 1793 | ENDFUNC |
| 170 : | Isibaar | 1790 | |
| 171 : | %ifidn __OUTPUT_FORMAT__,elf | ||
| 172 : | section ".note.GNU-stack" noalloc noexec nowrite progbits | ||
| 173 : | %endif | ||
| 174 : |
| No admin address has been configured | ViewVC Help |
| Powered by ViewVC 1.0.4 |