Parent Directory
| 
Revision Log
Revision 1.4 - (view) (download)
| 1 : | edgomez | 1.3 | ;/**************************************************************************** |
| 2 : | edgomez | 1.2 | ; * |
| 3 : | edgomez | 1.3 | ; * XVID MPEG-4 VIDEO CODEC |
| 4 : | ; * - 3dne CBP computation - | ||
| 5 : | edgomez | 1.2 | ; * |
| 6 : | edgomez | 1.3 | ; * Copyright (C) 2002 Jaan Kalda |
| 7 : | ; * | ||
| 8 : | ; * This program is free software ; you can redistribute it and/or modify | ||
| 9 : | ; * it under the terms of the GNU General Public License as published by | ||
| 10 : | ; * the Free Software Foundation ; either version 2 of the License, or | ||
| 11 : | ; * (at your option) any later version. | ||
| 12 : | ; * | ||
| 13 : | ; * This program 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 | ||
| 16 : | ; * GNU General Public License for more details. | ||
| 17 : | ; * | ||
| 18 : | ; * You should have received a copy of the GNU General Public License | ||
| 19 : | ; * along with this program ; if not, write to the Free Software | ||
| 20 : | ; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | ||
| 21 : | ; * | ||
| 22 : | ; * $Id$ | ||
| 23 : | ; * | ||
| 24 : | ; ***************************************************************************/ | ||
| 25 : | |||
| 26 : | ; these 3dne functions are compatible with iSSE, but are optimized | ||
| 27 : | ; specifically for K7 pipelines | ||
| 28 : | edgomez | 1.2 | |
| 29 : | edgomez | 1.3 | BITS 32 |
| 30 : | |||
| 31 : | ;============================================================================= | ||
| 32 : | ; Macros | ||
| 33 : | ;============================================================================= | ||
| 34 : | edgomez | 1.2 | |
| 35 : | %macro cglobal 1 | ||
| 36 : | %ifdef PREFIX | ||
| 37 : | edgomez | 1.4 | %ifdef MARK_FUNCS |
| 38 : | global _%1:function | ||
| 39 : | %define %1 _%1:function | ||
| 40 : | %else | ||
| 41 : | global _%1 | ||
| 42 : | %define %1 _%1 | ||
| 43 : | %endif | ||
| 44 : | edgomez | 1.2 | %else |
| 45 : | edgomez | 1.4 | %ifdef MARK_FUNCS |
| 46 : | global %1:function | ||
| 47 : | %else | ||
| 48 : | global %1 | ||
| 49 : | %endif | ||
| 50 : | edgomez | 1.2 | %endif |
| 51 : | %endmacro | ||
| 52 : | |||
| 53 : | %macro calc_cbp 1 | ||
| 54 : | edgomez | 1.3 | pshufw mm0, [eax], 229 ; =11100101 |
| 55 : | movq mm1, [eax+8] | ||
| 56 : | por mm0, [eax+64] | ||
| 57 : | por mm1, [eax+72] | ||
| 58 : | movq mm2, [eax+16] | ||
| 59 : | movq mm3, [eax+24] | ||
| 60 : | por mm2, [eax+80] | ||
| 61 : | por mm3, [eax+88] | ||
| 62 : | movq mm4, [eax+32] | ||
| 63 : | movq mm5, [eax+40] | ||
| 64 : | por mm4, [eax+96] | ||
| 65 : | por mm5, [eax+104] | ||
| 66 : | movq mm6, [eax+48] | ||
| 67 : | movq mm7, [eax+56] | ||
| 68 : | por mm6, [eax+112] | ||
| 69 : | por mm7, [eax+120] | ||
| 70 : | por mm1, mm0 | ||
| 71 : | %if %1 | ||
| 72 : | sub eax, byte -128 ;ecx ;+= 128; needed 3 bytes for alignment | ||
| 73 : | edgomez | 1.2 | %else |
| 74 : | edgomez | 1.3 | xor eax, eax |
| 75 : | xor edx, edx | ||
| 76 : | %endif | ||
| 77 : | por mm3, mm2 | ||
| 78 : | por mm5, mm4 | ||
| 79 : | por mm7, mm6 | ||
| 80 : | por mm3, mm1 | ||
| 81 : | por mm7, mm5 | ||
| 82 : | por mm7, mm3 | ||
| 83 : | packsswb mm7, mm7 | ||
| 84 : | movd [esp+%1*4], mm7 | ||
| 85 : | edgomez | 1.2 | %endmacro |
| 86 : | |||
| 87 : | edgomez | 1.3 | ;============================================================================= |
| 88 : | ; Code | ||
| 89 : | ;============================================================================= | ||
| 90 : | |||
| 91 : | SECTION .text | ||
| 92 : | |||
| 93 : | cglobal calc_cbp_3dne | ||
| 94 : | edgomez | 1.2 | |
| 95 : | edgomez | 1.3 | ;----------------------------------------------------------------------------- |
| 96 : | ; uint32_t calc_cbp_3dne(const int16_t coeff[6*64]); | ||
| 97 : | ;----------------------------------------------------------------------------- | ||
| 98 : | ;AMD K7, in cache: ca 80 clk | ||
| 99 : | edgomez | 1.2 | |
| 100 : | edgomez | 1.3 | ALIGN 16 |
| 101 : | edgomez | 1.2 | calc_cbp_3dne: |
| 102 : | edgomez | 1.3 | mov eax, [esp+ 4] ; coeff |
| 103 : | lea esp, [esp-24] | ||
| 104 : | edgomez | 1.2 | calc_cbp 5 ;bit 5 |
| 105 : | edgomez | 1.3 | calc_cbp 4 ;b4 |
| 106 : | edgomez | 1.2 | calc_cbp 3 ;b3 |
| 107 : | calc_cbp 2 ;b2 | ||
| 108 : | calc_cbp 1 ;b1 | ||
| 109 : | calc_cbp 0 ;b0 | ||
| 110 : | edgomez | 1.3 | cmp eax, [esp+5*4] |
| 111 : | adc eax, eax | ||
| 112 : | cmp edx, [esp+4*4] | ||
| 113 : | adc eax, eax | ||
| 114 : | cmp edx, [esp+3*4] | ||
| 115 : | adc eax, eax | ||
| 116 : | cmp edx, [esp+2*4] | ||
| 117 : | adc eax, eax | ||
| 118 : | cmp edx, [esp+1*4] | ||
| 119 : | adc eax, eax | ||
| 120 : | cmp edx, [esp+0*4] | ||
| 121 : | adc eax, eax | ||
| 122 : | add esp, byte 24 | ||
| 123 : | ret |
| No admin address has been configured | ViewVC Help |
| Powered by ViewVC 1.0.4 |