Parent Directory
| 
Revision Log
Revision 1793 - (view) (download)
| 1 : | edgomez | 1382 | ;/**************************************************************************** |
| 2 : | Isibaar | 262 | ; * |
| 3 : | edgomez | 1382 | ; * XVID MPEG-4 VIDEO CODEC |
| 4 : | ; * - SSE2 CBP computation - | ||
| 5 : | Isibaar | 262 | ; * |
| 6 : | edgomez | 1382 | ; * Copyright (C) 2002 Daniel Smith <danielsmith@astroboymail.com> |
| 7 : | ; * 2002 Pascal Massimino <skal@planet-d.net> | ||
| 8 : | Isibaar | 262 | ; * |
| 9 : | edgomez | 1382 | ; * 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 : | Isibaar | 262 | ; * |
| 14 : | edgomez | 1382 | ; * 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 : | Isibaar | 262 | ; * |
| 19 : | edgomez | 1382 | ; * 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 : | Isibaar | 262 | ; * |
| 23 : | Isibaar | 1793 | ; * $Id: cbp_sse2.asm,v 1.9 2008-11-11 20:46:24 Isibaar Exp $ |
| 24 : | Isibaar | 262 | ; * |
| 25 : | edgomez | 1382 | ; ***************************************************************************/ |
| 26 : | Isibaar | 262 | |
| 27 : | edgomez | 1382 | BITS 32 |
| 28 : | edgomez | 851 | |
| 29 : | edgomez | 1382 | ;============================================================================= |
| 30 : | ; Macros | ||
| 31 : | ;============================================================================= | ||
| 32 : | Isibaar | 262 | |
| 33 : | %macro cglobal 1 | ||
| 34 : | %ifdef PREFIX | ||
| 35 : | edgomez | 1535 | %ifdef MARK_FUNCS |
| 36 : | edgomez | 1540 | global _%1:function %1.endfunc-%1 |
| 37 : | %define %1 _%1:function %1.endfunc-%1 | ||
| 38 : | Isibaar | 1793 | %define ENDFUNC .endfunc |
| 39 : | edgomez | 1535 | %else |
| 40 : | global _%1 | ||
| 41 : | %define %1 _%1 | ||
| 42 : | Isibaar | 1793 | %define ENDFUNC |
| 43 : | edgomez | 1535 | %endif |
| 44 : | Isibaar | 262 | %else |
| 45 : | edgomez | 1535 | %ifdef MARK_FUNCS |
| 46 : | edgomez | 1540 | global %1:function %1.endfunc-%1 |
| 47 : | Isibaar | 1793 | %define ENDFUNC .endfunc |
| 48 : | edgomez | 1535 | %else |
| 49 : | global %1 | ||
| 50 : | Isibaar | 1793 | %define ENDFUNC |
| 51 : | edgomez | 1535 | %endif |
| 52 : | Isibaar | 262 | %endif |
| 53 : | %endmacro | ||
| 54 : | |||
| 55 : | edgomez | 1382 | %macro LOOP_SSE2 1 |
| 56 : | movdqa xmm0, [edx+(%1)*128] | ||
| 57 : | pand xmm0, xmm7 | ||
| 58 : | movdqa xmm1, [edx+(%1)*128+16] | ||
| 59 : | Isibaar | 262 | |
| 60 : | edgomez | 1382 | por xmm0, [edx+(%1)*128+32] |
| 61 : | por xmm1, [edx+(%1)*128+48] | ||
| 62 : | por xmm0, [edx+(%1)*128+64] | ||
| 63 : | por xmm1, [edx+(%1)*128+80] | ||
| 64 : | por xmm0, [edx+(%1)*128+96] | ||
| 65 : | por xmm1, [edx+(%1)*128+112] | ||
| 66 : | Isibaar | 262 | |
| 67 : | edgomez | 1382 | por xmm0, xmm1 ; xmm0 = xmm1 = 128 bits worth of info |
| 68 : | psadbw xmm0, xmm6 ; contains 2 dwords with sums | ||
| 69 : | movhlps xmm1, xmm0 ; move high dword from xmm0 to low xmm1 | ||
| 70 : | por xmm0, xmm1 ; combine | ||
| 71 : | movd ecx, xmm0 ; if ecx set, values were found | ||
| 72 : | test ecx, ecx | ||
| 73 : | %endmacro | ||
| 74 : | Isibaar | 262 | |
| 75 : | edgomez | 1382 | ;============================================================================= |
| 76 : | ; Data (Read Only) | ||
| 77 : | ;============================================================================= | ||
| 78 : | Isibaar | 262 | |
| 79 : | edgomez | 1382 | %ifdef FORMAT_COFF |
| 80 : | edgomez | 1519 | SECTION .rodata |
| 81 : | edgomez | 1382 | %else |
| 82 : | edgomez | 1519 | SECTION .rodata align=16 |
| 83 : | edgomez | 1382 | %endif |
| 84 : | Isibaar | 262 | |
| 85 : | edgomez | 1382 | ALIGN 16 |
| 86 : | ignore_dc: | ||
| 87 : | dw 0, -1, -1, -1, -1, -1, -1, -1 | ||
| 88 : | Isibaar | 262 | |
| 89 : | edgomez | 1382 | ;============================================================================= |
| 90 : | ; Code | ||
| 91 : | ;============================================================================= | ||
| 92 : | Isibaar | 262 | |
| 93 : | edgomez | 1382 | SECTION .text |
| 94 : | Isibaar | 262 | |
| 95 : | edgomez | 1382 | ;----------------------------------------------------------------------------- |
| 96 : | ; uint32_t calc_cbp_sse2(const int16_t coeff[6*64]); | ||
| 97 : | ;----------------------------------------------------------------------------- | ||
| 98 : | Isibaar | 262 | |
| 99 : | edgomez | 1382 | ALIGN 16 |
| 100 : | cglobal calc_cbp_sse2 | ||
| 101 : | Isibaar | 262 | calc_cbp_sse2: |
| 102 : | edgomez | 1382 | mov edx, [esp+4] ; coeff[] |
| 103 : | xor eax, eax ; cbp = 0 | ||
| 104 : | Isibaar | 262 | |
| 105 : | edgomez | 1382 | movdqu xmm7, [ignore_dc] ; mask to ignore dc value |
| 106 : | pxor xmm6, xmm6 ; zero | ||
| 107 : | Isibaar | 262 | |
| 108 : | LOOP_SSE2 0 | ||
| 109 : | edgomez | 1382 | test ecx, ecx |
| 110 : | jz .blk2 | ||
| 111 : | or eax, (1<<5) | ||
| 112 : | |||
| 113 : | Isibaar | 1793 | .blk2: |
| 114 : | Isibaar | 262 | LOOP_SSE2 1 |
| 115 : | edgomez | 1382 | test ecx, ecx |
| 116 : | jz .blk3 | ||
| 117 : | Isibaar | 262 | or eax, (1<<4) |
| 118 : | edgomez | 1382 | |
| 119 : | Isibaar | 1793 | .blk3: |
| 120 : | Isibaar | 262 | LOOP_SSE2 2 |
| 121 : | edgomez | 1382 | test ecx, ecx |
| 122 : | jz .blk4 | ||
| 123 : | Isibaar | 262 | or eax, (1<<3) |
| 124 : | edgomez | 1382 | |
| 125 : | Isibaar | 1793 | .blk4: |
| 126 : | Isibaar | 262 | LOOP_SSE2 3 |
| 127 : | edgomez | 1382 | test ecx, ecx |
| 128 : | jz .blk5 | ||
| 129 : | Isibaar | 262 | or eax, (1<<2) |
| 130 : | edgomez | 1382 | |
| 131 : | Isibaar | 1793 | .blk5: |
| 132 : | Isibaar | 262 | LOOP_SSE2 4 |
| 133 : | edgomez | 1382 | test ecx, ecx |
| 134 : | jz .blk6 | ||
| 135 : | Isibaar | 262 | or eax, (1<<1) |
| 136 : | edgomez | 1382 | |
| 137 : | Isibaar | 1793 | .blk6: |
| 138 : | Isibaar | 262 | LOOP_SSE2 5 |
| 139 : | edgomez | 1382 | test ecx, ecx |
| 140 : | jz .finished | ||
| 141 : | Isibaar | 262 | or eax, (1<<0) |
| 142 : | edgomez | 1382 | |
| 143 : | Isibaar | 1793 | .finished: |
| 144 : | edgomez | 1382 | ret |
| 145 : | Isibaar | 1793 | ENDFUNC |
| 146 : | edgomez | 1540 | |
| 147 : | Isibaar | 1790 | |
| 148 : | %ifidn __OUTPUT_FORMAT__,elf | ||
| 149 : | section ".note.GNU-stack" noalloc noexec nowrite progbits | ||
| 150 : | %endif | ||
| 151 : |
| No admin address has been configured | ViewVC Help |
| Powered by ViewVC 1.0.4 |