Parent Directory
| 
Revision Log
Revision 1.3.2.2 - (view) (download)
| 1 : | edgomez | 1.3.2.1 | ;/**************************************************************************** |
| 2 : | Isibaar | 1.1 | ; * |
| 3 : | edgomez | 1.3.2.1 | ; * XVID MPEG-4 VIDEO CODEC |
| 4 : | ; * - SSE2 CBP computation - | ||
| 5 : | Isibaar | 1.1 | ; * |
| 6 : | edgomez | 1.3.2.1 | ; * Copyright (C) 2002 Daniel Smith <danielsmith@astroboymail.com> |
| 7 : | ; * 2002 Pascal Massimino <skal@planet-d.net> | ||
| 8 : | edgomez | 1.3 | ; * |
| 9 : | edgomez | 1.3.2.1 | ; * 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 : | edgomez | 1.3 | ; * |
| 14 : | edgomez | 1.3.2.1 | ; * 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 : | edgomez | 1.3 | ; * |
| 19 : | edgomez | 1.3.2.1 | ; * 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 | 1.1 | ; * |
| 23 : | edgomez | 1.3.2.1 | ; * $Id$ |
| 24 : | Isibaar | 1.1 | ; * |
| 25 : | edgomez | 1.3.2.1 | ; ***************************************************************************/ |
| 26 : | Isibaar | 1.1 | |
| 27 : | edgomez | 1.3.2.1 | BITS 32 |
| 28 : | Isibaar | 1.1 | |
| 29 : | edgomez | 1.3.2.1 | ;============================================================================= |
| 30 : | ; Macros | ||
| 31 : | ;============================================================================= | ||
| 32 : | Isibaar | 1.1 | |
| 33 : | %macro cglobal 1 | ||
| 34 : | %ifdef PREFIX | ||
| 35 : | global _%1 | ||
| 36 : | %define %1 _%1 | ||
| 37 : | %else | ||
| 38 : | global %1 | ||
| 39 : | %endif | ||
| 40 : | %endmacro | ||
| 41 : | |||
| 42 : | edgomez | 1.3.2.1 | %macro LOOP_SSE2 1 |
| 43 : | movdqa xmm0, [edx+(%1)*128] | ||
| 44 : | pand xmm0, xmm7 | ||
| 45 : | movdqa xmm1, [edx+(%1)*128+16] | ||
| 46 : | |||
| 47 : | por xmm0, [edx+(%1)*128+32] | ||
| 48 : | por xmm1, [edx+(%1)*128+48] | ||
| 49 : | por xmm0, [edx+(%1)*128+64] | ||
| 50 : | por xmm1, [edx+(%1)*128+80] | ||
| 51 : | por xmm0, [edx+(%1)*128+96] | ||
| 52 : | por xmm1, [edx+(%1)*128+112] | ||
| 53 : | |||
| 54 : | por xmm0, xmm1 ; xmm0 = xmm1 = 128 bits worth of info | ||
| 55 : | psadbw xmm0, xmm6 ; contains 2 dwords with sums | ||
| 56 : | movhlps xmm1, xmm0 ; move high dword from xmm0 to low xmm1 | ||
| 57 : | por xmm0, xmm1 ; combine | ||
| 58 : | movd ecx, xmm0 ; if ecx set, values were found | ||
| 59 : | test ecx, ecx | ||
| 60 : | %endmacro | ||
| 61 : | Isibaar | 1.1 | |
| 62 : | edgomez | 1.3.2.1 | ;============================================================================= |
| 63 : | ; Data (Read Only) | ||
| 64 : | ;============================================================================= | ||
| 65 : | Isibaar | 1.1 | |
| 66 : | edgomez | 1.3.2.2 | %ifdef FORMAT_COFF |
| 67 : | SECTION .rodata data | ||
| 68 : | %else | ||
| 69 : | SECTION .rodata data align=16 | ||
| 70 : | %endif | ||
| 71 : | Isibaar | 1.1 | |
| 72 : | edgomez | 1.3.2.1 | ALIGN 16 |
| 73 : | ignore_dc: | ||
| 74 : | dw 0, -1, -1, -1, -1, -1, -1, -1 | ||
| 75 : | Isibaar | 1.1 | |
| 76 : | edgomez | 1.3.2.1 | ;============================================================================= |
| 77 : | ; Code | ||
| 78 : | ;============================================================================= | ||
| 79 : | Isibaar | 1.1 | |
| 80 : | edgomez | 1.3.2.1 | SECTION .text |
| 81 : | Isibaar | 1.1 | |
| 82 : | edgomez | 1.3.2.1 | ;----------------------------------------------------------------------------- |
| 83 : | ; uint32_t calc_cbp_sse2(const int16_t coeff[6*64]); | ||
| 84 : | ;----------------------------------------------------------------------------- | ||
| 85 : | Isibaar | 1.1 | |
| 86 : | edgomez | 1.3.2.1 | ALIGN 16 |
| 87 : | cglobal calc_cbp_sse2 | ||
| 88 : | Isibaar | 1.1 | calc_cbp_sse2: |
| 89 : | edgomez | 1.3.2.1 | mov edx, [esp+4] ; coeff[] |
| 90 : | xor eax, eax ; cbp = 0 | ||
| 91 : | Isibaar | 1.1 | |
| 92 : | edgomez | 1.3.2.1 | movdqu xmm7, [ignore_dc] ; mask to ignore dc value |
| 93 : | pxor xmm6, xmm6 ; zero | ||
| 94 : | Isibaar | 1.1 | |
| 95 : | LOOP_SSE2 0 | ||
| 96 : | edgomez | 1.3.2.1 | test ecx, ecx |
| 97 : | jz .blk2 | ||
| 98 : | or eax, (1<<5) | ||
| 99 : | |||
| 100 : | Isibaar | 1.1 | .blk2 |
| 101 : | LOOP_SSE2 1 | ||
| 102 : | edgomez | 1.3.2.1 | test ecx, ecx |
| 103 : | jz .blk3 | ||
| 104 : | Isibaar | 1.1 | or eax, (1<<4) |
| 105 : | edgomez | 1.3.2.1 | |
| 106 : | Isibaar | 1.1 | .blk3 |
| 107 : | LOOP_SSE2 2 | ||
| 108 : | edgomez | 1.3.2.1 | test ecx, ecx |
| 109 : | jz .blk4 | ||
| 110 : | Isibaar | 1.1 | or eax, (1<<3) |
| 111 : | edgomez | 1.3.2.1 | |
| 112 : | Isibaar | 1.1 | .blk4 |
| 113 : | LOOP_SSE2 3 | ||
| 114 : | edgomez | 1.3.2.1 | test ecx, ecx |
| 115 : | jz .blk5 | ||
| 116 : | Isibaar | 1.1 | or eax, (1<<2) |
| 117 : | edgomez | 1.3.2.1 | |
| 118 : | Isibaar | 1.1 | .blk5 |
| 119 : | LOOP_SSE2 4 | ||
| 120 : | edgomez | 1.3.2.1 | test ecx, ecx |
| 121 : | jz .blk6 | ||
| 122 : | Isibaar | 1.1 | or eax, (1<<1) |
| 123 : | edgomez | 1.3.2.1 | |
| 124 : | Isibaar | 1.1 | .blk6 |
| 125 : | LOOP_SSE2 5 | ||
| 126 : | edgomez | 1.3.2.1 | test ecx, ecx |
| 127 : | jz .finished | ||
| 128 : | Isibaar | 1.1 | or eax, (1<<0) |
| 129 : | edgomez | 1.3.2.1 | |
| 130 : | Isibaar | 1.1 | .finished |
| 131 : | edgomez | 1.3.2.1 | ret |
| No admin address has been configured | ViewVC Help |
| Powered by ViewVC 1.0.4 |