Parent Directory
| 
Revision Log
Revision 1.14 - (view) (download)
| 1 : | edgomez | 1.4 | ;/**************************************************************************** |
| 2 : | Isibaar | 1.1 | ; * |
| 3 : | edgomez | 1.4 | ; * XVID MPEG-4 VIDEO CODEC |
| 4 : | ; * - SSE2 CBP computation - | ||
| 5 : | Isibaar | 1.1 | ; * |
| 6 : | edgomez | 1.4 | ; * Copyright (C) 2002 Daniel Smith <danielsmith@astroboymail.com> |
| 7 : | ; * 2002 Pascal Massimino <skal@planet-d.net> | ||
| 8 : | edgomez | 1.3 | ; * |
| 9 : | edgomez | 1.4 | ; * 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 | 1.4 | ; * 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 | 1.4 | ; * 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 : | Isibaar | 1.14 | ; * $Id: cbp_sse2.asm,v 1.13 2008/12/04 14:41:50 Isibaar Exp $ |
| 24 : | Isibaar | 1.1 | ; * |
| 25 : | edgomez | 1.4 | ; ***************************************************************************/ |
| 26 : | edgomez | 1.3 | |
| 27 : | edgomez | 1.4 | ;============================================================================= |
| 28 : | ; Macros | ||
| 29 : | ;============================================================================= | ||
| 30 : | Isibaar | 1.1 | |
| 31 : | Isibaar | 1.10 | %include "nasm.inc" |
| 32 : | Isibaar | 1.1 | |
| 33 : | Isibaar | 1.10 | %macro LOOP_SSE2 2 |
| 34 : | movdqa xmm0, [%2+(%1)*128] | ||
| 35 : | Isibaar | 1.12 | pand xmm0, xmm3 |
| 36 : | Isibaar | 1.10 | movdqa xmm1, [%2+(%1)*128+16] |
| 37 : | edgomez | 1.4 | |
| 38 : | Isibaar | 1.10 | por xmm0, [%2+(%1)*128+32] |
| 39 : | por xmm1, [%2+(%1)*128+48] | ||
| 40 : | por xmm0, [%2+(%1)*128+64] | ||
| 41 : | por xmm1, [%2+(%1)*128+80] | ||
| 42 : | por xmm0, [%2+(%1)*128+96] | ||
| 43 : | por xmm1, [%2+(%1)*128+112] | ||
| 44 : | edgomez | 1.4 | |
| 45 : | por xmm0, xmm1 ; xmm0 = xmm1 = 128 bits worth of info | ||
| 46 : | Isibaar | 1.12 | psadbw xmm0, xmm2 ; contains 2 dwords with sums |
| 47 : | edgomez | 1.4 | movhlps xmm1, xmm0 ; move high dword from xmm0 to low xmm1 |
| 48 : | por xmm0, xmm1 ; combine | ||
| 49 : | movd ecx, xmm0 ; if ecx set, values were found | ||
| 50 : | Isibaar | 1.10 | test _ECX, _ECX |
| 51 : | Isibaar | 1.1 | %endmacro |
| 52 : | |||
| 53 : | edgomez | 1.4 | ;============================================================================= |
| 54 : | ; Data (Read Only) | ||
| 55 : | ;============================================================================= | ||
| 56 : | |||
| 57 : | Isibaar | 1.10 | DATA |
| 58 : | edgomez | 1.4 | |
| 59 : | Isibaar | 1.10 | ALIGN SECTION_ALIGN |
| 60 : | edgomez | 1.4 | ignore_dc: |
| 61 : | dw 0, -1, -1, -1, -1, -1, -1, -1 | ||
| 62 : | |||
| 63 : | ;============================================================================= | ||
| 64 : | ; Code | ||
| 65 : | ;============================================================================= | ||
| 66 : | |||
| 67 : | Isibaar | 1.13 | TEXT |
| 68 : | edgomez | 1.4 | |
| 69 : | ;----------------------------------------------------------------------------- | ||
| 70 : | ; uint32_t calc_cbp_sse2(const int16_t coeff[6*64]); | ||
| 71 : | ;----------------------------------------------------------------------------- | ||
| 72 : | Isibaar | 1.1 | |
| 73 : | Isibaar | 1.10 | ALIGN SECTION_ALIGN |
| 74 : | edgomez | 1.4 | cglobal calc_cbp_sse2 |
| 75 : | Isibaar | 1.1 | calc_cbp_sse2: |
| 76 : | Isibaar | 1.10 | mov _EDX, prm1 ; coeff[] |
| 77 : | xor _EAX, _EAX ; cbp = 0 | ||
| 78 : | Isibaar | 1.1 | |
| 79 : | Isibaar | 1.12 | movdqu xmm3, [ignore_dc] ; mask to ignore dc value |
| 80 : | pxor xmm2, xmm2 ; zero | ||
| 81 : | Isibaar | 1.1 | |
| 82 : | Isibaar | 1.10 | LOOP_SSE2 0, _EDX |
| 83 : | edgomez | 1.4 | jz .blk2 |
| 84 : | Isibaar | 1.10 | or _EAX, (1<<5) |
| 85 : | edgomez | 1.4 | |
| 86 : | Isibaar | 1.9 | .blk2: |
| 87 : | Isibaar | 1.10 | LOOP_SSE2 1, _EDX |
| 88 : | edgomez | 1.4 | jz .blk3 |
| 89 : | Isibaar | 1.10 | or _EAX, (1<<4) |
| 90 : | edgomez | 1.4 | |
| 91 : | Isibaar | 1.9 | .blk3: |
| 92 : | Isibaar | 1.10 | LOOP_SSE2 2, _EDX |
| 93 : | edgomez | 1.4 | jz .blk4 |
| 94 : | Isibaar | 1.10 | or _EAX, (1<<3) |
| 95 : | edgomez | 1.4 | |
| 96 : | Isibaar | 1.9 | .blk4: |
| 97 : | Isibaar | 1.10 | LOOP_SSE2 3, _EDX |
| 98 : | edgomez | 1.4 | jz .blk5 |
| 99 : | Isibaar | 1.10 | or _EAX, (1<<2) |
| 100 : | edgomez | 1.4 | |
| 101 : | Isibaar | 1.9 | .blk5: |
| 102 : | Isibaar | 1.10 | LOOP_SSE2 4, _EDX |
| 103 : | edgomez | 1.4 | jz .blk6 |
| 104 : | Isibaar | 1.10 | or _EAX, (1<<1) |
| 105 : | edgomez | 1.4 | |
| 106 : | Isibaar | 1.9 | .blk6: |
| 107 : | Isibaar | 1.10 | LOOP_SSE2 5, _EDX |
| 108 : | edgomez | 1.4 | jz .finished |
| 109 : | Isibaar | 1.10 | or _EAX, (1<<0) |
| 110 : | edgomez | 1.4 | |
| 111 : | Isibaar | 1.9 | .finished: |
| 112 : | Isibaar | 1.11 | |
| 113 : | ret | ||
| 114 : | Isibaar | 1.9 | ENDFUNC |
| 115 : | edgomez | 1.7 | |
| 116 : | Isibaar | 1.14 | NON_EXEC_STACK |
| No admin address has been configured | ViewVC Help |
| Powered by ViewVC 1.0.4 |