Parent Directory
| 
Revision Log
Revision 1519 - (view) (download)
| 1 : | suxen_drol | 1493 | ;/***************************************************************************** |
| 2 : | ; * | ||
| 3 : | ; * XVID MPEG-4 VIDEO CODEC | ||
| 4 : | ; * - sse2 post processing - | ||
| 5 : | ; * | ||
| 6 : | ; * Copyright(C) 2004 Peter Ross <pross@xvid.org> | ||
| 7 : | ; * 2004 Dcoder <dcoder@alexandria.cc> | ||
| 8 : | ; * | ||
| 9 : | ; * XviD is free software; you can redistribute it and/or modify it | ||
| 10 : | ; * 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 : | ; * | ||
| 14 : | ; * 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 : | ; * | ||
| 19 : | ; * 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 : | ; * | ||
| 23 : | ; *************************************************************************/ | ||
| 24 : | |||
| 25 : | BITS 32 | ||
| 26 : | |||
| 27 : | %macro cglobal 1 | ||
| 28 : | %ifdef PREFIX | ||
| 29 : | global _%1 | ||
| 30 : | %define %1 _%1 | ||
| 31 : | %else | ||
| 32 : | global %1 | ||
| 33 : | %endif | ||
| 34 : | %endmacro | ||
| 35 : | |||
| 36 : | %macro FILLBYTES 2 | ||
| 37 : | |||
| 38 : | mov [%1], %2 | ||
| 39 : | mov [%1 + 1], %2 | ||
| 40 : | mov [%1 + 2], %2 | ||
| 41 : | mov [%1 + 3], %2 | ||
| 42 : | mov [%1 + 4], %2 | ||
| 43 : | mov [%1 + 5], %2 | ||
| 44 : | mov [%1 + 6], %2 | ||
| 45 : | mov [%1 + 7], %2 | ||
| 46 : | mov [%1 + 8], %2 | ||
| 47 : | mov [%1 + 9], %2 | ||
| 48 : | mov [%1 + 10], %2 | ||
| 49 : | mov [%1 + 11], %2 | ||
| 50 : | mov [%1 + 12], %2 | ||
| 51 : | mov [%1 + 13], %2 | ||
| 52 : | mov [%1 + 14], %2 | ||
| 53 : | mov [%1 + 15], %2 | ||
| 54 : | |||
| 55 : | %endmacro | ||
| 56 : | |||
| 57 : | |||
| 58 : | ;=========================================================================== | ||
| 59 : | ; read only data | ||
| 60 : | ;=========================================================================== | ||
| 61 : | |||
| 62 : | %ifdef FORMAT_COFF | ||
| 63 : | edgomez | 1519 | SECTION .rodata |
| 64 : | suxen_drol | 1493 | %else |
| 65 : | edgomez | 1519 | SECTION .rodata align=16 |
| 66 : | suxen_drol | 1493 | %endif |
| 67 : | |||
| 68 : | xmm_0x80: | ||
| 69 : | times 16 db 0x80 | ||
| 70 : | |||
| 71 : | offset_xmm: | ||
| 72 : | times 16 db 0x00 | ||
| 73 : | |||
| 74 : | ;============================================================================= | ||
| 75 : | ; Code | ||
| 76 : | ;============================================================================= | ||
| 77 : | |||
| 78 : | SECTION .text | ||
| 79 : | |||
| 80 : | cglobal image_brightness_sse2 | ||
| 81 : | |||
| 82 : | |||
| 83 : | ;////////////////////////////////////////////////////////////////////// | ||
| 84 : | ;// image_brightness_sse2 | ||
| 85 : | ;////////////////////////////////////////////////////////////////////// | ||
| 86 : | |||
| 87 : | align 16 | ||
| 88 : | image_brightness_sse2: | ||
| 89 : | |||
| 90 : | push esi | ||
| 91 : | push edi | ||
| 92 : | |||
| 93 : | movdqa xmm6, [xmm_0x80] | ||
| 94 : | |||
| 95 : | mov eax, [esp+8+20] ; offset | ||
| 96 : | |||
| 97 : | FILLBYTES offset_xmm, al | ||
| 98 : | |||
| 99 : | movdqa xmm7, [offset_xmm] | ||
| 100 : | |||
| 101 : | mov edx, [esp+8+4] ; Dst | ||
| 102 : | mov ecx, [esp+8+8] ; stride | ||
| 103 : | mov esi, [esp+8+12] ; width | ||
| 104 : | mov edi, [esp+8+16] ; height | ||
| 105 : | |||
| 106 : | .yloop | ||
| 107 : | xor eax, eax | ||
| 108 : | |||
| 109 : | .xloop | ||
| 110 : | movdqa xmm0, [edx + eax] | ||
| 111 : | movdqa xmm1, [edx + eax + 16] ; xmm0 = [dst] | ||
| 112 : | |||
| 113 : | paddb xmm0, xmm6 ; unsigned -> signed domain | ||
| 114 : | paddb xmm1, xmm6 | ||
| 115 : | paddsb xmm0, xmm7 | ||
| 116 : | paddsb xmm1, xmm7 ; xmm0 += offset | ||
| 117 : | psubb xmm0, xmm6 | ||
| 118 : | psubb xmm1, xmm6 ; signed -> unsigned domain | ||
| 119 : | |||
| 120 : | movdqa [edx + eax], xmm0 | ||
| 121 : | movdqa [edx + eax + 16], xmm1 ; [dst] = xmm0 | ||
| 122 : | |||
| 123 : | add eax,32 | ||
| 124 : | cmp eax,esi | ||
| 125 : | jl .xloop | ||
| 126 : | |||
| 127 : | add edx, ecx ; dst += stride | ||
| 128 : | sub edi, 1 | ||
| 129 : | jg .yloop | ||
| 130 : | |||
| 131 : | pop edi | ||
| 132 : | pop esi | ||
| 133 : | |||
| 134 : | ret | ||
| 135 : | ;////////////////////////////////////////////////////////////////////// |
| No admin address has been configured | ViewVC Help |
| Powered by ViewVC 1.0.4 |