Parent Directory
| 
Revision Log
Revision 1.15 - (view) (download)
| 1 : | suxen_drol | 1.1 | ;/***************************************************************************** |
| 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 : | Isibaar | 1.8 | %include "nasm.inc" |
| 26 : | suxen_drol | 1.1 | |
| 27 : | ;=========================================================================== | ||
| 28 : | ; read only data | ||
| 29 : | ;=========================================================================== | ||
| 30 : | |||
| 31 : | Isibaar | 1.8 | DATA |
| 32 : | suxen_drol | 1.1 | |
| 33 : | xmm_0x80: | ||
| 34 : | times 16 db 0x80 | ||
| 35 : | |||
| 36 : | ;============================================================================= | ||
| 37 : | ; Code | ||
| 38 : | ;============================================================================= | ||
| 39 : | |||
| 40 : | Isibaar | 1.14 | TEXT |
| 41 : | suxen_drol | 1.1 | |
| 42 : | cglobal image_brightness_sse2 | ||
| 43 : | |||
| 44 : | ;////////////////////////////////////////////////////////////////////// | ||
| 45 : | ;// image_brightness_sse2 | ||
| 46 : | ;////////////////////////////////////////////////////////////////////// | ||
| 47 : | |||
| 48 : | edgomez | 1.4 | %macro CREATE_OFFSET_VECTOR 2 |
| 49 : | mov [%1 + 0], %2 | ||
| 50 : | mov [%1 + 1], %2 | ||
| 51 : | mov [%1 + 2], %2 | ||
| 52 : | mov [%1 + 3], %2 | ||
| 53 : | mov [%1 + 4], %2 | ||
| 54 : | mov [%1 + 5], %2 | ||
| 55 : | mov [%1 + 6], %2 | ||
| 56 : | mov [%1 + 7], %2 | ||
| 57 : | mov [%1 + 8], %2 | ||
| 58 : | mov [%1 + 9], %2 | ||
| 59 : | mov [%1 + 10], %2 | ||
| 60 : | mov [%1 + 11], %2 | ||
| 61 : | mov [%1 + 12], %2 | ||
| 62 : | mov [%1 + 13], %2 | ||
| 63 : | mov [%1 + 14], %2 | ||
| 64 : | mov [%1 + 15], %2 | ||
| 65 : | %endmacro | ||
| 66 : | |||
| 67 : | Isibaar | 1.8 | ALIGN SECTION_ALIGN |
| 68 : | suxen_drol | 1.1 | image_brightness_sse2: |
| 69 : | Isibaar | 1.9 | %ifdef ARCH_IS_X86_64 |
| 70 : | Isibaar | 1.15 | XVID_MOVSXD _EAX, prm5d |
| 71 : | Isibaar | 1.9 | %else |
| 72 : | mov eax, prm5 ; brightness offset value | ||
| 73 : | %endif | ||
| 74 : | Isibaar | 1.8 | mov TMP1, prm1 ; Dst |
| 75 : | mov TMP0, prm2 ; stride | ||
| 76 : | |||
| 77 : | push _ESI | ||
| 78 : | push _EDI ; 8 bytes offset for push | ||
| 79 : | sub _ESP, 32 ; 32 bytes for local data (16bytes will be used, 16bytes more to align correctly mod 16) | ||
| 80 : | edgomez | 1.4 | |
| 81 : | Isibaar | 1.13 | movdqa xmm2, [xmm_0x80] |
| 82 : | edgomez | 1.4 | |
| 83 : | ; Create a offset...offset vector | ||
| 84 : | Isibaar | 1.10 | mov _ESI, _ESP ; TMP1 will be esp aligned mod 16 |
| 85 : | add _ESI, 15 ; TMP1 = esp + 15 | ||
| 86 : | and _ESI, ~15 ; TMP1 = (esp + 15)&(~15) | ||
| 87 : | CREATE_OFFSET_VECTOR _ESI, al | ||
| 88 : | Isibaar | 1.13 | movdqa xmm3, [_ESI] |
| 89 : | Isibaar | 1.8 | |
| 90 : | %ifdef ARCH_IS_X86_64 | ||
| 91 : | mov _ESI, prm3 | ||
| 92 : | mov _EDI, prm4 | ||
| 93 : | %else | ||
| 94 : | mov _ESI, [_ESP+8+32+12] ; width | ||
| 95 : | mov _EDI, [_ESP+8+32+16] ; height | ||
| 96 : | %endif | ||
| 97 : | suxen_drol | 1.1 | |
| 98 : | Isibaar | 1.7 | .yloop: |
| 99 : | Isibaar | 1.8 | xor _EAX, _EAX |
| 100 : | suxen_drol | 1.1 | |
| 101 : | Isibaar | 1.7 | .xloop: |
| 102 : | Isibaar | 1.8 | movdqa xmm0, [TMP1 + _EAX] |
| 103 : | movdqa xmm1, [TMP1 + _EAX + 16] ; xmm0 = [dst] | ||
| 104 : | suxen_drol | 1.1 | |
| 105 : | Isibaar | 1.13 | paddb xmm0, xmm2 ; unsigned -> signed domain |
| 106 : | paddb xmm1, xmm2 | ||
| 107 : | paddsb xmm0, xmm3 | ||
| 108 : | paddsb xmm1, xmm3 ; xmm0 += offset | ||
| 109 : | psubb xmm0, xmm2 | ||
| 110 : | psubb xmm1, xmm2 ; signed -> unsigned domain | ||
| 111 : | edgomez | 1.4 | |
| 112 : | Isibaar | 1.8 | movdqa [TMP1 + _EAX], xmm0 |
| 113 : | movdqa [TMP1 + _EAX + 16], xmm1 ; [dst] = xmm0 | ||
| 114 : | edgomez | 1.4 | |
| 115 : | Isibaar | 1.8 | add _EAX,32 |
| 116 : | cmp _EAX,_ESI | ||
| 117 : | edgomez | 1.4 | jl .xloop |
| 118 : | |||
| 119 : | Isibaar | 1.8 | add TMP1, TMP0 ; dst += stride |
| 120 : | Isibaar | 1.12 | dec _EDI |
| 121 : | edgomez | 1.4 | jg .yloop |
| 122 : | |||
| 123 : | Isibaar | 1.8 | add _ESP, 32 |
| 124 : | pop _EDI | ||
| 125 : | pop _ESI | ||
| 126 : | suxen_drol | 1.1 | |
| 127 : | edgomez | 1.4 | ret |
| 128 : | Isibaar | 1.7 | ENDFUNC |
| 129 : | suxen_drol | 1.1 | ;////////////////////////////////////////////////////////////////////// |
| 130 : | Isibaar | 1.6 | |
| 131 : | %ifidn __OUTPUT_FORMAT__,elf | ||
| 132 : | section ".note.GNU-stack" noalloc noexec nowrite progbits | ||
| 133 : | %endif | ||
| 134 : |
| No admin address has been configured | ViewVC Help |
| Powered by ViewVC 1.0.4 |