| 1 |
;/************************************************************************** |
;/**************************************************************************** |
| 2 |
; * |
; * |
| 3 |
; * XVID MPEG-4 VIDEO CODEC |
; * XVID MPEG-4 VIDEO CODEC |
| 4 |
; * mmx 8bit<->16bit transfers |
; * - 8<->16 bit transfer functions - |
| 5 |
; * |
; * |
| 6 |
; * This program is an implementation of a part of one or more MPEG-4 |
; * Copyright (C) 2001 Peter Ross <pross@xvid.org> |
| 7 |
; * Video tools as specified in ISO/IEC 14496-2 standard. Those intending |
; * 2001 Michael Militzer <isibaar@xvid.org> |
| 8 |
; * to use this software module in hardware or software products are |
; * 2002 Pascal Massimino <skal@planet-d.net> |
|
; * advised that its use may infringe existing patents or copyrights, and |
|
|
; * any such use would be at such party's own risk. The original |
|
|
; * developer of this software module and his/her company, and subsequent |
|
|
; * editors and their companies, will have no liability for use of this |
|
|
; * software or modifications or derivatives thereof. |
|
| 9 |
; * |
; * |
| 10 |
; * This program is free software; you can redistribute it and/or modify |
; * This program is free software; you can redistribute it and/or modify |
| 11 |
; * it under the terms of the GNU General Public License as published by |
; * it under the terms of the GNU General Public License as published by |
| 19 |
; * |
; * |
| 20 |
; * You should have received a copy of the GNU General Public License |
; * You should have received a copy of the GNU General Public License |
| 21 |
; * along with this program; if not, write to the Free Software |
; * along with this program; if not, write to the Free Software |
| 22 |
; * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. |
; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
| 23 |
; * |
; * |
| 24 |
; *************************************************************************/ |
; * $Id$ |
|
|
|
|
;/************************************************************************** |
|
|
; * |
|
|
; * History: |
|
|
; * |
|
|
; * 07.01.2002 merge functions from compensate_mmx; rename functions |
|
|
; * 07.11.2001 initial version; (c)2001 peter ross <pross@cs.rmit.edu.au> |
|
| 25 |
; * |
; * |
| 26 |
; *************************************************************************/ |
; ***************************************************************************/ |
| 27 |
|
|
| 28 |
|
BITS 32 |
|
bits 32 |
|
| 29 |
|
|
| 30 |
%macro cglobal 1 |
%macro cglobal 1 |
| 31 |
%ifdef PREFIX |
%ifdef PREFIX |
| 32 |
|
%ifdef MARK_FUNCS |
| 33 |
|
global _%1:function %1.endfunc-%1 |
| 34 |
|
%define %1 _%1:function %1.endfunc-%1 |
| 35 |
|
%else |
| 36 |
global _%1 |
global _%1 |
| 37 |
%define %1 _%1 |
%define %1 _%1 |
| 38 |
|
%endif |
| 39 |
|
%else |
| 40 |
|
%ifdef MARK_FUNCS |
| 41 |
|
global %1:function %1.endfunc-%1 |
| 42 |
%else |
%else |
| 43 |
global %1 |
global %1 |
| 44 |
%endif |
%endif |
| 45 |
|
%endif |
| 46 |
%endmacro |
%endmacro |
| 47 |
|
|
| 48 |
|
;============================================================================= |
| 49 |
|
; Read only data |
| 50 |
|
;============================================================================= |
| 51 |
|
|
| 52 |
|
%ifdef FORMAT_COFF |
| 53 |
|
SECTION .rodata |
| 54 |
|
%else |
| 55 |
|
SECTION .rodata align=16 |
| 56 |
|
%endif |
| 57 |
|
|
| 58 |
|
ALIGN 16 |
| 59 |
|
mmx_one: |
| 60 |
|
dw 1, 1, 1, 1 |
| 61 |
|
|
| 62 |
|
;============================================================================= |
| 63 |
|
; Code |
| 64 |
|
;============================================================================= |
| 65 |
|
|
| 66 |
section .text |
SECTION .text |
| 67 |
|
|
| 68 |
|
cglobal transfer_8to16copy_mmx |
| 69 |
|
cglobal transfer_16to8copy_mmx |
| 70 |
|
cglobal transfer_8to16sub_mmx |
| 71 |
|
cglobal transfer_8to16subro_mmx |
| 72 |
|
cglobal transfer_8to16sub2_mmx |
| 73 |
|
cglobal transfer_8to16sub2_xmm |
| 74 |
|
cglobal transfer_8to16sub2ro_xmm |
| 75 |
|
cglobal transfer_16to8add_mmx |
| 76 |
|
cglobal transfer8x8_copy_mmx |
| 77 |
|
cglobal transfer8x4_copy_mmx |
| 78 |
|
|
| 79 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
| 80 |
; |
; |
| 81 |
; void transfer_8to16copy_mmx(int16_t * const dst, |
; void transfer_8to16copy_mmx(int16_t * const dst, |
| 82 |
; const uint8_t * const src, |
; const uint8_t * const src, |
| 83 |
; uint32_t stride); |
; uint32_t stride); |
| 84 |
; |
; |
| 85 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
|
|
|
align 16 |
|
|
cglobal transfer_8to16copy_mmx |
|
|
transfer_8to16copy_mmx |
|
|
|
|
|
push esi |
|
|
push edi |
|
|
|
|
|
mov edi, [esp + 8 + 4] ; dst |
|
|
mov esi, [esp + 8 + 8] ; src |
|
|
mov ecx, [esp + 8 + 12] ; stride |
|
|
|
|
|
pxor mm7, mm7 ; mm7 = zero |
|
|
|
|
|
mov eax, 8 |
|
| 86 |
|
|
| 87 |
.loop |
%macro COPY_8_TO_16 1 |
| 88 |
movq mm0, [esi] |
movq mm0, [eax] |
| 89 |
movq mm1, mm0 |
movq mm1, [eax+edx] |
| 90 |
punpcklbw mm0, mm7 ; mm01 = unpack([src]) |
movq mm2, mm0 |
| 91 |
punpckhbw mm1, mm7 |
movq mm3, mm1 |
| 92 |
|
punpcklbw mm0, mm7 |
| 93 |
movq [edi], mm0 ; [dst] = mm01 |
movq [ecx+%1*32], mm0 |
| 94 |
movq [edi + 8], mm1 |
punpcklbw mm1, mm7 |
| 95 |
|
movq [ecx+%1*32+16], mm1 |
| 96 |
|
punpckhbw mm2, mm7 |
| 97 |
|
punpckhbw mm3, mm7 |
| 98 |
|
lea eax, [eax+2*edx] |
| 99 |
|
movq [ecx+%1*32+8], mm2 |
| 100 |
|
movq [ecx+%1*32+24], mm3 |
| 101 |
|
%endmacro |
| 102 |
|
|
| 103 |
add edi, 16 |
ALIGN 16 |
| 104 |
add esi, ecx |
transfer_8to16copy_mmx: |
|
dec eax |
|
|
jnz .loop |
|
| 105 |
|
|
| 106 |
pop edi |
mov ecx, [esp+ 4] ; Dst |
| 107 |
pop esi |
mov eax, [esp+ 8] ; Src |
| 108 |
|
mov edx, [esp+12] ; Stride |
| 109 |
|
pxor mm7, mm7 |
| 110 |
|
|
| 111 |
|
COPY_8_TO_16 0 |
| 112 |
|
COPY_8_TO_16 1 |
| 113 |
|
COPY_8_TO_16 2 |
| 114 |
|
COPY_8_TO_16 3 |
| 115 |
ret |
ret |
| 116 |
|
.endfunc |
| 117 |
|
|
| 118 |
|
;----------------------------------------------------------------------------- |
|
|
|
|
;=========================================================================== |
|
| 119 |
; |
; |
| 120 |
; void transfer_16to8copy_mmx(uint8_t * const dst, |
; void transfer_16to8copy_mmx(uint8_t * const dst, |
| 121 |
; const int16_t * const src, |
; const int16_t * const src, |
| 122 |
; uint32_t stride); |
; uint32_t stride); |
| 123 |
; |
; |
| 124 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
|
|
|
align 16 |
|
|
cglobal transfer_16to8copy_mmx |
|
|
transfer_16to8copy_mmx |
|
| 125 |
|
|
| 126 |
push esi |
%macro COPY_16_TO_8 1 |
| 127 |
push edi |
movq mm0, [eax+%1*32] |
| 128 |
|
movq mm1, [eax+%1*32+8] |
| 129 |
mov edi, [esp + 8 + 4] ; dst |
packuswb mm0, mm1 |
| 130 |
mov esi, [esp + 8 + 8] ; src |
movq [ecx], mm0 |
| 131 |
mov ecx, [esp + 8 + 12] ; stride |
movq mm2, [eax+%1*32+16] |
| 132 |
|
movq mm3, [eax+%1*32+24] |
| 133 |
mov eax, 8 |
packuswb mm2, mm3 |
| 134 |
|
movq [ecx+edx], mm2 |
| 135 |
.loop |
%endmacro |
|
movq mm0, [esi] |
|
|
packuswb mm0, [esi + 8] ; mm0 = pack([src]) |
|
|
|
|
|
movq [edi], mm0 ; [dst] = mm0 |
|
|
|
|
|
add esi, 16 |
|
|
add edi, ecx |
|
|
dec eax |
|
|
jnz .loop |
|
| 136 |
|
|
| 137 |
pop edi |
ALIGN 16 |
| 138 |
pop esi |
transfer_16to8copy_mmx: |
| 139 |
|
|
| 140 |
|
mov ecx, [esp+ 4] ; Dst |
| 141 |
|
mov eax, [esp+ 8] ; Src |
| 142 |
|
mov edx, [esp+12] ; Stride |
| 143 |
|
|
| 144 |
|
COPY_16_TO_8 0 |
| 145 |
|
lea ecx,[ecx+2*edx] |
| 146 |
|
COPY_16_TO_8 1 |
| 147 |
|
lea ecx,[ecx+2*edx] |
| 148 |
|
COPY_16_TO_8 2 |
| 149 |
|
lea ecx,[ecx+2*edx] |
| 150 |
|
COPY_16_TO_8 3 |
| 151 |
ret |
ret |
| 152 |
|
.endfunc |
| 153 |
|
|
| 154 |
|
;----------------------------------------------------------------------------- |
|
;=========================================================================== |
|
| 155 |
; |
; |
| 156 |
; void transfer_8to16sub_mmx(int16_t * const dct, |
; void transfer_8to16sub_mmx(int16_t * const dct, |
| 157 |
; uint8_t * const cur, |
; uint8_t * const cur, |
| 158 |
; const uint8_t * const ref, |
; const uint8_t * const ref, |
| 159 |
; const uint32_t stride); |
; const uint32_t stride); |
| 160 |
; |
; |
| 161 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
;/************************************************************************** |
|
|
; * |
|
|
; * History: |
|
|
; * |
|
|
; * 27.12.2001 renamed from 'compensate' to 'transfer_8to16sub' |
|
|
; * 02.12.2001 loop unrolled, code runs 10% faster now (Isibaar) |
|
|
; * 30.11.2001 16 pixels are processed per iteration (Isibaar) |
|
|
; * 30.11.2001 .text missing |
|
|
; * 06.11.2001 inital version; (c)2001 peter ross <pross@cs.rmit.edu.au> |
|
|
; * |
|
|
; *************************************************************************/ |
|
| 162 |
|
|
| 163 |
align 16 |
; when second argument == 1, reference (ebx) block is to current (eax) |
| 164 |
cglobal transfer_8to16sub_mmx |
%macro COPY_8_TO_16_SUB 2 |
| 165 |
transfer_8to16sub_mmx |
movq mm0, [eax] ; cur |
| 166 |
push esi |
movq mm2, [eax+edx] |
|
push edi |
|
|
push ebx |
|
|
|
|
|
mov edi, [esp + 12 + 4] ; dct [out] |
|
|
mov edx, [esp + 12 + 8] ; cur [in/out] |
|
|
mov esi, [esp + 12 + 12] ; ref [in] |
|
|
mov ecx, [esp + 12 + 16] ; stride [in] |
|
|
|
|
|
mov eax, edx ; cur -> eax |
|
|
mov ebx, esi ; ref -> ebx |
|
|
add eax, ecx ; cur + stride |
|
|
add ebx, ecx ; ref + stride |
|
|
|
|
|
shl ecx, 1 |
|
|
|
|
|
pxor mm7, mm7 ; mm7 = zero |
|
|
|
|
|
movq mm0, [edx] ; mm01 = [cur] |
|
| 167 |
movq mm1, mm0 |
movq mm1, mm0 |
|
|
|
|
punpcklbw mm0, mm7 |
|
|
punpckhbw mm1, mm7 |
|
|
|
|
|
movq mm4, [eax] |
|
|
movq mm5, mm4 |
|
|
|
|
|
punpcklbw mm4, mm7 |
|
|
punpckhbw mm5, mm7 |
|
|
|
|
|
movq mm2, [esi] ; mm23 = [ref] |
|
| 168 |
movq mm3, mm2 |
movq mm3, mm2 |
| 169 |
|
|
| 170 |
movq mm6, [ebx] |
punpcklbw mm0, mm7 |
|
|
|
|
movq [edx], mm2 ; [cur] = [ref] |
|
|
movq [eax], mm6 |
|
|
|
|
| 171 |
punpcklbw mm2, mm7 |
punpcklbw mm2, mm7 |
| 172 |
|
movq mm4, [ebx] ; ref |
| 173 |
|
punpckhbw mm1, mm7 |
| 174 |
punpckhbw mm3, mm7 |
punpckhbw mm3, mm7 |
| 175 |
|
movq mm5, [ebx+edx] ; ref |
| 176 |
|
|
| 177 |
psubsw mm0, mm2 ; mm01 -= mm23 |
movq mm6, mm4 |
| 178 |
|
%if %2 == 1 |
| 179 |
|
movq [eax], mm4 |
| 180 |
|
movq [eax+edx], mm5 |
| 181 |
|
%endif |
| 182 |
|
punpcklbw mm4, mm7 |
| 183 |
|
punpckhbw mm6, mm7 |
| 184 |
|
psubsw mm0, mm4 |
| 185 |
|
psubsw mm1, mm6 |
| 186 |
|
movq mm6, mm5 |
| 187 |
|
punpcklbw mm5, mm7 |
| 188 |
|
punpckhbw mm6, mm7 |
| 189 |
|
psubsw mm2, mm5 |
| 190 |
|
lea eax, [eax+2*edx] |
| 191 |
|
psubsw mm3, mm6 |
| 192 |
|
lea ebx,[ebx+2*edx] |
| 193 |
|
|
| 194 |
|
movq [ecx+%1*32+ 0], mm0 ; dst |
| 195 |
|
movq [ecx+%1*32+ 8], mm1 |
| 196 |
|
movq [ecx+%1*32+16], mm2 |
| 197 |
|
movq [ecx+%1*32+24], mm3 |
| 198 |
|
%endmacro |
| 199 |
|
|
| 200 |
movq mm2, mm6 |
ALIGN 16 |
| 201 |
|
transfer_8to16sub_mmx: |
| 202 |
|
mov ecx, [esp + 4] ; Dst |
| 203 |
|
mov eax, [esp + 8] ; Cur |
| 204 |
|
push ebx |
| 205 |
|
mov ebx, [esp+4+12] ; Ref |
| 206 |
|
mov edx, [esp+4+16] ; Stride |
| 207 |
|
pxor mm7, mm7 |
| 208 |
|
|
| 209 |
punpcklbw mm2, mm7 |
COPY_8_TO_16_SUB 0, 1 |
| 210 |
punpckhbw mm6, mm7 |
COPY_8_TO_16_SUB 1, 1 |
| 211 |
|
COPY_8_TO_16_SUB 2, 1 |
| 212 |
|
COPY_8_TO_16_SUB 3, 1 |
| 213 |
|
|
| 214 |
psubsw mm1, mm3 |
pop ebx |
| 215 |
|
ret |
| 216 |
|
.endfunc |
| 217 |
|
|
|
psubsw mm4, mm2 |
|
|
psubsw mm5, mm6 |
|
| 218 |
|
|
| 219 |
movq [edi], mm0 ; dct[] = mm01 |
ALIGN 16 |
| 220 |
movq [edi + 8], mm1 |
transfer_8to16subro_mmx: |
| 221 |
movq [edi + 16], mm4 |
mov ecx, [esp + 4] ; Dst |
| 222 |
movq [edi + 24], mm5 |
mov eax, [esp + 8] ; Cur |
| 223 |
|
push ebx |
| 224 |
add edx, ecx |
mov ebx, [esp+4+12] ; Ref |
| 225 |
add esi, ecx |
mov edx, [esp+4+16] ; Stride |
| 226 |
add eax, ecx |
pxor mm7, mm7 |
|
add ebx, ecx |
|
| 227 |
|
|
| 228 |
movq mm0, [edx] ; mm01 = [cur] |
COPY_8_TO_16_SUB 0, 0 |
| 229 |
movq mm1, mm0 |
COPY_8_TO_16_SUB 1, 0 |
| 230 |
|
COPY_8_TO_16_SUB 2, 0 |
| 231 |
|
COPY_8_TO_16_SUB 3, 0 |
| 232 |
|
|
| 233 |
punpcklbw mm0, mm7 |
pop ebx |
| 234 |
punpckhbw mm1, mm7 |
ret |
| 235 |
|
.endfunc |
| 236 |
|
|
|
movq mm4, [eax] |
|
|
movq mm5, mm4 |
|
| 237 |
|
|
| 238 |
|
;----------------------------------------------------------------------------- |
| 239 |
|
; |
| 240 |
|
; void transfer_8to16sub2_mmx(int16_t * const dct, |
| 241 |
|
; uint8_t * const cur, |
| 242 |
|
; const uint8_t * ref1, |
| 243 |
|
; const uint8_t * ref2, |
| 244 |
|
; const uint32_t stride) |
| 245 |
|
; |
| 246 |
|
;----------------------------------------------------------------------------- |
| 247 |
|
|
| 248 |
|
%macro COPY_8_TO_16_SUB2_MMX 1 |
| 249 |
|
movq mm0, [eax] ; cur |
| 250 |
|
movq mm2, [eax+edx] |
| 251 |
|
|
| 252 |
|
; mm4 <- (ref1+ref2+1) / 2 |
| 253 |
|
movq mm4, [ebx] ; ref1 |
| 254 |
|
movq mm1, [esi] ; ref2 |
| 255 |
|
movq mm6, mm4 |
| 256 |
|
movq mm3, mm1 |
| 257 |
punpcklbw mm4, mm7 |
punpcklbw mm4, mm7 |
| 258 |
punpckhbw mm5, mm7 |
punpcklbw mm1, mm7 |
| 259 |
|
punpckhbw mm6, mm7 |
| 260 |
|
punpckhbw mm3, mm7 |
| 261 |
|
paddusw mm4, mm1 |
| 262 |
|
paddusw mm6, mm3 |
| 263 |
|
paddusw mm4, [mmx_one] |
| 264 |
|
paddusw mm6, [mmx_one] |
| 265 |
|
psrlw mm4, 1 |
| 266 |
|
psrlw mm6, 1 |
| 267 |
|
packuswb mm4, mm6 |
| 268 |
|
movq [eax], mm4 |
| 269 |
|
|
| 270 |
|
; mm5 <- (ref1+ref2+1) / 2 |
| 271 |
|
movq mm5, [ebx+edx] ; ref1 |
| 272 |
|
movq mm1, [esi+edx] ; ref2 |
| 273 |
|
movq mm6, mm5 |
| 274 |
|
movq mm3, mm1 |
| 275 |
|
punpcklbw mm5, mm7 |
| 276 |
|
punpcklbw mm1, mm7 |
| 277 |
|
punpckhbw mm6, mm7 |
| 278 |
|
punpckhbw mm3, mm7 |
| 279 |
|
paddusw mm5, mm1 |
| 280 |
|
paddusw mm6, mm3 |
| 281 |
|
paddusw mm5, [mmx_one] |
| 282 |
|
paddusw mm6, [mmx_one] |
| 283 |
|
lea esi, [esi+2*edx] |
| 284 |
|
psrlw mm5, 1 |
| 285 |
|
psrlw mm6, 1 |
| 286 |
|
packuswb mm5, mm6 |
| 287 |
|
movq [eax+edx], mm5 |
| 288 |
|
|
| 289 |
movq mm2, [esi] ; mm23 = [ref] |
movq mm1, mm0 |
| 290 |
movq mm3, mm2 |
movq mm3, mm2 |
| 291 |
|
punpcklbw mm0, mm7 |
|
movq mm6, [ebx] |
|
|
|
|
|
movq [edx], mm2 ; [cur] = [ref] |
|
|
movq [eax], mm6 |
|
|
|
|
| 292 |
punpcklbw mm2, mm7 |
punpcklbw mm2, mm7 |
| 293 |
|
punpckhbw mm1, mm7 |
| 294 |
punpckhbw mm3, mm7 |
punpckhbw mm3, mm7 |
| 295 |
|
|
| 296 |
psubsw mm0, mm2 ; mm01 -= mm23 |
movq mm6, mm4 |
| 297 |
|
punpcklbw mm4, mm7 |
| 298 |
movq mm2, mm6 |
punpckhbw mm6, mm7 |
| 299 |
|
psubsw mm0, mm4 |
| 300 |
punpcklbw mm2, mm7 |
psubsw mm1, mm6 |
| 301 |
|
movq mm6, mm5 |
| 302 |
|
punpcklbw mm5, mm7 |
| 303 |
punpckhbw mm6, mm7 |
punpckhbw mm6, mm7 |
| 304 |
|
psubsw mm2, mm5 |
| 305 |
|
lea eax, [eax+2*edx] |
| 306 |
|
psubsw mm3, mm6 |
| 307 |
|
lea ebx, [ebx+2*edx] |
| 308 |
|
|
| 309 |
|
movq [ecx+%1*32+ 0], mm0 ; dst |
| 310 |
|
movq [ecx+%1*32+ 8], mm1 |
| 311 |
|
movq [ecx+%1*32+16], mm2 |
| 312 |
|
movq [ecx+%1*32+24], mm3 |
| 313 |
|
%endmacro |
| 314 |
|
|
| 315 |
psubsw mm1, mm3 |
ALIGN 16 |
| 316 |
|
transfer_8to16sub2_mmx: |
| 317 |
|
mov ecx, [esp + 4] ; Dst |
| 318 |
|
mov eax, [esp + 8] ; Cur |
| 319 |
|
push ebx |
| 320 |
|
mov ebx, [esp+4+12] ; Ref1 |
| 321 |
|
push esi |
| 322 |
|
mov esi, [esp+8+16] ; Ref2 |
| 323 |
|
mov edx, [esp+8+20] ; Stride |
| 324 |
|
pxor mm7, mm7 |
| 325 |
|
|
| 326 |
psubsw mm4, mm2 |
COPY_8_TO_16_SUB2_MMX 0 |
| 327 |
psubsw mm5, mm6 |
COPY_8_TO_16_SUB2_MMX 1 |
| 328 |
|
COPY_8_TO_16_SUB2_MMX 2 |
| 329 |
|
COPY_8_TO_16_SUB2_MMX 3 |
| 330 |
|
|
| 331 |
movq [edi + 32], mm0 ; dct[] = mm01 |
pop esi |
| 332 |
movq [edi + 40], mm1 |
pop ebx |
| 333 |
movq [edi + 48], mm4 |
ret |
| 334 |
movq [edi + 56], mm5 |
.endfunc |
|
|
|
|
add edx, ecx |
|
|
add esi, ecx |
|
|
add eax, ecx |
|
|
add ebx, ecx |
|
| 335 |
|
|
| 336 |
movq mm0, [edx] ; mm01 = [cur] |
;----------------------------------------------------------------------------- |
| 337 |
|
; |
| 338 |
|
; void transfer_8to16sub2_xmm(int16_t * const dct, |
| 339 |
|
; uint8_t * const cur, |
| 340 |
|
; const uint8_t * ref1, |
| 341 |
|
; const uint8_t * ref2, |
| 342 |
|
; const uint32_t stride) |
| 343 |
|
; |
| 344 |
|
;----------------------------------------------------------------------------- |
| 345 |
|
|
| 346 |
|
%macro COPY_8_TO_16_SUB2_SSE 1 |
| 347 |
|
movq mm0, [eax] ; cur |
| 348 |
|
movq mm2, [eax+edx] |
| 349 |
movq mm1, mm0 |
movq mm1, mm0 |
|
|
|
|
punpcklbw mm0, mm7 |
|
|
punpckhbw mm1, mm7 |
|
|
|
|
|
movq mm4, [eax] |
|
|
movq mm5, mm4 |
|
|
|
|
|
punpcklbw mm4, mm7 |
|
|
punpckhbw mm5, mm7 |
|
|
|
|
|
movq mm2, [esi] ; mm23 = [ref] |
|
| 350 |
movq mm3, mm2 |
movq mm3, mm2 |
| 351 |
|
|
| 352 |
movq mm6, [ebx] |
punpcklbw mm0, mm7 |
|
|
|
|
movq [edx], mm2 ; [cur] = [ref] |
|
|
movq [eax], mm6 |
|
|
|
|
| 353 |
punpcklbw mm2, mm7 |
punpcklbw mm2, mm7 |
| 354 |
|
movq mm4, [ebx] ; ref1 |
| 355 |
|
pavgb mm4, [esi] ; ref2 |
| 356 |
|
movq [eax], mm4 |
| 357 |
|
punpckhbw mm1, mm7 |
| 358 |
punpckhbw mm3, mm7 |
punpckhbw mm3, mm7 |
| 359 |
|
movq mm5, [ebx+edx] ; ref |
| 360 |
|
pavgb mm5, [esi+edx] ; ref2 |
| 361 |
|
movq [eax+edx], mm5 |
| 362 |
|
|
| 363 |
psubsw mm0, mm2 ; mm01 -= mm23 |
movq mm6, mm4 |
| 364 |
|
punpcklbw mm4, mm7 |
|
movq mm2, mm6 |
|
|
|
|
|
punpcklbw mm2, mm7 |
|
| 365 |
punpckhbw mm6, mm7 |
punpckhbw mm6, mm7 |
| 366 |
|
psubsw mm0, mm4 |
| 367 |
|
psubsw mm1, mm6 |
| 368 |
|
lea esi, [esi+2*edx] |
| 369 |
|
movq mm6, mm5 |
| 370 |
|
punpcklbw mm5, mm7 |
| 371 |
|
punpckhbw mm6, mm7 |
| 372 |
|
psubsw mm2, mm5 |
| 373 |
|
lea eax, [eax+2*edx] |
| 374 |
|
psubsw mm3, mm6 |
| 375 |
|
lea ebx, [ebx+2*edx] |
| 376 |
|
|
| 377 |
|
movq [ecx+%1*32+ 0], mm0 ; dst |
| 378 |
|
movq [ecx+%1*32+ 8], mm1 |
| 379 |
|
movq [ecx+%1*32+16], mm2 |
| 380 |
|
movq [ecx+%1*32+24], mm3 |
| 381 |
|
%endmacro |
| 382 |
|
|
| 383 |
psubsw mm1, mm3 |
ALIGN 16 |
| 384 |
|
transfer_8to16sub2_xmm: |
| 385 |
psubsw mm4, mm2 |
mov ecx, [esp + 4] ; Dst |
| 386 |
psubsw mm5, mm6 |
mov eax, [esp + 8] ; Cur |
| 387 |
|
push ebx |
| 388 |
movq [edi + 64], mm0 ; dct[] = mm01 |
mov ebx, [esp+4+12] ; Ref1 |
| 389 |
movq [edi + 72], mm1 |
push esi |
| 390 |
movq [edi + 80], mm4 |
mov esi, [esp+8+16] ; Ref2 |
| 391 |
movq [edi + 88], mm5 |
mov edx, [esp+8+20] ; Stride |
| 392 |
|
pxor mm7, mm7 |
|
add edx, ecx |
|
|
add esi, ecx |
|
|
add eax, ecx |
|
|
add ebx, ecx |
|
| 393 |
|
|
| 394 |
movq mm0, [edx] ; mm01 = [cur] |
COPY_8_TO_16_SUB2_SSE 0 |
| 395 |
movq mm1, mm0 |
COPY_8_TO_16_SUB2_SSE 1 |
| 396 |
|
COPY_8_TO_16_SUB2_SSE 2 |
| 397 |
|
COPY_8_TO_16_SUB2_SSE 3 |
| 398 |
|
|
| 399 |
punpcklbw mm0, mm7 |
pop esi |
| 400 |
punpckhbw mm1, mm7 |
pop ebx |
| 401 |
|
ret |
| 402 |
|
.endfunc |
| 403 |
|
|
|
movq mm4, [eax] |
|
|
movq mm5, mm4 |
|
| 404 |
|
|
| 405 |
punpcklbw mm4, mm7 |
;----------------------------------------------------------------------------- |
| 406 |
punpckhbw mm5, mm7 |
; |
| 407 |
|
; void transfer_8to16sub2ro_xmm(int16_t * const dct, |
| 408 |
|
; const uint8_t * const cur, |
| 409 |
|
; const uint8_t * ref1, |
| 410 |
|
; const uint8_t * ref2, |
| 411 |
|
; const uint32_t stride) |
| 412 |
|
; |
| 413 |
|
;----------------------------------------------------------------------------- |
| 414 |
|
|
| 415 |
movq mm2, [esi] ; mm23 = [ref] |
%macro COPY_8_TO_16_SUB2RO_SSE 1 |
| 416 |
|
movq mm0, [eax] ; cur |
| 417 |
|
movq mm2, [eax+edx] |
| 418 |
|
movq mm1, mm0 |
| 419 |
movq mm3, mm2 |
movq mm3, mm2 |
| 420 |
|
|
| 421 |
movq mm6, [ebx] |
punpcklbw mm0, mm7 |
|
|
|
|
movq [edx], mm2 ; [cur] = [ref] |
|
|
movq [eax], mm6 |
|
|
|
|
| 422 |
punpcklbw mm2, mm7 |
punpcklbw mm2, mm7 |
| 423 |
|
movq mm4, [ebx] ; ref1 |
| 424 |
|
pavgb mm4, [esi] ; ref2 |
| 425 |
|
punpckhbw mm1, mm7 |
| 426 |
punpckhbw mm3, mm7 |
punpckhbw mm3, mm7 |
| 427 |
|
movq mm5, [ebx+edx] ; ref |
| 428 |
|
pavgb mm5, [esi+edx] ; ref2 |
| 429 |
|
|
| 430 |
psubsw mm0, mm2 ; mm01 -= mm23 |
movq mm6, mm4 |
| 431 |
|
punpcklbw mm4, mm7 |
|
movq mm2, mm6 |
|
|
|
|
|
punpcklbw mm2, mm7 |
|
| 432 |
punpckhbw mm6, mm7 |
punpckhbw mm6, mm7 |
| 433 |
|
psubsw mm0, mm4 |
| 434 |
|
psubsw mm1, mm6 |
| 435 |
|
lea esi, [esi+2*edx] |
| 436 |
|
movq mm6, mm5 |
| 437 |
|
punpcklbw mm5, mm7 |
| 438 |
|
punpckhbw mm6, mm7 |
| 439 |
|
psubsw mm2, mm5 |
| 440 |
|
lea eax, [eax+2*edx] |
| 441 |
|
psubsw mm3, mm6 |
| 442 |
|
lea ebx, [ebx+2*edx] |
| 443 |
|
|
| 444 |
|
movq [ecx+%1*32+ 0], mm0 ; dst |
| 445 |
|
movq [ecx+%1*32+ 8], mm1 |
| 446 |
|
movq [ecx+%1*32+16], mm2 |
| 447 |
|
movq [ecx+%1*32+24], mm3 |
| 448 |
|
%endmacro |
| 449 |
|
|
| 450 |
psubsw mm1, mm3 |
ALIGN 16 |
| 451 |
|
transfer_8to16sub2ro_xmm: |
| 452 |
psubsw mm4, mm2 |
pxor mm7, mm7 |
| 453 |
psubsw mm5, mm6 |
mov ecx, [esp + 4] ; Dst |
| 454 |
|
mov eax, [esp + 8] ; Cur |
| 455 |
|
push ebx |
| 456 |
|
mov ebx, [esp+4+12] ; Ref1 |
| 457 |
|
push esi |
| 458 |
|
mov esi, [esp+8+16] ; Ref2 |
| 459 |
|
mov edx, [esp+8+20] ; Stride |
| 460 |
|
|
| 461 |
movq [edi + 96], mm0 ; dct[] = mm01 |
COPY_8_TO_16_SUB2RO_SSE 0 |
| 462 |
movq [edi + 104], mm1 |
COPY_8_TO_16_SUB2RO_SSE 1 |
| 463 |
movq [edi + 112], mm4 |
COPY_8_TO_16_SUB2RO_SSE 2 |
| 464 |
movq [edi + 120], mm5 |
COPY_8_TO_16_SUB2RO_SSE 3 |
| 465 |
|
|
|
pop ebx |
|
|
pop edi |
|
| 466 |
pop esi |
pop esi |
| 467 |
|
pop ebx |
| 468 |
ret |
ret |
| 469 |
|
.endfunc |
| 470 |
|
|
| 471 |
|
|
| 472 |
|
;----------------------------------------------------------------------------- |
|
;=========================================================================== |
|
| 473 |
; |
; |
| 474 |
; void transfer_16to8add_mmx(uint8_t * const dst, |
; void transfer_16to8add_mmx(uint8_t * const dst, |
| 475 |
; const int16_t * const src, |
; const int16_t * const src, |
| 476 |
; uint32_t stride); |
; uint32_t stride); |
| 477 |
; |
; |
| 478 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
|
|
|
|
align 16 |
|
|
cglobal transfer_16to8add_mmx |
|
|
transfer_16to8add_mmx |
|
|
|
|
|
push esi |
|
|
push edi |
|
|
|
|
|
mov edi, [esp + 8 + 4] ; dst |
|
|
mov esi, [esp + 8 + 8] ; src |
|
|
mov ecx, [esp + 8 + 12] ; stride |
|
|
|
|
|
pxor mm7, mm7 |
|
| 479 |
|
|
| 480 |
mov eax, 8 |
%macro COPY_16_TO_8_ADD 1 |
| 481 |
|
movq mm0, [ecx] |
| 482 |
.loop |
movq mm2, [ecx+edx] |
|
movq mm0, [edi] |
|
| 483 |
movq mm1, mm0 |
movq mm1, mm0 |
| 484 |
punpcklbw mm0, mm7 ; mm23 = unpack([dst]) |
movq mm3, mm2 |
| 485 |
|
punpcklbw mm0, mm7 |
| 486 |
|
punpcklbw mm2, mm7 |
| 487 |
punpckhbw mm1, mm7 |
punpckhbw mm1, mm7 |
| 488 |
|
punpckhbw mm3, mm7 |
| 489 |
|
paddsw mm0, [eax+%1*32+ 0] |
| 490 |
|
paddsw mm1, [eax+%1*32+ 8] |
| 491 |
|
paddsw mm2, [eax+%1*32+16] |
| 492 |
|
paddsw mm3, [eax+%1*32+24] |
| 493 |
|
packuswb mm0, mm1 |
| 494 |
|
movq [ecx], mm0 |
| 495 |
|
packuswb mm2, mm3 |
| 496 |
|
movq [ecx+edx], mm2 |
| 497 |
|
%endmacro |
| 498 |
|
|
|
movq mm2, [esi] ; mm01 = [src] |
|
|
movq mm3, [esi + 8] |
|
|
|
|
|
paddsw mm0, mm2 ; mm01 += mm23 |
|
|
paddsw mm1, mm3 |
|
|
|
|
|
packuswb mm0, mm1 ; [dst] = pack(mm01) |
|
|
movq [edi], mm0 |
|
|
|
|
|
add esi, 16 |
|
|
add edi, ecx |
|
|
dec eax |
|
|
jnz .loop |
|
| 499 |
|
|
| 500 |
pop edi |
ALIGN 16 |
| 501 |
pop esi |
transfer_16to8add_mmx: |
| 502 |
|
mov ecx, [esp+ 4] ; Dst |
| 503 |
|
mov eax, [esp+ 8] ; Src |
| 504 |
|
mov edx, [esp+12] ; Stride |
| 505 |
|
pxor mm7, mm7 |
| 506 |
|
|
| 507 |
|
COPY_16_TO_8_ADD 0 |
| 508 |
|
lea ecx,[ecx+2*edx] |
| 509 |
|
COPY_16_TO_8_ADD 1 |
| 510 |
|
lea ecx,[ecx+2*edx] |
| 511 |
|
COPY_16_TO_8_ADD 2 |
| 512 |
|
lea ecx,[ecx+2*edx] |
| 513 |
|
COPY_16_TO_8_ADD 3 |
| 514 |
ret |
ret |
| 515 |
|
.endfunc |
| 516 |
|
|
| 517 |
|
;----------------------------------------------------------------------------- |
|
;=========================================================================== |
|
| 518 |
; |
; |
| 519 |
; void transfer8x8_copy_mmx(uint8_t * const dst, |
; void transfer8x8_copy_mmx(uint8_t * const dst, |
| 520 |
; const uint8_t * const src, |
; const uint8_t * const src, |
| 521 |
; const uint32_t stride); |
; const uint32_t stride); |
| 522 |
; |
; |
| 523 |
; |
; |
| 524 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
| 525 |
|
|
| 526 |
align 16 |
%macro COPY_8_TO_8 0 |
| 527 |
cglobal transfer8x8_copy_mmx |
movq mm0, [eax] |
| 528 |
transfer8x8_copy_mmx |
movq mm1, [eax+edx] |
| 529 |
push esi |
movq [ecx], mm0 |
| 530 |
push edi |
lea eax, [eax+2*edx] |
| 531 |
|
movq [ecx+edx], mm1 |
| 532 |
|
%endmacro |
| 533 |
|
|
| 534 |
mov edi, [esp + 8 + 4] ; dst [out] |
ALIGN 16 |
| 535 |
mov esi, [esp + 8 + 8] ; src [in] |
transfer8x8_copy_mmx: |
| 536 |
mov eax, [esp + 8 + 12] ; stride [in] |
mov ecx, [esp+ 4] ; Dst |
| 537 |
|
mov eax, [esp+ 8] ; Src |
| 538 |
movq mm0, [esi] |
mov edx, [esp+12] ; Stride |
| 539 |
movq mm1, [esi+eax] |
|
| 540 |
movq [edi], mm0 |
COPY_8_TO_8 |
| 541 |
movq [edi+eax], mm1 |
lea ecx,[ecx+2*edx] |
| 542 |
|
COPY_8_TO_8 |
| 543 |
add esi, eax |
lea ecx,[ecx+2*edx] |
| 544 |
add edi, eax |
COPY_8_TO_8 |
| 545 |
add esi, eax |
lea ecx,[ecx+2*edx] |
| 546 |
add edi, eax |
COPY_8_TO_8 |
| 547 |
|
ret |
| 548 |
movq mm0, [esi] |
.endfunc |
|
movq mm1, [esi+eax] |
|
|
movq [edi], mm0 |
|
|
movq [edi+eax], mm1 |
|
|
|
|
|
add esi, eax |
|
|
add edi, eax |
|
|
add esi, eax |
|
|
add edi, eax |
|
|
|
|
|
movq mm0, [esi] |
|
|
movq mm1, [esi+eax] |
|
|
movq [edi], mm0 |
|
|
movq [edi+eax], mm1 |
|
|
|
|
|
add esi, eax |
|
|
add edi, eax |
|
|
add esi, eax |
|
|
add edi, eax |
|
|
|
|
|
movq mm0, [esi] |
|
|
movq mm1, [esi+eax] |
|
|
movq [edi], mm0 |
|
|
movq [edi+eax], mm1 |
|
|
|
|
|
add esi, eax |
|
|
add edi, eax |
|
|
add esi, eax |
|
|
add edi, eax |
|
| 549 |
|
|
| 550 |
pop edi |
;----------------------------------------------------------------------------- |
| 551 |
pop esi |
; |
| 552 |
|
; void transfer8x4_copy_mmx(uint8_t * const dst, |
| 553 |
|
; const uint8_t * const src, |
| 554 |
|
; const uint32_t stride); |
| 555 |
|
; |
| 556 |
|
; |
| 557 |
|
;----------------------------------------------------------------------------- |
| 558 |
|
|
| 559 |
|
ALIGN 16 |
| 560 |
|
transfer8x4_copy_mmx: |
| 561 |
|
mov ecx, [esp+ 4] ; Dst |
| 562 |
|
mov eax, [esp+ 8] ; Src |
| 563 |
|
mov edx, [esp+12] ; Stride |
| 564 |
|
|
| 565 |
|
COPY_8_TO_8 |
| 566 |
|
lea ecx,[ecx+2*edx] |
| 567 |
|
COPY_8_TO_8 |
| 568 |
ret |
ret |
| 569 |
|
.endfunc |
| 570 |
|
|
Parent Directory
| 
Revision Log
| 
Patch