1 |
;/************************************************************************** |
;/**************************************************************************** |
2 |
; * |
; * |
3 |
; * XVID MPEG-4 VIDEO CODEC |
; * XVID MPEG-4 VIDEO CODEC |
4 |
; * colorspace |
; * - MMX and XMM YUYV<->YV12 conversion - |
5 |
; * |
; * |
6 |
; * This program is free software; you can redistribute it and/or modify |
; * Copyright(C) 2002 Peter Ross <pross@xvid.org> |
7 |
; * it under the terms of the GNU General Public License as published by |
; * |
8 |
|
; * This program is free software; you can redistribute it and/or modify it |
9 |
|
; * under the terms of the GNU General Public License as published by |
10 |
; * the Free Software Foundation; either version 2 of the License, or |
; * the Free Software Foundation; either version 2 of the License, or |
11 |
; * (at your option) any later version. |
; * (at your option) any later version. |
12 |
; * |
; * |
17 |
; * |
; * |
18 |
; * You should have received a copy of the GNU General Public License |
; * You should have received a copy of the GNU General Public License |
19 |
; * along with this program; if not, write to the Free Software |
; * along with this program; if not, write to the Free Software |
20 |
; * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. |
; * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
|
; * |
|
|
; *************************************************************************/ |
|
|
|
|
|
;/************************************************************************** |
|
21 |
; * |
; * |
22 |
; * History: |
; * $Id: colorspace_yuyv_mmx.asm,v 1.7 2006-06-05 21:27:36 Skal Exp $ |
23 |
; * |
; * |
24 |
; * 10.10.2001 initial version; (c)2002 peter ross <pross@xvid.org> |
; ***************************************************************************/ |
|
; * |
|
|
; *************************************************************************/ |
|
|
|
|
25 |
|
|
26 |
bits 32 |
BITS 32 |
27 |
|
|
28 |
%macro cglobal 1 |
%macro cglobal 1 |
29 |
%ifdef PREFIX |
%ifdef PREFIX |
30 |
|
%ifdef MARK_FUNCS |
31 |
|
global _%1:function %1.endfunc-%1 |
32 |
|
%define %1 _%1:function %1.endfunc-%1 |
33 |
|
%else |
34 |
global _%1 |
global _%1 |
35 |
%define %1 _%1 |
%define %1 _%1 |
36 |
|
%endif |
37 |
|
%else |
38 |
|
%ifdef MARK_FUNCS |
39 |
|
global %1:function %1.endfunc-%1 |
40 |
%else |
%else |
41 |
global %1 |
global %1 |
42 |
%endif |
%endif |
43 |
|
%endif |
44 |
%endmacro |
%endmacro |
45 |
|
|
46 |
|
;============================================================================= |
47 |
|
; Read only data |
48 |
|
;============================================================================= |
49 |
|
|
50 |
section .data |
%ifdef FORMAT_COFF |
51 |
align 16 |
SECTION .rodata |
52 |
|
%else |
53 |
|
SECTION .rodata align=16 |
54 |
|
%endif |
55 |
|
|
56 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
57 |
; yuyv/uyvy mask for extracting yuv components |
; yuyv/uyvy mask for extracting yuv components |
58 |
;=========================================================================== |
;----------------------------------------------------------------------------- |
59 |
; y u y v y u y v |
; y u y v y u y v |
|
yuyv_mask db 0xff, 0, 0xff, 0, 0xff, 0, 0xff, 0 |
|
|
mmx_one dw 1, 1, 1, 1 |
|
|
|
|
|
|
|
60 |
|
|
61 |
section .text |
ALIGN 16 |
62 |
|
yuyv_mask: db 0xff, 0, 0xff, 0, 0xff, 0, 0xff, 0 |
63 |
|
mmx_one: dw 1, 1, 1, 1 |
64 |
|
|
65 |
|
;============================================================================= |
66 |
|
; helper macros used with colorspace_mmx.inc |
67 |
|
;============================================================================= |
68 |
|
|
69 |
%include "colorspace_mmx.inc" |
;----------------------------------------------------------------------------- |
|
|
|
|
|
|
|
;==================================================================== |
|
70 |
; YUYV_TO_YV12( TYPE, PAVG ) |
; YUYV_TO_YV12( TYPE, PAVG ) |
71 |
; |
; |
72 |
; TYPE 0=yuyv, 1=uyvy |
; TYPE 0=yuyv, 1=uyvy |
73 |
; PAVG 0=mmx, pavgusb=3dnow, pavgb=xmm |
; PAVG 0=mmx, pavgusb=3dnow, pavgb=xmm |
74 |
; |
; |
75 |
; bytes=2, pixels = 8, vpixels=2 |
; bytes=2, pixels = 8, vpixels=2 |
76 |
;==================================================================== |
;----------------------------------------------------------------------------- |
77 |
|
|
78 |
%macro YUYV_TO_YV12_INIT 2 |
%macro YUYV_TO_YV12_INIT 2 |
79 |
movq mm7, [yuyv_mask] |
movq mm7, [yuyv_mask] |
80 |
%endmacro |
%endmacro |
168 |
movd [ebx],mm5 |
movd [ebx],mm5 |
169 |
movd [ecx],mm4 |
movd [ecx],mm4 |
170 |
%endmacro |
%endmacro |
|
;==================================================================== |
|
|
|
|
171 |
|
|
172 |
;------------------------------------------------------------------------------ |
;----------------------------------------------------------------------------- |
173 |
; YV12_TO_YUYV( TYPE ) |
; YV12_TO_YUYV( TYPE ) |
174 |
; |
; |
175 |
; TYPE 0=yuyv, 1=uyvy |
; bytes=2, pixels = 16, vpixels=2 |
176 |
; |
;----------------------------------------------------------------------------- |
177 |
; bytes=2, pixels = 8, vpixels=2 |
|
|
;------------------------------------------------------------------------------ |
|
178 |
%macro YV12_TO_YUYV_INIT 2 |
%macro YV12_TO_YUYV_INIT 2 |
179 |
%endmacro |
%endmacro |
180 |
|
|
181 |
|
|
182 |
%macro YV12_TO_YUYV 2 |
%macro YV12_TO_YUYV 2 |
183 |
movd mm4, [ebx] ; [ |uuuu] |
movq mm6, [ebx] ; [ |uuuu] |
184 |
movd mm5, [ecx] ; [ |vvvv] |
movq mm2, [ecx] ; [ |vvvv] |
185 |
movq mm0, [esi] ; [yyyy|yyyy] ; y row 0 |
movq mm0, [esi] ; [yyyy|yyyy] ; y row 0 |
186 |
movq mm1, [esi+eax] ; [yyyy|yyyy] ; y row 1 |
movq mm1, [esi+eax] ; [yyyy|yyyy] ; y row 1 |
187 |
punpcklbw mm4, mm5 ; [vuvu|vuvu] ; uv row 0 |
movq mm7, mm6 |
188 |
|
punpcklbw mm6, mm2 ; [vuvu|vuvu] ; uv[0..3] |
189 |
|
punpckhbw mm7, mm2 ; [vuvu|vuvu] ; uv[4..7] |
190 |
|
|
191 |
%if %1 == 0 ; YUYV |
%if %1 == 0 ; YUYV |
192 |
movq mm2, mm0 |
movq mm2, mm0 |
193 |
movq mm3, mm1 |
movq mm3, mm1 |
194 |
punpcklbw mm0, mm4 ; [vyuy|vyuy] ; y row 0 + 0 |
movq mm4, [esi +8] ; [yyyy|yyyy] ; y[8..15] row 0 |
195 |
punpckhbw mm2, mm4 ; [vyuy|vyuy] ; y row 0 + 8 |
movq mm5, [esi+eax+8] ; [yyyy|yyyy] ; y[8..15] row 1 |
196 |
punpcklbw mm1, mm4 ; [vyuy|vyuy] ; y row 1 + 0 |
punpcklbw mm0, mm6 ; [vyuy|vyuy] ; y row 0 + 0 |
197 |
punpckhbw mm3, mm4 ; [vyuy|vyuy] ; y row 1 + 8 |
punpckhbw mm2, mm6 ; [vyuy|vyuy] ; y row 0 + 8 |
198 |
|
punpcklbw mm1, mm6 ; [vyuy|vyuy] ; y row 1 + 0 |
199 |
|
punpckhbw mm3, mm6 ; [vyuy|vyuy] ; y row 1 + 8 |
200 |
movq [edi], mm0 |
movq [edi], mm0 |
201 |
movq [edi+8], mm2 |
movq [edi+8], mm2 |
202 |
movq [edi+edx], mm1 |
movq [edi+edx], mm1 |
203 |
movq [edi+edx+8], mm3 |
movq [edi+edx+8], mm3 |
204 |
|
movq mm0, mm4 |
205 |
|
movq mm2, mm5 |
206 |
|
punpcklbw mm0, mm7 ; [vyuy|vyuy] ; y row 0 + 16 |
207 |
|
punpckhbw mm4, mm7 ; [vyuy|vyuy] ; y row 0 + 24 |
208 |
|
punpcklbw mm2, mm7 ; [vyuy|vyuy] ; y row 1 + 16 |
209 |
|
punpckhbw mm5, mm7 ; [vyuy|vyuy] ; y row 1 + 24 |
210 |
|
movq [edi +16], mm0 |
211 |
|
movq [edi +24], mm4 |
212 |
|
movq [edi+edx+16], mm2 |
213 |
|
movq [edi+edx+24], mm5 |
214 |
%else ; UYVY |
%else ; UYVY |
215 |
movq mm5, mm4 |
movq mm2, mm6 |
216 |
movq mm6, mm4 |
movq mm3, mm6 |
217 |
movq mm7, mm4 |
movq mm4, mm6 |
218 |
punpcklbw mm4, mm0 ; [yvyu|yvyu] ; y row 0 + 0 |
punpcklbw mm2, mm0 ; [yvyu|yvyu] ; y row 0 + 0 |
219 |
punpckhbw mm5, mm0 ; [yvyu|yvyu] ; y row 0 + 8 |
punpckhbw mm3, mm0 ; [yvyu|yvyu] ; y row 0 + 8 |
220 |
punpcklbw mm6, mm1 ; [yvyu|yvyu] ; y row 1 + 0 |
movq mm0, [esi +8] ; [yyyy|yyyy] ; y[8..15] row 0 |
221 |
punpckhbw mm7, mm1 ; [yvyu|yvyu] ; y row 1 + 8 |
movq mm5, [esi+eax+8] ; [yyyy|yyyy] ; y[8..15] row 1 |
222 |
movq [edi], mm4 |
punpcklbw mm4, mm1 ; [yvyu|yvyu] ; y row 1 + 0 |
223 |
movq [edi+8], mm5 |
punpckhbw mm6, mm1 ; [yvyu|yvyu] ; y row 1 + 8 |
224 |
movq [edi+edx], mm6 |
movq [edi ], mm2 |
225 |
movq [edi+edx+8], mm7 |
movq [edi +8], mm3 |
226 |
|
movq [edi+edx ], mm4 |
227 |
|
movq [edi+edx+8], mm6 |
228 |
|
movq mm2, mm7 |
229 |
|
movq mm3, mm7 |
230 |
|
movq mm6, mm7 |
231 |
|
punpcklbw mm2, mm0 ; [yvyu|yvyu] ; y row 0 + 0 |
232 |
|
punpckhbw mm3, mm0 ; [yvyu|yvyu] ; y row 0 + 8 |
233 |
|
punpcklbw mm6, mm5 ; [yvyu|yvyu] ; y row 1 + 0 |
234 |
|
punpckhbw mm7, mm5 ; [yvyu|yvyu] ; y row 1 + 8 |
235 |
|
movq [edi +16], mm2 |
236 |
|
movq [edi +24], mm3 |
237 |
|
movq [edi+edx+16], mm6 |
238 |
|
movq [edi+edx+24], mm7 |
239 |
%endif |
%endif |
240 |
%endmacro |
%endmacro |
|
;------------------------------------------------------------------------------ |
|
|
|
|
241 |
|
|
242 |
;------------------------------------------------------------------------------ |
;------------------------------------------------------------------------------ |
243 |
; YV12_TO_YUYVI( TYPE ) |
; YV12_TO_YUYVI( TYPE ) |
246 |
; |
; |
247 |
; bytes=2, pixels = 8, vpixels=4 |
; bytes=2, pixels = 8, vpixels=4 |
248 |
;------------------------------------------------------------------------------ |
;------------------------------------------------------------------------------ |
249 |
|
|
250 |
%macro YV12_TO_YUYVI_INIT 2 |
%macro YV12_TO_YUYVI_INIT 2 |
251 |
%endmacro |
%endmacro |
252 |
|
|
328 |
pop esi |
pop esi |
329 |
%endif |
%endif |
330 |
%endmacro |
%endmacro |
|
;------------------------------------------------------------------------------ |
|
331 |
|
|
332 |
|
;============================================================================= |
333 |
|
; Code |
334 |
|
;============================================================================= |
335 |
|
|
336 |
|
SECTION .text |
337 |
|
|
338 |
|
%include "colorspace_mmx.inc" |
339 |
|
|
340 |
; input |
; input |
341 |
|
|
349 |
|
|
350 |
; output |
; output |
351 |
|
|
352 |
MAKE_COLORSPACE yv12_to_yuyv_mmx,0, 2,8,2, YV12_TO_YUYV, 0, -1 |
MAKE_COLORSPACE yv12_to_yuyv_mmx,0, 2,16,2, YV12_TO_YUYV, 0, -1 |
353 |
MAKE_COLORSPACE yv12_to_uyvy_mmx,0, 2,8,2, YV12_TO_YUYV, 1, -1 |
MAKE_COLORSPACE yv12_to_uyvy_mmx,0, 2,16,2, YV12_TO_YUYV, 1, -1 |
354 |
|
|
355 |
MAKE_COLORSPACE yv12_to_yuyvi_mmx,0, 2,8,4, YV12_TO_YUYVI, 0, -1 |
MAKE_COLORSPACE yv12_to_yuyvi_mmx,0, 2,8,4, YV12_TO_YUYVI, 0, -1 |
356 |
MAKE_COLORSPACE yv12_to_uyvyi_mmx,0, 2,8,4, YV12_TO_YUYVI, 1, -1 |
MAKE_COLORSPACE yv12_to_uyvyi_mmx,0, 2,8,4, YV12_TO_YUYVI, 1, -1 |