1 /*****************************************************************************
2 * i420_ymga.c : YUV to YUV conversion module for vlc
3 *****************************************************************************
4 * Copyright (C) 2000, 2001 VideoLAN
5 * $Id: i420_ymga.c,v 1.7 2002/07/23 00:39:16 sam Exp $
7 * Authors: Samuel Hocevar <sam@zoy.org>
9 * 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.
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.
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, USA.
22 *****************************************************************************/
24 /*****************************************************************************
26 *****************************************************************************/
27 #include <errno.h> /* ENOMEM */
28 #include <string.h> /* strerror() */
29 #include <stdlib.h> /* malloc(), free() */
34 #define SRC_FOURCC "I420,IYUV,YV12"
35 #define DEST_FOURCC "YMGA"
37 /*****************************************************************************
38 * Local and extern prototypes.
39 *****************************************************************************/
40 static void chroma_getfunctions ( function_list_t * p_function_list );
42 static int chroma_Init ( vout_thread_t *p_vout );
43 static void chroma_End ( vout_thread_t *p_vout );
45 static void I420_YMGA ( vout_thread_t *, picture_t *, picture_t * );
47 /*****************************************************************************
48 * Build configuration tree.
49 *****************************************************************************/
54 #if defined (MODULE_NAME_IS_chroma_i420_ymga)
55 SET_DESCRIPTION( _("conversions from " SRC_FOURCC " to " DEST_FOURCC) )
56 ADD_CAPABILITY( CHROMA, 80 )
57 #elif defined (MODULE_NAME_IS_chroma_i420_ymga_mmx)
58 SET_DESCRIPTION( _("MMX conversions from " SRC_FOURCC " to " DEST_FOURCC) )
59 ADD_CAPABILITY( CHROMA, 100 )
60 ADD_REQUIREMENT( MMX )
65 chroma_getfunctions( &p_module->p_functions->chroma );
68 MODULE_DEACTIVATE_START
69 MODULE_DEACTIVATE_STOP
71 /*****************************************************************************
72 * Functions exported as capabilities. They are declared as static so that
73 * we don't pollute the namespace too much.
74 *****************************************************************************/
75 static void chroma_getfunctions( function_list_t * p_function_list )
77 p_function_list->functions.chroma.pf_init = chroma_Init;
78 p_function_list->functions.chroma.pf_end = chroma_End;
81 /*****************************************************************************
82 * chroma_Init: allocate a chroma function
83 *****************************************************************************
84 * This function allocates and initializes a chroma function
85 *****************************************************************************/
86 static int chroma_Init( vout_thread_t *p_vout )
88 if( p_vout->render.i_width & 1 || p_vout->render.i_height & 1 )
93 switch( p_vout->render.i_chroma )
95 case VLC_FOURCC('Y','V','1','2'):
96 case VLC_FOURCC('I','4','2','0'):
97 case VLC_FOURCC('I','Y','U','V'):
98 switch( p_vout->output.i_chroma )
100 case VLC_FOURCC('Y','M','G','A'):
101 p_vout->chroma.pf_convert = I420_YMGA;
116 /*****************************************************************************
117 * chroma_End: free the chroma function
118 *****************************************************************************
119 * This function frees the previously allocated chroma function
120 *****************************************************************************/
121 static void chroma_End( vout_thread_t *p_vout )
126 /* Following functions are local */
128 /*****************************************************************************
129 * I420_YMGA: planar YUV 4:2:0 to Matrox's planar/packed YUV 4:2:0
130 *****************************************************************************/
131 static void I420_YMGA( vout_thread_t *p_vout, picture_t *p_source,
134 u8 *p_uv = p_dest->U_PIXELS;
135 u8 *p_u = p_source->U_PIXELS;
136 u8 *p_v = p_source->V_PIXELS;
140 /* Copy the Y part */
141 p_vout->p_vlc->pf_memcpy( p_dest->Y_PIXELS, p_source->Y_PIXELS,
142 p_dest->p[Y_PLANE].i_pitch * p_dest->p[Y_PLANE].i_lines );
144 /* Copy the U:V part */
145 for( i_x = p_dest->p[U_PLANE].i_pitch * p_dest->p[U_PLANE].i_lines / 64;
148 #if defined (MODULE_NAME_IS_chroma_i420_ymga)
149 *p_uv++ = *p_u++; *p_uv++ = *p_v++; *p_uv++ = *p_u++; *p_uv++ = *p_v++;
150 *p_uv++ = *p_u++; *p_uv++ = *p_v++; *p_uv++ = *p_u++; *p_uv++ = *p_v++;
151 *p_uv++ = *p_u++; *p_uv++ = *p_v++; *p_uv++ = *p_u++; *p_uv++ = *p_v++;
152 *p_uv++ = *p_u++; *p_uv++ = *p_v++; *p_uv++ = *p_u++; *p_uv++ = *p_v++;
153 *p_uv++ = *p_u++; *p_uv++ = *p_v++; *p_uv++ = *p_u++; *p_uv++ = *p_v++;
154 *p_uv++ = *p_u++; *p_uv++ = *p_v++; *p_uv++ = *p_u++; *p_uv++ = *p_v++;
155 *p_uv++ = *p_u++; *p_uv++ = *p_v++; *p_uv++ = *p_u++; *p_uv++ = *p_v++;
156 *p_uv++ = *p_u++; *p_uv++ = *p_v++; *p_uv++ = *p_u++; *p_uv++ = *p_v++;
158 __asm__( ".align 32 \n\
159 movd (%0), %%mm0 # Load 4 Cr 00 00 00 00 v3 v2 v1 v0 \n\
160 movd 4(%0), %%mm2 # Load 4 Cr 00 00 00 00 v3 v2 v1 v0 \n\
161 movd 8(%0), %%mm4 # Load 4 Cr 00 00 00 00 v3 v2 v1 v0 \n\
162 movd 12(%0), %%mm6 # Load 4 Cr 00 00 00 00 v3 v2 v1 v0 \n\
163 movd (%1), %%mm1 # Load 4 Cb 00 00 00 00 u3 u2 u1 u0 \n\
164 movd 4(%1), %%mm3 # Load 4 Cb 00 00 00 00 u3 u2 u1 u0 \n\
165 movd 8(%1), %%mm5 # Load 4 Cb 00 00 00 00 u3 u2 u1 u0 \n\
166 movd 12(%1), %%mm7 # Load 4 Cb 00 00 00 00 u3 u2 u1 u0 \n\
167 punpcklbw %%mm1, %%mm0 # u3 v3 u2 v2 u1 v1 u0 v0 \n\
168 punpcklbw %%mm3, %%mm2 # u3 v3 u2 v2 u1 v1 u0 v0 \n\
169 punpcklbw %%mm5, %%mm4 # u3 v3 u2 v2 u1 v1 u0 v0 \n\
170 punpcklbw %%mm7, %%mm6 # u3 v3 u2 v2 u1 v1 u0 v0 \n\
171 movq %%mm0, (%2) # Store CrCb \n\
172 movq %%mm2, 8(%2) # Store CrCb \n\
173 movq %%mm4, 16(%2) # Store CrCb \n\
174 movq %%mm6, 24(%2) # Store CrCb"
175 : : "r" (p_v), "r" (p_u), "r" (p_uv) );
177 p_v += 16; p_u += 16; p_uv += 32;