1 /*****************************************************************************
2 * i420_ymga.c : YUV to YUV conversion module for vlc
3 *****************************************************************************
4 * Copyright (C) 2000, 2001 the VideoLAN team
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., 51 Franklin Street, Fifth Floor, Boston MA 02110-1301, 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 int Activate ( vlc_object_t * );
41 static void I420_YMGA ( vout_thread_t *, picture_t *, picture_t * );
43 /*****************************************************************************
45 *****************************************************************************/
47 #if defined (MODULE_NAME_IS_i420_ymga)
48 set_description( _("Conversions from " SRC_FOURCC " to " DEST_FOURCC) );
49 set_capability( "chroma", 80 );
50 #elif defined (MODULE_NAME_IS_i420_ymga_mmx)
51 set_description( _("MMX conversions from " SRC_FOURCC " to " DEST_FOURCC) );
52 set_capability( "chroma", 100 );
53 add_requirement( MMX );
55 set_callbacks( Activate, NULL );
58 /*****************************************************************************
59 * Activate: allocate a chroma function
60 *****************************************************************************
61 * This function allocates and initializes a chroma function
62 *****************************************************************************/
63 static int Activate( vlc_object_t *p_this )
65 vout_thread_t *p_vout = (vout_thread_t *)p_this;
67 if( p_vout->render.i_width & 1 || p_vout->render.i_height & 1 )
72 switch( p_vout->render.i_chroma )
74 case VLC_FOURCC('Y','V','1','2'):
75 case VLC_FOURCC('I','4','2','0'):
76 case VLC_FOURCC('I','Y','U','V'):
77 switch( p_vout->output.i_chroma )
79 case VLC_FOURCC('Y','M','G','A'):
80 p_vout->chroma.pf_convert = I420_YMGA;
95 /* Following functions are local */
97 /*****************************************************************************
98 * I420_YMGA: planar YUV 4:2:0 to Matrox's planar/packed YUV 4:2:0
99 *****************************************************************************/
100 static void I420_YMGA( vout_thread_t *p_vout, picture_t *p_source,
103 uint8_t *p_uv = p_dest->U_PIXELS;
104 uint8_t *p_u = p_source->U_PIXELS;
105 uint8_t *p_v = p_source->V_PIXELS;
109 /* Copy the Y part */
110 p_vout->p_vlc->pf_memcpy( p_dest->Y_PIXELS, p_source->Y_PIXELS,
111 p_dest->p[Y_PLANE].i_pitch * p_dest->p[Y_PLANE].i_visible_lines );
113 /* Copy the U:V part */
114 for( i_x = p_dest->p[U_PLANE].i_pitch * p_dest->p[U_PLANE].i_visible_lines / 64;
117 #if defined (MODULE_NAME_IS_i420_ymga)
118 *p_uv++ = *p_u++; *p_uv++ = *p_v++; *p_uv++ = *p_u++; *p_uv++ = *p_v++;
119 *p_uv++ = *p_u++; *p_uv++ = *p_v++; *p_uv++ = *p_u++; *p_uv++ = *p_v++;
120 *p_uv++ = *p_u++; *p_uv++ = *p_v++; *p_uv++ = *p_u++; *p_uv++ = *p_v++;
121 *p_uv++ = *p_u++; *p_uv++ = *p_v++; *p_uv++ = *p_u++; *p_uv++ = *p_v++;
122 *p_uv++ = *p_u++; *p_uv++ = *p_v++; *p_uv++ = *p_u++; *p_uv++ = *p_v++;
123 *p_uv++ = *p_u++; *p_uv++ = *p_v++; *p_uv++ = *p_u++; *p_uv++ = *p_v++;
124 *p_uv++ = *p_u++; *p_uv++ = *p_v++; *p_uv++ = *p_u++; *p_uv++ = *p_v++;
125 *p_uv++ = *p_u++; *p_uv++ = *p_v++; *p_uv++ = *p_u++; *p_uv++ = *p_v++;
127 __asm__( ".p2align 5 \n\
128 movd (%0), %%mm0 # Load 4 Cr 00 00 00 00 v3 v2 v1 v0 \n\
129 movd 4(%0), %%mm2 # Load 4 Cr 00 00 00 00 v3 v2 v1 v0 \n\
130 movd 8(%0), %%mm4 # Load 4 Cr 00 00 00 00 v3 v2 v1 v0 \n\
131 movd 12(%0), %%mm6 # Load 4 Cr 00 00 00 00 v3 v2 v1 v0 \n\
132 movd (%1), %%mm1 # Load 4 Cb 00 00 00 00 u3 u2 u1 u0 \n\
133 movd 4(%1), %%mm3 # Load 4 Cb 00 00 00 00 u3 u2 u1 u0 \n\
134 movd 8(%1), %%mm5 # Load 4 Cb 00 00 00 00 u3 u2 u1 u0 \n\
135 movd 12(%1), %%mm7 # Load 4 Cb 00 00 00 00 u3 u2 u1 u0 \n\
136 punpcklbw %%mm1, %%mm0 # u3 v3 u2 v2 u1 v1 u0 v0 \n\
137 punpcklbw %%mm3, %%mm2 # u3 v3 u2 v2 u1 v1 u0 v0 \n\
138 punpcklbw %%mm5, %%mm4 # u3 v3 u2 v2 u1 v1 u0 v0 \n\
139 punpcklbw %%mm7, %%mm6 # u3 v3 u2 v2 u1 v1 u0 v0 \n\
140 movq %%mm0, (%2) # Store CrCb \n\
141 movq %%mm2, 8(%2) # Store CrCb \n\
142 movq %%mm4, 16(%2) # Store CrCb \n\
143 movq %%mm6, 24(%2) # Store CrCb"
144 : : "r" (p_v), "r" (p_u), "r" (p_uv) );
146 p_v += 16; p_u += 16; p_uv += 32;