+ deinterlace.c : added an Altivec version of Merge() - makes the

     filter more than 100% faster here (blend mode). Feel free to check
     if I haven't broken anything, first time doing Altivec ;)

diff --git a/configure.ac b/configure.ac
index 8f0a8c4534a85871dd42711b405a002c332639e4..7f1396f9b3a4d080555dfef22ea4935fec469cea 100644 (file)
--- a/configure.ac
+++ b/configure.ac
@@ -1,5 +1,5 @@
 dnl Autoconf settings for vlc
-dnl $Id: configure.ac,v 1.152 2004/01/22 01:20:39 jlj Exp $
+dnl $Id: configure.ac,v 1.153 2004/01/23 15:36:23 titer Exp $
 
 AC_INIT(vlc,0.7.1-cvs)
 
@@ -995,7 +995,7 @@ AC_CACHE_CHECK([if \$CC groks AltiVec C extensions],
      CFLAGS="${CFLAGS_save}"])
 if test "${ac_cv_c_altivec}" != "no"; then
   AC_DEFINE(CAN_COMPILE_C_ALTIVEC, 1, Define if your compiler groks C AltiVec extensions.)
-  AX_ADD_CFLAGS([vlc idctaltivec motionaltivec memcpyaltivec],[${ac_cv_c_altivec}])
+  AX_ADD_CFLAGS([vlc idctaltivec motionaltivec memcpyaltivec deinterlace],[${ac_cv_c_altivec}])
   ACCEL_MODULES="${ACCEL_MODULES} ${ALTIVEC_MODULES}"
 fi
 

diff --git a/modules/video_filter/deinterlace/deinterlace.c b/modules/video_filter/deinterlace/deinterlace.c
index 35c1e2291bdcb50cfdd26f6215a733990e4a7a2f..aa58eb00b54722ad4245f22782dfcf19d9e310c4 100644 (file)
--- a/modules/video_filter/deinterlace/deinterlace.c
+++ b/modules/video_filter/deinterlace/deinterlace.c
@@ -2,7 +2,7 @@
  * deinterlace.c : deinterlacer plugin for vlc
  *****************************************************************************
  * Copyright (C) 2000, 2001, 2002, 2003 VideoLAN
- * $Id: deinterlace.c,v 1.18 2003/12/22 14:32:56 sam Exp $
+ * $Id: deinterlace.c,v 1.19 2004/01/23 15:36:23 titer Exp $
  *
  * Author: Sam Hocevar <sam@zoy.org>
  *
@@ -55,7 +55,8 @@ static void RenderMean   ( vout_thread_t *, picture_t *, picture_t * );
 static void RenderBlend  ( vout_thread_t *, picture_t *, picture_t * );
 static void RenderLinear ( vout_thread_t *, picture_t *, picture_t *, int );
 
-static void Merge        ( void *, const void *, const void *, size_t );
+static void MergeGeneric ( void *, const void *, const void *, size_t );
+static void MergeAltivec ( void *, const void *, const void *, size_t );
 
 static int  SendEvents   ( vlc_object_t *, char const *,
                            vlc_value_t, vlc_value_t, void * );
@@ -107,6 +108,8 @@ struct vout_sys_t
     vout_thread_t *p_vout;
 
     vlc_mutex_t filter_lock;
+
+    void (*pf_merge) ( void *, const void *, const void *, size_t );
 };
 
 /*****************************************************************************
@@ -138,6 +141,15 @@ static int Create( vlc_object_t *p_this )
     p_vout->p_sys->last_date = 0;
     vlc_mutex_init( p_vout, &p_vout->p_sys->filter_lock );
 
+    if( p_vout->p_libvlc->i_cpu & CPU_CAPABILITY_ALTIVEC )
+    {
+        p_vout->p_sys->pf_merge = MergeAltivec;
+    }
+    else
+    {
+        p_vout->p_sys->pf_merge = MergeGeneric;
+    }
+
     /* Look what method was requested */
     var_Create( p_vout, "deinterlace-mode", VLC_VAR_STRING );
     var_Change( p_vout, "deinterlace-mode", VLC_VAR_INHERITVALUE, &val, NULL );
@@ -617,6 +629,8 @@ static void RenderBob( vout_thread_t *p_vout,
     }
 }
 
+#define Merge p_vout->p_sys->pf_merge
+
 /*****************************************************************************
  * RenderLinear: BOB with linear interpolation
  *****************************************************************************/
@@ -774,8 +788,10 @@ static void RenderBlend( vout_thread_t *p_vout,
     }
 }
 
-static void Merge( void *_p_dest, const void *_p_s1,
-                   const void *_p_s2, size_t i_bytes )
+#undef Merge
+
+static void MergeGeneric( void *_p_dest, const void *_p_s1,
+                          const void *_p_s2, size_t i_bytes )
 {
     uint8_t* p_dest = (uint8_t*)_p_dest;
     const uint8_t *p_s1 = (const uint8_t *)_p_s1;
@@ -802,6 +818,41 @@ static void Merge( void *_p_dest, const void *_p_s1,
     }
 }
 
+static void MergeAltivec( void *_p_dest, const void *_p_s1,
+                          const void *_p_s2, size_t i_bytes )
+{
+#ifdef CAN_COMPILE_C_ALTIVEC
+    uint8_t *p_dest = (uint8_t*)_p_dest;
+    const uint8_t *p_s1 = (const uint8_t *)_p_s1;
+    const uint8_t *p_s2 = (const uint8_t *)_p_s2;
+    uint8_t *p_end = p_dest + i_bytes - 16;
+
+    if( ( (int)p_s1 & 0xF ) | ( (int)p_s2 & 0xF ) |
+        ( (int)p_dest & 0xF ) )
+    {
+        /* TODO Handle non 16-bytes aligned planes */
+        MergeGeneric( _p_dest, _p_s1, _p_s2, i_bytes );
+        return;
+    }
+
+    while( p_dest < p_end )
+    {
+        vec_st( vec_avg( vec_ld( 0, p_s1 ), vec_ld( 0, p_s2 ) ),
+                0, p_dest );
+        p_s1   += 16;
+        p_s2   += 16;
+        p_dest += 16;
+    }
+
+    p_end += 16;
+
+    while( p_dest < p_end )
+    {
+        *p_dest++ = ( (uint16_t)(*p_s1++) + (uint16_t)(*p_s2++) ) >> 1;
+    }
+#endif
+}
+
 /*****************************************************************************
  * SendEvents: forward mouse and keyboard events to the parent p_vout
  *****************************************************************************/