Merge svn changes up to r30419

5 files changed, 88 insertions, 22 deletions

diff --git a/libmpcodecs/vd.c b/libmpcodecs/vd.c
index 1910af62ee..1821498865 100644
--- a/libmpcodecs/vd.c
+++ b/libmpcodecs/vd.c
@@ -189,7 +189,7 @@ int mpcodecs_config_vo(sh_video_t *sh, int w, int h,
         } else {
             // sws failed, if the last filter (vf_vo) support MPEGPES try
             // to append vf_lavc
-            vf_instance_t *vo, *vp = NULL, *ve;
+            vf_instance_t *vo, *vp = NULL, *ve, *vpp = NULL;
             // Remove the scale filter if we added it ourselves
             if (vf == sc) {
                 ve = vf;
@@ -197,8 +197,10 @@ int mpcodecs_config_vo(sh_video_t *sh, int w, int h,
                 vf_uninit_filter(ve);
             }
             // Find the last filter (vf_vo)
-            for (vo = vf; vo->next; vo = vo->next)
+            for (vo = vf; vo->next; vo = vo->next) {
+                vpp = vp;
                 vp = vo;
+            }
             if (vo->query_format(vo, IMGFMT_MPEGPES)
                 && (!vp || (vp && strcmp(vp->info->name, "lavc")))) {
                 ve = vf_open_filter(opts, vo, "lavc", NULL);
@@ -208,6 +210,13 @@ int mpcodecs_config_vo(sh_video_t *sh, int w, int h,
                     vf = ve;
                 goto csp_again;
             }
+            if (vp && !strcmp(vp->info->name,"lavc")) {
+		if (vpp)
+                    vpp->next = vo;
+		else
+                    vf = vo;
+		vf_uninit_filter(vp);
+            }
         }
         mp_tmsg(MSGT_CPLAYER, MSGL_WARN,
             "The selected video_out device is incompatible with this codec.\n"\
diff --git a/libmpcodecs/vf_fspp.c b/libmpcodecs/vf_fspp.c
index 7264199c21..8ef33381b1 100644
--- a/libmpcodecs/vf_fspp.c
+++ b/libmpcodecs/vf_fspp.c
@@ -533,9 +533,15 @@ static int put_image(struct vf_instance* vf, mp_image_t *mpi, double pts)
 
     vf->priv->mpeg2= mpi->qscale_type;
     if(mpi->pict_type != 3 && mpi->qscale && !vf->priv->qp){
-	if(!vf->priv->non_b_qp)
-	    vf->priv->non_b_qp= malloc(mpi->qstride * ((mpi->h + 15) >> 4));
-	fast_memcpy(vf->priv->non_b_qp, mpi->qscale, mpi->qstride * ((mpi->h + 15) >> 4));
+        int w = mpi->qstride;
+        int h = (mpi->h + 15) >> 4;
+        if (!w) {
+            w = (mpi->w + 15) >> 4;
+            h = 1;
+        }
+        if(!vf->priv->non_b_qp)
+            vf->priv->non_b_qp= malloc(w*h);
+        fast_memcpy(vf->priv->non_b_qp, mpi->qscale, w*h);
     }
     if(vf->priv->log2_count || !(mpi->flags&MP_IMGFLAG_DIRECT)){
 	char *qp_tab= vf->priv->non_b_qp;
diff --git a/libmpcodecs/vf_halfpack.c b/libmpcodecs/vf_halfpack.c
index 682b80a91e..4a3bdb578f 100644
--- a/libmpcodecs/vf_halfpack.c
+++ b/libmpcodecs/vf_halfpack.c
@@ -10,11 +10,14 @@
 #include "img_format.h"
 #include "mp_image.h"
 #include "vf.h"
+#include "vf_scale.h"
 
-#include "libswscale/rgb2rgb.h"
+#include "libswscale/swscale.h"
+#include "fmt-conversion.h"
 
 struct vf_priv_s {
 	int field;
+	struct SwsContext *ctx;
 };
 
 #if HAVE_MMX
@@ -144,6 +147,10 @@ static void (*halfpack)(unsigned char *dst, unsigned char *src[3],
 
 static int put_image(struct vf_instance* vf, mp_image_t *mpi, double pts)
 {
+	const uint8_t *src[MP_MAX_PLANES] = {
+		mpi->planes[0] + mpi->stride[0]*vf->priv->field,
+		mpi->planes[1], mpi->planes[2], NULL};
+	int src_stride[MP_MAX_PLANES] = {mpi->stride[0]*2, mpi->stride[1], mpi->stride[2], 0};
 	mp_image_t *dmpi;
 
 	// hope we'll get DR buffer:
@@ -154,9 +161,8 @@ static int put_image(struct vf_instance* vf, mp_image_t *mpi, double pts)
 	switch(vf->priv->field) {
 	case 0:
 	case 1:
-		yuv422ptoyuy2(mpi->planes[0] + mpi->stride[0]*vf->priv->field,
-			mpi->planes[1], mpi->planes[2], dmpi->planes[0],
-			mpi->w, mpi->h/2, mpi->stride[0]*2, mpi->stride[1], dmpi->stride[0]);
+		sws_scale(vf->priv->ctx, src, src_stride,
+		          0, mpi->h/2, dmpi->planes, dmpi->stride);
 		break;
 	default:
 		halfpack(dmpi->planes[0], mpi->planes, dmpi->stride[0],
@@ -170,6 +176,15 @@ static int config(struct vf_instance* vf,
 		  int width, int height, int d_width, int d_height,
 		  unsigned int flags, unsigned int outfmt)
 {
+	if (vf->priv->field < 2) {
+		sws_freeContext(vf->priv->ctx);
+		// get unscaled 422p -> yuy2 conversion
+		vf->priv->ctx =
+			sws_getContext(width, height / 2, PIX_FMT_YUV422P,
+			               width, height / 2, PIX_FMT_YUYV422,
+			               SWS_POINT | SWS_PRINT_INFO | get_sws_cpuflags(),
+			               NULL, NULL, NULL);
+	}
 	/* FIXME - also support UYVY output? */
 	return vf_next_config(vf, width, height/2, d_width, d_height, flags, IMGFMT_YUY2);
 }
@@ -189,6 +204,7 @@ static int query_format(struct vf_instance* vf, unsigned int fmt)
 
 static void uninit(struct vf_instance* vf)
 {
+	sws_freeContext(vf->priv->ctx);
 	free(vf->priv);
 }
 
diff --git a/libmpcodecs/vf_scale.c b/libmpcodecs/vf_scale.c
index a38cf895d9..82a56535bb 100644
--- a/libmpcodecs/vf_scale.c
+++ b/libmpcodecs/vf_scale.c
@@ -31,7 +31,6 @@ static struct vf_priv_s {
     int interlaced;
     int noup;
     int accurate_rnd;
-    int query_format_cache[64];
 } const vf_priv_dflt = {
   -1,-1,
   0,
@@ -92,18 +91,48 @@ static const unsigned int outfmt_list[]={
     0
 };
 
-static unsigned int find_best_out(vf_instance_t *vf){
+/**
+ * A list of preferred conversions, in order of preference.
+ * This should be used for conversions that e.g. involve no scaling
+ * or to stop vf_scale from choosing a conversion that has no
+ * fast assembler implementation.
+ */
+static int preferred_conversions[][2] = {
+    {IMGFMT_YUY2, IMGFMT_UYVY},
+    {IMGFMT_YUY2, IMGFMT_422P},
+    {IMGFMT_UYVY, IMGFMT_YUY2},
+    {IMGFMT_UYVY, IMGFMT_422P},
+    {IMGFMT_422P, IMGFMT_YUY2},
+    {IMGFMT_422P, IMGFMT_UYVY},
+    {0, 0}
+};
+
+static unsigned int find_best_out(vf_instance_t *vf, int in_format){
     unsigned int best=0;
-    int i;
+    int i = -1;
+    int j = -1;
+    int format = 0;
 
     // find the best outfmt:
-    for(i=0; i<sizeof(outfmt_list)/sizeof(int)-1; i++){
-        const int format= outfmt_list[i];
-        int ret= vf->priv->query_format_cache[i]-1;
-        if(ret == -1){
-            ret= vf_next_query_format(vf, outfmt_list[i]);
-            vf->priv->query_format_cache[i]= ret+1;
+    while (1) {
+        int ret;
+        if (j < 0) {
+            format = in_format;
+            j = 0;
+        } else if (i < 0) {
+            while (preferred_conversions[j][0] &&
+                   preferred_conversions[j][0] != in_format)
+                j++;
+            format = preferred_conversions[j++][1];
+            // switch to standard list
+            if (!format)
+                i = 0;
         }
+        if (i >= 0)
+            format = outfmt_list[i++];
+        if (!format)
+            break;
+        ret = vf_next_query_format(vf, format);
 
 	mp_msg(MSGT_VFILTER,MSGL_DBG2,"scale: query(%s) -> %d\n",vo_format_name(format),ret&3);
 	if(ret&VFCAP_CSP_SUPPORTED_BY_HW){
@@ -120,7 +149,7 @@ static int config(struct vf_instance* vf,
         int width, int height, int d_width, int d_height,
 	unsigned int flags, unsigned int outfmt){
     struct MPOpts *opts = vf->opts;
-    unsigned int best=find_best_out(vf);
+    unsigned int best=find_best_out(vf, outfmt);
     int vo_flags;
     int int_sws_flags=0;
     int round_w=0, round_h=0;
@@ -494,7 +523,7 @@ static int query_format(struct vf_instance* vf, unsigned int fmt){
     case IMGFMT_RGB48LE:
     case IMGFMT_RGB48BE:
     {
-	unsigned int best=find_best_out(vf);
+	unsigned int best=find_best_out(vf, fmt);
 	int flags;
 	if(!best) return 0;	 // no matching out-fmt
 	flags=vf_next_query_format(vf,best);
diff --git a/libmpcodecs/vf_spp.c b/libmpcodecs/vf_spp.c
index 85d7ff9fc1..8dfcf41b5c 100644
--- a/libmpcodecs/vf_spp.c
+++ b/libmpcodecs/vf_spp.c
@@ -475,9 +475,15 @@ static int put_image(struct vf_instance* vf, mp_image_t *mpi, double pts){
 
         vf->priv->mpeg2= mpi->qscale_type;
         if(mpi->pict_type != 3 && mpi->qscale && !vf->priv->qp){
+            int w = mpi->qstride;
+            int h = (mpi->h + 15) >> 4;
+            if (!w) {
+                w = (mpi->w + 15) >> 4;
+                h = 1;
+            }
             if(!vf->priv->non_b_qp)
-                vf->priv->non_b_qp= malloc(mpi->qstride * ((mpi->h + 15) >> 4));
-            fast_memcpy(vf->priv->non_b_qp, mpi->qscale, mpi->qstride * ((mpi->h + 15) >> 4));
+                vf->priv->non_b_qp= malloc(w*h);
+            fast_memcpy(vf->priv->non_b_qp, mpi->qscale, w*h);
         }
 	if(vf->priv->log2_count || !(mpi->flags&MP_IMGFLAG_DIRECT)){
             char *qp_tab= vf->priv->non_b_qp;