From ca47f85bf3274fc970650763398795fc5e8e666c Mon Sep 17 00:00:00 2001
From: deva <deva>
Date: Mon, 9 May 2005 16:40:20 +0000
Subject: Added optimize yuv conversion code

---
 src/mov_encoder.cc   | 102 +++++++++++++++++++++++++++++++++++----------------
 src/mov_encoder.h    |   6 +++
 src/server.cc        |  14 ++++---
 src/server_status.cc |   6 ++-
 4 files changed, 90 insertions(+), 38 deletions(-)

diff --git a/src/mov_encoder.cc b/src/mov_encoder.cc
index f9d47e7..0e2b935 100644
--- a/src/mov_encoder.cc
+++ b/src/mov_encoder.cc
@@ -39,6 +39,10 @@
 
 /*
  * $Log$
+ * Revision 1.15  2005/05/09 16:40:20  deva
+ *
+ * Added optimize yuv conversion code
+ *
  * Revision 1.14  2005/05/07 10:56:18  deva
  * Changed print outs
  *
@@ -80,9 +84,9 @@ MovEncoder::MovEncoder(const char *filename)
   yuv.w = w;
   yuv.h = h;
   yuv.p = w;
-  yuv.y = new unsigned char [w*h];
-  yuv.u = new unsigned char [w*h/4];
-  yuv.v = new unsigned char [w*h/4];
+  yuv.y = new unsigned char [w*h * 2];
+  yuv.u = new unsigned char [w*h];// [w*h/4]
+  yuv.v = new unsigned char [w*h];// [w*h/4]
 
   ////////////LIBDV STUFF///////////////
   
@@ -132,7 +136,7 @@ MovEncoder::MovEncoder(const char *filename)
   fame_par.coding = coding;
 
   // quality is a percentage, which controls compression versus quality.
-  fame_par.quality = 100;
+  fame_par.quality = 100; // FIXME: This should be in config file!
 
   // slices_per_frame is the number of frame slices per frame. More slices provide 
   // better error recovery. There must be at least one slice per frame, and at most 
@@ -150,7 +154,7 @@ MovEncoder::MovEncoder(const char *filename)
 
   // shape_quality is percentage determing the average binary shape accuracy in 
   // video with arbitrary shape.
-  fame_par.shape_quality = 100;
+  fame_par.shape_quality = 1;
 
   // search_range specifies the motion estimation search range in pixel unit. 
   // Small search ranges work best with slow motion videos, whereas larger search 
@@ -159,7 +163,7 @@ MovEncoder::MovEncoder(const char *filename)
 
   // verbose when set to 1 outputs information on copyright, modules used and 
   // current frame on standard error.
-  fame_par.verbose = 1;
+  fame_par.verbose = 0;
 
   fame_init(fame_context, &fame_par, fame_buffer, FAME_BUFFER_SIZE);
   // FIXME: fame_init return a new context, or NULL if an error occurred.
@@ -182,7 +186,9 @@ void MovEncoder::encode(Frame *dvframe)
   encode_audio(dvframe);
 }
 
+//#define RGB
 
+#ifdef RGB
 #define _CLAMP(x) (unsigned char) ((x)<0?0:(((x)>255)?255:(x)))
 
 inline void rgb_to_yuv(unsigned char *rgb, unsigned char &y, unsigned char &u, unsigned char &v)
@@ -191,17 +197,18 @@ inline void rgb_to_yuv(unsigned char *rgb, unsigned char &y, unsigned char &u, u
   v = _CLAMP( 0.439 * rgb[0] - 0.368 * rgb[1] - 0.071 * rgb[2] + 128);
   u = _CLAMP(-0.148 * rgb[0] - 0.291 * rgb[1] + 0.439 * rgb[2] + 128);
 }
+#endif
 
 void MovEncoder::encode_video(Frame *dvframe)
 {
   // Decode DV Frame to YUV
   int w = 720;
   int h = 576;
-  unsigned char rgb[720*576*4];
 
   unsigned char *pixels[3];
   int pitches[3];
 
+#ifdef RGB
   pixels[ 0 ] = rgb;
   pixels[ 1 ] = NULL;
   pixels[ 2 ] = NULL;
@@ -209,6 +216,7 @@ void MovEncoder::encode_video(Frame *dvframe)
   pitches[ 0 ] = 720 * 3;
   pitches[ 1 ] = 0;
   pitches[ 2 ] = 0;
+#endif
 
   if(!dvdecoder) {
     dvdecoder = dv_decoder_new(FALSE/*this value is unused*/, FALSE, FALSE);
@@ -223,39 +231,69 @@ void MovEncoder::encode_video(Frame *dvframe)
     dvdecoder->num_dif_seqs = 12;
   }
 
+#ifdef RGB
   dv_decode_full_frame(dvdecoder, 
                        dvframe->data, 
                        e_dv_color_rgb,
                        pixels,
                        pitches);
 
-  /*
-                       e_dv_color_yuv,
-                       (uint8_t**) &yuv.y, //pixels,
-                       (int*) &yuv.w);//pitches);
-  */  
   // cvt rgb to yuv
-  for (int y=0; y<h; y+=2)
-    {
-      for (int x=0; x<w; x+=2)
-        {
-          unsigned char vy[4],vu[4],vv[4];
-          
-          rgb_to_yuv(rgb + 3*((y+0)*w + (x+0)), vy[0],vu[0],vv[0]);
-          rgb_to_yuv(rgb + 3*((y+0)*w + (x+1)), vy[1],vu[1],vv[1]);
-          rgb_to_yuv(rgb + 3*((y+1)*w + (x+0)), vy[2],vu[2],vv[2]);
-          rgb_to_yuv(rgb + 3*((y+1)*w + (x+1)), vy[3],vu[3],vv[3]);
-          // Y
-          yuv.y[(y+0)*w+(x+0)] = vy[0];
-          yuv.y[(y+0)*w+(x+1)] = vy[1];
-          yuv.y[(y+1)*w+(x+0)] = vy[2];
-          yuv.y[(y+1)*w+(x+1)] = vy[3];
-          // Cb
-          yuv.u[y*w/4+x/2] = (vu[0]+vu[1]+vu[2]+vu[3])/4;
-          // Cr
-          yuv.v[y*w/4+x/2] = (vv[0]+vv[1]+vv[2]+vv[3])/4;
-        }
+  for (int y=0; y<h; y+=2) {
+    for (int x=0; x<w; x+=2) {
+      unsigned char vy[4],vu[4],vv[4];
+      
+      rgb_to_yuv(rgb + 3*((y+0)*w + (x+0)), vy[0],vu[0],vv[0]);
+      rgb_to_yuv(rgb + 3*((y+0)*w + (x+1)), vy[1],vu[1],vv[1]);
+      rgb_to_yuv(rgb + 3*((y+1)*w + (x+0)), vy[2],vu[2],vv[2]);
+      rgb_to_yuv(rgb + 3*((y+1)*w + (x+1)), vy[3],vu[3],vv[3]);
+      // Y
+      yuv.y[(y+0)*w+(x+0)] = vy[0];
+      yuv.y[(y+0)*w+(x+1)] = vy[1];
+      yuv.y[(y+1)*w+(x+0)] = vy[2];
+      yuv.y[(y+1)*w+(x+1)] = vy[3];
+      // Cb
+      yuv.u[y*w/4+x/2] = (vu[0]+vu[1]+vu[2]+vu[3])/4;
+      // Cr
+      yuv.v[y*w/4+x/2] = (vv[0]+vv[1]+vv[2]+vv[3])/4;
     }
+  }
+#else
+  pixels[ 0 ] = rgb; // We use this as the output buffer
+  pitches[ 0 ] = w * 2;
+ 
+  dv_decode_full_frame(dvdecoder, 
+                       dvframe->data, 
+                       e_dv_color_yuv,
+                       pixels,
+                       pitches);
+
+  int w2 = w / 2;
+  uint8_t *y = yuv.y;
+  uint8_t *cb = yuv.u;
+  uint8_t *cr = yuv.v;
+  uint8_t *p = rgb;
+ 
+  for ( int i = 0; i < h; i += 2 ) {
+    // process two scanlines (one from each field, interleaved)
+    for ( int j = 0; j < w2; j++ ) {
+      // packed YUV 422 is: Y[i] U[i] Y[i+1] V[i]
+      *( y++ ) = *( p++ );
+      *( cb++ ) = *( p++ );
+      *( y++ ) = *( p++ );
+      *( cr++ ) = *( p++ );
+    }
+
+    // process next two scanlines (one from each field, interleaved)
+    for ( int j = 0; j < w2; j++ ) {
+      // skip every second line for U and V
+      *( y++ ) = *( p++ );
+      p++;
+      *( y++ ) = *( p++ );
+      p++;
+    }
+  }
+#endif
 
   // Encode YUV frame and write it to disk.
   fame_start_frame(fame_context, &yuv, 0);
diff --git a/src/mov_encoder.h b/src/mov_encoder.h
index 17412a2..457b577 100644
--- a/src/mov_encoder.h
+++ b/src/mov_encoder.h
@@ -36,6 +36,10 @@
 
 /*
  * $Log$
+ * Revision 1.7  2005/05/09 16:40:20  deva
+ *
+ * Added optimize yuv conversion code
+ *
  * Revision 1.6  2005/05/05 20:41:38  deva
  *
  * Removed the last pieces of ffmpeg... replaced it with libfame...
@@ -89,6 +93,8 @@ class MovEncoder {
 
   // libdv decoder
  	dv_decoder_t *dvdecoder;
+
+  unsigned char rgb[720*576*4];
 };
 
 #endif
diff --git a/src/server.cc b/src/server.cc
index 7d6127d..76ba8c9 100644
--- a/src/server.cc
+++ b/src/server.cc
@@ -31,6 +31,10 @@
 
 /*
  * $Log$
+ * Revision 1.13  2005/05/09 16:40:20  deva
+ *
+ * Added optimize yuv conversion code
+ *
  * Revision 1.12  2005/05/07 10:56:18  deva
  * Changed print outs
  *
@@ -294,11 +298,11 @@ void newConnection(Socket *socket)
     
     //    printf("Read: %d bytes ", ret);
     //    printf("typ: %d ", h.header_type);
-    fprintf(stdout, "cpr: %s ", cpr);
-    fprintf(stdout, "frz: %d ", h.header.h_data.freeze);
-    fprintf(stdout, "sht: %d ", h.header.h_data.snapshot);
-    fprintf(stdout, "save: %d ", h.header.h_data.savestate);
-    fflush(stdout);
+//    fprintf(stdout, "cpr: %s ", cpr);
+//    fprintf(stdout, "frz: %d ", h.header.h_data.freeze);
+//    fprintf(stdout, "sht: %d ", h.header.h_data.snapshot);
+//    fprintf(stdout, "save: %d ", h.header.h_data.savestate);
+//    fflush(stdout);
     
     if(h.header.h_data.snapshot) {
       if(freeze_frame) {
diff --git a/src/server_status.cc b/src/server_status.cc
index 66cc34c..1e58ee7 100644
--- a/src/server_status.cc
+++ b/src/server_status.cc
@@ -31,6 +31,10 @@
 
 /*
  * $Log$
+ * Revision 1.4  2005/05/09 16:40:20  deva
+ *
+ * Added optimize yuv conversion code
+ *
  * Revision 1.3  2005/05/03 08:31:59  deva
  * Removed the error object, and replaced it with a more generic info object.
  *
@@ -72,7 +76,7 @@ void ServerStatus::checkPoint()
   for(int cnt = 0; cnt < BUFFERSIZE; cnt++) {
     total += (double)frametime[cnt];
   }
-  fprintf(stderr, "[ms: %d, fps: %f]\t", frametime[0], 1000000.0 / (total / (double)BUFFERSIZE) );
+  fprintf(stderr, "[ms: %d, fps: %f]\n", frametime[0], 1000000.0 / (total / (double)BUFFERSIZE) );
   
 }
 
-- 
cgit v1.2.3