libtremor: use render_line from ffmpeg (libavcodec/vorbis.c), speedup ~1% on both arm and coldfire, output unchanged.

git-svn-id: svn://svn.rockbox.org/rockbox/trunk@27857 a1c6a512-1295-4272-9138-f99709370657

1 files changed, 49 insertions, 26 deletions

diff --git a/apps/codecs/libtremor/floor1.c b/apps/codecs/libtremor/floor1.c
index 138f718651..21bae2820b 100644
--- a/apps/codecs/libtremor/floor1.c
+++ b/apps/codecs/libtremor/floor1.c
@@ -274,32 +274,55 @@ static const ogg_int32_t FLOOR_fromdB_LOOKUP[256] ICONST_ATTR = {
   XdB(0x69f80e9a), XdB(0x70dafda8), XdB(0x78307d76), XdB(0x7fffffff),
 };
 
-static void render_line(int n, int x0,register int x1,int y0,int y1,ogg_int32_t *d){
-  int dy=y1-y0;
-  register int x=x0;
-  register int y=y0;
-  register int adx=x1-x0;
-  register int ady=abs(dy);
-  register int base=dy/adx;
-  register int sy=(dy<0?base-1:base+1);
-  int err=0;
-
-  if(n>x1)n=x1;
-  ady-=abs(base*adx);
-
-  if(LIKELY(x<n))
-    d[x]= MULT31_SHIFT15(d[x],FLOOR_fromdB_LOOKUP[y]);
-
-  while(++x<n){
-    err=err+ady;
-    if(err>=adx){
-      err-=adx;
-      y+=sy;
-    }else{
-      y+=base;
+static inline void render_line_unrolled(int x, int y, int x1,
+                                        int sy, int ady, int adx,
+                                        ogg_int32_t *buf)
+{
+    int err = -adx;
+    x -= x1 - 1;
+    buf += x1 - 1;
+    while (++x < 0) {
+        err += ady;
+        if (err >= 0) {
+            err += ady - adx;
+            y   += sy;
+            buf[x] = MULT31_SHIFT15(buf[x],FLOOR_fromdB_LOOKUP[y]);
+            x++;
+        }
+        buf[x] = MULT31_SHIFT15(buf[x],FLOOR_fromdB_LOOKUP[y]);
+    }
+    if (x <= 0) {
+        if (err + ady >= 0)
+            y += sy;
+        buf[x] = MULT31_SHIFT15(buf[x],FLOOR_fromdB_LOOKUP[y]);
+    }
+}
+
+static void render_line(int x0, int y0, int x1, int y1, ogg_int32_t *buf)
+{
+    int dy  = y1 - y0;
+    int adx = x1 - x0;
+    int ady = abs(dy);
+    int sy  = dy < 0 ? -1 : 1;
+    buf[x0] = MULT31_SHIFT15(buf[x0],FLOOR_fromdB_LOOKUP[y0]);
+    if (ady*2 <= adx) { // optimized common case
+        render_line_unrolled(x0, y0, x1, sy, ady, adx, buf);
+    } else {
+        int base = dy / adx;
+        int x    = x0;
+        int y    = y0;
+        int err  = -adx;
+        ady -= abs(base) * adx;
+        while (++x < x1) {
+            y += base;
+            err += ady;
+            if (err >= 0) {
+                err -= adx;
+                y   += sy;
+            }
+            buf[x] = MULT31_SHIFT15(buf[x],FLOOR_fromdB_LOOKUP[y]);
+        }
     }
-    d[x]= MULT31_SHIFT15(d[x],FLOOR_fromdB_LOOKUP[y]);
-  }
 }
 
 static void *floor1_inverse1(vorbis_block *vb,vorbis_look_floor *in)
@@ -414,7 +437,7 @@ static int floor1_inverse2(vorbis_block *vb,vorbis_look_floor *in,void *memo,
         hy*=info->mult;
         hx=info->postlist[current];
         
-        render_line(n,lx,hx,ly,hy,out);
+        render_line(lx, ly, hx, hy, out);
         
         lx=hx;
         ly=hy;