From 7eb5b20a87c1b786759de6b61cd6655edbb3e320 Mon Sep 17 00:00:00 2001
From: Christophe Gisquet <christophe.gisquet@gmail.com>
Date: Thu, 5 Jun 2014 23:37:00 +0200
Subject: [PATCH] huffyuvdec: trick for plane decoding <= 14bits

Refactor the code to minimize code duplication.

Before:
130870 decicycles in g, 1048139 runs, 437 skips
10bits:  9.048
12bits: 10.733

After:
126960 decicycles in g, 1048136 runs, 440 skips
10bits:  8.642
12bits:  9.656

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
---
 libavcodec/huffyuvdec.c | 38 ++++++++++++++++----------------------
 1 file changed, 16 insertions(+), 22 deletions(-)

diff --git a/libavcodec/huffyuvdec.c b/libavcodec/huffyuvdec.c
index 6c925d35f6..bc9d038dce 100644
--- a/libavcodec/huffyuvdec.c
+++ b/libavcodec/huffyuvdec.c
@@ -586,15 +586,14 @@ static av_cold int decode_init_thread_copy(AVCodecContext *avctx)
 
 
 #define GET_VLC_DUAL(dst0, dst1, name, gb, dtable, table1, table2,  \
-                     bits, max_depth)                               \
+                     bits, max_depth, OP)                           \
     do {                                                            \
         unsigned int index = SHOW_UBITS(name, gb, bits);            \
         int          code, n = dtable[index][1];                    \
                                                                     \
         if (n>0) {                                                  \
             code = dtable[index][0];                                \
-            dst0 = code>>8;                                         \
-            dst1 = code;                                            \
+            OP(dst0, dst1, code);                                   \
             LAST_SKIP_BITS(name, gb, n);                            \
         } else {                                                    \
             int nb_bits;                                            \
@@ -606,10 +605,12 @@ static av_cold int decode_init_thread_copy(AVCodecContext *avctx)
         }                                                           \
     } while (0)
 
+#define OP8bits(dst0, dst1, code) dst0 = code>>8; dst1 = code
+
 #define READ_2PIX(dst0, dst1, plane1)\
     UPDATE_CACHE(re, &s->gb); \
     GET_VLC_DUAL(dst0, dst1, re, &s->gb, s->vlc[4+plane1].table, \
-                 s->vlc[0].table, s->vlc[plane1].table, VLC_BITS, 3)
+                 s->vlc[0].table, s->vlc[plane1].table, VLC_BITS, 3, OP8bits)
 
 static void decode_422_bitstream(HYuvContext *s, int count)
 {
@@ -634,24 +635,15 @@ static void decode_422_bitstream(HYuvContext *s, int count)
     CLOSE_READER(re, &s->gb);
 }
 
-/* TODO instead of restarting the read when the code isn't in the first level
- * of the joint table, jump into the 2nd level of the individual table. */
-#define READ_2PIX_PLANE(dst0, dst1, plane) \
+#define READ_2PIX_PLANE(dst0, dst1, plane, OP) \
     UPDATE_CACHE(re, &s->gb); \
     GET_VLC_DUAL(dst0, dst1, re, &s->gb, s->vlc[4+plane].table, \
-                 s->vlc[plane].table, s->vlc[plane].table, VLC_BITS, 3)
+                 s->vlc[plane].table, s->vlc[plane].table, VLC_BITS, 3, OP)
 
-#define READ_2PIX_PLANE14(dst0, dst1, plane){\
-    int16_t code = get_vlc2(&s->gb, s->vlc[4+plane].table, VLC_BITS, 1);\
-    if(code != (int16_t)0xffff){\
-        dst0 = code>>8;\
-        dst1 = sign_extend(code, 8);\
-    }else{\
-        dst0 = get_vlc2(&s->gb, s->vlc[plane].table, VLC_BITS, 3);\
-        dst1 = get_vlc2(&s->gb, s->vlc[plane].table, VLC_BITS, 3);\
-    }\
-}
+#define OP14bits(dst0, dst1, code) dst0 = code>>8; dst1 = sign_extend(code, 8)
 
+/* TODO instead of restarting the read when the code isn't in the first level
+ * of the joint table, jump into the 2nd level of the individual table. */
 #define READ_2PIX_PLANE16(dst0, dst1, plane){\
     dst0 = get_vlc2(&s->gb, s->vlc[plane].table, VLC_BITS, 3)<<2;\
     dst0 += get_bits(&s->gb, 2);\
@@ -668,24 +660,26 @@ static void decode_plane_bitstream(HYuvContext *s, int count, int plane)
         OPEN_READER(re, &s->gb);
         if (count >= (get_bits_left(&s->gb)) / (31 * 2)) {
             for (i = 0; i < count && get_bits_left(&s->gb) > 0; i++) {
-                READ_2PIX_PLANE(s->temp[0][2 * i], s->temp[0][2 * i + 1], plane);
+                READ_2PIX_PLANE(s->temp[0][2 * i], s->temp[0][2 * i + 1], plane, OP8bits);
             }
         } else {
             for(i=0; i<count; i++){
-                READ_2PIX_PLANE(s->temp[0][2 * i], s->temp[0][2 * i + 1], plane);
+                READ_2PIX_PLANE(s->temp[0][2 * i], s->temp[0][2 * i + 1], plane, OP8bits);
             }
         }
         CLOSE_READER(re, &s->gb);
     } else if (s->bps <= 14) {
+        OPEN_READER(re, &s->gb);
         if (count >= (get_bits_left(&s->gb)) / (31 * 2)) {
             for (i = 0; i < count && get_bits_left(&s->gb) > 0; i++) {
-                READ_2PIX_PLANE14(s->temp16[0][2 * i], s->temp16[0][2 * i + 1], plane);
+                READ_2PIX_PLANE(s->temp16[0][2 * i], s->temp16[0][2 * i + 1], plane, OP14bits);
             }
         } else {
             for(i=0; i<count; i++){
-                READ_2PIX_PLANE14(s->temp16[0][2 * i], s->temp16[0][2 * i + 1], plane);
+                READ_2PIX_PLANE(s->temp16[0][2 * i], s->temp16[0][2 * i + 1], plane, OP14bits);
             }
         }
+        CLOSE_READER(re, &s->gb);
     } else {
         if (count >= (get_bits_left(&s->gb)) / (31 * 2)) {
             for (i = 0; i < count && get_bits_left(&s->gb) > 0; i++) {