put_bits: fix invalid shift by 32 in flush_put_bits()
authorMans Rullgard <mans@mansr.com>
Sat, 8 Oct 2011 01:09:42 +0000 (02:09 +0100)
committerMichael Niedermayer <michaelni@gmx.at>
Thu, 3 Nov 2011 23:59:53 +0000 (00:59 +0100)
If flush_put_bits() is called when the 32-bit buffer is empty,
e.g. after writing a multiple of 32 bits, and invalid shift by
32 is performed.  Since flush_put_bits() is called infrequently,
this additional check should have negligible performance impact.

Signed-off-by: Mans Rullgard <mans@mansr.com>
(cherry picked from commit ac6eab1496aad6f8b09deabbef4fe5fd829e142d)

Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
libavcodec/put_bits.h

index 7901691..ccd2565 100644 (file)
@@ -100,7 +100,8 @@ static inline void flush_put_bits(PutBitContext *s)
     align_put_bits(s);
 #else
 #ifndef BITSTREAM_WRITER_LE
-    s->bit_buf<<= s->bit_left;
+    if (s->bit_left < 32)
+        s->bit_buf<<= s->bit_left;
 #endif
     while (s->bit_left < 32) {
         /* XXX: should test end of buffer */