Fix float-cast-overflow in x264_ratecontrol_end function
authorAlexey Samsonov <vonosmas@gmail.com>
Tue, 26 Jan 2016 00:05:25 +0000 (16:05 -0800)
committerHenrik Gramner <henrik@gramner.com>
Fri, 5 Feb 2016 17:48:13 +0000 (18:48 +0100)
According to the C standard, it is undefined behavior to cast a negative
floating point number to an unsigned integer. Float-cast-overflow in
general is known to produce different results on different architectures.

Building x264 code with Clang and -fsanitize=float-cast-overflow
(http://clang.llvm.org/docs/UndefinedBehaviorSanitizer.html#availablle-checks)
and running it on some real-life examples occasionally produces errors
of the form:

encoder/ratecontrol.c:1892: runtime error: value -5011.14 is outside the
range of representable values of type 'unsigned short'

Fix these errors by explicitly coding the de-facto x86 behavior: casting
float to uint16_t through int16_t.

encoder/ratecontrol.c

index d8d06b3..0c6913a 100644 (file)
@@ -1889,7 +1889,7 @@ int x264_ratecontrol_end( x264_t *h, int bits, int *filler )
             uint8_t i_type = h->sh.i_type;
             /* Values are stored as big-endian FIX8.8 */
             for( int i = 0; i < h->mb.i_mb_count; i++ )
-                rc->mbtree.qp_buffer[0][i] = endian_fix16( h->fenc->f_qp_offset[i]*256.0 );
+                rc->mbtree.qp_buffer[0][i] = endian_fix16( (int16_t)(h->fenc->f_qp_offset[i]*256.0) );
             if( fwrite( &i_type, 1, 1, rc->p_mbtree_stat_file_out ) < 1 )
                 goto fail;
             if( fwrite( rc->mbtree.qp_buffer[0], sizeof(uint16_t), h->mb.i_mb_count, rc->p_mbtree_stat_file_out ) < h->mb.i_mb_count )