This patch series draws upon remarks of Mark and Ronald regarding the lack of safety of the common idiom return a - b for qsort comparators.
I made an observation that the interesting (x > y) - (x < y) idiom works well: it not only avoids branches commonly not optimized by compilers when the ternary operator method is used, but is also safe with respect to overflow. See e.g the link: https://stackoverflow.com/questions/10996418/efficient-integer-compare-function/10997428#10997428 On a suggestion from Nicolas, slightly modified by me, 1/2 adds FFDIFFSIGN macro to avutil/common, and documents it and its rationale. Note that is quite generic, and in particular works for all of the standard built in integral and numerical types. NaN's will compare equal with everything, which seems reasonable and matches C++'s std::sort AFAIK. 2/2 utilizes FFDIFFSIGN in comparators across the codebase. A few were left out due to reasons mentioned in a comment for 2/2. Ganesh Ajjanagadde (2): avutil/common: add FFDIFFSIGN macro all: use FFDIFFSIGN to resolve possible undefined behavior in comparators cmdutils.c | 2 +- cmdutils_opencl.c | 2 +- ffmpeg.c | 3 +-- libavcodec/aacsbr_template.c | 2 +- libavcodec/motion_est.c | 2 +- libavfilter/f_sendcmd.c | 8 +++----- libavfilter/vf_deshake.c | 3 +-- libavfilter/vf_palettegen.c | 2 +- libavfilter/vf_removegrain.c | 5 +---- libavformat/subtitles.c | 9 +++------ libavutil/common.h | 11 +++++++++++ libswresample/swresample-test.c | 2 +- 12 files changed, 26 insertions(+), 25 deletions(-) -- 2.6.2 _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
