[FFmpeg-devel] [PATCH 1/5] avcodec/ppc/hpeldsp_altivec: Fix left-shift of negative number
Andreas Rheinhardt
andreas.rheinhardt at outlook.com
Mon Apr 1 21:41:00 EEST 2024
Andreas Rheinhardt:
> It is UB and affected e.g. the vp5 and vp61 FATE tests:
> https://fate.ffmpeg.org/report.cgi?time=20240327083327&slot=ppc-linux-gcc-13.2-ubsan-altivec-qemu
>
> Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt at outlook.com>
> ---
> libavcodec/ppc/hpeldsp_altivec.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/libavcodec/ppc/hpeldsp_altivec.c b/libavcodec/ppc/hpeldsp_altivec.c
> index a531b6b6ec..4bf6b28ed6 100644
> --- a/libavcodec/ppc/hpeldsp_altivec.c
> +++ b/libavcodec/ppc/hpeldsp_altivec.c
> @@ -41,9 +41,9 @@ void ff_put_pixels16_altivec(uint8_t *block, const uint8_t *pixels, ptrdiff_t li
> register vector unsigned char pixelsv1D;
>
> int i;
> - register ptrdiff_t line_size_2 = line_size << 1;
> + register ptrdiff_t line_size_2 = line_size * (1 << 1);
> register ptrdiff_t line_size_3 = line_size + line_size_2;
> - register ptrdiff_t line_size_4 = line_size << 2;
> + register ptrdiff_t line_size_4 = line_size * (1 << 2);
>
> // hand-unrolling the loop by 4 gains about 15%
> // mininum execution time goes from 74 to 60 cycles
Will apply the remainder of this patchset tomorrow unless there are
objections.
- Andreas
More information about the ffmpeg-devel
mailing list