[FFmpeg-devel] [PATCH] avutil/mem: limit alignment to maximum simg align

[Timo Rothenpieler]

On 13.01.2024 01:57, Timo Rothenpieler wrote:
> FFmpeg has instances of DECLARE_ALIGNED(32, ...) in a lot of structs,
> which then end up heap-allocated.
> By declaring any variable in a struct, or tree of structs, to be 32 byte
> aligned, it allows the compiler to safely assume the entire struct
> itself is also 32 byte aligned.
> 
> This might make the compiler emit code which straight up crashes or
> misbehaves in other ways, and at least in one instances is now
> documented to actually do (see ticket 10549 on trac).
> The issue there is that an unrelated variable in SingleChannelElement is
> declared to have an alignment of 32 bytes. So if the compiler does a copy
> in decode_cpe() with avx instructions, but ffmpeg is built with
> --disable-avx, this results in a crash, since the memory is only 16 byte
> aligned.
> 
> Mind you, even if the compiler does not emit avx instructions, the code
> is still invalid and could misbehave. It just happens not to. Declaring
> any variable in a struct with a 32 byte alignment promises 32 byte
> alignment of the whole struct to the compiler.
> 
> This patch limits the maximum alignment to the maximum possible simd
> alignment according to configure.
> While not perfect, it at the very least gets rid of a lot of UB, by
> matching up the maximum DECLARE_ALIGNED value with the alignment of heap
> allocations done by lavu.
> ---
>   libavutil/mem.c          |  2 +-
>   libavutil/mem_internal.h | 20 +++++++++++---------
>   2 files changed, 12 insertions(+), 10 deletions(-)
> 
> diff --git a/libavutil/mem.c b/libavutil/mem.c
> index 36b8940a0c..62163b4cb3 100644
> --- a/libavutil/mem.c
> +++ b/libavutil/mem.c
> @@ -62,7 +62,7 @@ void  free(void *ptr);
>   
>   #endif /* MALLOC_PREFIX */
>   
> -#define ALIGN (HAVE_AVX512 ? 64 : (HAVE_AVX ? 32 : 16))
> +#define ALIGN (HAVE_SIMD_ALIGN_64 ? 64 : (HAVE_SIMD_ALIGN_32 ? 32 : 16))
>   
>   /* NOTE: if you want to override these functions with your own
>    * implementations (not recommended) you have to link libav* as
> diff --git a/libavutil/mem_internal.h b/libavutil/mem_internal.h
> index 2448c606f1..ddd3c24806 100644
> --- a/libavutil/mem_internal.h
> +++ b/libavutil/mem_internal.h
> @@ -75,22 +75,24 @@
>    * @param v Name of the variable
>    */
>   
> +#define MAX_ALIGNMENT (HAVE_SIMD_ALIGN_64 ? 64 : (HAVE_SIMD_ALIGN_32 ? 32 : 16))

The issue with this approach is that code that previously allowed the 
compiler to optimize it heavily will now only be optimized if ffmpeg was 
built without disabling avx.
Probably a fair tradeoff though, since that's a very niche case.

I also did not test this with MSVC or ICC, so I have no idea if they 
allow FFMIN in the middle of an alignment attribute.