Re: [FFmpeg-devel] [PATCH] checkasm/v210enc.c: Use checkasm_check_type

["Martin Storsjö" <martin@martin.st>]

On Sat, 8 Feb 2025, Kieran Kunhya via ffmpeg-devel wrote:

> $subj

> -            if (memcmp(y0, y1, BUF_SIZE * sizeof(type))                            \
> -                    || memcmp(u0, u1, BUF_SIZE * sizeof(type) / 2)                 \
> -                    || memcmp(v0, v1, BUF_SIZE * sizeof(type) / 2)                 \
> -                    || memcmp(dst0, dst1, width * 8 / 3))                          \
> +            if (checkasm_check_##type(NULL, 0, y0, 0, y1, 0, BUF_SIZE, 0, NULL)                     \
> +                    || checkasm_check_##type(NULL, 0, u0, 0, u1, 0, BUF_SIZE / 2, 0, NULL)          \
> +                    || checkasm_check_##type(NULL, 0, v0, 0, v1, 0, BUF_SIZE / 2, 0, NULL)          \
> +                    || checkasm_check_uint8_t(NULL, 0, dst0, 0, dst1, 0, (width * 8 / 3), 0, NULL)) \

This actually doesn't detect any failures at all; you're passing it 
parameters for checking a buffer of BUF_SIZE width and 0 height, so it 
doesn't check anything.

By passing height 1, it does seem to work as intended (detecting an 
intentionally added error in the asm).

It feels a little bit unwieldy to check (and print out, if checkasm is run 
with "-v") the whole BUF_SIZE, but it's good to have testing for potential 
writes out of bounds at least, like before.

Further in dav1d there are more improvements to these checkasm helpers, 
which we haven't backported to ffmpeg yet, that helps with allocating 
padded buffers and checking that the padding isn't overwritten; that would 
allow reducing the size of the checked area, to make it easier to read on 
errors too.

But until that's backported, I guess this is fine, if you change the 
height to 1.

// Martin

_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".