On Sun, 13 Jul 2014, Martin Storsjö wrote:
On Fri, 11 Jul 2014, Ben Avison wrote:
.align 3
Is there any particular reason why this is aligned to 8 bytes instead of 4 -
shouldn't 4 (aka .align 2) be enough for float constants? (Yes, I know this
isn't added by this patch though.)
The extra alignment here triggered a bug in MS armasm where the jump table
offsets were miscalculated, which seemed to be triggered by this extra
alignment here. (I haven't reduced the bug to a reportable testcase yet.)
Actually, changing ".align 3" to ".align 2" doesn't really help much -
armasm seems to be quite broken when it comes to calculating label
differences - in this case it can be triggered when there's any sort of
alignment directives between the two labels.
So this is another argument for going with a plain pointer list (as in
fft_neon.S), as would solve issues with apple's tools as well.
// Martin
_______________________________________________
libav-devel mailing list
[email protected]
https://lists.libav.org/mailman/listinfo/libav-devel
