[FFmpeg-devel] lurking bugs in the mmx-related assembler code (?)

[Ronald S. Bultje]

Hi,

On Sun, Oct 2, 2016 at 4:24 PM, Carl Eugen Hoyos <ceffmpeg at gmail.com> wrote:

> 2016-10-02 16:01 GMT+02:00 Carl Eugen Hoyos <ceffmpeg at gmail.com>:
>
> > PS: There are several unrelated build issues.
>
> The build issues were just seen here because I
> had to add "-melf_i386" to the musl spec file.
> After adding it, building works perfectly smooth.
>
> I tested with --enable-debug for musl and still get
> no backtrace, valgrind cannot help.
>
> No issue on x86_64: The musl spec file works
> fine and fate passes with asm optimizations.


Floating point operations on x86-64 (which is by definition >= SSE2) are
typically done using scalar SSE instructions, not using x87. Therefore, if
you don't clean up the floating point state, things will usually still work
fine. On x86-32, you can't do this because you can't be sure that the CPU
supports SSE.

Ronald