[vlc-devel] [vlc-commits] packetizer/startcode_helper: enhance with AVX2

[jnqnfe at gmail.com]

On Thu, 2019-03-07 at 07:15 +0200, Remi Denis-Courmont wrote:
> Le 2019-03-07 01:07, Jean-Baptiste Kempf a écrit :
> > On Wed, 6 Mar 2019, at 21:35, Rémi Denis-Courmont wrote:
> > > Le keskiviikkona 6. maaliskuuta 2019, 18.13.42 EET Lyndon Brown
> > > a 
> > > écrit :
> > > > vlc | branch: master | Lyndon Brown <jnqnfe at gmail.com> | Fri
> > > > Jan 25 06:05:54
> > > > 2019 +0000| [c29fcde3887e4fbccc6af4a860a59febfe837f4d] |
> > > > committer:
> > > > Jean-Baptiste Kempf
> > > > 
> > > > packetizer/startcode_helper: enhance with AVX2
> > > 
> > > This does not compile with GCC 8. What did you compile it with?
> > 
> > gcc -v
> 
> I cannot compile with gcc-8 -march=core-avx2.
> 
> And this patchset seems to crash the buildbot.
> 

Per something queried in a private email from JBK sent yesterday, it
seems that there has been a misunderstanding with respect to my use of
nasm and a couple of the latest patches merged of mine need to be
removed (the AVX2 chroma conversion enhancement and the packetizer one
referenced), which may likely be contributing or responsible for the
issue here.

It seems that use of nasm syntax requires that the asm be moved from
inline to separate files, which I did not know.

With the former patch, JBK had previously suggested that I move it to
nasm syntax, something for which I lacked familiarity. Nothing that
turned up in research on that suggested it would not work inline. I
thus carefully converted the syntax and resubmitted, along with
converting various other bits of inline asm to nasm syntax in
additional patches, and recently resubmitted a rebased version on
request.

I was careful to point out that I had not made any change to the build
files that might be necessary to indicate nasm syntax (earlier research
had been unhelpful in suggesting what if anything was needed, and I
have no home internet connection, thus little time to do further such
research when I can be online and spend a little time on it). I was
rather relying upon review of my submission to notice and indicate what
if any such change was needed to complete it.

Unfortunately this seems to have been overlooked and the patches merged
despite this fundamental issue :(.

A little fresh research now suggests that in fact gcc cannot handle
inline nasm, and I need to go back and revise/reconsider things.