[FFmpeg-devel] [PATCH] avcodec/svq1: zero initialize entries array

[Clément Bœsch]

On Mon, Apr 10, 2017 at 02:48:51PM +0200, Michael Niedermayer wrote:
> On Mon, Apr 10, 2017 at 02:42:42PM +0200, Clément Bœsch wrote:
> > On Mon, Apr 10, 2017 at 02:19:54PM +0200, Michael Niedermayer wrote:
> > > On Mon, Apr 10, 2017 at 02:04:30PM +0200, Clément Bœsch wrote:
> > > > On Mon, Apr 10, 2017 at 01:37:54PM +0200, Michael Niedermayer wrote:
> > > > > On Mon, Apr 10, 2017 at 08:02:34AM +0200, Clément Bœsch wrote:
> > > > > > On Sun, Apr 09, 2017 at 11:10:16PM -0300, James Almer wrote:
> > > > > > [...]
> > > > > > > > is this a regression ?
> > > > > > > 
> > > > > > > Fate didn't always complain about this, so it's either something
> > > > > > > introduced by a change in our tree, or a valgrind bug introduced in a
> > > > > > > relatively recent version.
> > > > > > > 
> > > > > > 
> > > > > > > The reports in
> > > > > > > http://fate.ffmpeg.org/history.cgi?slot=x86_64-archlinux-gcc-valgrindundef
> > > > > > > are kinda broken and report a nonsense commit as the "last known good
> > > > > > > ref", so i can't say when it started failing.
> > > > > > 
> > > > > > That was because I couldn't upgrade for a long time and bumped both GCC
> > > > > > and Valgrind (and actually the whole system) at once. Sorry, I couldn't do
> > > > > > it gradually.
> > > > > 
> > > > > does your fate client have optimizations enabled ?
> > > > > i dont see them being disabled, and iam not sure valgrind + full
> > > > > compiler optimizations are a good idea
> > > > > 
> > > > > if they are enabled, please try to disable them to check if that
> > > > > causes the false positives
> > > > > 
> > > > 
> > > > Yeah actually, only --extra-cflags=-fno-tree-vectorize is needed to make
> > > > fate-vsynth3-svq1 pass. I thought it was supposed to be disabled?
> > > > 
> > > > > my local valgrind shows nothing for svq1
> > > > 
> > > 
> > > > With aed84ee4d reverted? --toolchain=valgrind-memcheck
> > > > --disable-stripping?
> > > 
> > > reverted, yes
> > > no toolchain, no stripping just tried it under valgrind
> > > 
> > > but thats with a relativly old gcc
> > > 
> > 
> > Actually i was wrong: the GCC option that makes valgrind go crazy is
> > unswitch-loops, not tree-vectorize. Do you have it in your GCC?
> 
> its in man gcc on the box i tried on.
> maybe gcc has improved between version 4 an 6
> but it doesnt really matter
> 
> if it fixes all the false positives, please add it and then lets
> revert the workaround (s?)
> 

I reverted and I'm in the process of reporting to valgrind.

-- 
Clément B.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 488 bytes
Desc: not available
URL: <http://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20170410/43eb1fcf/attachment.sig>