Closed fancycode closed 3 weeks ago
I don't think that's expected but also I don't think it should use AVX2 for an i386 platform so something weird going on 😅
Are you using automake or cmake for the building?
Are you using automake or cmake for the building?
I'm using automake, see here for the output from a run on the Debian buildd: https://buildd.debian.org/status/fetch.php?pkg=kvazaar&arch=i386&ver=2.3.0-1&stamp=1715809980&raw=0
I'll try to find steps how this can be reproduced locally easily, i.e. without building the Debian package.
I don't think that's expected but also I don't think it should use AVX2 for an i386 platform so something weird going on 😅
The logs above were from my local (amd64) machine where I built and ran the 32bit package, maybe that's why it detected AVX2? Running the 32bit compiled code on amd64 should work fine, at least the other tests are passing.
Here are the steps to reproduce this locally in a 32bit Docker container:
docker run --rm -ti i386/debian:sid /bin/bash
Then inside Docker:
apt-get update && apt-get install -y --no-install-recommends \
autoconf \
automake \
build-essential \
ca-certificates \
ffmpeg \
g++ \
gcc \
git \
hm \
libtool \
make
git clone https://github.com/ultravideo/kvazaar.git
cd kvazaar
./autogen.sh
./configure --build=i686-linux-gnu
make -j$(nproc)
make check
While this hangs, you can connect to the running container from a second terminal and use gdb
to inspect the process.
Slightly related: is there a flag to show the output of the tests while they are running instead of just writing to the test log files?
I think Automake uses only logs for the tests because of parallel test running. But you can always run cat tests/*.log
afterwards =)
I pushed a fix 6bd23734070655610cd415ae3873a5b043198e33 that might solve the issue. It should not have been a problem to align buffers but for some reason it failed on the i386 in some cases..
Thanks, unfortunately this didn't help and the test is still running with 100% CPU until cancelled. Do the steps described above work for you with latest master (i.e. running in 32bit Docker)?
strange, I did test things with the Docker container but I was using --disable-shared and it was causing some other similar problem 😅 Now when testing again I did found another problem with comparing double values, I'm pushing a fix soon
Now there's another fix 2f9a21423a1ef666c93f15e2068bcde979d20f7e
Now there's another fix 2f9a214
Thanks! With only this applied on an otherwise unmodified 2.3.1, I can no longer reproduce the problem locally. I'll prepare an update for the Debian packaging to see if this also fixes the problem on the other architectures.
I'll prepare an update for the Debian packaging to see if this also fixes the problem on the other architectures.
The update is building fine now on the Debian build infrastructure, is available in Debian unstable
and should be available in testing
in the next couple of days.
Thanks for the fix!
While packaging kvazaar 2.3.0 / 2.3.1 for Debian I noticed the test "test_weird_shapes.sh" takes forever (i.e. must be killed after taking 100% CPU for multiple minutes without visible progress) when running on i386.
This is the output of
tests/test_weird_shapes.sh.log
:The input and output files in '/tmp` are small:
Two stack traces while interrupted from running in
gdb
:The problem doesn't happen on amd64 or most other platforms (see https://buildd.debian.org/status/logs.php?pkg=kvazaar&ver=2.3.0-1&suite=sid).
Is this expected or do you have any suggestions on how to resolve this?