volk 3.1.1 regression test failure
chenrui333 opened this issue · comments
👋 trying to build the latest release, but run into some build issue. The error log is as below:
error build log
==> /opt/homebrew/Cellar/volk/3.1.1/bin/volk_profile --iter 10
RUN_VOLK_TESTS: volk_64u_popcntpuppet_64u(131071,10)
generic completed in 0.712 ms
neon completed in 0.643 ms
Best aligned arch: neon
Best unaligned arch: neon
RUN_VOLK_TESTS: volk_16u_byteswappuppet_16u(131071,10)
generic completed in 0.619 ms
neon completed in 0.273 ms
neon_table completed in 0.136 ms
Error: volk: failed
An exception occurred within a child process:
BuildError: Failed executing: /opt/homebrew/Cellar/volk/3.1.1/bin/volk_profile --iter 10
full build log, https://github.com/Homebrew/homebrew-core/actions/runs/7704559273/job/20997188097?pr=161268
relates to Homebrew/homebrew-core#161268
this is my local debug run
$ lldb /opt/homebrew/Cellar/volk/3.1.1/bin/volk_profile
(lldb) target create "/opt/homebrew/Cellar/volk/3.1.1/bin/volk_profile"
Current executable set to '/opt/homebrew/Cellar/volk/3.1.1/bin/volk_profile' (arm64).
(lldb) run --update
Process 74547 launched: '/opt/homebrew/Cellar/volk/3.1.1/bin/volk_profile' (arm64)
No prior test results found ...
RUN_VOLK_TESTS: volk_64u_popcntpuppet_64u(131071,1987)
generic completed in 74.997 ms
neon completed in 117.34 ms
Best aligned arch: generic
Best unaligned arch: generic
RUN_VOLK_TESTS: volk_16u_byteswappuppet_16u(131071,1987)
generic completed in 30.298 ms
neon completed in 51.535 ms
neon_table completed in 27.018 ms
Process 74547 stopped
* thread #1, queue = 'com.apple.main-thread', stop reason = EXC_BAD_ACCESS (code=2, address=0x100044000)
frame #0: 0x0000000183b896f8 libsystem_platform.dylib`_platform_memmove + 168
libsystem_platform.dylib`:
-> 0x183b896f8 <+168>: stp q2, q3, [x0]
0x183b896fc <+172>: subs x2, x2, #0x40
0x183b89700 <+176>: b.ls 0x183b8971c ; <+204>
0x183b89704 <+180>: stp q0, q1, [x3]
Target 0: (volk_profile) stopped.
Thanks for the bug report.
From the log, it would appear the crash occurs in volk_16_byteswap_u_orc. That protokernel was previously inaccessible, and re-enabled in #719. It runs correctly in CI and locally on a Raspberry Pi, but it would seem there is trouble on Apple M1 for some reason.
@chenrui333 I opened #747 which I expect will fix this. Are you able to test it on your system?
I added a test with FlyCI. It builds on their systems with an M2:
https://github.com/jdemel/volk/actions/runs/7717048987/job/21035255427
It looks like that built without ORC support:
ORC support not found, Overruled arch orc
It looks like that built without ORC support:
ORC support not found, Overruled arch orc
It's time to figure out how we enable ORC on MacOS.
GitHub runners have Homebrew, so you could probably do brew install orc.
@chenrui333 I opened #747 which I expect will fix this. Are you able to test it on your system?
yeah, totally, testing it now.
trie to apply the patch, the build was not successful this time
[ 20%] Generating volk_32f_s32f_add_32f_a_orc_impl.c
cd /tmp/volk-20240131-22114-h6wzw/volk-3.1.1/build/lib && /opt/homebrew/bin/orcc --include math.h --implementation --include volk/volk_complex.h -o /tmp/volk-20240131-22114-h6wzw/volk-3.1.1/build/lib/volk_32f_s32f_add_32f_a_orc_impl.c /tmp/volk-20240131-22114-h6wzw/volk-3.1.1/kernels/volk/asm/orc/volk_32f_s32f_add_32f_a_orc_impl.orc
gmake[2]: *** [lib/CMakeFiles/volk_obj.dir/build.make:89: lib/volk_16u_byteswap_a_orc_impl.c] Error 1
You'll need to run cmake again (and maybe clear out the build directory if that doesn't work).
I realize I have to also drop orc dependency, the build works fine for me now. Thanks folks! Closing this issue. :)
I realize I have to also drop
orcdependency
Why?
I realize I have to also drop
orcdependencyWhy?
HAVE_ORC detection?
You'll need to run cmake again (and maybe clear out the build directory if that doesn't work).
we always do the clean build, so that actually does not apply.
Hmm, if VOLK fails to build with ORC present that sounds like a bug. Could you open an issue for that?
Also, how did you build earlier when you encountered the error in volk_profile? Was ORC enabled at that time?
Hmm, if VOLK fails to build with ORC present that sounds like a bug. Could you open an issue for that?
totally. this is the success https://github.com/Homebrew/homebrew-core/actions/runs/7727239095/job/21065529883
Also, how did you build earlier when you encountered the error in
volk_profile? Was ORC enabled at that time?
yes, that was when the orc present (do I need to explicitly enable any option to turn it on?) this is the build log, https://github.com/Homebrew/homebrew-core/actions/runs/7704559273/job/20997188097?pr=161268
do I need to explicitly enable any option to turn it on?
I don't think so. It should be automatically detected.
I'm not sure disabling ORC is a great idea.
Are you sure that you correctly applied the patch from #747 when you tested it? It builds fine for me, and also in our CI. Note that the patch deletes kernels/volk/asm/orc/volk_16u_byteswap_a_orc_impl.orc.
Are you sure that you correctly applied the patch from #747 when you tested it? It builds fine for me, and also in our CI. Note that the patch deletes
kernels/volk/asm/orc/volk_16u_byteswap_a_orc_impl.orc.
yeah, I did, you can see this linked PR, Homebrew/homebrew-core#161268
Can you show me the build that failed with just the patch in place (and not the ORC removal)?
I found only this one, but it appears that it was cancelled: https://github.com/Homebrew/homebrew-core/actions/runs/7727191280
The PR #747 was only intended to locate the source of the problem, so I don't think it is a good idea for Homebrew to ship it. Since Homebrew has disabled ORC, the patch does nothing.