-
Notifications
You must be signed in to change notification settings - Fork 34
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Illegal instruction error #3
Comments
Hi Ed, Best regards, |
AMD cpus support sse instructions. I use a lot of intel compiled software that use sse running on amd hardware, minimap2 for example. I also compiled from source in compatibility mode (x86-64). I can’t say for certain but I would guess it’s not an sse instruction that’s causing the problem.
I’m going to disassemble the binary and find the problem myself when I get the chance.
… On 4 Sep 2019, at 16:10, Robert Vaser ***@***.***> wrote:
Hi Ed,
we are using Intel SIMD instructions (packages SSE4.1 and above) which if are not supported on machines different from the machine you compiled the code on will cause the error you got. If you compile on a machine without support, the code will fall back to unvectorized algorithms which will be slower. If you have a cluster with mismatching CPUs, try picking nodes that are compatible with each other.
Best regards,
Robert
—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub <#3?email_source=notifications&email_token=AHHI7IJVT4QRGSFTBCOX3KLQH7FXTA5CNFSM4ITTDARKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOD535ASQ#issuecomment-527945802>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AHHI7IJRYETQUSQFSX2B2C3QH7FXTANCNFSM4ITTDARA>.
|
I'm seeing this same error with a Bioconda-installed racon on an Intel I7-3770. It runs without error on a Xeon E7-8867 v4 server, using the same installation/binaries. |
Weird as I7-3770 should support SSE4.1. Maybe your libstdc++ is outdated? You can always try and compile from source. |
Yeah could be a CentOS outdated-libraries thing. It is a bit weird, since both my workstation (the I7) and the server (the Xeon) are running CentOS 7. It's possible my workstation is just a bit out of date, though. I'll put in a request to get it updated and see if that helps. |
I think I've found the problem, the spoa module you use has a cmake build option "spoa_optimize_for_native" that defaults to ON, that I think is overriding the CPP flags to set "-march=native". Turn that option off to get cross compilation working. |
The Bioconda recipe for racon adds |
I’m pretty sure native compilation will essentially lock you in to the specific CPU model you compiled on, or newer models in the same line. For the bioconda release at least, I would strongly suggest turning native off (the current bioconda release doesn’t work on any AMD hardware, for example).
Without doing any tests myself, I would guess that native compilation won’t give you any significant performance boost anyway, since the SIMD instructions are all hardcoded.
With regards to AVX2 on or off for the bioconda release. May I suggest something like the following: you compile two versions (called say, racon.avx2 and racon.noavx2 for example) and then have the main racon executable be a shell script that does a runtime hardware determination (using say, lscpu) and dispatches to the appropriate binary. Do you think that something like that would be feasible?
… On 2 Oct 2019, at 08:43, Robert Vaser ***@***.***> wrote:
The Bioconda recipe for racon adds -DCMAKE_CXX_FLAGS="-mno-avx2" to the compiler list (see here <https://github.com/bioconda/bioconda-recipes/blob/master/recipes/racon/build.sh>), which is propagated to Spoa where only the SSE4.1 part of the code is compiled. Does -march=native add some other flags that are problematic?
—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub <#3?email_source=notifications&email_token=AHHI7IKEHYLPUKE2XK5FCD3QMRGIPA5CNFSM4ITTDARKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEAD3OWY#issuecomment-537376603>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AHHI7IIB6BMCNHHZ5W6LWJDQMRGIPANCNFSM4ITTDARA>.
|
I'll see what I can do to sort this out. |
Hi, are there any news regarding a working conda installation? I am able to run racon when compiling from source, but conda installation gives core dump error. |
Unfortunately, not yet. |
Sure, I can try on Monday
…Sent from my iPhone
On 8 Mar 2020, at 13:23, Robert Vaser ***@***.***> wrote:
@EdHarry and @MaestSi, can you please try version v1.4.11 from bioconda? I have replaced -march=native with -msse4.1 in Spoa and propagated that to Racon.
I am really sorry for the delay!
Best regards,
Robert
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub, or unsubscribe.
|
Hi, I just tried racon v1.4.11 from bioconda, but the issue persists. I tested it on the same machine where I previously tested spoa, but differently to spoa the issue is not solved for racon yet. |
Does |
I think -march=x86-64 might need to be specified in the build options
…Sent from my iPhone
On 8 Mar 2020, at 14:07, Simone Maestri ***@***.***> wrote:
Hi, I just tried racon v1.4.11 from bioconda, but the issue persists. I tested it on the same machine where I previously tested spoa, but differently to spoa the issue is not solved for racon yet.
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub, or unsubscribe.
|
Yes, it looks like.
|
Hmmm, no idea why :/ |
Yes, same issue on AMD hardware for me.
… On 8 Mar 2020, at 15:45, Robert Vaser ***@***.***> wrote:
Hmmm, no idea why :/
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub <#3?email_source=notifications&email_token=AHHI7IMWCPWSMRMQ6FBNURTRGO4RZA5CNFSM4ITTDARKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEOEZLQA#issuecomment-596219328>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AHHI7IM5EGPGEOM2BPT7GV3RGO4RZANCNFSM4ITTDARA>.
|
@EdHarry, can you please try if |
Yes, spoa 3.0.2 from bioconda works
… On 10 Mar 2020, at 13:28, Robert Vaser ***@***.***> wrote:
@EdHarry <https://github.com/EdHarry>, can you please try if spoa is working on your machine? It is in bioconda as well.
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub <#3?email_source=notifications&email_token=AHHI7IMGU4BJ355Q4CYFM3LRGY6BDA5CNFSM4ITTDARKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEOLM2TY#issuecomment-597085519>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AHHI7IOM6F2RAH2SJXIZ6K3RGY6BDANCNFSM4ITTDARA>.
|
Hmm, maybe something else is the problem in racon executable. I'll try to find out. |
Hi @rvaser , conda installation for racon v1.4.12 is now working for me! |
Confirmed, version 1.4.12 from bioconda works for me on both Intel and AMD hardware.
… On 20 Mar 2020, at 09:42, Simone Maestri ***@***.***> wrote:
Hi @rvaser <https://github.com/rvaser> , conda installation for racon v1.4.12 is now working for me!
Thanks,
Simone
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub <#3 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AHHI7IOEV3F2L4EBLDXG2EDRIM3BDANCNFSM4ITTDARA>.
|
Finally! I found |
Reopening previous issue here
isovic/racon#119
Seems to happen on AMD cpus only, at least from my experience.
The text was updated successfully, but these errors were encountered: