Skip to content
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

Can not load trace when using current trace-cmd master #50

Open
subdiff opened this issue Jun 15, 2020 · 10 comments
Open

Can not load trace when using current trace-cmd master #50

subdiff opened this issue Jun 15, 2020 · 10 comments

Comments

@subdiff
Copy link
Contributor

subdiff commented Jun 15, 2020

Following error message appears in GPUVis log:

[Error] handle_options: unknown option 11
[Error] read_trace_file: setjmp error called for trace_06-15-2020_20-11-53.dat.
[Error] load_trace_file(trace_06-15-2020_20-11-53.dat) failed.
@subdiff
Copy link
Contributor Author

subdiff commented Jun 15, 2020

Repo: https://git.kernel.org/pub/scm/utils/trace-cmd/trace-cmd.git
Checkout at: 27ee7e41e5f5e3f134ce985fec03b1386f30e45f

@Oschowa
Copy link

Oschowa commented Apr 20, 2022

This issue seems to back with a different error message when trying to open traces captured with trace-cmd 3.0.3:

[Error] read_header_files: header_page not found.
[Error] read_trace_file: setjmp error called for trace.dat

Rolling back to trace-cmd 2.9.7 makes it work again.

@mikesart
Copy link
Owner

mikesart commented May 1, 2022

Can you try the libtraceevent branch and see if it still happens? And if it does, is it possible to send me a trace that fails? Thanks Oschowa.

@Oschowa
Copy link

Oschowa commented May 6, 2022

I'm still getting the same error on the libtraceevent branch, here's a sample trace:
trace_05-06-2022_17-19-23.zip

@mikesart
Copy link
Owner

mikesart commented Jun 14, 2022

Apologies for the late reply - been a bit slammed at work. So it looks like they added a compressed page format so this bit of code:

337├───> if ( memcmp( buf, "header_page", 12 ) != 0 )
338│ die( handle, "%s: header_page not found.\n", func );

Is failing because buf is "zstd". Do you have a way to disable this compression with the latest trace-cmd? I'll look at adding this but will take a bit of time. Thanks Oschowa.

@mikesart
Copy link
Owner

Got Oschowa's zstd sample trace loading in gpuvis. This is all on Linux only right now cause I had to pull in a lot of the libtracefs, trace-cmd, and libtraceevent code and all this stuff relies a lot on gcc extensions and Linux header files, functions, etc.

Lot of work still to go to verify older traces (some of which aren't loading right now) and make sure new ones load correctly and with decent performance.

Then have to get all this working on Windows. 😭

image

@YellowOnion
Copy link

Can you push the branch? would be nice for us linux users to be able to use :P

mikesart added a commit that referenced this issue Nov 12, 2022
@mikesart
Copy link
Owner

Pushed it to zstd branch:

https://github.com/mikesart/gpuvis/tree/zstd

And sorry - just recently started a new job and have been kinda slammed...

@lostgoat
Copy link
Collaborator

trace-cmd Version 3.1.5 has been released with file version 7 as the default for recordings.

As a workaround, a tracefile can be converted to version 6 to be opened with gpuvis using the command:

trace-cmd convert -i trace-7.dat -o trace-6.dat --file-version 6

lostgoat added a commit to lostgoat/gpu-trace that referenced this issue Dec 30, 2022
Current gpuvis only supports trace-cmd file format 6. So we convert
to it after capture.

Reference:
mikesart/gpuvis#50
@uis246
Copy link

uis246 commented Feb 6, 2024

Wow, zstd makes trace files 6 times smaller. I hope it will be supported here soon. Thanks lostgoat for workaround.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

5 participants