Open zamazan4ik opened 1 year ago
When I reduce perf profile size (to approximately 400 Mib) - the issue does not appear. I guess it's something with processing huge profiles but I am not sure.
@llvm/issue-subscribers-bolt
Try to run df -h
(or if you want to see what's happening htop
in parallel tab) in a parallel tab when you run perf2bolt
, from what I know perf2bolt
runs perf script
on your perf.data
file and since the file is huge, it ends up consuming a lot of disk space. I am not sure what might be the solution other than increasing the disk space.
Something which might help is removing /tmp/perf.script-*
files.
If there is insufficient memory during linking, try:
export TMPDIR=/path/to/your/tmp
During the BOLT evaluation, I met an error with
perf2bolt
. I collectedperf
profile withperf record -b -p my_pid -o perf.data -- sleep 630
, got aperf.data
for 2.2 Gib. Then, I transform it to BOLT format withperf2bolt -nl -strict=0 -p perf.data -o perf.fdata my_binary
. During the transformation process I get the following log:And
perf2bolt
fails to proceed with my perf report. Is it a known issue? How could I fix it?My setup:
llvm-bolt
,perf2bolt
versions: from Fedora 38 repositories - 16.0.0