20:05 <irker074> systemtap: smakarov systemtap.git:refs/heads/stapbpf/pr22330 * release-4.0-138-g57d177c / bpf-translate.cxx: stapbpf PR22330 :: cleanup round 3 of n (sprintf typo fix) http://tinyurl.com/y55yurxd

20:05 <irker074> systemtap: smakarov systemtap.git:refs/heads/stapbpf/pr22330 * release-4.0-139-gc497891 / testsuite/systemtap.bpf/bpf_tests/string3.stp testsuite/systemtap.bpf/bpf_tests/string4.stp: stapbpf PR22330,PR23816 :: add testcases http://tinyurl.com/y5a6nxu7

20:05 <irker074> systemtap: smakarov systemtap.git:refs/heads/stapbpf/pr22330 * release-4.0-140-gf1e1d05 / stapbpf/stapbpf.cxx: stapbpf PR22330 fix :: support for non-contiguous active cpus http://tinyurl.com/y5nud8ao

20:55 <agentzh> hi folks, i noted that varuse_collecting_visitor::visit_embeddedcode() is painfully slow when translating large .stp scripts (taking >90% of the total CPU time in my samples). i wonder if we could just cache the result for the same "current_function"?

20:56 <agentzh> the interned_string::find() calls in that visit_embeddedcode() method is so expensive.

22:24 <agentzh> I tried adding a varuse_collected boolean field to the embeddedcode class and our most complex stap script's compilation time (phase 2) reduces from 28s to 6s on my machine. yay

22:25 <agentzh> interestingly this bottleneck was spotted by stap itself, via timer.profile + reversed cpu flame graphs.

22:27 <agentzh> the reversed flame graph is here: http://openresty.org/misc/flamegraph/stap-varuse-embeddedcode-inv.svg

22:30 <agentzh> my mistake, the compilation time is from 28s to 3s. not 6s. it's even more dramastic :)

23:29 <mjw> nice speedup!