Chris Matthews via llvm-dev
2021-Jan-19 21:55 UTC
[llvm-dev] Some questions re tracking LLD for Mach-O's performance using LNT
I forget when that was setup, but here is a copy of the schema from the lnt.llvm.org <http://lnt.llvm.org/> server: ``` format_version: '2' name: link run_fields: - name: llvm_project_revision order: true machine_fields: - name: hardware - name: os metrics: - name: branch-misses bigger_is_better: false type: Real - name: stalled-cycles-frontend bigger_is_better: false type: Real - name: branches bigger_is_better: false type: Real - name: context-switches bigger_is_better: false type: Real - name: cpu-migrations bigger_is_better: false type: Real - name: cycles bigger_is_better: false type: Real - name: instructions bigger_is_better: false type: Real - name: seconds-elapsed bigger_is_better: false type: Real - name: page-faults bigger_is_better: false type: Real - name: task-clock bigger_is_better: false type: Real ```> On Jan 19, 2021, at 1:49 PM, Jez via llvm-dev <llvm-dev at lists.llvm.org> wrote: > > Hi, > > I understand that the `link` test suite schema at https://lnt.llvm.org/ <https://lnt.llvm.org/> is being used to record profiling results for LLD for ELF. I would like to do the same with LLD's Mach-O implementation by running the profiling jobs on an internal server & submitting the results to the public LNT instance. A few questions: > > 1. I don't see the `link` schema under `schemas/` in the `lnt` repo. Where can I find it? > > 2. Should I reuse the schema or create a new one? I expect that we can stick to the same set of perf data fields, so reusing should be fine as long as we create one more column to indicate whether we're running the ELF or Mach-O suite. > > 3. I would also like to reuse the lld/utils/benchmark.py script. Currently, it invokes the LLD for ELF process via a hard-coded name, so I would like to make that configurable via a CLI flag. But that would involve updating the current job on the `as-worker` machines to specify the ELF back-end. Who owns those machines and would be able to do this update? (That said, it's hardly a blocker, since I can still have the script default the ELF backend for now.) > > Thanks, > Jez > _______________________________________________ > LLVM Developers mailing list > llvm-dev at lists.llvm.org > https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev-------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20210119/891a3d5f/attachment.html>
Jez via llvm-dev
2021-Jan-21 17:14 UTC
[llvm-dev] Who owns the as-worker-5:2 machine? (Was: Some questions re tracking LLD for Mach-O's performance using LNT)
Thanks Chris! I belatedly found that file under lld/utils/. A bit surprising that it isn't in the LNT repo itself but I guess it was copied over into the deployed server by hand... I would like to re-surface question #3 in my original message: who owns as-worker-5:2? I've created a handful of diffs modifying benchmark.py (D95082, D95084), which I believe is being run on that machine. I don't think my changes will break anything, but it would be nice to get a stamp from whomever still depends on it. (The original author was Rafael Espindola.) Jez On Tue, Jan 19, 2021 at 4:55 PM Chris Matthews <chris.matthews at apple.com> wrote:> > I forget when that was setup, but here is a copy of the schema from the lnt.llvm.org server: > > ``` > format_version: '2' > name: link > run_fields: > - name: llvm_project_revision > order: true > machine_fields: > - name: hardware > - name: os > metrics: > - name: branch-misses > bigger_is_better: false > type: Real > - name: stalled-cycles-frontend > bigger_is_better: false > type: Real > - name: branches > bigger_is_better: false > type: Real > - name: context-switches > bigger_is_better: false > type: Real > - name: cpu-migrations > bigger_is_better: false > type: Real > - name: cycles > bigger_is_better: false > type: Real > - name: instructions > bigger_is_better: false > type: Real > - name: seconds-elapsed > bigger_is_better: false > type: Real > - name: page-faults > bigger_is_better: false > type: Real > - name: task-clock > bigger_is_better: false > type: Real > ``` > > On Jan 19, 2021, at 1:49 PM, Jez via llvm-dev <llvm-dev at lists.llvm.org> wrote: > > Hi, > > I understand that the `link` test suite schema at https://lnt.llvm.org/ is being used to record profiling results for LLD for ELF. I would like to do the same with LLD's Mach-O implementation by running the profiling jobs on an internal server & submitting the results to the public LNT instance. A few questions: > > 1. I don't see the `link` schema under `schemas/` in the `lnt` repo. Where can I find it? > > 2. Should I reuse the schema or create a new one? I expect that we can stick to the same set of perf data fields, so reusing should be fine as long as we create one more column to indicate whether we're running the ELF or Mach-O suite. > > 3. I would also like to reuse the lld/utils/benchmark.py script. Currently, it invokes the LLD for ELF process via a hard-coded name, so I would like to make that configurable via a CLI flag. But that would involve updating the current job on the `as-worker` machines to specify the ELF back-end. Who owns those machines and would be able to do this update? (That said, it's hardly a blocker, since I can still have the script default the ELF backend for now.) > > Thanks, > Jez > _______________________________________________ > LLVM Developers mailing list > llvm-dev at lists.llvm.org > https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev > >