Mircea Trofin via llvm-dev
2021-Oct-14 05:28 UTC
[llvm-dev] Consolidating copies of google/benchmark in the repo (Was: Proposal: introduce dependency on abseil when building benchmarks)
On Wed, Oct 13, 2021 at 6:45 AM Renato Golin <rengolin at gmail.com> wrote:> On Tue, 12 Oct 2021 at 21:20, Mircea Trofin <mtrofin at google.com> wrote: > >> One thing that does come to mind, llvm-test-suite can be cloned >> separately from llvm-project/llvm. We'd need to point to llvm (or just >> llvm/utils/benchmark), and require that it be available on bots, for >> example. Does that cause an issue for anyone? >> > > Because libc++ is now in the monorepo, unless it needs a different > (patched?) version from llvm's main one, it should not need its own copy. > > AFAIK, the usual benchmark infrastructure compiles clang on demand and > then use that on the test-suite repo. However, that doesn't mean the repos > are on the same machine. For example, if you compile clang on a larger > cluster, install+pack and send to a more focued machine to do the > single-threaded-single-use-no-services benchmarking. > > So, I'd say it should be safe to merge libc++ and llvm one, but not > necessarily the test-suite one. > > If we don't have any local patches on those, we could also just clone it > from some known hash at runtime, or $deity forbid, as a submodule. :D >Putting it together, it seems like we can start with unifying the 2 in llvm/libcxx, then figure llvm-test-suite along the lines discussed so far. For the former, libcxx seems to have been updated most recently, I'll copy that one over to llvm/utils, seems benign, unless I'm missing something.>cheers,> --renato >-------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20211013/fd26d6f1/attachment.html>
Mircea Trofin via llvm-dev
2021-Oct-25 15:18 UTC
[llvm-dev] Consolidating copies of google/benchmark in the repo (Was: Proposal: introduce dependency on abseil when building benchmarks)
Circling back from https://reviews.llvm.org/D112012 <https://reviews.llvm.org/D112012#3078815> - would having google/benchmark under a new `llvm-project/third-party` work, instead of `llvm-project/llvm/utils`? (rationale there, TL;DR; libcxx (and other runtimes) wants to detach itself from anything `llvm-project/llvm`) On Wed, Oct 13, 2021 at 10:28 PM Mircea Trofin <mtrofin at google.com> wrote:> > > On Wed, Oct 13, 2021 at 6:45 AM Renato Golin <rengolin at gmail.com> wrote: > >> On Tue, 12 Oct 2021 at 21:20, Mircea Trofin <mtrofin at google.com> wrote: >> >>> One thing that does come to mind, llvm-test-suite can be cloned >>> separately from llvm-project/llvm. We'd need to point to llvm (or just >>> llvm/utils/benchmark), and require that it be available on bots, for >>> example. Does that cause an issue for anyone? >>> >> >> Because libc++ is now in the monorepo, unless it needs a different >> (patched?) version from llvm's main one, it should not need its own copy. >> >> AFAIK, the usual benchmark infrastructure compiles clang on demand and >> then use that on the test-suite repo. However, that doesn't mean the repos >> are on the same machine. For example, if you compile clang on a larger >> cluster, install+pack and send to a more focued machine to do the >> single-threaded-single-use-no-services benchmarking. >> >> So, I'd say it should be safe to merge libc++ and llvm one, but not >> necessarily the test-suite one. >> >> If we don't have any local patches on those, we could also just clone it >> from some known hash at runtime, or $deity forbid, as a submodule. :D >> > > Putting it together, it seems like we can start with unifying the 2 in > llvm/libcxx, then figure llvm-test-suite along the lines discussed so far. > > For the former, libcxx seems to have been updated most recently, I'll copy > that one over to llvm/utils, seems benign, unless I'm missing something. > > > >> > cheers, >> --renato >> >-------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20211025/ce60e6fb/attachment.html>