diff options
| author | bors <bors@rust-lang.org> | 2021-10-17 22:29:31 +0000 |
|---|---|---|
| committer | bors <bors@rust-lang.org> | 2021-10-17 22:29:31 +0000 |
| commit | 5e02151318ddd431aea6d58e23948246c1446044 (patch) | |
| tree | e9fcb11ec84636b922fec5a652d376a4d8e7d96e | |
| parent | 1f12ac87296ac61ec002e0243e7ad5a50364da35 (diff) | |
| parent | 86608f1796c9335b103b596df53bef03f7fcc303 (diff) | |
| download | rust-5e02151318ddd431aea6d58e23948246c1446044.tar.gz rust-5e02151318ddd431aea6d58e23948246c1446044.zip | |
Auto merge of #89499 - Mark-Simulacrum:with-llvm-13, r=nikic
Split out LLVM PGO step and use clang 13 to compile LLVM
We're seeing a PGO version mismatch error in CI logs:
LLVM Profile Error: Runtime and instrumentation version mismatch : expected 5, but get 7
which is likely due to the version bumped here differing from that used by
rustc.
This PR fixes this by splitting out the PGO step for LLVM into a separate phase of the pgo.sh script, which nets no change to performance (see [these results](https://perf.rust-lang.org/compare.html?start=c34ac8747ca96d09cb08b8f5adddead826e77c06&end=e272c2af45f40c74dab83948235903ffbe3ad57f)). Then, it follows that up with an upgrade to LLVM/clang version 13 as our bootstrap compiler, which yields the performance improvements for this PR -- around 5%. This depends on the first step here, because otherwise we end up somehow clobbering or otherwise hurting our ability to effectively collect performance data, yielding reductions in performance for a subset of benchmarks -- it is not clear what the cause here was precisely, but the split only costs ~10 minutes and seems worthwhile.
| -rwxr-xr-x | src/ci/docker/host-x86_64/dist-x86_64-linux/build-clang.sh | 2 | ||||
| -rwxr-xr-x | src/ci/pgo.sh | 38 |
2 files changed, 31 insertions, 9 deletions
diff --git a/src/ci/docker/host-x86_64/dist-x86_64-linux/build-clang.sh b/src/ci/docker/host-x86_64/dist-x86_64-linux/build-clang.sh index ed5edfec4e1..562be752f84 100755 --- a/src/ci/docker/host-x86_64/dist-x86_64-linux/build-clang.sh +++ b/src/ci/docker/host-x86_64/dist-x86_64-linux/build-clang.sh @@ -4,7 +4,7 @@ set -ex source shared.sh -LLVM=llvmorg-12.0.1 +LLVM=llvmorg-13.0.0 mkdir llvm-project cd llvm-project diff --git a/src/ci/pgo.sh b/src/ci/pgo.sh index e35e3e670cc..29ef13a60fb 100755 --- a/src/ci/pgo.sh +++ b/src/ci/pgo.sh @@ -4,9 +4,13 @@ set -euxo pipefail rm -rf /tmp/rustc-pgo +# We collect LLVM profiling information and rustc profiling information in +# separate phases. This increases build time -- though not by a huge amount -- +# but prevents any problems from arising due to different profiling runtimes +# being simultaneously linked in. + python3 ../x.py build --target=$PGO_HOST --host=$PGO_HOST \ --stage 2 library/std \ - --rust-profile-generate=/tmp/rustc-pgo \ --llvm-profile-generate # Profile libcore compilation in opt-level=0 and opt-level=3 @@ -15,6 +19,29 @@ RUSTC_BOOTSTRAP=1 ./build/$PGO_HOST/stage2/bin/rustc --edition=2018 \ RUSTC_BOOTSTRAP=1 ./build/$PGO_HOST/stage2/bin/rustc --edition=2018 \ --crate-type=lib -Copt-level=3 ../library/core/src/lib.rs +# Merge the profile data we gathered for LLVM +# Note that this uses the profdata from the clang we used to build LLVM, +# which likely has a different version than our in-tree clang. +/rustroot/bin/llvm-profdata \ + merge -o /tmp/llvm-pgo.profdata ./build/$PGO_HOST/llvm/build/profiles + +# Rustbuild currently doesn't support rebuilding LLVM when PGO options +# change (or any other llvm-related options); so just clear out the relevant +# directories ourselves. +rm -r ./build/$PGO_HOST/llvm ./build/$PGO_HOST/lld + +# Okay, LLVM profiling is done, switch to rustc PGO. + +python3 ../x.py build --target=$PGO_HOST --host=$PGO_HOST \ + --stage 2 library/std \ + --rust-profile-generate=/tmp/rustc-pgo + +# Profile libcore compilation in opt-level=0 and opt-level=3 +RUSTC_BOOTSTRAP=1 ./build/$PGO_HOST/stage2/bin/rustc --edition=2018 \ + --crate-type=lib ../library/core/src/lib.rs +RUSTC_BOOTSTRAP=1 ./build/$PGO_HOST/stage2/bin/rustc --edition=2018 \ + --crate-type=lib -Copt-level=3 ../library/core/src/lib.rs + cp -r /tmp/rustc-perf ./ chown -R $(whoami): ./rustc-perf cd rustc-perf @@ -46,18 +73,13 @@ cd /checkout/obj ./build/$PGO_HOST/llvm/bin/llvm-profdata \ merge -o /tmp/rustc-pgo.profdata /tmp/rustc-pgo -# Merge the profile data we gathered for LLVM -# Note that this uses the profdata from the clang we used to build LLVM, -# which likely has a different version than our in-tree clang. -/rustroot/bin/llvm-profdata \ - merge -o /tmp/llvm-pgo.profdata ./build/$PGO_HOST/llvm/build/profiles - # Rustbuild currently doesn't support rebuilding LLVM when PGO options # change (or any other llvm-related options); so just clear out the relevant # directories ourselves. rm -r ./build/$PGO_HOST/llvm ./build/$PGO_HOST/lld -# This produces the actual final set of artifacts. +# This produces the actual final set of artifacts, using both the LLVM and rustc +# collected profiling data. $@ \ --rust-profile-use=/tmp/rustc-pgo.profdata \ --llvm-profile-use=/tmp/llvm-pgo.profdata |
