Utilize PGO for rustc linux dist builds

This implements support for applying PGO to the rustc compilation step (not
standard library or any tooling, including rustdoc). Expanding PGO to more tools
is not terribly difficult but will involve more work and greater CI time
commitment.

For the same reason of avoiding greater time commitment, this currently avoids
implementing for platforms outside of x86_64-unknown-linux-gnu, though in
practice it should be quite simple to extend over time to more platforms. The
initial implementation is intentionally minimal here to avoid too much work
investment before we start seeing wins for a subset of Rust users.

The choice of workloads to profile here is somewhat arbitrary, but the general
rationale was to aim for a small set that largely avoided time regressions on
perf.rust-lang.org's full suite of crates. The set chosen is libcore, cargo (and
its dependencies), and a few ad-hoc stress tests from perf.rlo. The stress tests
are arguably the most controversial, but they benefit those cases (avoiding
regressions) and do not really remove wins from other benchmarks.

The primary next step after this PR lands is to implement support for PGO in
LLVM. It is unclear whether we can afford a full LLVM rebuild in CI, though, so
the approach taken there may need to be more staggered. rustc-only PGO seems
well affordable on linux at least, giving us up to 20% wall time wins on some
crates for 15 minutes of extra CI time (1 hour up from 45 minutes).

Author: Mark-Simulacrum

src/bootstrap/compile.rs

@@ -501,6 +501,20 @@ impl Step for Rustc {
         let mut cargo = builder.cargo(compiler, Mode::Rustc, SourceType::InTree, target, "build");
         rustc_cargo(builder, &mut cargo, target);
 
+        if let Some(path) = &builder.config.rust_profile_generate {
+            if compiler.stage == 1 {
+                cargo.rustflag(&format!("-Cprofile-generate={}", path));
+                // Apparently necessary to avoid overflowing the counters during
+                // a Cargo build profile
+                cargo.rustflag("-Cllvm-args=-vp-counters-per-site=4");
+            }
+        }
+        if let Some(path) = &builder.config.rust_profile_use {
+            if compiler.stage == 1 {
+                cargo.rustflag(&format!("-Cprofile-use={}", path));
+            }
+        }
+
         builder.info(&format!(
             "Building stage{} compiler artifacts ({} -> {})",
             compiler.stage, &compiler.host, target
@@ -752,7 +766,7 @@ fn copy_codegen_backends_to_sysroot(
     // Here we're looking for the output dylib of the `CodegenBackend` step and
     // we're copying that into the `codegen-backends` folder.
     let dst = builder.sysroot_codegen_backends(target_compiler);
-    t!(fs::create_dir_all(&dst));
+    t!(fs::create_dir_all(&dst), dst);
 
     if builder.config.dry_run {
         return;

src/bootstrap/config.rs

@@ -133,6 +133,8 @@ pub struct Config {
     pub rust_thin_lto_import_instr_limit: Option<u32>,
     pub rust_remap_debuginfo: bool,
     pub rust_new_symbol_mangling: bool,
+    pub rust_profile_use: Option<String>,
+    pub rust_profile_generate: Option<String>,
 
     pub build: TargetSelection,
     pub hosts: Vec<TargetSelection>,
@@ -494,6 +496,8 @@ struct Rust {
     llvm_libunwind: Option<String>,
     control_flow_guard: Option<bool>,
     new_symbol_mangling: Option<bool>,
+    profile_generate: Option<String>,
+    profile_use: Option<String>,
 }
 
 /// TOML representation of how each build target is configured.
@@ -871,6 +875,11 @@ impl Config {
 
             config.rust_codegen_units = rust.codegen_units.map(threads_from_config);
             config.rust_codegen_units_std = rust.codegen_units_std.map(threads_from_config);
+            config.rust_profile_use = flags.rust_profile_use.or(rust.profile_use);
+            config.rust_profile_generate = flags.rust_profile_generate.or(rust.profile_generate);
+        } else {
+            config.rust_profile_use = flags.rust_profile_use;
+            config.rust_profile_generate = flags.rust_profile_generate;
         }
 
         if let Some(t) = toml.target {

src/bootstrap/flags.rs

@@ -68,6 +68,9 @@ pub struct Flags {
     pub deny_warnings: Option<bool>,
 
     pub llvm_skip_rebuild: Option<bool>,
+
+    pub rust_profile_use: Option<String>,
+    pub rust_profile_generate: Option<String>,
 }
 
 pub enum Subcommand {
@@ -219,6 +222,8 @@ To learn more about a subcommand, run `./x.py <subcommand> -h`",
              VALUE overrides the skip-rebuild option in config.toml.",
             "VALUE",
         );
+        opts.optopt("", "rust-profile-generate", "rustc error format", "FORMAT");
+        opts.optopt("", "rust-profile-use", "rustc error format", "FORMAT");
 
         // We can't use getopt to parse the options until we have completed specifying which
         // options are valid, but under the current implementation, some options are conditional on
@@ -674,6 +679,8 @@ Arguments:
             color: matches
                 .opt_get_default("color", Color::Auto)
                 .expect("`color` should be `always`, `never`, or `auto`"),
+            rust_profile_use: matches.opt_str("rust-profile-use"),
+            rust_profile_generate: matches.opt_str("rust-profile-generate"),
         }
     }
 }

src/ci/docker/host-x86_64/dist-x86_64-linux/Dockerfile

@@ -85,6 +85,7 @@ ENV CC=clang CXX=clang++
 COPY scripts/sccache.sh /scripts/
 RUN sh /scripts/sccache.sh
 
+
 ENV HOSTS=x86_64-unknown-linux-gnu
 
 ENV RUST_CONFIGURE_ARGS \
@@ -98,9 +99,8 @@ ENV RUST_CONFIGURE_ARGS \
       --set llvm.thin-lto=true \
       --set llvm.ninja=false \
       --set rust.jemalloc
-ENV SCRIPT python2.7 ../x.py dist --host $HOSTS --target $HOSTS \
-        --include-default-paths \
-        src/tools/build-manifest
+COPY host-x86_64/dist-x86_64-linux/pgo.sh /tmp/
+ENV SCRIPT /tmp/pgo.sh
 ENV CARGO_TARGET_X86_64_UNKNOWN_LINUX_GNU_LINKER=clang
 
 # This is the only builder which will create source tarballs

src/ci/docker/host-x86_64/dist-x86_64-linux/pgo.sh

@@ -0,0 +1,50 @@
+#!/bin/bash
+
+set -euxo pipefail
+
+rm -rf /tmp/rustc-pgo
+
+python2.7 ../x.py build --stage 2 library/std --rust-profile-generate=/tmp/rustc-pgo
+
+./build/x86_64-unknown-linux-gnu/stage2/bin/rustc --edition=2018 \
+    --crate-type=lib ../library/core/src/lib.rs
+
+PERF=e095f5021bf01cf3800f50b3a9f14a9683eb3e4e
+
+curl -o /tmp/externs.rs \
+https://raw.githubusercontent.com/rust-lang/rustc-perf/$PERF/collector/benchmarks/externs/src/lib.rs
+./build/x86_64-unknown-linux-gnu/stage2/bin/rustc --edition=2018 --crate-type=lib /tmp/externs.rs
+
+curl -o /tmp/ctfe.rs \
+https://raw.githubusercontent.com/rust-lang/rustc-perf/$PERF/collector/benchmarks/ctfe-stress-4/src/lib.rs
+./build/x86_64-unknown-linux-gnu/stage2/bin/rustc --edition=2018 --crate-type=lib /tmp/ctfe.rs
+
+cp -pri ../src/tools/cargo /tmp/cargo
+
+RUSTC=./build/x86_64-unknown-linux-gnu/stage2/bin/rustc CARGO_INCREMENTAL=1 \
+    ./build/x86_64-unknown-linux-gnu/stage0/bin/cargo check \
+    --manifest-path /tmp/cargo/Cargo.toml
+echo 'pub fn barbarbar() {}' >> /tmp/cargo/src/cargo/lib.rs
+RUSTC=./build/x86_64-unknown-linux-gnu/stage2/bin/rustc CARGO_INCREMENTAL=1 \
+    ./build/x86_64-unknown-linux-gnu/stage0/bin/cargo check \
+    --manifest-path /tmp/cargo/Cargo.toml
+touch /tmp/cargo/src/cargo/lib.rs
+RUSTC=./build/x86_64-unknown-linux-gnu/stage2/bin/rustc CARGO_INCREMENTAL=1 \
+    ./build/x86_64-unknown-linux-gnu/stage0/bin/cargo check \
+    --manifest-path /tmp/cargo/Cargo.toml
+RUSTC=./build/x86_64-unknown-linux-gnu/stage2/bin/rustc CARGO_INCREMENTAL=1 \
+    ./build/x86_64-unknown-linux-gnu/stage0/bin/cargo check \
+    --manifest-path /tmp/cargo/Cargo.toml
+RUSTC=./build/x86_64-unknown-linux-gnu/stage2/bin/rustc \
+    ./build/x86_64-unknown-linux-gnu/stage0/bin/cargo build --release \
+    --manifest-path /tmp/cargo/Cargo.toml
+./build/x86_64-unknown-linux-gnu/llvm/bin/llvm-profdata \
+    merge -o /tmp/rustc-pgo.profdata /tmp/rustc-pgo
+
+cp /tmp/rustc-pgo.profdata ./build/dist/rustc-pgo.profdata
+
+# This produces the actual final set of artifacts
+python2.7 ../x.py dist --rust-profile-use=/tmp/rustc-pgo.profdata \
+    --host $HOSTS --target $HOSTS \
+    --include-default-paths \
+    src/tools/build-manifest

src/ci/docker/scripts/pgo.rs

@@ -0,0 +1 @@
+fn main() {}