Auto merge of rust-lang#116352 - Kobzol:rustc-driver-bolt, r=<try>

Optimize `librustc_driver.so` with BOLT

This PR optimizes `librustc_driver.so` on 64-bit Linux CI with BOLT.

### Code
One thing that's not clear yet to me how to resolve is how to best pass a linker flag that we need for BOLT (the second commit). It is currently passed unconditionally, which is not a good idea. We somehow have to:
1) Only pass it when we actually plan to use BOLT. How to best do that? `config.toml` entry? Environment variable? CLI flag for bootstrap? BOLT optimization is done by `opt-dist`, therefore bootstrap doesn't know about it by default.
2) Only pass it to `librustc_driver.so` (see performance below).

Some discussion of this flag already happened on [Zulip](https://rust-lang.zulipchat.com/#narrow/stream/326414-t-infra.2Fbootstrap/topic/Adding.20a.20one-off.20linker.20flag).

### Performance
Latest perf. results can be found [here](rust-lang#102487 (comment)). Note that instruction counts are not very interesting here, there are only regressions on hello world programs. Probably caused by a larger C++ libstd (?).

Summary:
- ✔️ `-1.8%` mean improvement in cycle counts across many primary benchmarks.
- ✔️ `-1.8%` mean Max-RSS improvement.
- ✖️ 34 MiB (+48%) artifact size regression of `librustc_driver.so`.
  - This is caused by building `librustc_driver.so` with relocations (which are required for BOLT). Hopefully, it will be [fixed](https://discourse.llvm.org/t/bolt-rfc-a-new-mode-to-rewrite-entire-binary/68674) in the future with BOLT improvements, but now trying to reduce this size increase is [tricky](rust-lang#114649).
  - Note that the size of this file was recently reduced in rust-lang#115554 by pretty much the same amount (33 MiB). So the size after this PR is basically the same as it was for the last ~year.
- ✖️ 1.4 MiB (+53%) artifact size regression of `rustc`.
  - This is annoying and pretty much unnecessary. It is caused by the way relocations are currently applied in this PR, because they are applied both to `librustc_driver.so` (where they are needed) and for `rustc` (where they aren't needed), since both are built with a single cargo invocation. We might need e.g. some tricks in the bootstrap `rustc` shim to only apply the relocation flag for the shared library and not for `rustc`.

### CI time
CI (try build) got slower by ~5 minutes, which is fine, IMO. It can be further reduced by running LLVM and `librustc_driver` BOLT profile gathering at the same time (now they are gathered separately for LLVM and `librustc_driver`).

r? `@Mark-Simulacrum`

Also CC `@onur-ozkan,` primarily for the bootstrap linker flag issue.

Author: bors

src/bootstrap/bin/rustc.rs

@@ -217,6 +217,12 @@ fn main() {
         eprintln!("{prefix} libdir: {libdir:?}");
     }
 
+    if env::var_os("RUSTC_BOLT_LINK_FLAGS").is_some() {
+        if let Some("rustc_driver") = crate_name {
+            cmd.arg("-Clink-args=-Wl,-q");
+        }
+    }
+
     let start = Instant::now();
     let (child, status) = {
         let errmsg = format!("\nFailed to run:\n{cmd:?}\n-------------");

src/bootstrap/compile.rs

@@ -906,6 +906,11 @@ impl Step for Rustc {
             cargo.arg("-p").arg(krate);
         }
 
+        if builder.build.config.enable_bolt_settings && compiler.stage == 1 {
+            // Relocations are required for BOLT to work.k
+            cargo.env("RUSTC_BOLT_LINK_FLAGS", "1");
+        }
+
         let _guard = builder.msg_sysroot_tool(
             Kind::Build,
             compiler.stage,

src/bootstrap/config.rs

@@ -234,6 +234,7 @@ pub struct Config {
     pub llvm_profile_use: Option<String>,
     pub llvm_profile_generate: bool,
     pub llvm_libunwind_default: Option<LlvmLibunwind>,
+    pub enable_bolt_settings: bool,
 
     pub reproducible_artifacts: Vec<String>,
 
@@ -1128,6 +1129,7 @@ impl Config {
         config.free_args = std::mem::take(&mut flags.free_args);
         config.llvm_profile_use = flags.llvm_profile_use;
         config.llvm_profile_generate = flags.llvm_profile_generate;
+        config.enable_bolt_settings = flags.enable_bolt_settings;
 
         // Infer the rest of the configuration.
 

src/bootstrap/flags.rs

@@ -152,6 +152,9 @@ pub struct Flags {
     /// generate PGO profile with llvm built for rustc
     #[arg(global(true), long)]
     pub llvm_profile_generate: bool,
+    /// Enable BOLT link flags
+    #[arg(global(true), long)]
+    pub enable_bolt_settings: bool,
     /// Additional reproducible artifacts that should be added to the reproducible artifacts archive.
     #[arg(global(true), long)]
     pub reproducible_artifact: Vec<String>,

src/ci/docker/host-x86_64/dist-x86_64-linux/build-gcc.sh

@@ -27,10 +27,15 @@ sed -i'' 's|ftp://gcc\.gnu\.org/|https://gcc.gnu.org/|g' ./contrib/download_prer
 ./contrib/download_prerequisites
 mkdir ../gcc-build
 cd ../gcc-build
+
+# '-fno-reorder-blocks-and-partition' is required to
+# enable BOLT optimization of the C++ standard library,
+# which is included in librustc_driver.so
 hide_output ../gcc-$GCC/configure \
     --prefix=/rustroot \
     --enable-languages=c,c++ \
-    --disable-gnu-unique-object
+    --disable-gnu-unique-object \
+    --enable-cxx-flags='-fno-reorder-blocks-and-partition'
 hide_output make -j$(nproc)
 hide_output make install
 ln -s gcc /rustroot/bin/cc

src/etc/completions/x.py.fish

@@ -27,6 +27,7 @@ complete -c x.py -n "__fish_use_subcommand" -l include-default-paths -d 'include
 complete -c x.py -n "__fish_use_subcommand" -l dry-run -d 'dry run; don\'t build anything'
 complete -c x.py -n "__fish_use_subcommand" -l json-output -d 'use message-format=json'
 complete -c x.py -n "__fish_use_subcommand" -l llvm-profile-generate -d 'generate PGO profile with llvm built for rustc'
+complete -c x.py -n "__fish_use_subcommand" -l enable-bolt-settings -d 'Enable BOLT link flags'
 complete -c x.py -n "__fish_use_subcommand" -s h -l help -d 'Print help'
 complete -c x.py -n "__fish_use_subcommand" -f -a "build" -d 'Compile either the compiler or libraries'
 complete -c x.py -n "__fish_use_subcommand" -f -a "check" -d 'Compile either the compiler or libraries, using cargo check'
@@ -71,6 +72,7 @@ complete -c x.py -n "__fish_seen_subcommand_from build" -l include-default-paths
 complete -c x.py -n "__fish_seen_subcommand_from build" -l dry-run -d 'dry run; don\'t build anything'
 complete -c x.py -n "__fish_seen_subcommand_from build" -l json-output -d 'use message-format=json'
 complete -c x.py -n "__fish_seen_subcommand_from build" -l llvm-profile-generate -d 'generate PGO profile with llvm built for rustc'
+complete -c x.py -n "__fish_seen_subcommand_from build" -l enable-bolt-settings -d 'Enable BOLT link flags'
 complete -c x.py -n "__fish_seen_subcommand_from build" -s h -l help -d 'Print help (see more with \'--help\')'
 complete -c x.py -n "__fish_seen_subcommand_from check" -l config -d 'TOML configuration file for build' -r -F
 complete -c x.py -n "__fish_seen_subcommand_from check" -l build-dir -d 'Build directory, overrides `build.build-dir` in `config.toml`' -r -f -a "(__fish_complete_directories)"
@@ -102,6 +104,7 @@ complete -c x.py -n "__fish_seen_subcommand_from check" -l include-default-paths
 complete -c x.py -n "__fish_seen_subcommand_from check" -l dry-run -d 'dry run; don\'t build anything'
 complete -c x.py -n "__fish_seen_subcommand_from check" -l json-output -d 'use message-format=json'
 complete -c x.py -n "__fish_seen_subcommand_from check" -l llvm-profile-generate -d 'generate PGO profile with llvm built for rustc'
+complete -c x.py -n "__fish_seen_subcommand_from check" -l enable-bolt-settings -d 'Enable BOLT link flags'
 complete -c x.py -n "__fish_seen_subcommand_from check" -s h -l help -d 'Print help (see more with \'--help\')'
 complete -c x.py -n "__fish_seen_subcommand_from clippy" -s A -d 'clippy lints to allow' -r
 complete -c x.py -n "__fish_seen_subcommand_from clippy" -s D -d 'clippy lints to deny' -r
@@ -137,6 +140,7 @@ complete -c x.py -n "__fish_seen_subcommand_from clippy" -l include-default-path
 complete -c x.py -n "__fish_seen_subcommand_from clippy" -l dry-run -d 'dry run; don\'t build anything'
 complete -c x.py -n "__fish_seen_subcommand_from clippy" -l json-output -d 'use message-format=json'
 complete -c x.py -n "__fish_seen_subcommand_from clippy" -l llvm-profile-generate -d 'generate PGO profile with llvm built for rustc'
+complete -c x.py -n "__fish_seen_subcommand_from clippy" -l enable-bolt-settings -d 'Enable BOLT link flags'
 complete -c x.py -n "__fish_seen_subcommand_from clippy" -s h -l help -d 'Print help (see more with \'--help\')'
 complete -c x.py -n "__fish_seen_subcommand_from fix" -l config -d 'TOML configuration file for build' -r -F
 complete -c x.py -n "__fish_seen_subcommand_from fix" -l build-dir -d 'Build directory, overrides `build.build-dir` in `config.toml`' -r -f -a "(__fish_complete_directories)"
@@ -167,6 +171,7 @@ complete -c x.py -n "__fish_seen_subcommand_from fix" -l include-default-paths -
 complete -c x.py -n "__fish_seen_subcommand_from fix" -l dry-run -d 'dry run; don\'t build anything'
 complete -c x.py -n "__fish_seen_subcommand_from fix" -l json-output -d 'use message-format=json'
 complete -c x.py -n "__fish_seen_subcommand_from fix" -l llvm-profile-generate -d 'generate PGO profile with llvm built for rustc'
+complete -c x.py -n "__fish_seen_subcommand_from fix" -l enable-bolt-settings -d 'Enable BOLT link flags'
 complete -c x.py -n "__fish_seen_subcommand_from fix" -s h -l help -d 'Print help (see more with \'--help\')'
 complete -c x.py -n "__fish_seen_subcommand_from fmt" -l config -d 'TOML configuration file for build' -r -F
 complete -c x.py -n "__fish_seen_subcommand_from fmt" -l build-dir -d 'Build directory, overrides `build.build-dir` in `config.toml`' -r -f -a "(__fish_complete_directories)"
@@ -198,6 +203,7 @@ complete -c x.py -n "__fish_seen_subcommand_from fmt" -l include-default-paths -
 complete -c x.py -n "__fish_seen_subcommand_from fmt" -l dry-run -d 'dry run; don\'t build anything'
 complete -c x.py -n "__fish_seen_subcommand_from fmt" -l json-output -d 'use message-format=json'
 complete -c x.py -n "__fish_seen_subcommand_from fmt" -l llvm-profile-generate -d 'generate PGO profile with llvm built for rustc'
+complete -c x.py -n "__fish_seen_subcommand_from fmt" -l enable-bolt-settings -d 'Enable BOLT link flags'
 complete -c x.py -n "__fish_seen_subcommand_from fmt" -s h -l help -d 'Print help (see more with \'--help\')'
 complete -c x.py -n "__fish_seen_subcommand_from doc" -l config -d 'TOML configuration file for build' -r -F
 complete -c x.py -n "__fish_seen_subcommand_from doc" -l build-dir -d 'Build directory, overrides `build.build-dir` in `config.toml`' -r -f -a "(__fish_complete_directories)"
@@ -230,6 +236,7 @@ complete -c x.py -n "__fish_seen_subcommand_from doc" -l include-default-paths -
 complete -c x.py -n "__fish_seen_subcommand_from doc" -l dry-run -d 'dry run; don\'t build anything'
 complete -c x.py -n "__fish_seen_subcommand_from doc" -l json-output -d 'use message-format=json'
 complete -c x.py -n "__fish_seen_subcommand_from doc" -l llvm-profile-generate -d 'generate PGO profile with llvm built for rustc'
+complete -c x.py -n "__fish_seen_subcommand_from doc" -l enable-bolt-settings -d 'Enable BOLT link flags'
 complete -c x.py -n "__fish_seen_subcommand_from doc" -s h -l help -d 'Print help (see more with \'--help\')'
 complete -c x.py -n "__fish_seen_subcommand_from test" -l skip -d 'skips tests matching SUBSTRING, if supported by test tool. May be passed multiple times' -r -F
 complete -c x.py -n "__fish_seen_subcommand_from test" -l test-args -d 'extra arguments to be passed for the test tool being used (e.g. libtest, compiletest or rustdoc)' -r
@@ -273,6 +280,7 @@ complete -c x.py -n "__fish_seen_subcommand_from test" -l include-default-paths
 complete -c x.py -n "__fish_seen_subcommand_from test" -l dry-run -d 'dry run; don\'t build anything'
 complete -c x.py -n "__fish_seen_subcommand_from test" -l json-output -d 'use message-format=json'
 complete -c x.py -n "__fish_seen_subcommand_from test" -l llvm-profile-generate -d 'generate PGO profile with llvm built for rustc'
+complete -c x.py -n "__fish_seen_subcommand_from test" -l enable-bolt-settings -d 'Enable BOLT link flags'
 complete -c x.py -n "__fish_seen_subcommand_from test" -s h -l help -d 'Print help (see more with \'--help\')'
 complete -c x.py -n "__fish_seen_subcommand_from bench" -l test-args -r
 complete -c x.py -n "__fish_seen_subcommand_from bench" -l config -d 'TOML configuration file for build' -r -F
@@ -304,6 +312,7 @@ complete -c x.py -n "__fish_seen_subcommand_from bench" -l include-default-paths
 complete -c x.py -n "__fish_seen_subcommand_from bench" -l dry-run -d 'dry run; don\'t build anything'
 complete -c x.py -n "__fish_seen_subcommand_from bench" -l json-output -d 'use message-format=json'
 complete -c x.py -n "__fish_seen_subcommand_from bench" -l llvm-profile-generate -d 'generate PGO profile with llvm built for rustc'
+complete -c x.py -n "__fish_seen_subcommand_from bench" -l enable-bolt-settings -d 'Enable BOLT link flags'
 complete -c x.py -n "__fish_seen_subcommand_from bench" -s h -l help -d 'Print help'
 complete -c x.py -n "__fish_seen_subcommand_from clean" -l stage -d 'Clean a specific stage without touching other artifacts. By default, every stage is cleaned if this option is not used' -r
 complete -c x.py -n "__fish_seen_subcommand_from clean" -l config -d 'TOML configuration file for build' -r -F
@@ -335,6 +344,7 @@ complete -c x.py -n "__fish_seen_subcommand_from clean" -l include-default-paths
 complete -c x.py -n "__fish_seen_subcommand_from clean" -l dry-run -d 'dry run; don\'t build anything'
 complete -c x.py -n "__fish_seen_subcommand_from clean" -l json-output -d 'use message-format=json'
 complete -c x.py -n "__fish_seen_subcommand_from clean" -l llvm-profile-generate -d 'generate PGO profile with llvm built for rustc'
+complete -c x.py -n "__fish_seen_subcommand_from clean" -l enable-bolt-settings -d 'Enable BOLT link flags'
 complete -c x.py -n "__fish_seen_subcommand_from clean" -s h -l help -d 'Print help'
 complete -c x.py -n "__fish_seen_subcommand_from dist" -l config -d 'TOML configuration file for build' -r -F
 complete -c x.py -n "__fish_seen_subcommand_from dist" -l build-dir -d 'Build directory, overrides `build.build-dir` in `config.toml`' -r -f -a "(__fish_complete_directories)"
@@ -365,6 +375,7 @@ complete -c x.py -n "__fish_seen_subcommand_from dist" -l include-default-paths
 complete -c x.py -n "__fish_seen_subcommand_from dist" -l dry-run -d 'dry run; don\'t build anything'
 complete -c x.py -n "__fish_seen_subcommand_from dist" -l json-output -d 'use message-format=json'
 complete -c x.py -n "__fish_seen_subcommand_from dist" -l llvm-profile-generate -d 'generate PGO profile with llvm built for rustc'
+complete -c x.py -n "__fish_seen_subcommand_from dist" -l enable-bolt-settings -d 'Enable BOLT link flags'
 complete -c x.py -n "__fish_seen_subcommand_from dist" -s h -l help -d 'Print help'
 complete -c x.py -n "__fish_seen_subcommand_from install" -l config -d 'TOML configuration file for build' -r -F
 complete -c x.py -n "__fish_seen_subcommand_from install" -l build-dir -d 'Build directory, overrides `build.build-dir` in `config.toml`' -r -f -a "(__fish_complete_directories)"
@@ -395,6 +406,7 @@ complete -c x.py -n "__fish_seen_subcommand_from install" -l include-default-pat
 complete -c x.py -n "__fish_seen_subcommand_from install" -l dry-run -d 'dry run; don\'t build anything'
 complete -c x.py -n "__fish_seen_subcommand_from install" -l json-output -d 'use message-format=json'
 complete -c x.py -n "__fish_seen_subcommand_from install" -l llvm-profile-generate -d 'generate PGO profile with llvm built for rustc'
+complete -c x.py -n "__fish_seen_subcommand_from install" -l enable-bolt-settings -d 'Enable BOLT link flags'
 complete -c x.py -n "__fish_seen_subcommand_from install" -s h -l help -d 'Print help'
 complete -c x.py -n "__fish_seen_subcommand_from run" -l args -d 'arguments for the tool' -r
 complete -c x.py -n "__fish_seen_subcommand_from run" -l config -d 'TOML configuration file for build' -r -F
@@ -426,6 +438,7 @@ complete -c x.py -n "__fish_seen_subcommand_from run" -l include-default-paths -
 complete -c x.py -n "__fish_seen_subcommand_from run" -l dry-run -d 'dry run; don\'t build anything'
 complete -c x.py -n "__fish_seen_subcommand_from run" -l json-output -d 'use message-format=json'
 complete -c x.py -n "__fish_seen_subcommand_from run" -l llvm-profile-generate -d 'generate PGO profile with llvm built for rustc'
+complete -c x.py -n "__fish_seen_subcommand_from run" -l enable-bolt-settings -d 'Enable BOLT link flags'
 complete -c x.py -n "__fish_seen_subcommand_from run" -s h -l help -d 'Print help (see more with \'--help\')'
 complete -c x.py -n "__fish_seen_subcommand_from setup" -l config -d 'TOML configuration file for build' -r -F
 complete -c x.py -n "__fish_seen_subcommand_from setup" -l build-dir -d 'Build directory, overrides `build.build-dir` in `config.toml`' -r -f -a "(__fish_complete_directories)"
@@ -456,6 +469,7 @@ complete -c x.py -n "__fish_seen_subcommand_from setup" -l include-default-paths
 complete -c x.py -n "__fish_seen_subcommand_from setup" -l dry-run -d 'dry run; don\'t build anything'
 complete -c x.py -n "__fish_seen_subcommand_from setup" -l json-output -d 'use message-format=json'
 complete -c x.py -n "__fish_seen_subcommand_from setup" -l llvm-profile-generate -d 'generate PGO profile with llvm built for rustc'
+complete -c x.py -n "__fish_seen_subcommand_from setup" -l enable-bolt-settings -d 'Enable BOLT link flags'
 complete -c x.py -n "__fish_seen_subcommand_from setup" -s h -l help -d 'Print help (see more with \'--help\')'
 complete -c x.py -n "__fish_seen_subcommand_from suggest" -l config -d 'TOML configuration file for build' -r -F
 complete -c x.py -n "__fish_seen_subcommand_from suggest" -l build-dir -d 'Build directory, overrides `build.build-dir` in `config.toml`' -r -f -a "(__fish_complete_directories)"
@@ -487,4 +501,5 @@ complete -c x.py -n "__fish_seen_subcommand_from suggest" -l include-default-pat
 complete -c x.py -n "__fish_seen_subcommand_from suggest" -l dry-run -d 'dry run; don\'t build anything'
 complete -c x.py -n "__fish_seen_subcommand_from suggest" -l json-output -d 'use message-format=json'
 complete -c x.py -n "__fish_seen_subcommand_from suggest" -l llvm-profile-generate -d 'generate PGO profile with llvm built for rustc'
+complete -c x.py -n "__fish_seen_subcommand_from suggest" -l enable-bolt-settings -d 'Enable BOLT link flags'
 complete -c x.py -n "__fish_seen_subcommand_from suggest" -s h -l help -d 'Print help (see more with \'--help\')'

src/etc/completions/x.py.ps1

@@ -53,6 +53,7 @@ Register-ArgumentCompleter -Native -CommandName 'x.py' -ScriptBlock {
             [CompletionResult]::new('--dry-run', 'dry-run', [CompletionResultType]::ParameterName, 'dry run; don''t build anything')
             [CompletionResult]::new('--json-output', 'json-output', [CompletionResultType]::ParameterName, 'use message-format=json')
             [CompletionResult]::new('--llvm-profile-generate', 'llvm-profile-generate', [CompletionResultType]::ParameterName, 'generate PGO profile with llvm built for rustc')
+            [CompletionResult]::new('--enable-bolt-settings', 'enable-bolt-settings', [CompletionResultType]::ParameterName, 'Enable BOLT link flags')
             [CompletionResult]::new('-h', 'h', [CompletionResultType]::ParameterName, 'Print help')
             [CompletionResult]::new('--help', 'help', [CompletionResultType]::ParameterName, 'Print help')
             [CompletionResult]::new('build', 'build', [CompletionResultType]::ParameterValue, 'Compile either the compiler or libraries')
@@ -104,6 +105,7 @@ Register-ArgumentCompleter -Native -CommandName 'x.py' -ScriptBlock {
             [CompletionResult]::new('--dry-run', 'dry-run', [CompletionResultType]::ParameterName, 'dry run; don''t build anything')
             [CompletionResult]::new('--json-output', 'json-output', [CompletionResultType]::ParameterName, 'use message-format=json')
             [CompletionResult]::new('--llvm-profile-generate', 'llvm-profile-generate', [CompletionResultType]::ParameterName, 'generate PGO profile with llvm built for rustc')
+            [CompletionResult]::new('--enable-bolt-settings', 'enable-bolt-settings', [CompletionResultType]::ParameterName, 'Enable BOLT link flags')
             [CompletionResult]::new('-h', 'h', [CompletionResultType]::ParameterName, 'Print help (see more with ''--help'')')
             [CompletionResult]::new('--help', 'help', [CompletionResultType]::ParameterName, 'Print help (see more with ''--help'')')
             break
@@ -142,6 +144,7 @@ Register-ArgumentCompleter -Native -CommandName 'x.py' -ScriptBlock {
             [CompletionResult]::new('--dry-run', 'dry-run', [CompletionResultType]::ParameterName, 'dry run; don''t build anything')
             [CompletionResult]::new('--json-output', 'json-output', [CompletionResultType]::ParameterName, 'use message-format=json')
             [CompletionResult]::new('--llvm-profile-generate', 'llvm-profile-generate', [CompletionResultType]::ParameterName, 'generate PGO profile with llvm built for rustc')
+            [CompletionResult]::new('--enable-bolt-settings', 'enable-bolt-settings', [CompletionResultType]::ParameterName, 'Enable BOLT link flags')
             [CompletionResult]::new('-h', 'h', [CompletionResultType]::ParameterName, 'Print help (see more with ''--help'')')
             [CompletionResult]::new('--help', 'help', [CompletionResultType]::ParameterName, 'Print help (see more with ''--help'')')
             break
@@ -184,6 +187,7 @@ Register-ArgumentCompleter -Native -CommandName 'x.py' -ScriptBlock {
             [CompletionResult]::new('--dry-run', 'dry-run', [CompletionResultType]::ParameterName, 'dry run; don''t build anything')
             [CompletionResult]::new('--json-output', 'json-output', [CompletionResultType]::ParameterName, 'use message-format=json')
             [CompletionResult]::new('--llvm-profile-generate', 'llvm-profile-generate', [CompletionResultType]::ParameterName, 'generate PGO profile with llvm built for rustc')
+            [CompletionResult]::new('--enable-bolt-settings', 'enable-bolt-settings', [CompletionResultType]::ParameterName, 'Enable BOLT link flags')
             [CompletionResult]::new('-h', 'h', [CompletionResultType]::ParameterName, 'Print help (see more with ''--help'')')
             [CompletionResult]::new('--help', 'help', [CompletionResultType]::ParameterName, 'Print help (see more with ''--help'')')
             break
@@ -221,6 +225,7 @@ Register-ArgumentCompleter -Native -CommandName 'x.py' -ScriptBlock {
             [CompletionResult]::new('--dry-run', 'dry-run', [CompletionResultType]::ParameterName, 'dry run; don''t build anything')
             [CompletionResult]::new('--json-output', 'json-output', [CompletionResultType]::ParameterName, 'use message-format=json')
             [CompletionResult]::new('--llvm-profile-generate', 'llvm-profile-generate', [CompletionResultType]::ParameterName, 'generate PGO profile with llvm built for rustc')
+            [CompletionResult]::new('--enable-bolt-settings', 'enable-bolt-settings', [CompletionResultType]::ParameterName, 'Enable BOLT link flags')
             [CompletionResult]::new('-h', 'h', [CompletionResultType]::ParameterName, 'Print help (see more with ''--help'')')
             [CompletionResult]::new('--help', 'help', [CompletionResultType]::ParameterName, 'Print help (see more with ''--help'')')
             break
@@ -259,6 +264,7 @@ Register-ArgumentCompleter -Native -CommandName 'x.py' -ScriptBlock {
             [CompletionResult]::new('--dry-run', 'dry-run', [CompletionResultType]::ParameterName, 'dry run; don''t build anything')
             [CompletionResult]::new('--json-output', 'json-output', [CompletionResultType]::ParameterName, 'use message-format=json')
             [CompletionResult]::new('--llvm-profile-generate', 'llvm-profile-generate', [CompletionResultType]::ParameterName, 'generate PGO profile with llvm built for rustc')
+            [CompletionResult]::new('--enable-bolt-settings', 'enable-bolt-settings', [CompletionResultType]::ParameterName, 'Enable BOLT link flags')
             [CompletionResult]::new('-h', 'h', [CompletionResultType]::ParameterName, 'Print help (see more with ''--help'')')
             [CompletionResult]::new('--help', 'help', [CompletionResultType]::ParameterName, 'Print help (see more with ''--help'')')
             break
@@ -298,6 +304,7 @@ Register-ArgumentCompleter -Native -CommandName 'x.py' -ScriptBlock {
             [CompletionResult]::new('--dry-run', 'dry-run', [CompletionResultType]::ParameterName, 'dry run; don''t build anything')
             [CompletionResult]::new('--json-output', 'json-output', [CompletionResultType]::ParameterName, 'use message-format=json')
             [CompletionResult]::new('--llvm-profile-generate', 'llvm-profile-generate', [CompletionResultType]::ParameterName, 'generate PGO profile with llvm built for rustc')
+            [CompletionResult]::new('--enable-bolt-settings', 'enable-bolt-settings', [CompletionResultType]::ParameterName, 'Enable BOLT link flags')
             [CompletionResult]::new('-h', 'h', [CompletionResultType]::ParameterName, 'Print help (see more with ''--help'')')
             [CompletionResult]::new('--help', 'help', [CompletionResultType]::ParameterName, 'Print help (see more with ''--help'')')
             break
@@ -348,6 +355,7 @@ Register-ArgumentCompleter -Native -CommandName 'x.py' -ScriptBlock {
             [CompletionResult]::new('--dry-run', 'dry-run', [CompletionResultType]::ParameterName, 'dry run; don''t build anything')
             [CompletionResult]::new('--json-output', 'json-output', [CompletionResultType]::ParameterName, 'use message-format=json')
             [CompletionResult]::new('--llvm-profile-generate', 'llvm-profile-generate', [CompletionResultType]::ParameterName, 'generate PGO profile with llvm built for rustc')
+            [CompletionResult]::new('--enable-bolt-settings', 'enable-bolt-settings', [CompletionResultType]::ParameterName, 'Enable BOLT link flags')
             [CompletionResult]::new('-h', 'h', [CompletionResultType]::ParameterName, 'Print help (see more with ''--help'')')
             [CompletionResult]::new('--help', 'help', [CompletionResultType]::ParameterName, 'Print help (see more with ''--help'')')
             break
@@ -386,6 +394,7 @@ Register-ArgumentCompleter -Native -CommandName 'x.py' -ScriptBlock {
             [CompletionResult]::new('--dry-run', 'dry-run', [CompletionResultType]::ParameterName, 'dry run; don''t build anything')
             [CompletionResult]::new('--json-output', 'json-output', [CompletionResultType]::ParameterName, 'use message-format=json')
             [CompletionResult]::new('--llvm-profile-generate', 'llvm-profile-generate', [CompletionResultType]::ParameterName, 'generate PGO profile with llvm built for rustc')
+            [CompletionResult]::new('--enable-bolt-settings', 'enable-bolt-settings', [CompletionResultType]::ParameterName, 'Enable BOLT link flags')
             [CompletionResult]::new('-h', 'h', [CompletionResultType]::ParameterName, 'Print help')
             [CompletionResult]::new('--help', 'help', [CompletionResultType]::ParameterName, 'Print help')
             break
@@ -424,6 +433,7 @@ Register-ArgumentCompleter -Native -CommandName 'x.py' -ScriptBlock {
             [CompletionResult]::new('--dry-run', 'dry-run', [CompletionResultType]::ParameterName, 'dry run; don''t build anything')
             [CompletionResult]::new('--json-output', 'json-output', [CompletionResultType]::ParameterName, 'use message-format=json')
             [CompletionResult]::new('--llvm-profile-generate', 'llvm-profile-generate', [CompletionResultType]::ParameterName, 'generate PGO profile with llvm built for rustc')
+            [CompletionResult]::new('--enable-bolt-settings', 'enable-bolt-settings', [CompletionResultType]::ParameterName, 'Enable BOLT link flags')
             [CompletionResult]::new('-h', 'h', [CompletionResultType]::ParameterName, 'Print help')
             [CompletionResult]::new('--help', 'help', [CompletionResultType]::ParameterName, 'Print help')
             break
@@ -461,6 +471,7 @@ Register-ArgumentCompleter -Native -CommandName 'x.py' -ScriptBlock {
             [CompletionResult]::new('--dry-run', 'dry-run', [CompletionResultType]::ParameterName, 'dry run; don''t build anything')
             [CompletionResult]::new('--json-output', 'json-output', [CompletionResultType]::ParameterName, 'use message-format=json')
             [CompletionResult]::new('--llvm-profile-generate', 'llvm-profile-generate', [CompletionResultType]::ParameterName, 'generate PGO profile with llvm built for rustc')
+            [CompletionResult]::new('--enable-bolt-settings', 'enable-bolt-settings', [CompletionResultType]::ParameterName, 'Enable BOLT link flags')
             [CompletionResult]::new('-h', 'h', [CompletionResultType]::ParameterName, 'Print help')
             [CompletionResult]::new('--help', 'help', [CompletionResultType]::ParameterName, 'Print help')
             break
@@ -498,6 +509,7 @@ Register-ArgumentCompleter -Native -CommandName 'x.py' -ScriptBlock {
             [CompletionResult]::new('--dry-run', 'dry-run', [CompletionResultType]::ParameterName, 'dry run; don''t build anything')
             [CompletionResult]::new('--json-output', 'json-output', [CompletionResultType]::ParameterName, 'use message-format=json')
             [CompletionResult]::new('--llvm-profile-generate', 'llvm-profile-generate', [CompletionResultType]::ParameterName, 'generate PGO profile with llvm built for rustc')
+            [CompletionResult]::new('--enable-bolt-settings', 'enable-bolt-settings', [CompletionResultType]::ParameterName, 'Enable BOLT link flags')
             [CompletionResult]::new('-h', 'h', [CompletionResultType]::ParameterName, 'Print help')
             [CompletionResult]::new('--help', 'help', [CompletionResultType]::ParameterName, 'Print help')
             break
@@ -536,6 +548,7 @@ Register-ArgumentCompleter -Native -CommandName 'x.py' -ScriptBlock {
             [CompletionResult]::new('--dry-run', 'dry-run', [CompletionResultType]::ParameterName, 'dry run; don''t build anything')
             [CompletionResult]::new('--json-output', 'json-output', [CompletionResultType]::ParameterName, 'use message-format=json')
             [CompletionResult]::new('--llvm-profile-generate', 'llvm-profile-generate', [CompletionResultType]::ParameterName, 'generate PGO profile with llvm built for rustc')
+            [CompletionResult]::new('--enable-bolt-settings', 'enable-bolt-settings', [CompletionResultType]::ParameterName, 'Enable BOLT link flags')
             [CompletionResult]::new('-h', 'h', [CompletionResultType]::ParameterName, 'Print help (see more with ''--help'')')
             [CompletionResult]::new('--help', 'help', [CompletionResultType]::ParameterName, 'Print help (see more with ''--help'')')
             break
@@ -573,6 +586,7 @@ Register-ArgumentCompleter -Native -CommandName 'x.py' -ScriptBlock {
             [CompletionResult]::new('--dry-run', 'dry-run', [CompletionResultType]::ParameterName, 'dry run; don''t build anything')
             [CompletionResult]::new('--json-output', 'json-output', [CompletionResultType]::ParameterName, 'use message-format=json')
             [CompletionResult]::new('--llvm-profile-generate', 'llvm-profile-generate', [CompletionResultType]::ParameterName, 'generate PGO profile with llvm built for rustc')
+            [CompletionResult]::new('--enable-bolt-settings', 'enable-bolt-settings', [CompletionResultType]::ParameterName, 'Enable BOLT link flags')
             [CompletionResult]::new('-h', 'h', [CompletionResultType]::ParameterName, 'Print help (see more with ''--help'')')
             [CompletionResult]::new('--help', 'help', [CompletionResultType]::ParameterName, 'Print help (see more with ''--help'')')
             break
@@ -611,6 +625,7 @@ Register-ArgumentCompleter -Native -CommandName 'x.py' -ScriptBlock {
             [CompletionResult]::new('--dry-run', 'dry-run', [CompletionResultType]::ParameterName, 'dry run; don''t build anything')
             [CompletionResult]::new('--json-output', 'json-output', [CompletionResultType]::ParameterName, 'use message-format=json')
             [CompletionResult]::new('--llvm-profile-generate', 'llvm-profile-generate', [CompletionResultType]::ParameterName, 'generate PGO profile with llvm built for rustc')
+            [CompletionResult]::new('--enable-bolt-settings', 'enable-bolt-settings', [CompletionResultType]::ParameterName, 'Enable BOLT link flags')
             [CompletionResult]::new('-h', 'h', [CompletionResultType]::ParameterName, 'Print help (see more with ''--help'')')
             [CompletionResult]::new('--help', 'help', [CompletionResultType]::ParameterName, 'Print help (see more with ''--help'')')
             break

src/etc/completions/x.py.sh

@@ -1,14 +1,14 @@
 use anyhow::Context;
 
 use crate::exec::cmd;
-use crate::training::LlvmBoltProfile;
+use crate::training::BoltProfile;
 use camino::{Utf8Path, Utf8PathBuf};
 
 use crate::utils::io::copy_file;
 
 /// Instruments an artifact at the given `path` (in-place) with BOLT and then calls `func`.
 /// After this function finishes, the original file will be restored.
-pub fn with_bolt_instrumented<F: FnOnce() -> anyhow::Result<R>, R>(
+pub fn with_bolt_instrumented<F: FnOnce(&Utf8Path) -> anyhow::Result<R>, R>(
     path: &Utf8Path,
     func: F,
 ) -> anyhow::Result<R> {
@@ -20,10 +20,16 @@ pub fn with_bolt_instrumented<F: FnOnce() -> anyhow::Result<R>, R>(
 
     let instrumented_path = tempfile::NamedTempFile::new()?.into_temp_path();
 
+    let profile_dir =
+        tempfile::TempDir::new().context("Could not create directory for BOLT profiles")?;
+    let profile_prefix = profile_dir.path().join("prof.fdata");
+    let profile_prefix = Utf8Path::from_path(&profile_prefix).unwrap();
+
     // Instrument the original file with BOLT, saving the result into `instrumented_path`
     cmd(&["llvm-bolt"])
         .arg("-instrument")
         .arg(path)
+        .arg(&format!("--instrumentation-file={profile_prefix}"))
         // Make sure that each process will write its profiles into a separate file
         .arg("--instrumentation-file-append-pid")
         .arg("-o")
@@ -36,11 +42,11 @@ pub fn with_bolt_instrumented<F: FnOnce() -> anyhow::Result<R>, R>(
 
     // Run the function that will make use of the instrumented artifact.
     // The original file will be restored when `_backup_file` is dropped.
-    func()
+    func(profile_prefix)
 }
 
 /// Optimizes the file at `path` with BOLT in-place using the given `profile`.
-pub fn bolt_optimize(path: &Utf8Path, profile: &LlvmBoltProfile) -> anyhow::Result<()> {
+pub fn bolt_optimize(path: &Utf8Path, profile: &BoltProfile) -> anyhow::Result<()> {
     // Copy the artifact to a new location, so that we do not use the same input and output file.
     // BOLT cannot handle optimizing when the input and output is the same file, because it performs
     // in-place patching.

src/tools/opt-dist/src/bolt.rs

@@ -1,7 +1,7 @@
 use crate::environment::Environment;
 use crate::metrics::{load_metrics, record_metrics};
 use crate::timer::TimerSection;
-use crate::training::{LlvmBoltProfile, LlvmPGOProfile, RustcPGOProfile};
+use crate::training::{BoltProfile, LlvmPGOProfile, RustcPGOProfile};
 use camino::{Utf8Path, Utf8PathBuf};
 use std::collections::BTreeMap;
 use std::fs::File;
@@ -159,7 +159,12 @@ impl Bootstrap {
         self
     }
 
-    pub fn with_bolt_profile(mut self, profile: LlvmBoltProfile) -> Self {
+    pub fn with_rustc_bolt_ldflags(mut self) -> Self {
+        self.cmd = self.cmd.arg("--enable-bolt-settings");
+        self
+    }
+
+    pub fn with_bolt_profile(mut self, profile: BoltProfile) -> Self {
         self.cmd = self.cmd.arg("--reproducible-artifact").arg(profile.0.as_str());
         self
     }

src/tools/opt-dist/src/exec.rs

@@ -12,7 +12,10 @@ use crate::environment::{Environment, EnvironmentBuilder};
 use crate::exec::{cmd, Bootstrap};
 use crate::tests::run_tests;
 use crate::timer::Timer;
-use crate::training::{gather_llvm_bolt_profiles, gather_llvm_profiles, gather_rustc_profiles};
+use crate::training::{
+    gather_bolt_profiles, gather_llvm_profiles, gather_rustc_profiles, llvm_benchmarks,
+    rustc_benchmarks,
+};
 use crate::utils::artifact_size::print_binary_sizes;
 use crate::utils::io::{copy_directory, move_directory, reset_directory};
 use crate::utils::{
@@ -246,31 +249,33 @@ fn execute_pipeline(
         Ok(profile)
     })?;
 
-    let llvm_bolt_profile = if env.use_bolt() {
+    let bolt_profiles = if env.use_bolt() {
         // Stage 3: Build BOLT instrumented LLVM
         // We build a PGO optimized LLVM in this step, then instrument it with BOLT and gather BOLT profiles.
         // Note that we don't remove LLVM artifacts after this step, so that they are reused in the final dist build.
         // BOLT instrumentation is performed "on-the-fly" when the LLVM library is copied to the sysroot of rustc,
         // therefore the LLVM artifacts on disk are not "tainted" with BOLT instrumentation and they can be reused.
-        timer.section("Stage 3 (LLVM BOLT)", |stage| {
+        timer.section("Stage 3 (BOLT)", |stage| {
             stage.section("Build PGO optimized LLVM", |stage| {
                 Bootstrap::build(env)
                     .with_llvm_bolt_ldflags()
+                    .with_rustc_bolt_ldflags()
                     .llvm_pgo_optimize(&llvm_pgo_profile)
                     .avoid_rustc_rebuild()
                     .run(stage)
             })?;
 
-            // Find the path to the `libLLVM.so` file
-            let llvm_lib = io::find_file_in_dir(
-                &env.build_artifacts().join("stage2").join("lib"),
-                "libLLVM",
-                ".so",
-            )?;
+            let libdir = env.build_artifacts().join("stage2").join("lib");
+            let llvm_lib = io::find_file_in_dir(&libdir, "libLLVM", ".so")?;
 
-            // Instrument it and gather profiles
-            let profile = with_bolt_instrumented(&llvm_lib, || {
-                stage.section("Gather profiles", |_| gather_llvm_bolt_profiles(env))
+            log::info!("Optimizing {llvm_lib} with BOLT");
+
+            // FIXME(kobzol: try gather profiles together, at once for LLVM and rustc
+            // Instrument the libraries and gather profiles
+            let llvm_profile = with_bolt_instrumented(&llvm_lib, |llvm_profile_dir| {
+                stage.section("Gather profiles", |_| {
+                    gather_bolt_profiles(env, "LLVM", llvm_benchmarks(env), llvm_profile_dir)
+                })
             })?;
             print_free_disk_space()?;
 
@@ -279,27 +284,43 @@ fn execute_pipeline(
             // the final dist build. However, when BOLT optimizes an artifact, it does so *in-place*,
             // therefore it will actually optimize all the hard links, which means that the final
             // packaged `libLLVM.so` file *will* be BOLT optimized.
-            bolt_optimize(&llvm_lib, &profile).context("Could not optimize LLVM with BOLT")?;
+            bolt_optimize(&llvm_lib, &llvm_profile).context("Could not optimize LLVM with BOLT")?;
+
+            let rustc_lib = io::find_file_in_dir(&libdir, "librustc_driver", ".so")?;
+
+            log::info!("Optimizing {rustc_lib} with BOLT");
+
+            // Instrument it and gather profiles
+            let rustc_profile = with_bolt_instrumented(&rustc_lib, |rustc_profile_dir| {
+                stage.section("Gather profiles", |_| {
+                    gather_bolt_profiles(env, "rustc", rustc_benchmarks(env), rustc_profile_dir)
+                })
+            })?;
+            print_free_disk_space()?;
+
+            // Now optimize the library with BOLT.
+            bolt_optimize(&rustc_lib, &rustc_profile)
+                .context("Could not optimize rustc with BOLT")?;
 
             // LLVM is not being cleared here, we want to use the BOLT-optimized LLVM
-            Ok(Some(profile))
+            Ok(vec![llvm_profile, rustc_profile])
         })?
     } else {
-        None
+        vec![]
     };
 
     let mut dist = Bootstrap::dist(env, &dist_args)
         .llvm_pgo_optimize(&llvm_pgo_profile)
         .rustc_pgo_optimize(&rustc_pgo_profile)
         .avoid_rustc_rebuild();
 
-    if let Some(llvm_bolt_profile) = llvm_bolt_profile {
-        dist = dist.with_bolt_profile(llvm_bolt_profile);
+    for bolt_profile in bolt_profiles {
+        dist = dist.with_bolt_profile(bolt_profile);
     }
 
     // Final stage: Assemble the dist artifacts
     // The previous PGO optimized rustc build and PGO optimized LLVM builds should be reused.
-    timer.section("Stage 4 (final build)", |stage| dist.run(stage))?;
+    timer.section("Stage 5 (final build)", |stage| dist.run(stage))?;
 
     // After dist has finished, run a subset of the test suite on the optimized artifacts to discover
     // possible regressions.

src/tools/opt-dist/src/main.rs

@@ -27,8 +27,6 @@ const RUSTC_PGO_CRATES: &[&str] = &[
     "bitmaps-3.1.0",
 ];
 
-const LLVM_BOLT_CRATES: &[&str] = LLVM_PGO_CRATES;
-
 fn init_compiler_benchmarks(
     env: &Environment,
     profiles: &[&str],
@@ -113,6 +111,14 @@ fn log_profile_stats(
     Ok(())
 }
 
+pub fn llvm_benchmarks(env: &Environment) -> CmdBuilder {
+    init_compiler_benchmarks(env, &["Debug", "Opt"], &["Full"], LLVM_PGO_CRATES)
+}
+
+pub fn rustc_benchmarks(env: &Environment) -> CmdBuilder {
+    init_compiler_benchmarks(env, &["Check", "Debug", "Opt"], &["All"], RUSTC_PGO_CRATES)
+}
+
 pub struct LlvmPGOProfile(pub Utf8PathBuf);
 
 pub fn gather_llvm_profiles(
@@ -122,9 +128,7 @@ pub fn gather_llvm_profiles(
     log::info!("Running benchmarks with PGO instrumented LLVM");
 
     with_log_group("Running benchmarks", || {
-        init_compiler_benchmarks(env, &["Debug", "Opt"], &["Full"], LLVM_PGO_CRATES)
-            .run()
-            .context("Cannot gather LLVM PGO profiles")
+        llvm_benchmarks(env).run().context("Cannot gather LLVM PGO profiles")
     })?;
 
     let merged_profile = env.artifact_dir().join("llvm-pgo.profdata");
@@ -157,7 +161,7 @@ pub fn gather_rustc_profiles(
     // Here we're profiling the `rustc` frontend, so we also include `Check`.
     // The benchmark set includes various stress tests that put the frontend under pressure.
     with_log_group("Running benchmarks", || {
-        init_compiler_benchmarks(env, &["Check", "Debug", "Opt"], &["All"], RUSTC_PGO_CRATES)
+        rustc_benchmarks(env)
             .env("LLVM_PROFILE_FILE", profile_template.as_str())
             .run()
             .context("Cannot gather rustc PGO profiles")
@@ -176,23 +180,25 @@ pub fn gather_rustc_profiles(
     Ok(RustcPGOProfile(merged_profile))
 }
 
-pub struct LlvmBoltProfile(pub Utf8PathBuf);
+pub struct BoltProfile(pub Utf8PathBuf);
 
-pub fn gather_llvm_bolt_profiles(env: &Environment) -> anyhow::Result<LlvmBoltProfile> {
-    log::info!("Running benchmarks with BOLT instrumented LLVM");
+pub fn gather_bolt_profiles(
+    env: &Environment,
+    name: &str,
+    benchmarks: CmdBuilder,
+    profile_prefix: &Utf8Path,
+) -> anyhow::Result<BoltProfile> {
+    log::info!("Running benchmarks with BOLT instrumented {name}");
 
     with_log_group("Running benchmarks", || {
-        init_compiler_benchmarks(env, &["Check", "Debug", "Opt"], &["Full"], LLVM_BOLT_CRATES)
-            .run()
-            .context("Cannot gather LLVM BOLT profiles")
+        benchmarks.run().with_context(|| "Cannot gather {name} BOLT profiles")
     })?;
 
-    let merged_profile = env.artifact_dir().join("llvm-bolt.profdata");
-    let profile_root = Utf8PathBuf::from("/tmp/prof.fdata");
-    log::info!("Merging LLVM BOLT profiles to {merged_profile}");
+    let merged_profile = env.artifact_dir().join(format!("{name}-bolt.profdata"));
+    log::info!("Merging {name} BOLT profiles from {profile_prefix} to {merged_profile}");
 
     let profiles: Vec<_> =
-        glob::glob(&format!("{profile_root}*"))?.collect::<Result<Vec<_>, _>>()?;
+        glob::glob(&format!("{profile_prefix}*"))?.collect::<Result<Vec<_>, _>>()?;
 
     let mut merge_args = vec!["merge-fdata"];
     merge_args.extend(profiles.iter().map(|p| p.to_str().unwrap()));
@@ -204,7 +210,7 @@ pub fn gather_llvm_bolt_profiles(env: &Environment) -> anyhow::Result<LlvmBoltPr
             .context("Cannot merge BOLT profiles")
     })?;
 
-    log::info!("LLVM BOLT statistics");
+    log::info!("{name} BOLT statistics");
     log::info!(
         "{merged_profile}: {}",
         humansize::format_size(std::fs::metadata(merged_profile.as_std_path())?.len(), BINARY)
@@ -216,8 +222,17 @@ pub fn gather_llvm_bolt_profiles(env: &Environment) -> anyhow::Result<LlvmBoltPr
         .collect::<Result<Vec<_>, _>>()?
         .into_iter()
         .sum::<u64>();
-    log::info!("{profile_root}: {}", humansize::format_size(size, BINARY));
+    log::info!("{profile_prefix}: {}", humansize::format_size(size, BINARY));
     log::info!("Profile file count: {}", profiles.len());
 
-    Ok(LlvmBoltProfile(merged_profile))
+    // Delete the gathered profiles
+    for profile in glob::glob(&format!("{profile_prefix}*"))?.into_iter() {
+        if let Ok(profile) = profile {
+            if let Err(error) = std::fs::remove_file(&profile) {
+                log::error!("Cannot delete BOLT profile {}: {error:?}", profile.display());
+            }
+        }
+    }
+
+    Ok(BoltProfile(merged_profile))
 }