Auto merge of #118568 - DianQK:no-builtins-symbols, r=pnkfelix

Avoid adding builtin functions to `symbols.o`

We found performance regressions in #113923. The problem seems to be that `--gc-sections` does not remove these symbols. I tested that lld removes these symbols, but ld and gold do not.

I found that `used` adds symbols to `symbols.o` at https://github.com/rust-lang/rust/blob/3e202ead604be31f4c1a5798a296953d3159da7e/compiler/rustc_codegen_ssa/src/back/linker.rs#L1786-L1791.
The PR removes builtin functions.

Note that under LTO, ld still preserves these symbols. (lld will still remove them.)

The first commit also fixes #118559. But I think the second commit also makes sense.

Author: bors

compiler/rustc_codegen_llvm/src/back/lto.rs

@@ -60,7 +60,7 @@ fn prepare_lto(
     };
 
     let symbol_filter = &|&(ref name, info): &(String, SymbolExportInfo)| {
-        if info.level.is_below_threshold(export_threshold) || info.used {
+        if info.level.is_below_threshold(export_threshold) || info.used || info.used_compiler {
             Some(CString::new(name.as_str()).unwrap())
         } else {
             None

compiler/rustc_codegen_ssa/src/back/symbol_export.rs

@@ -105,20 +105,21 @@ fn reachable_non_generics_provider(tcx: TyCtxt<'_>, _: LocalCrate) -> DefIdMap<S
             }
         })
         .map(|def_id| {
+            let codegen_attrs = tcx.codegen_fn_attrs(def_id.to_def_id());
             // We won't link right if this symbol is stripped during LTO.
             let name = tcx.symbol_name(Instance::mono(tcx, def_id.to_def_id())).name;
             // We have to preserve the symbols of the built-in functions during LTO.
             let is_builtin_fn = is_compiler_builtins
                 && symbol_export_level(tcx, def_id.to_def_id())
-                    .is_below_threshold(SymbolExportLevel::C);
-            let used = is_builtin_fn || name == "rust_eh_personality";
+                    .is_below_threshold(SymbolExportLevel::C)
+                && codegen_attrs.flags.contains(CodegenFnAttrFlags::NO_MANGLE);
+            let used = name == "rust_eh_personality";
 
             let export_level = if special_runtime_crate {
                 SymbolExportLevel::Rust
             } else {
                 symbol_export_level(tcx, def_id.to_def_id())
             };
-            let codegen_attrs = tcx.codegen_fn_attrs(def_id.to_def_id());
             debug!(
                 "EXPORTED SYMBOL (local): {} ({:?})",
                 tcx.symbol_name(Instance::mono(tcx, def_id.to_def_id())),
@@ -138,6 +139,7 @@ fn reachable_non_generics_provider(tcx: TyCtxt<'_>, _: LocalCrate) -> DefIdMap<S
                 used: codegen_attrs.flags.contains(CodegenFnAttrFlags::USED)
                     || codegen_attrs.flags.contains(CodegenFnAttrFlags::USED_LINKER)
                     || used,
+                used_compiler: is_builtin_fn,
             };
             (def_id.to_def_id(), info)
         })
@@ -150,6 +152,7 @@ fn reachable_non_generics_provider(tcx: TyCtxt<'_>, _: LocalCrate) -> DefIdMap<S
                 level: SymbolExportLevel::C,
                 kind: SymbolExportKind::Data,
                 used: false,
+                used_compiler: false,
             },
         );
     }
@@ -198,6 +201,7 @@ fn exported_symbols_provider_local(
                         level: info.level,
                         kind: SymbolExportKind::Text,
                         used: info.used,
+                        used_compiler: false,
                     },
                 )
             })
@@ -214,6 +218,7 @@ fn exported_symbols_provider_local(
                 level: SymbolExportLevel::C,
                 kind: SymbolExportKind::Text,
                 used: false,
+                used_compiler: false,
             },
         ));
     }
@@ -233,6 +238,7 @@ fn exported_symbols_provider_local(
                     level: SymbolExportLevel::Rust,
                     kind: SymbolExportKind::Text,
                     used: false,
+                    used_compiler: false,
                 },
             ));
         }
@@ -245,6 +251,7 @@ fn exported_symbols_provider_local(
                 level: SymbolExportLevel::Rust,
                 kind: SymbolExportKind::Data,
                 used: false,
+                used_compiler: false,
             },
         ))
     }
@@ -264,6 +271,7 @@ fn exported_symbols_provider_local(
                     level: SymbolExportLevel::C,
                     kind: SymbolExportKind::Data,
                     used: false,
+                    used_compiler: false,
                 },
             )
         }));
@@ -289,6 +297,7 @@ fn exported_symbols_provider_local(
                     level: SymbolExportLevel::C,
                     kind: SymbolExportKind::Data,
                     used: false,
+                    used_compiler: false,
                 },
             )
         }));
@@ -306,6 +315,7 @@ fn exported_symbols_provider_local(
                 level: SymbolExportLevel::C,
                 kind: SymbolExportKind::Data,
                 used: true,
+                used_compiler: false,
             },
         ));
     }
@@ -346,6 +356,7 @@ fn exported_symbols_provider_local(
                                 level: SymbolExportLevel::Rust,
                                 kind: SymbolExportKind::Text,
                                 used: false,
+                                used_compiler: false,
                             },
                         ));
                     }
@@ -362,6 +373,7 @@ fn exported_symbols_provider_local(
                             level: SymbolExportLevel::Rust,
                             kind: SymbolExportKind::Text,
                             used: false,
+                            used_compiler: false,
                         },
                     ));
                 }

compiler/rustc_middle/src/middle/exported_symbols.rs

@@ -35,7 +35,12 @@ pub enum SymbolExportKind {
 pub struct SymbolExportInfo {
     pub level: SymbolExportLevel,
     pub kind: SymbolExportKind,
+    /// Used to mark these symbols not to be internalized by LTO. These symbols
+    /// are also added to `symbols.o` to avoid circular dependencies when linking.
     pub used: bool,
+    /// Also used to mark these symbols not to be internalized by LTO. But will
+    /// not be added to `symbols.o`. Currently there are only builtin functions.
+    pub used_compiler: bool,
 }
 
 #[derive(Eq, PartialEq, Debug, Copy, Clone, TyEncodable, TyDecodable, HashStable)]

src/tools/miri/src/bin/miri.rs

@@ -165,6 +165,7 @@ impl rustc_driver::Callbacks for MiriBeRustCompilerCalls {
                                     level: SymbolExportLevel::C,
                                     kind: SymbolExportKind::Text,
                                     used: false,
+                                    used_compiler: false,
                                 },
                             ))
                         }),

tests/run-make/no-builtins-symbols/Makefile

@@ -0,0 +1,7 @@
+include ../tools.mk
+
+# only-x86_64-unknown-linux-gnu
+
+all:
+	$(RUSTC) main.rs -o $(TMPDIR)/main
+	[ "$$("$(LLVM_BIN_DIR)"/llvm-nm -U $(TMPDIR)/main | grep -c __fixunssfti)" -eq "0" ]

tests/run-make/no-builtins-symbols/main.rs

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