Rollup merge of #95202 - Urgau:check-cfg-perf-well-known-values, r=petrochenkov

Reduce the cost of loading all built-ins targets

This PR started by measuring the exact slowdown of checking of well known conditional values.
Than this PR implemented some technics to reduce the cost of loading all built-ins targets.

cf. #82450 (comment)

Author: Dylan-DPC

compiler/rustc_codegen_cranelift/src/driver/aot.rs

@@ -304,8 +304,12 @@ pub(crate) fn run_aot(
     };
 
     // FIXME handle `-Ctarget-cpu=native`
-    let target_cpu =
-        tcx.sess.opts.cg.target_cpu.as_ref().unwrap_or(&tcx.sess.target.cpu).to_owned();
+    let target_cpu = match tcx.sess.opts.cg.target_cpu {
+        Some(ref name) => name,
+        None => tcx.sess.target.cpu.as_ref(),
+    }
+    .to_owned();
+
     Box::new((
         CodegenResults {
             modules,

compiler/rustc_codegen_gcc/src/lib.rs

@@ -287,8 +287,10 @@ fn handle_native(name: &str) -> &str {
 }
 
 pub fn target_cpu(sess: &Session) -> &str {
-    let name = sess.opts.cg.target_cpu.as_ref().unwrap_or(&sess.target.cpu);
-    handle_native(name)
+    match sess.opts.cg.target_cpu {
+        Some(ref name) => handle_native(name),
+        None => handle_native(sess.target.cpu.as_ref()),
+    }
 }
 
 pub fn target_features(sess: &Session) -> Vec<Symbol> {

compiler/rustc_codegen_llvm/src/attributes.rs

@@ -116,7 +116,7 @@ fn instrument_function_attr<'ll>(cx: &CodegenCx<'ll, '_>) -> Option<&'ll Attribu
 
         // The function name varies on platforms.
         // See test/CodeGen/mcount.c in clang.
-        let mcount_name = cx.sess().target.mcount.as_str();
+        let mcount_name = cx.sess().target.mcount.as_ref();
 
         Some(llvm::CreateAttrStringValue(
             cx.llcx,

compiler/rustc_codegen_llvm/src/builder.rs

@@ -1452,7 +1452,7 @@ impl<'a, 'll, 'tcx> Builder<'a, 'll, 'tcx> {
     }
 
     fn fptoint_sat_broken_in_llvm(&self) -> bool {
-        match self.tcx.sess.target.arch.as_str() {
+        match self.tcx.sess.target.arch.as_ref() {
             // FIXME - https://bugs.llvm.org/show_bug.cgi?id=50083
             "riscv64" => llvm_util::get_version() < (13, 0, 0),
             _ => false,

compiler/rustc_codegen_llvm/src/context.rs

@@ -134,7 +134,7 @@ pub unsafe fn create_module<'ll>(
     let mod_name = SmallCStr::new(mod_name);
     let llmod = llvm::LLVMModuleCreateWithNameInContext(mod_name.as_ptr(), llcx);
 
-    let mut target_data_layout = sess.target.data_layout.clone();
+    let mut target_data_layout = sess.target.data_layout.to_string();
     let llvm_version = llvm_util::get_version();
     if llvm_version < (13, 0, 0) {
         if sess.target.arch == "powerpc64" {
@@ -859,7 +859,7 @@ impl<'ll> CodegenCx<'ll, '_> {
 
         // This isn't an "LLVM intrinsic", but LLVM's optimization passes
         // recognize it like one and we assume it exists in `core::slice::cmp`
-        match self.sess().target.arch.as_str() {
+        match self.sess().target.arch.as_ref() {
             "avr" | "msp430" => ifn!("memcmp", fn(i8p, i8p, t_isize) -> t_i16),
             _ => ifn!("memcmp", fn(i8p, i8p, t_isize) -> t_i32),
         }

compiler/rustc_codegen_llvm/src/intrinsic.rs

@@ -329,7 +329,7 @@ impl<'ll, 'tcx> IntrinsicCallMethods<'tcx> for Builder<'_, 'll, 'tcx> {
                     let b_ptr = self.bitcast(b, i8p_ty);
                     let n = self.const_usize(layout.size().bytes());
                     let cmp = self.call_intrinsic("memcmp", &[a_ptr, b_ptr, n]);
-                    match self.cx.sess().target.arch.as_str() {
+                    match self.cx.sess().target.arch.as_ref() {
                         "avr" | "msp430" => self.icmp(IntPredicate::IntEQ, cmp, self.const_i16(0)),
                         _ => self.icmp(IntPredicate::IntEQ, cmp, self.const_i32(0)),
                     }

compiler/rustc_codegen_llvm/src/llvm_util.rs

@@ -61,8 +61,8 @@ unsafe fn configure_llvm(sess: &Session) {
         full_arg.trim().split(|c: char| c == '=' || c.is_whitespace()).next().unwrap_or("")
     }
 
-    let cg_opts = sess.opts.cg.llvm_args.iter();
-    let tg_opts = sess.target.llvm_args.iter();
+    let cg_opts = sess.opts.cg.llvm_args.iter().map(AsRef::as_ref);
+    let tg_opts = sess.target.llvm_args.iter().map(AsRef::as_ref);
     let sess_args = cg_opts.chain(tg_opts);
 
     let user_specified_args: FxHashSet<_> =
@@ -375,8 +375,10 @@ fn handle_native(name: &str) -> &str {
 }
 
 pub fn target_cpu(sess: &Session) -> &str {
-    let name = sess.opts.cg.target_cpu.as_ref().unwrap_or(&sess.target.cpu);
-    handle_native(name)
+    match sess.opts.cg.target_cpu {
+        Some(ref name) => handle_native(name),
+        None => handle_native(sess.target.cpu.as_ref()),
+    }
 }
 
 /// The list of LLVM features computed from CLI flags (`-Ctarget-cpu`, `-Ctarget-feature`,

compiler/rustc_codegen_ssa/src/back/link.rs

@@ -40,6 +40,7 @@ use std::ffi::OsString;
 use std::fs::{File, OpenOptions};
 use std::io::{BufWriter, Write};
 use std::lazy::OnceCell;
+use std::ops::Deref;
 use std::path::{Path, PathBuf};
 use std::process::{ExitStatus, Output, Stdio};
 use std::{ascii, char, env, fmt, fs, io, mem, str};
@@ -674,11 +675,11 @@ fn link_natively<'a, B: ArchiveBuilder<'a>>(
 
     linker::disable_localization(&mut cmd);
 
-    for &(ref k, ref v) in &sess.target.link_env {
-        cmd.env(k, v);
+    for &(ref k, ref v) in sess.target.link_env.as_ref() {
+        cmd.env(k.as_ref(), v.as_ref());
     }
-    for k in &sess.target.link_env_remove {
-        cmd.env_remove(k);
+    for k in sess.target.link_env_remove.as_ref() {
+        cmd.env_remove(k.as_ref());
     }
 
     if sess.opts.prints.contains(&PrintRequest::LinkArgs) {
@@ -1216,7 +1217,7 @@ pub fn linker_and_flavor(sess: &Session) -> (PathBuf, LinkerFlavor) {
 
     if let Some(ret) = infer_from(
         sess,
-        sess.target.linker.clone().map(PathBuf::from),
+        sess.target.linker.as_deref().map(PathBuf::from),
         Some(sess.target.linker_flavor),
     ) {
         return ret;
@@ -1586,7 +1587,7 @@ fn add_post_link_objects(
 /// FIXME: Determine where exactly these args need to be inserted.
 fn add_pre_link_args(cmd: &mut dyn Linker, sess: &Session, flavor: LinkerFlavor) {
     if let Some(args) = sess.target.pre_link_args.get(&flavor) {
-        cmd.args(args);
+        cmd.args(args.iter().map(Deref::deref));
     }
     cmd.args(&sess.opts.debugging_opts.pre_link_args);
 }
@@ -1602,7 +1603,7 @@ fn add_link_script(cmd: &mut dyn Linker, sess: &Session, tmpdir: &Path, crate_ty
             let file_name = ["rustc", &sess.target.llvm_target, "linkfile.ld"].join("-");
 
             let path = tmpdir.join(file_name);
-            if let Err(e) = fs::write(&path, script) {
+            if let Err(e) = fs::write(&path, script.as_ref()) {
                 sess.fatal(&format!("failed to write link script to {}: {}", path.display(), e));
             }
 
@@ -1634,23 +1635,23 @@ fn add_late_link_args(
         });
     if any_dynamic_crate {
         if let Some(args) = sess.target.late_link_args_dynamic.get(&flavor) {
-            cmd.args(args);
+            cmd.args(args.iter().map(Deref::deref));
         }
     } else {
         if let Some(args) = sess.target.late_link_args_static.get(&flavor) {
-            cmd.args(args);
+            cmd.args(args.iter().map(Deref::deref));
         }
     }
     if let Some(args) = sess.target.late_link_args.get(&flavor) {
-        cmd.args(args);
+        cmd.args(args.iter().map(Deref::deref));
     }
 }
 
 /// Add arbitrary "post-link" args defined by the target spec.
 /// FIXME: Determine where exactly these args need to be inserted.
 fn add_post_link_args(cmd: &mut dyn Linker, sess: &Session, flavor: LinkerFlavor) {
     if let Some(args) = sess.target.post_link_args.get(&flavor) {
-        cmd.args(args);
+        cmd.args(args.iter().map(Deref::deref));
     }
 }
 
@@ -1960,8 +1961,8 @@ fn add_order_independent_options(
         cmd.arg(&codegen_results.crate_info.target_cpu);
         cmd.arg("--cpu-features");
         cmd.arg(match &sess.opts.cg.target_feature {
-            feat if !feat.is_empty() => feat,
-            _ => &sess.target.options.features,
+            feat if !feat.is_empty() => feat.as_ref(),
+            _ => sess.target.options.features.as_ref(),
         });
     }
 
@@ -2478,12 +2479,12 @@ fn add_apple_sdk(cmd: &mut dyn Linker, sess: &Session, flavor: LinkerFlavor) {
     let os = &sess.target.os;
     let llvm_target = &sess.target.llvm_target;
     if sess.target.vendor != "apple"
-        || !matches!(os.as_str(), "ios" | "tvos")
+        || !matches!(os.as_ref(), "ios" | "tvos")
         || flavor != LinkerFlavor::Gcc
     {
         return;
     }
-    let sdk_name = match (arch.as_str(), os.as_str()) {
+    let sdk_name = match (arch.as_ref(), os.as_ref()) {
         ("aarch64", "tvos") => "appletvos",
         ("x86_64", "tvos") => "appletvsimulator",
         ("arm", "ios") => "iphoneos",

compiler/rustc_codegen_ssa/src/back/linker.rs

@@ -75,7 +75,7 @@ pub fn get_linker<'a>(
         if let Some(ref tool) = msvc_tool {
             let original_path = tool.path();
             if let Some(ref root_lib_path) = original_path.ancestors().nth(4) {
-                let arch = match t.arch.as_str() {
+                let arch = match t.arch.as_ref() {
                     "x86_64" => Some("x64"),
                     "x86" => Some("x86"),
                     "aarch64" => Some("arm64"),
@@ -1520,7 +1520,7 @@ impl<'a> L4Bender<'a> {
 
 pub(crate) fn exported_symbols(tcx: TyCtxt<'_>, crate_type: CrateType) -> Vec<String> {
     if let Some(ref exports) = tcx.sess.target.override_export_symbols {
-        return exports.clone();
+        return exports.iter().map(ToString::to_string).collect();
     }
 
     let mut symbols = Vec::new();

compiler/rustc_codegen_ssa/src/back/write.rs

@@ -218,7 +218,7 @@ impl ModuleConfig {
                 false
             ),
             emit_obj,
-            bc_cmdline: sess.target.bitcode_llvm_cmdline.clone(),
+            bc_cmdline: sess.target.bitcode_llvm_cmdline.to_string(),
 
             verify_llvm_ir: sess.verify_llvm_ir(),
             no_prepopulate_passes: sess.opts.cg.no_prepopulate_passes,
@@ -1061,7 +1061,7 @@ fn start_executing_work<B: ExtraBackendMethods>(
         is_pe_coff: tcx.sess.target.is_like_windows,
         target_can_use_split_dwarf: tcx.sess.target_can_use_split_dwarf(),
         target_pointer_width: tcx.sess.target.pointer_width,
-        target_arch: tcx.sess.target.arch.clone(),
+        target_arch: tcx.sess.target.arch.to_string(),
         debuginfo: tcx.sess.opts.debuginfo,
         split_debuginfo: tcx.sess.split_debuginfo(),
         split_dwarf_kind: tcx.sess.opts.debugging_opts.split_dwarf_kind,