Merging upstream version 1.75.0+dfsg1.

Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>

1 files changed, 125 insertions, 71 deletions

diff --git a/compiler/rustc_codegen_gcc/src/lib.rs b/compiler/rustc_codegen_gcc/src/lib.rs
index ce7e31682..a530fc994 100644
--- a/compiler/rustc_codegen_gcc/src/lib.rs
+++ b/compiler/rustc_codegen_gcc/src/lib.rs
@@ -2,10 +2,19 @@
  * TODO(antoyo): implement equality in libgccjit based on https://zpz.github.io/blog/overloading-equality-operator-in-cpp-class-hierarchy/ (for type equality?)
  * TODO(antoyo): support #[inline] attributes.
  * TODO(antoyo): support LTO (gcc's equivalent to Full LTO is -flto -flto-partition=one — https://documentation.suse.com/sbp/all/html/SBP-GCC-10/index.html).
+ * For Thin LTO, this might be helpful:
+ * In gcc 4.6 -fwhopr was removed and became default with -flto. The non-whopr path can still be executed via -flto-partition=none.
+ *
+ * Maybe some missing optizations enabled by rustc's LTO is in there: https://gcc.gnu.org/onlinedocs/gcc/Optimize-Options.html
+ * Like -fipa-icf (should be already enabled) and maybe -fdevirtualize-at-ltrans.
+ * TODO: disable debug info always being emitted. Perhaps this slows down things?
  *
  * TODO(antoyo): remove the patches.
  */
 
+#![cfg_attr(not(bootstrap), allow(internal_features))]
+#![cfg_attr(not(bootstrap), doc(rust_logo))]
+#![cfg_attr(not(bootstrap), feature(rustdoc_internals))]
 #![feature(
     rustc_private,
     decl_macro,
@@ -28,6 +37,7 @@ extern crate rustc_codegen_ssa;
 extern crate rustc_data_structures;
 extern crate rustc_errors;
 extern crate rustc_fluent_macro;
+extern crate rustc_fs_util;
 extern crate rustc_hir;
 extern crate rustc_macros;
 extern crate rustc_metadata;
@@ -35,7 +45,8 @@ extern crate rustc_middle;
 extern crate rustc_session;
 extern crate rustc_span;
 extern crate rustc_target;
-extern crate tempfile;
+#[macro_use]
+extern crate tracing;
 
 // This prevents duplicating functions and statics that are already part of the host rustc process.
 #[allow(unused_extern_crates)]
@@ -57,6 +68,7 @@ mod coverageinfo;
 mod debuginfo;
 mod declare;
 mod errors;
+mod gcc_util;
 mod int;
 mod intrinsic;
 mod mono_item;
@@ -64,18 +76,29 @@ mod type_;
 mod type_of;
 
 use std::any::Any;
-use std::sync::{Arc, Mutex};
-
-use crate::errors::LTONotSupported;
-use gccjit::{Context, OptimizationLevel, CType};
+use std::fmt::Debug;
+use std::sync::Arc;
+use std::sync::Mutex;
+#[cfg(not(feature="master"))]
+use std::sync::atomic::AtomicBool;
+#[cfg(not(feature="master"))]
+use std::sync::atomic::Ordering;
+
+use gccjit::{Context, OptimizationLevel};
+#[cfg(feature="master")]
+use gccjit::TargetInfo;
+#[cfg(not(feature="master"))]
+use gccjit::CType;
+use errors::LTONotSupported;
 use rustc_ast::expand::allocator::AllocatorKind;
 use rustc_codegen_ssa::{CodegenResults, CompiledModule, ModuleCodegen};
 use rustc_codegen_ssa::base::codegen_crate;
 use rustc_codegen_ssa::back::write::{CodegenContext, FatLtoInput, ModuleConfig, TargetMachineFactoryFn};
 use rustc_codegen_ssa::back::lto::{LtoModuleCodegen, SerializedModule, ThinModule};
 use rustc_codegen_ssa::target_features::supported_target_features;
-use rustc_codegen_ssa::traits::{CodegenBackend, ExtraBackendMethods, ModuleBufferMethods, ThinBufferMethods, WriteBackendMethods};
 use rustc_data_structures::fx::FxIndexMap;
+use rustc_data_structures::sync::IntoDynSyncSend;
+use rustc_codegen_ssa::traits::{CodegenBackend, ExtraBackendMethods, ThinBufferMethods, WriteBackendMethods};
 use rustc_errors::{DiagnosticMessage, ErrorGuaranteed, Handler, SubdiagnosticMessage};
 use rustc_fluent_macro::fluent_messages;
 use rustc_metadata::EncodedMetadata;
@@ -88,6 +111,9 @@ use rustc_span::Symbol;
 use rustc_span::fatal_error::FatalError;
 use tempfile::TempDir;
 
+use crate::back::lto::ModuleBuffer;
+use crate::gcc_util::target_cpu;
+
 fluent_messages! { "../messages.ftl" }
 
 pub struct PrintOnPanic<F: Fn() -> String>(pub F);
@@ -100,9 +126,47 @@ impl<F: Fn() -> String> Drop for PrintOnPanic<F> {
     }
 }
 
+#[cfg(not(feature="master"))]
+#[derive(Debug)]
+pub struct TargetInfo {
+    supports_128bit_integers: AtomicBool,
+}
+
+#[cfg(not(feature="master"))]
+impl TargetInfo {
+    fn cpu_supports(&self, _feature: &str) -> bool {
+        false
+    }
+
+    fn supports_128bit_int(&self) -> bool {
+        self.supports_128bit_integers.load(Ordering::SeqCst)
+    }
+}
+
+#[derive(Clone)]
+pub struct LockedTargetInfo {
+    info: Arc<Mutex<IntoDynSyncSend<TargetInfo>>>,
+}
+
+impl Debug for LockedTargetInfo {
+    fn fmt(&self, formatter: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        self.info.lock().expect("lock").fmt(formatter)
+    }
+}
+
+impl LockedTargetInfo {
+    fn cpu_supports(&self, feature: &str) -> bool {
+        self.info.lock().expect("lock").cpu_supports(feature)
+    }
+
+    fn supports_128bit_int(&self) -> bool {
+        self.info.lock().expect("lock").supports_128bit_int()
+    }
+}
+
 #[derive(Clone)]
 pub struct GccCodegenBackend {
-    supports_128bit_integers: Arc<Mutex<bool>>,
+    target_info: LockedTargetInfo,
 }
 
 impl CodegenBackend for GccCodegenBackend {
@@ -112,24 +176,40 @@ impl CodegenBackend for GccCodegenBackend {
 
     fn init(&self, sess: &Session) {
         #[cfg(feature="master")]
+        {
+            let target_cpu = target_cpu(sess);
+
+            // Get the second TargetInfo with the correct CPU features by setting the arch.
+            let context = Context::default();
+            if target_cpu != "generic" {
+                context.add_command_line_option(&format!("-march={}", target_cpu));
+            }
+
+            **self.target_info.info.lock().expect("lock") = context.get_target_info();
+        }
+
+        #[cfg(feature="master")]
         gccjit::set_global_personality_function_name(b"rust_eh_personality\0");
-        if sess.lto() != Lto::No {
+        if sess.lto() == Lto::Thin {
             sess.emit_warning(LTONotSupported {});
         }
 
-        let temp_dir = TempDir::new().expect("cannot create temporary directory");
-        let temp_file = temp_dir.into_path().join("result.asm");
-        let check_context = Context::default();
-        check_context.set_print_errors_to_stderr(false);
-        let _int128_ty = check_context.new_c_type(CType::UInt128t);
-        // NOTE: we cannot just call compile() as this would require other files than libgccjit.so.
-        check_context.compile_to_file(gccjit::OutputKind::Assembler, temp_file.to_str().expect("path to str"));
-        *self.supports_128bit_integers.lock().expect("lock") = check_context.get_last_error() == Ok(None);
+        #[cfg(not(feature="master"))]
+        {
+            let temp_dir = TempDir::new().expect("cannot create temporary directory");
+            let temp_file = temp_dir.into_path().join("result.asm");
+            let check_context = Context::default();
+            check_context.set_print_errors_to_stderr(false);
+            let _int128_ty = check_context.new_c_type(CType::UInt128t);
+            // NOTE: we cannot just call compile() as this would require other files than libgccjit.so.
+            check_context.compile_to_file(gccjit::OutputKind::Assembler, temp_file.to_str().expect("path to str"));
+            self.target_info.info.lock().expect("lock").supports_128bit_integers.store(check_context.get_last_error() == Ok(None), Ordering::SeqCst);
+        }
     }
 
     fn provide(&self, providers: &mut Providers) {
-        // FIXME(antoyo) compute list of enabled features from cli flags
-        providers.global_backend_features = |_tcx, ()| vec![];
+        providers.global_backend_features =
+            |tcx, ()| gcc_util::global_gcc_features(tcx.sess, true)
     }
 
     fn codegen_crate<'tcx>(&self, tcx: TyCtxt<'tcx>, metadata: EncodedMetadata, need_metadata_module: bool) -> Box<dyn Any> {
@@ -160,7 +240,7 @@ impl CodegenBackend for GccCodegenBackend {
     }
 
     fn target_features(&self, sess: &Session, allow_unstable: bool) -> Vec<Symbol> {
-        target_features(sess, allow_unstable)
+        target_features(sess, allow_unstable, &self.target_info)
     }
 }
 
@@ -168,13 +248,19 @@ impl ExtraBackendMethods for GccCodegenBackend {
     fn codegen_allocator<'tcx>(&self, tcx: TyCtxt<'tcx>, module_name: &str, kind: AllocatorKind, alloc_error_handler_kind: AllocatorKind) -> Self::Module {
         let mut mods = GccContext {
             context: Context::default(),
+            should_combine_object_files: false,
+            temp_dir: None,
         };
+
+        if tcx.sess.target.arch == "x86" || tcx.sess.target.arch == "x86_64" {
+            mods.context.add_command_line_option("-masm=intel");
+        }
         unsafe { allocator::codegen(tcx, &mut mods, module_name, kind, alloc_error_handler_kind); }
         mods
     }
 
     fn compile_codegen_unit(&self, tcx: TyCtxt<'_>, cgu_name: Symbol) -> (ModuleCodegen<Self::Module>, u64) {
-        base::compile_codegen_unit(tcx, cgu_name, *self.supports_128bit_integers.lock().expect("lock"))
+        base::compile_codegen_unit(tcx, cgu_name, self.target_info.clone())
     }
 
     fn target_machine_factory(&self, _sess: &Session, _opt_level: OptLevel, _features: &[String]) -> TargetMachineFactoryFn<Self> {
@@ -185,14 +271,6 @@ impl ExtraBackendMethods for GccCodegenBackend {
     }
 }
 
-pub struct ModuleBuffer;
-
-impl ModuleBufferMethods for ModuleBuffer {
-    fn data(&self) -> &[u8] {
-        unimplemented!();
-    }
-}
-
 pub struct ThinBuffer;
 
 impl ThinBufferMethods for ThinBuffer {
@@ -203,6 +281,9 @@ impl ThinBufferMethods for ThinBuffer {
 
 pub struct GccContext {
     context: Context<'static>,
+    should_combine_object_files: bool,
+    // Temporary directory used by LTO. We keep it here so that it's not removed before linking.
+    temp_dir: Option<TempDir>,
 }
 
 unsafe impl Send for GccContext {}
@@ -217,18 +298,8 @@ impl WriteBackendMethods for GccCodegenBackend {
     type ThinData = ();
     type ThinBuffer = ThinBuffer;
 
-    fn run_fat_lto(_cgcx: &CodegenContext<Self>, mut modules: Vec<FatLtoInput<Self>>, _cached_modules: Vec<(SerializedModule<Self::ModuleBuffer>, WorkProduct)>) -> Result<LtoModuleCodegen<Self>, FatalError> {
-        // TODO(antoyo): implement LTO by sending -flto to libgccjit and adding the appropriate gcc linker plugins.
-        // NOTE: implemented elsewhere.
-        // TODO(antoyo): what is implemented elsewhere ^ ?
-        let module =
-            match modules.remove(0) {
-                FatLtoInput::InMemory(module) => module,
-                FatLtoInput::Serialized { .. } => {
-                    unimplemented!();
-                }
-            };
-        Ok(LtoModuleCodegen::Fat { module, _serialized_bitcode: vec![] })
+    fn run_fat_lto(cgcx: &CodegenContext<Self>, modules: Vec<FatLtoInput<Self>>, cached_modules: Vec<(SerializedModule<Self::ModuleBuffer>, WorkProduct)>) -> Result<LtoModuleCodegen<Self>, FatalError> {
+        back::lto::run_fat(cgcx, modules, cached_modules)
     }
 
     fn run_thin_lto(_cgcx: &CodegenContext<Self>, _modules: Vec<(String, Self::ThinBuffer)>, _cached_modules: Vec<(SerializedModule<Self::ModuleBuffer>, WorkProduct)>) -> Result<(Vec<LtoModuleCodegen<Self>>, Vec<WorkProduct>), FatalError> {
@@ -277,8 +348,19 @@ impl WriteBackendMethods for GccCodegenBackend {
 /// This is the entrypoint for a hot plugged rustc_codegen_gccjit
 #[no_mangle]
 pub fn __rustc_codegen_backend() -> Box<dyn CodegenBackend> {
+    #[cfg(feature="master")]
+    let info = {
+        // Check whether the target supports 128-bit integers.
+        let context = Context::default();
+        Arc::new(Mutex::new(IntoDynSyncSend(context.get_target_info())))
+    };
+    #[cfg(not(feature="master"))]
+    let info = Arc::new(Mutex::new(IntoDynSyncSend(TargetInfo {
+        supports_128bit_integers: AtomicBool::new(false),
+    })));
+
     Box::new(GccCodegenBackend {
-        supports_128bit_integers: Arc::new(Mutex::new(false)),
+        target_info: LockedTargetInfo { info },
     })
 }
 
@@ -297,22 +379,7 @@ fn to_gcc_opt_level(optlevel: Option<OptLevel>) -> OptimizationLevel {
     }
 }
 
-fn handle_native(name: &str) -> &str {
-    if name != "native" {
-        return name;
-    }
-
-    unimplemented!();
-}
-
-pub fn target_cpu(sess: &Session) -> &str {
-    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, allow_unstable: bool) -> Vec<Symbol> {
+pub fn target_features(sess: &Session, allow_unstable: bool, target_info: &LockedTargetInfo) -> Vec<Symbol> {
     supported_target_features(sess)
         .iter()
         .filter_map(
@@ -321,26 +388,13 @@ pub fn target_features(sess: &Session, allow_unstable: bool) -> Vec<Symbol> {
             },
         )
         .filter(|_feature| {
-            // TODO(antoyo): implement a way to get enabled feature in libgccjit.
-            // Probably using the equivalent of __builtin_cpu_supports.
-            // TODO(antoyo): maybe use whatever outputs the following command:
-            // gcc -march=native -Q --help=target
-            #[cfg(feature="master")]
-            {
-                // NOTE: the CPU in the CI doesn't support sse4a, so disable it to make the stdarch tests pass in the CI.
-                (_feature.contains("sse") || _feature.contains("avx")) && !_feature.contains("avx512") && !_feature.contains("sse4a")
-            }
-            #[cfg(not(feature="master"))]
-            {
-                false
-            }
+            target_info.cpu_supports(_feature)
             /*
                adx, aes, avx, avx2, avx512bf16, avx512bitalg, avx512bw, avx512cd, avx512dq, avx512er, avx512f, avx512ifma,
                avx512pf, avx512vbmi, avx512vbmi2, avx512vl, avx512vnni, avx512vp2intersect, avx512vpopcntdq,
                bmi1, bmi2, cmpxchg16b, ermsb, f16c, fma, fxsr, gfni, lzcnt, movbe, pclmulqdq, popcnt, rdrand, rdseed, rtm,
                sha, sse, sse2, sse3, sse4.1, sse4.2, sse4a, ssse3, tbm, vaes, vpclmulqdq, xsave, xsavec, xsaveopt, xsaves
              */
-            //false
         })
         .map(|feature| Symbol::intern(feature))
         .collect()