Auto merge of #87570 - nikic:llvm-13, r=nagisa

Upgrade to LLVM 13

Work in progress update to LLVM 13. Main changes:

 * InlineAsm diagnostics reported using SrcMgr diagnostic kind are now handled. Previously these used a separate diag handler.
 * Codegen tests are updated for additional attributes.
 * Some data layouts have changed.
 * Switch `#[used]` attribute from `llvm.used` to `llvm.compiler.used` to avoid SHF_GNU_RETAIN flag introduced in https://reviews.llvm.org/D97448, which appears to trigger a bug in older versions of gold.
 * Set `LLVM_INCLUDE_TESTS=OFF` to avoid Python 3.6 requirement.

Upstream issues:

 * ~~https://bugs.llvm.org/show_bug.cgi?id=51210 (InlineAsm diagnostic reporting for module asm)~~ Fixed by https://github.com/llvm/llvm-project/commit/1558bb80c01b695ce12642527cbfccf16cf54ece.
 * ~~https://bugs.llvm.org/show_bug.cgi?id=51476 (Miscompile on AArch64 due to incorrect comparison elimination)~~ Fixed by https://github.com/llvm/llvm-project/commit/81b106584f2baf33e09be2362c35c1bf2f6bfe94.
 * https://bugs.llvm.org/show_bug.cgi?id=51207 (Can't set custom section flags anymore). Problematic change reverted in our fork, https://reviews.llvm.org/D107216 posted for upstream revert.
 * https://bugs.llvm.org/show_bug.cgi?id=51211 (Regression in codegen for #83623). This is an optimization regression that we may likely have to eat for this release. The fix for #83623 was based on an incorrect premise, and this needs to be properly addressed in the MergeICmps pass.

The [compile-time impact](https://perf.rust-lang.org/compare.html?start=ef9549b6c0efb7525c9b012148689c8d070f9bc0&end=0983094463497eec22d550dad25576a894687002) is mixed, but quite positive as LLVM upgrades go.

The LLVM 13 final release is scheduled for Sep 21st. The current nightly is scheduled for stable release on Oct 21st.

r? `@ghost`

7 files changed, 145 insertions, 79 deletions

diff --git a/compiler/rustc_codegen_llvm/src/back/write.rs b/compiler/rustc_codegen_llvm/src/back/write.rs
index 5b4a187a1d5..791604a1827 100644
--- a/compiler/rustc_codegen_llvm/src/back/write.rs
+++ b/compiler/rustc_codegen_llvm/src/back/write.rs
@@ -296,39 +296,8 @@ unsafe extern "C" fn inline_asm_handler(diag: &SMDiagnostic, user: *const c_void
     }
     let (cgcx, _) = *(user as *const (&CodegenContext<LlvmCodegenBackend>, &Handler));
 
-    // Recover the post-substitution assembly code from LLVM for better
-    // diagnostics.
-    let mut have_source = false;
-    let mut buffer = String::new();
-    let mut level = llvm::DiagnosticLevel::Error;
-    let mut loc = 0;
-    let mut ranges = [0; 8];
-    let mut num_ranges = ranges.len() / 2;
-    let msg = llvm::build_string(|msg| {
-        buffer = llvm::build_string(|buffer| {
-            have_source = llvm::LLVMRustUnpackSMDiagnostic(
-                diag,
-                msg,
-                buffer,
-                &mut level,
-                &mut loc,
-                ranges.as_mut_ptr(),
-                &mut num_ranges,
-            );
-        })
-        .expect("non-UTF8 inline asm");
-    })
-    .expect("non-UTF8 SMDiagnostic");
-
-    let source = have_source.then(|| {
-        let mut spans = vec![InnerSpan::new(loc as usize, loc as usize)];
-        for i in 0..num_ranges {
-            spans.push(InnerSpan::new(ranges[i * 2] as usize, ranges[i * 2 + 1] as usize));
-        }
-        (buffer, spans)
-    });
-
-    report_inline_asm(cgcx, msg, level, cookie, source);
+    let smdiag = llvm::diagnostic::SrcMgrDiagnostic::unpack(diag);
+    report_inline_asm(cgcx, smdiag.message, smdiag.level, cookie, smdiag.source);
 }
 
 unsafe extern "C" fn diagnostic_handler(info: &DiagnosticInfo, user: *mut c_void) {
@@ -339,13 +308,7 @@ unsafe extern "C" fn diagnostic_handler(info: &DiagnosticInfo, user: *mut c_void
 
     match llvm::diagnostic::Diagnostic::unpack(info) {
         llvm::diagnostic::InlineAsm(inline) => {
-            report_inline_asm(
-                cgcx,
-                llvm::twine_to_string(inline.message),
-                inline.level,
-                inline.cookie,
-                None,
-            );
+            report_inline_asm(cgcx, inline.message, inline.level, inline.cookie, inline.source);
         }
 
         llvm::diagnostic::Optimization(opt) => {
diff --git a/compiler/rustc_codegen_llvm/src/base.rs b/compiler/rustc_codegen_llvm/src/base.rs
index cc3cbea4def..a6bdbd11899 100644
--- a/compiler/rustc_codegen_llvm/src/base.rs
+++ b/compiler/rustc_codegen_llvm/src/base.rs
@@ -157,16 +157,18 @@ pub fn compile_codegen_unit(
             }
 
             // Finalize code coverage by injecting the coverage map. Note, the coverage map will
-            // also be added to the `llvm.used` variable, created next.
+            // also be added to the `llvm.compiler.used` variable, created next.
             if cx.sess().instrument_coverage() {
                 cx.coverageinfo_finalize();
             }
 
-            // Create the llvm.used variable
-            // This variable has type [N x i8*] and is stored in the llvm.metadata section
+            // Create the llvm.used and llvm.compiler.used variables.
             if !cx.used_statics().borrow().is_empty() {
                 cx.create_used_variable()
             }
+            if !cx.compiler_used_statics().borrow().is_empty() {
+                cx.create_compiler_used_variable()
+            }
 
             // Finalize debuginfo
             if cx.sess().opts.debuginfo != DebugInfo::None {
diff --git a/compiler/rustc_codegen_llvm/src/consts.rs b/compiler/rustc_codegen_llvm/src/consts.rs
index 3ca295f4a7e..e1baf95e1d9 100644
--- a/compiler/rustc_codegen_llvm/src/consts.rs
+++ b/compiler/rustc_codegen_llvm/src/consts.rs
@@ -474,7 +474,13 @@ impl StaticMethods for CodegenCx<'ll, 'tcx> {
             }
 
             if attrs.flags.contains(CodegenFnAttrFlags::USED) {
-                self.add_used_global(g);
+                // The semantics of #[used] in Rust only require the symbol to make it into the
+                // object file. It is explicitly allowed for the linker to strip the symbol if it
+                // is dead. As such, use llvm.compiler.used instead of llvm.used.
+                // Additionally, https://reviews.llvm.org/D97448 in LLVM 13 started emitting unique
+                // sections with SHF_GNU_RETAIN flag for llvm.used symbols, which may trigger bugs
+                // in some versions of the gold linker.
+                self.add_compiler_used_global(g);
             }
         }
     }
@@ -484,4 +490,11 @@ impl StaticMethods for CodegenCx<'ll, 'tcx> {
         let cast = unsafe { llvm::LLVMConstPointerCast(global, self.type_i8p()) };
         self.used_statics.borrow_mut().push(cast);
     }
+
+    /// Add a global value to a list to be stored in the `llvm.compiler.used` variable,
+    /// an array of i8*.
+    fn add_compiler_used_global(&self, global: &'ll Value) {
+        let cast = unsafe { llvm::LLVMConstPointerCast(global, self.type_i8p()) };
+        self.compiler_used_statics.borrow_mut().push(cast);
+    }
 }
diff --git a/compiler/rustc_codegen_llvm/src/context.rs b/compiler/rustc_codegen_llvm/src/context.rs
index 5d56c93f835..35c866d48a4 100644
--- a/compiler/rustc_codegen_llvm/src/context.rs
+++ b/compiler/rustc_codegen_llvm/src/context.rs
@@ -75,6 +75,10 @@ pub struct CodegenCx<'ll, 'tcx> {
     /// See <https://llvm.org/docs/LangRef.html#the-llvm-used-global-variable> for details
     pub used_statics: RefCell<Vec<&'ll Value>>,
 
+    /// Statics that will be placed in the llvm.compiler.used variable
+    /// See <https://llvm.org/docs/LangRef.html#the-llvm-compiler-used-global-variable> for details
+    pub compiler_used_statics: RefCell<Vec<&'ll Value>>,
+
     /// Mapping of non-scalar types to llvm types and field remapping if needed.
     pub type_lowering: RefCell<FxHashMap<(Ty<'tcx>, Option<VariantIdx>), TypeLowering<'ll>>>,
 
@@ -115,10 +119,6 @@ fn to_llvm_tls_model(tls_model: TlsModel) -> llvm::ThreadLocalMode {
     }
 }
 
-fn strip_powerpc64_vectors(data_layout: String) -> String {
-    data_layout.replace("-v256:256:256-v512:512:512", "")
-}
-
 pub unsafe fn create_module(
     tcx: TyCtxt<'_>,
     llcx: &'ll llvm::Context,
@@ -130,7 +130,18 @@ pub unsafe fn create_module(
 
     let mut target_data_layout = sess.target.data_layout.clone();
     if llvm_util::get_version() < (12, 0, 0) && sess.target.arch == "powerpc64" {
-        target_data_layout = strip_powerpc64_vectors(target_data_layout);
+        target_data_layout = target_data_layout.replace("-v256:256:256-v512:512:512", "");
+    }
+    if llvm_util::get_version() < (13, 0, 0) {
+        if sess.target.arch == "powerpc64" {
+            target_data_layout = target_data_layout.replace("-S128", "");
+        }
+        if sess.target.arch == "wasm32" {
+            target_data_layout = "e-m:e-p:32:32-i64:64-n32:64-S128".to_string();
+        }
+        if sess.target.arch == "wasm64" {
+            target_data_layout = "e-m:e-p:64:64-i64:64-n32:64-S128".to_string();
+        }
     }
 
     // Ensure the data-layout values hardcoded remain the defaults.
@@ -318,6 +329,7 @@ impl<'ll, 'tcx> CodegenCx<'ll, 'tcx> {
             const_globals: Default::default(),
             statics_to_rauw: RefCell::new(Vec::new()),
             used_statics: RefCell::new(Vec::new()),
+            compiler_used_statics: RefCell::new(Vec::new()),
             type_lowering: Default::default(),
             scalar_lltypes: Default::default(),
             pointee_infos: Default::default(),
@@ -340,6 +352,18 @@ impl<'ll, 'tcx> CodegenCx<'ll, 'tcx> {
     pub fn coverage_context(&'a self) -> Option<&'a coverageinfo::CrateCoverageContext<'ll, 'tcx>> {
         self.coverage_cx.as_ref()
     }
+
+    fn create_used_variable_impl(&self, name: &'static CStr, values: &[&'ll Value]) {
+        let section = cstr!("llvm.metadata");
+        let array = self.const_array(&self.type_ptr_to(self.type_i8()), values);
+
+        unsafe {
+            let g = llvm::LLVMAddGlobal(self.llmod, self.val_ty(array), name.as_ptr());
+            llvm::LLVMSetInitializer(g, array);
+            llvm::LLVMRustSetLinkage(g, llvm::Linkage::AppendingLinkage);
+            llvm::LLVMSetSection(g, section.as_ptr());
+        }
+    }
 }
 
 impl MiscMethods<'tcx> for CodegenCx<'ll, 'tcx> {
@@ -430,6 +454,10 @@ impl MiscMethods<'tcx> for CodegenCx<'ll, 'tcx> {
         &self.used_statics
     }
 
+    fn compiler_used_statics(&self) -> &RefCell<Vec<&'ll Value>> {
+        &self.compiler_used_statics
+    }
+
     fn set_frame_pointer_type(&self, llfn: &'ll Value) {
         attributes::set_frame_pointer_type(self, llfn)
     }
@@ -440,17 +468,14 @@ impl MiscMethods<'tcx> for CodegenCx<'ll, 'tcx> {
     }
 
     fn create_used_variable(&self) {
-        let name = cstr!("llvm.used");
-        let section = cstr!("llvm.metadata");
-        let array =
-            self.const_array(&self.type_ptr_to(self.type_i8()), &*self.used_statics.borrow());
+        self.create_used_variable_impl(cstr!("llvm.used"), &*self.used_statics.borrow());
+    }
 
-        unsafe {
-            let g = llvm::LLVMAddGlobal(self.llmod, self.val_ty(array), name.as_ptr());
-            llvm::LLVMSetInitializer(g, array);
-            llvm::LLVMRustSetLinkage(g, llvm::Linkage::AppendingLinkage);
-            llvm::LLVMSetSection(g, section.as_ptr());
-        }
+    fn create_compiler_used_variable(&self) {
+        self.create_used_variable_impl(
+            cstr!("llvm.compiler.used"),
+            &*self.compiler_used_statics.borrow(),
+        );
     }
 
     fn declare_c_main(&self, fn_type: Self::Type) -> Option<Self::Function> {
diff --git a/compiler/rustc_codegen_llvm/src/lib.rs b/compiler/rustc_codegen_llvm/src/lib.rs
index aa4db1622b2..1e6e5252b25 100644
--- a/compiler/rustc_codegen_llvm/src/lib.rs
+++ b/compiler/rustc_codegen_llvm/src/lib.rs
@@ -352,8 +352,8 @@ impl ModuleLlvm {
 impl Drop for ModuleLlvm {
     fn drop(&mut self) {
         unsafe {
-            llvm::LLVMContextDispose(&mut *(self.llcx as *mut _));
             llvm::LLVMRustDisposeTargetMachine(&mut *(self.tm as *mut _));
+            llvm::LLVMContextDispose(&mut *(self.llcx as *mut _));
         }
     }
 }
diff --git a/compiler/rustc_codegen_llvm/src/llvm/diagnostic.rs b/compiler/rustc_codegen_llvm/src/llvm/diagnostic.rs
index ccd3e42e458..36aa022d746 100644
--- a/compiler/rustc_codegen_llvm/src/llvm/diagnostic.rs
+++ b/compiler/rustc_codegen_llvm/src/llvm/diagnostic.rs
@@ -6,7 +6,8 @@ pub use self::OptimizationDiagnosticKind::*;
 use crate::value::Value;
 use libc::c_uint;
 
-use super::{DiagnosticInfo, Twine};
+use super::{DiagnosticInfo, SMDiagnostic};
+use rustc_span::InnerSpan;
 
 #[derive(Copy, Clone)]
 pub enum OptimizationDiagnosticKind {
@@ -86,36 +87,91 @@ impl OptimizationDiagnostic<'ll> {
     }
 }
 
-#[derive(Copy, Clone)]
-pub struct InlineAsmDiagnostic<'ll> {
+pub struct SrcMgrDiagnostic {
+    pub level: super::DiagnosticLevel,
+    pub message: String,
+    pub source: Option<(String, Vec<InnerSpan>)>,
+}
+
+impl SrcMgrDiagnostic {
+    pub unsafe fn unpack(diag: &SMDiagnostic) -> SrcMgrDiagnostic {
+        // Recover the post-substitution assembly code from LLVM for better
+        // diagnostics.
+        let mut have_source = false;
+        let mut buffer = String::new();
+        let mut level = super::DiagnosticLevel::Error;
+        let mut loc = 0;
+        let mut ranges = [0; 8];
+        let mut num_ranges = ranges.len() / 2;
+        let message = super::build_string(|message| {
+            buffer = super::build_string(|buffer| {
+                have_source = super::LLVMRustUnpackSMDiagnostic(
+                    diag,
+                    message,
+                    buffer,
+                    &mut level,
+                    &mut loc,
+                    ranges.as_mut_ptr(),
+                    &mut num_ranges,
+                );
+            })
+            .expect("non-UTF8 inline asm");
+        })
+        .expect("non-UTF8 SMDiagnostic");
+
+        SrcMgrDiagnostic {
+            message,
+            level,
+            source: have_source.then(|| {
+                let mut spans = vec![InnerSpan::new(loc as usize, loc as usize)];
+                for i in 0..num_ranges {
+                    spans.push(InnerSpan::new(ranges[i * 2] as usize, ranges[i * 2 + 1] as usize));
+                }
+                (buffer, spans)
+            }),
+        }
+    }
+}
+
+#[derive(Clone)]
+pub struct InlineAsmDiagnostic {
     pub level: super::DiagnosticLevel,
     pub cookie: c_uint,
-    pub message: &'ll Twine,
-    pub instruction: Option<&'ll Value>,
+    pub message: String,
+    pub source: Option<(String, Vec<InnerSpan>)>,
 }
 
-impl InlineAsmDiagnostic<'ll> {
-    unsafe fn unpack(di: &'ll DiagnosticInfo) -> Self {
+impl InlineAsmDiagnostic {
+    unsafe fn unpackInlineAsm(di: &'ll DiagnosticInfo) -> Self {
         let mut cookie = 0;
         let mut message = None;
-        let mut instruction = None;
         let mut level = super::DiagnosticLevel::Error;
 
-        super::LLVMRustUnpackInlineAsmDiagnostic(
-            di,
-            &mut level,
-            &mut cookie,
-            &mut message,
-            &mut instruction,
-        );
+        super::LLVMRustUnpackInlineAsmDiagnostic(di, &mut level, &mut cookie, &mut message);
 
-        InlineAsmDiagnostic { level, cookie, message: message.unwrap(), instruction }
+        InlineAsmDiagnostic {
+            level,
+            cookie,
+            message: super::twine_to_string(message.unwrap()),
+            source: None,
+        }
+    }
+
+    unsafe fn unpackSrcMgr(di: &'ll DiagnosticInfo) -> Self {
+        let mut cookie = 0;
+        let smdiag = SrcMgrDiagnostic::unpack(super::LLVMRustGetSMDiagnostic(di, &mut cookie));
+        InlineAsmDiagnostic {
+            level: smdiag.level,
+            cookie,
+            message: smdiag.message,
+            source: smdiag.source,
+        }
     }
 }
 
 pub enum Diagnostic<'ll> {
     Optimization(OptimizationDiagnostic<'ll>),
-    InlineAsm(InlineAsmDiagnostic<'ll>),
+    InlineAsm(InlineAsmDiagnostic),
     PGO(&'ll DiagnosticInfo),
     Linker(&'ll DiagnosticInfo),
     Unsupported(&'ll DiagnosticInfo),
@@ -130,7 +186,7 @@ impl Diagnostic<'ll> {
         let kind = super::LLVMRustGetDiagInfoKind(di);
 
         match kind {
-            Dk::InlineAsm => InlineAsm(InlineAsmDiagnostic::unpack(di)),
+            Dk::InlineAsm => InlineAsm(InlineAsmDiagnostic::unpackInlineAsm(di)),
 
             Dk::OptimizationRemark => {
                 Optimization(OptimizationDiagnostic::unpack(OptimizationRemark, di))
@@ -162,6 +218,8 @@ impl Diagnostic<'ll> {
             Dk::Linker => Linker(di),
             Dk::Unsupported => Unsupported(di),
 
+            Dk::SrcMgr => InlineAsm(InlineAsmDiagnostic::unpackSrcMgr(di)),
+
             _ => UnknownDiagnostic(di),
         }
     }
diff --git a/compiler/rustc_codegen_llvm/src/llvm/ffi.rs b/compiler/rustc_codegen_llvm/src/llvm/ffi.rs
index 57173a49107..3f2ed02d90d 100644
--- a/compiler/rustc_codegen_llvm/src/llvm/ffi.rs
+++ b/compiler/rustc_codegen_llvm/src/llvm/ffi.rs
@@ -490,6 +490,7 @@ pub enum DiagnosticKind {
     PGOProfile,
     Linker,
     Unsupported,
+    SrcMgr,
 }
 
 /// LLVMRustDiagnosticLevel
@@ -2264,13 +2265,17 @@ extern "C" {
         level_out: &mut DiagnosticLevel,
         cookie_out: &mut c_uint,
         message_out: &mut Option<&'a Twine>,
-        instruction_out: &mut Option<&'a Value>,
     );
 
     #[allow(improper_ctypes)]
     pub fn LLVMRustWriteDiagnosticInfoToString(DI: &DiagnosticInfo, s: &RustString);
     pub fn LLVMRustGetDiagInfoKind(DI: &DiagnosticInfo) -> DiagnosticKind;
 
+    pub fn LLVMRustGetSMDiagnostic(
+        DI: &'a DiagnosticInfo,
+        cookie_out: &mut c_uint,
+    ) -> &'a SMDiagnostic;
+
     pub fn LLVMRustSetInlineAsmDiagnosticHandler(
         C: &Context,
         H: InlineAsmDiagHandler,