rustc: Persist LLVM's Linker in Fat LTO

This commit updates our Fat LTO logic to tweak our custom wrapper around LLVM's
"link modules" functionality. Previously whenever the
`LLVMRustLinkInExternalBitcode` function was called it would call LLVM's
`Linker::linkModules` wrapper. Internally this would crate an instance of a
`Linker` which internally creates an instance of an `IRMover`. Unfortunately for
us the creation of `IRMover` is somewhat O(n) with the input module. This means
that every time we linked a module it was O(n) with respect to the entire module
we had built up!

Now the modules we build up during LTO are quite large, so this quickly started
creating an O(n^2) problem for us! Discovered in #48025 it turns out this has
always been a problem and we just haven't noticed it. It became particularly
worse recently though due to most libraries having 16x more object files than
they previously did (1 -> 16).

This commit fixes this performance issue by preserving the `Linker` instance
across all links into the main LLVM module. This means we only create one
`IRMover` and allows LTO to progress much speedier.

From the `cargo-cache` project in #48025 a **full build** locally when from
5m15s to 2m24s. Looking at the timing logs each object file was linked in in
single-digit millisecond rather than hundreds, clearly being a nice improvement!

Closes #48025

Author: alexcrichton

src/librustc_llvm/build.rs

@@ -155,6 +155,7 @@ fn main() {
     cfg.file("../rustllvm/PassWrapper.cpp")
        .file("../rustllvm/RustWrapper.cpp")
        .file("../rustllvm/ArchiveWrapper.cpp")
+       .file("../rustllvm/Linker.cpp")
        .cpp(true)
        .cpp_link_stdlib(None) // we handle this below
        .compile("rustllvm");

src/librustc_llvm/ffi.rs

@@ -444,6 +444,9 @@ pub type RustArchiveMemberRef = *mut RustArchiveMember_opaque;
 #[allow(missing_copy_implementations)]
 pub enum OperandBundleDef_opaque {}
 pub type OperandBundleDefRef = *mut OperandBundleDef_opaque;
+#[allow(missing_copy_implementations)]
+pub enum Linker_opaque {}
+pub type LinkerRef = *mut Linker_opaque;
 
 pub type DiagnosticHandler = unsafe extern "C" fn(DiagnosticInfoRef, *mut c_void);
 pub type InlineAsmDiagHandler = unsafe extern "C" fn(SMDiagnosticRef, *const c_void, c_uint);
@@ -1608,7 +1611,6 @@ extern "C" {
     pub fn LLVMRustPrintPasses();
     pub fn LLVMRustSetNormalizedTarget(M: ModuleRef, triple: *const c_char);
     pub fn LLVMRustAddAlwaysInlinePass(P: PassManagerBuilderRef, AddLifetimes: bool);
-    pub fn LLVMRustLinkInExternalBitcode(M: ModuleRef, bc: *const c_char, len: size_t) -> bool;
     pub fn LLVMRustRunRestrictionPass(M: ModuleRef, syms: *const *const c_char, len: size_t);
     pub fn LLVMRustMarkAllFunctionsNounwind(M: ModuleRef);
 
@@ -1724,4 +1726,10 @@ extern "C" {
                                            CU2: *mut *mut c_void);
     pub fn LLVMRustThinLTOPatchDICompileUnit(M: ModuleRef, CU: *mut c_void);
     pub fn LLVMRustThinLTORemoveAvailableExternally(M: ModuleRef);
+
+    pub fn LLVMRustLinkerNew(M: ModuleRef) -> LinkerRef;
+    pub fn LLVMRustLinkerAdd(linker: LinkerRef,
+                             bytecode: *const c_char,
+                             bytecode_len: usize) -> bool;
+    pub fn LLVMRustLinkerFree(linker: LinkerRef);
 }

src/librustc_trans/back/lto.rs

@@ -247,22 +247,20 @@ fn fat_lto(cgcx: &CodegenContext,
     // know much about the memory management here so we err on the side of being
     // save and persist everything with the original module.
     let mut serialized_bitcode = Vec::new();
+    let mut linker = Linker::new(llmod);
     for (bc_decoded, name) in serialized_modules {
         info!("linking {:?}", name);
-        time(cgcx.time_passes, &format!("ll link {:?}", name), || unsafe {
+        time(cgcx.time_passes, &format!("ll link {:?}", name), || {
             let data = bc_decoded.data();
-            if llvm::LLVMRustLinkInExternalBitcode(llmod,
-                                                   data.as_ptr() as *const libc::c_char,
-                                                   data.len() as libc::size_t) {
-                Ok(())
-            } else {
+            linker.add(&data).map_err(|()| {
                 let msg = format!("failed to load bc of {:?}", name);
-                Err(write::llvm_err(&diag_handler, msg))
-            }
+                write::llvm_err(&diag_handler, msg)
+            })
         })?;
         timeline.record(&format!("link {:?}", name));
         serialized_bitcode.push(bc_decoded);
     }
+    drop(linker);
     cgcx.save_temp_bitcode(&module, "lto.input");
 
     // Internalize everything that *isn't* in our whitelist to help strip out
@@ -289,6 +287,32 @@ fn fat_lto(cgcx: &CodegenContext,
     }])
 }
 
+struct Linker(llvm::LinkerRef);
+
+impl Linker {
+    fn new(llmod: ModuleRef) -> Linker {
+        unsafe { Linker(llvm::LLVMRustLinkerNew(llmod)) }
+    }
+
+    fn add(&mut self, bytecode: &[u8]) -> Result<(), ()> {
+        unsafe {
+            if llvm::LLVMRustLinkerAdd(self.0,
+                                       bytecode.as_ptr() as *const libc::c_char,
+                                       bytecode.len()) {
+                Ok(())
+            } else {
+                Err(())
+            }
+        }
+    }
+}
+
+impl Drop for Linker {
+    fn drop(&mut self) {
+        unsafe { llvm::LLVMRustLinkerFree(self.0); }
+    }
+}
+
 /// Prepare "thin" LTO to get run on these modules.
 ///
 /// The general structure of ThinLTO is quite different from the structure of

src/rustllvm/Linker.cpp

@@ -0,0 +1,72 @@
+// Copyright 2018 The Rust Project Developers. See the COPYRIGHT
+// file at the top-level directory of this distribution and at
+// http://rust-lang.org/COPYRIGHT.
+//
+// Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
+// http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
+// <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
+// option. This file may not be copied, modified, or distributed
+// except according to those terms.
+
+#include "llvm/Linker/Linker.h"
+
+#include "rustllvm.h"
+
+using namespace llvm;
+
+struct RustLinker {
+  Linker L;
+  LLVMContext &Ctx;
+
+  RustLinker(Module &M) :
+    L(M),
+    Ctx(M.getContext())
+  {}
+};
+
+extern "C" RustLinker*
+LLVMRustLinkerNew(LLVMModuleRef DstRef) {
+  Module *Dst = unwrap(DstRef);
+
+  auto Ret = llvm::make_unique<RustLinker>(*Dst);
+  return Ret.release();
+}
+
+extern "C" void
+LLVMRustLinkerFree(RustLinker *L) {
+  delete L;
+}
+
+extern "C" bool
+LLVMRustLinkerAdd(RustLinker *L, char *BC, size_t Len) {
+  std::unique_ptr<MemoryBuffer> Buf =
+      MemoryBuffer::getMemBufferCopy(StringRef(BC, Len));
+
+#if LLVM_VERSION_GE(4, 0)
+  Expected<std::unique_ptr<Module>> SrcOrError =
+      llvm::getLazyBitcodeModule(Buf->getMemBufferRef(), L->Ctx);
+  if (!SrcOrError) {
+    LLVMRustSetLastError(toString(SrcOrError.takeError()).c_str());
+    return false;
+  }
+
+  auto Src = std::move(*SrcOrError);
+#else
+  ErrorOr<std::unique_ptr<Module>> Src =
+      llvm::getLazyBitcodeModule(std::move(Buf), L->Ctx);
+  if (!Src) {
+    LLVMRustSetLastError(Src.getError().message().c_str());
+    return false;
+  }
+#endif
+
+#if LLVM_VERSION_GE(4, 0)
+  if (L->L.linkInModule(std::move(Src))) {
+#else
+  if (L->L.linkInModule(std::move(Src.get()))) {
+#endif
+    LLVMRustSetLastError("");
+    return false;
+  }
+  return true;
+}

src/rustllvm/RustWrapper.cpp

@@ -916,46 +916,6 @@ extern "C" void LLVMRustWriteValueToString(LLVMValueRef V,
   }
 }
 
-extern "C" bool LLVMRustLinkInExternalBitcode(LLVMModuleRef DstRef, char *BC,
-                                              size_t Len) {
-  Module *Dst = unwrap(DstRef);
-
-  std::unique_ptr<MemoryBuffer> Buf =
-      MemoryBuffer::getMemBufferCopy(StringRef(BC, Len));
-
-#if LLVM_VERSION_GE(4, 0)
-  Expected<std::unique_ptr<Module>> SrcOrError =
-      llvm::getLazyBitcodeModule(Buf->getMemBufferRef(), Dst->getContext());
-  if (!SrcOrError) {
-    LLVMRustSetLastError(toString(SrcOrError.takeError()).c_str());
-    return false;
-  }
-
-  auto Src = std::move(*SrcOrError);
-#else
-  ErrorOr<std::unique_ptr<Module>> Src =
-      llvm::getLazyBitcodeModule(std::move(Buf), Dst->getContext());
-  if (!Src) {
-    LLVMRustSetLastError(Src.getError().message().c_str());
-    return false;
-  }
-#endif
-
-  std::string Err;
-
-  raw_string_ostream Stream(Err);
-  DiagnosticPrinterRawOStream DP(Stream);
-#if LLVM_VERSION_GE(4, 0)
-  if (Linker::linkModules(*Dst, std::move(Src))) {
-#else
-  if (Linker::linkModules(*Dst, std::move(Src.get()))) {
-#endif
-    LLVMRustSetLastError(Err.c_str());
-    return false;
-  }
-  return true;
-}
-
 // Note that the two following functions look quite similar to the
 // LLVMGetSectionName function. Sadly, it appears that this function only
 // returns a char* pointer, which isn't guaranteed to be null-terminated. The