[LLD][COFF] Deduplicate common chunks when linking COFF files. (llvm#162553)

This fixes [issue
162148](llvm#162148).

Common symbols are intended to have only a single version of the data
present in the final executable. The MSVC linker is able to successfully
deduplicate these chunks. If you have an application with a large number
of translation units with a large block of common data (this is
possible, for example, with Fortran code), then failing to deduplicate
these chunks can make the data size so large that the resulting
executable fails to load.

The logic in this patch doesn't catch all of the potential cases for
deduplication, but it should catch the most common ones.

Author: jcranmer-intel

lld/COFF/Chunks.cpp

@@ -783,7 +783,7 @@ uint32_t SectionChunk::getSectionNumber() const {
   return s.getIndex() + 1;
 }
 
-CommonChunk::CommonChunk(const COFFSymbolRef s) : sym(s) {
+CommonChunk::CommonChunk(const COFFSymbolRef s) : live(false), sym(s) {
   // The value of a common symbol is its size. Align all common symbols smaller
   // than 32 bytes naturally, i.e. round the size up to the next power of two.
   // This is what MSVC link.exe does.

lld/COFF/Chunks.h

@@ -520,6 +520,8 @@ class CommonChunk : public NonSectionChunk {
   uint32_t getOutputCharacteristics() const override;
   StringRef getSectionName() const override { return ".bss"; }
 
+  bool live;
+
 private:
   const COFFSymbolRef sym;
 };

lld/COFF/Symbols.h

@@ -232,6 +232,8 @@ class DefinedCommon : public DefinedCOFF {
                 CommonChunk *c = nullptr)
       : DefinedCOFF(DefinedCommonKind, f, n, s), data(c), size(size) {
     this->isExternal = true;
+    if (c)
+      c->live = true;
   }
 
   static bool classof(const Symbol *s) {

lld/COFF/Writer.cpp

@@ -1101,6 +1101,10 @@ void Writer::createSections() {
         sc->printDiscardedMessage();
       continue;
     }
+    if (auto *cc = dyn_cast<CommonChunk>(c)) {
+      if (!cc->live)
+        continue;
+    }
     StringRef name = c->getSectionName();
     if (shouldStripSectionSuffix(sc, name, ctx.config.mingw))
       name = name.split('$').first;

lld/test/COFF/common-dedup.ll

@@ -0,0 +1,39 @@
+; REQUIRES: x86
+; RUN: rm -rf %t.dir
+; RUN: split-file %s %t.dir
+; RUN: llc %t.dir/t1.ll -o %t.dir/t1.obj --filetype=obj
+; RUN: llc %t.dir/t2.ll -o %t.dir/t2.obj --filetype=obj
+; RUN: lld-link %t.dir/t1.obj %t.dir/t2.obj -entry:main -out:%t.dir/out.exe
+; RUN: llvm-readobj --section-headers %t.dir/out.exe | FileCheck %s
+
+; Make sure that the data section contains just one copy of @a, not two.
+; CHECK: Name: .data
+; CHECK-NEXT: VirtualSize: 0x1000
+
+;--- t1.ll
+target triple = "x86_64-pc-windows-msvc"
+@a = common global [4096 x i8] zeroinitializer
+
+define i32 @usea() {
+  %ref_common = load i32, ptr @a
+  ret i32 %ref_common
+}
+
+;--- t2.ll
+target triple = "x86_64-pc-windows-msvc"
+@a = common global [4096 x i8] zeroinitializer
+
+define i32 @useb() {
+  %ref_common = load i32, ptr @a
+  ret i32 %ref_common
+}
+
+declare i32 @usea()
+
+define dso_local i32 @main() local_unnamed_addr {
+entry:
+  %a = tail call i32 @usea()
+  %b = tail call i32 @useb()
+  %add = add nsw i32 %a, %b
+  ret i32 %add
+}