[lto] Record whether a variable is TLS.

[silvas]

Summary

I'm not sure whether the if (GV) is needed; I'm just following the
code above which is checking the same thing. No tests fail when I remove
both if (GV) checks.
Rafael, any ideas?

[• rafael]

The !GV case is for symbols defined in inline assembly.

Now that I think of it, we should probably drop it for now. It will
be clear when adding inline assembly.

Cheers,
Rafael

[silvas]

Update for r262972

[• rafael]

LGTM with nit.

test/ELF/lto/Inputs/tls-mixed.s

4

Why do you need the alignment?

ELF/InputFiles.cpp

for (const BasicSymbolRef &Sym : Obj->symbols()) {
     } else if (Flags & BasicSymbolRef::SF_Common) {
       const DataLayout &DL = M.getDataLayout();
       uint64_t Size = DL.getTypeAllocSize(GV->getValueType());
       Body = new (Alloc)
           DefinedCommon(NameRef, Size, GV->getAlignment(), IsWeak, Visibility);
     } else {
       Body = new (Alloc) DefinedBitcode(NameRef, IsWeak, Visibility);
     }
+    Body->IsTls = GV->isThreadLocal();
     SymbolBodies.push_back(Body);
   }
 }
 
 template <typename T>
 static std::unique_ptr<InputFile> createELFFileAux(MemoryBufferRef MB) {
   std::unique_ptr<T> Ret = llvm::make_unique<T>(MB);
 

test/ELF/lto/Inputs/tls-mixed.s

+.globl foo
+.section .tbss,"awT",@nobits
+.p2align 2
+foo:
+.long 0

test/ELF/lto/tls-mixed.ll

+; REQUIRES: x86
+; RUN: llvm-as %s -o %t1.o
+; RUN: llvm-mc %p/Inputs/tls-mixed.s -o %t2.o -filetype=obj -triple=x86_64-pc-linux
+; RUN: ld.lld -m elf_x86_64 %t1.o %t2.o -o %t.so -shared
+
+target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+target triple = "x86_64-unknown-linux-gnu"
+
+; Should not encounter TLS-ness mismatch for @foo
+@foo = external thread_local global i32, align 4