[AsmPrinter][TLOF] Reintroduce AArch64 test

Follow up from r231505.

Fix the non-determinism by using a MapVector and reintroduce the AArch64
testcase. Defer deleting the got candidates up to the end and remove
them in a bulk, avoiding linear time removal of each element.

Thanks to Renato Golin for trying it out on other platforms.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@231830 91177308-0d34-0410-b5e6-96231b3b80d8

include/llvm/CodeGen/AsmPrinter.h

@@ -16,6 +16,7 @@
 #ifndef LLVM_CODEGEN_ASMPRINTER_H
 #define LLVM_CODEGEN_ASMPRINTER_H
 
+#include "llvm/ADT/MapVector.h"
 #include "llvm/ADT/Twine.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/IR/InlineAsm.h"
@@ -102,7 +103,7 @@ class AsmPrinter : public MachineFunctionPass {
   /// Map global GOT equivalent MCSymbols to GlobalVariables and keep track of
   /// its number of uses by other globals.
   typedef std::pair<const GlobalVariable *, unsigned> GOTEquivUsePair;
-  DenseMap<const MCSymbol *, GOTEquivUsePair> GlobalGOTEquivs;
+  MapVector<const MCSymbol *, GOTEquivUsePair> GlobalGOTEquivs;
 
 private:
   MCSymbol *CurrentFnBegin;

lib/CodeGen/AsmPrinter/AsmPrinter.cpp

@@ -983,14 +983,17 @@ void AsmPrinter::emitGlobalGOTEquivs() {
   if (!getObjFileLowering().supportIndirectSymViaGOTPCRel())
     return;
 
-  while (!GlobalGOTEquivs.empty()) {
-    DenseMap<const MCSymbol *, GOTEquivUsePair>::iterator I =
-      GlobalGOTEquivs.begin();
-    const MCSymbol *S = I->first;
-    const GlobalVariable *GV = I->second.first;
-    GlobalGOTEquivs.erase(S);
-    EmitGlobalVariable(GV);
+  SmallVector<const GlobalVariable *, 8> FailedCandidates;
+  for (auto &I : GlobalGOTEquivs) {
+    const GlobalVariable *GV = I.second.first;
+    unsigned Cnt = I.second.second;
+    if (Cnt)
+      FailedCandidates.push_back(GV);
   }
+  GlobalGOTEquivs.clear();
+
+  for (auto *GV : FailedCandidates)
+    EmitGlobalVariable(GV);
 }
 
 bool AsmPrinter::doFinalization(Module &M) {
@@ -2114,18 +2117,16 @@ static void handleIndirectSymViaGOTPCRel(AsmPrinter &AP, const MCExpr **ME,
   //
   AsmPrinter::GOTEquivUsePair Result = AP.GlobalGOTEquivs[GOTEquivSym];
   const GlobalVariable *GV = Result.first;
-  unsigned NumUses = Result.second;
+  int NumUses = (int)Result.second;
   const GlobalValue *FinalGV = dyn_cast<GlobalValue>(GV->getOperand(0));
   const MCSymbol *FinalSym = AP.getSymbol(FinalGV);
   *ME = AP.getObjFileLowering().getIndirectSymViaGOTPCRel(
       FinalSym, MV, Offset, AP.MMI, AP.OutStreamer);
 
   // Update GOT equivalent usage information
   --NumUses;
-  if (NumUses)
+  if (NumUses >= 0)
     AP.GlobalGOTEquivs[GOTEquivSym] = std::make_pair(GV, NumUses);
-  else
-    AP.GlobalGOTEquivs.erase(GOTEquivSym);
 }
 
 static void emitGlobalConstantImpl(const Constant *CV, AsmPrinter &AP,

test/MC/MachO/AArch64/cstexpr-gotpcrel.ll

@@ -0,0 +1,92 @@
+; RUN: llc -mtriple=arm64-apple-darwin %s -o - | FileCheck %s
+
+; GOT equivalent globals references can be replaced by the GOT entry of the
+; final symbol instead.
+
+%struct.data = type { i32, %struct.anon }
+%struct.anon = type { i32, i32 }
+
+@localfoo = global i32 42
+@localgotequiv = private unnamed_addr constant i32* @localfoo
+
+@extfoo = external global i32
+@extgotequiv = private unnamed_addr constant i32* @extfoo
+
+; Don't replace GOT equivalent usage within instructions and emit the GOT
+; equivalent since it can't be replaced by the GOT entry. @bargotequiv is
+; used by an instruction inside @t0.
+;
+; CHECK: l_bargotequiv:
+; CHECK-NEXT:  .quad   _extbar
+@extbar = external global i32
+@bargotequiv = private unnamed_addr constant i32* @extbar
+
+@table = global [4 x %struct.data] [
+; CHECK-LABEL: _table
+  %struct.data { i32 1, %struct.anon { i32 2, i32 3 } },
+; Test GOT equivalent usage inside nested constant arrays.
+
+; CHECK: .long   5
+; CHECK-NOT: .long   _localgotequiv-(_table+20)
+; CHECK-NEXT: Ltmp1:
+; CHECK-NEXT: .long _localfoo@GOT-Ltmp1
+  %struct.data { i32 4, %struct.anon { i32 5,
+    i32 trunc (i64 sub (i64 ptrtoint (i32** @localgotequiv to i64),
+                        i64 ptrtoint (i32* getelementptr inbounds ([4 x %struct.data]* @table, i32 0, i64 1, i32 1, i32 1) to i64))
+                        to i32)}
+  },
+
+; CHECK: .long   5
+; CHECK-NOT: _extgotequiv-(_table+32)
+; CHECK-NEXT: Ltmp2:
+; CHECK-NEXT: _extfoo@GOT-Ltmp2
+  %struct.data { i32 4, %struct.anon { i32 5,
+    i32 trunc (i64 sub (i64 ptrtoint (i32** @extgotequiv to i64),
+                        i64 ptrtoint (i32* getelementptr inbounds ([4 x %struct.data]* @table, i32 0, i64 2, i32 1, i32 1) to i64))
+                        to i32)}
+  },
+; Test support for arbitrary constants into the GOTPCREL offset, which is
+; supported on x86-64 but not on ARM64
+
+; CHECK: .long   5
+; CHECK-NEXT: .long (l_extgotequiv-(_table+44))+24
+  %struct.data { i32 4, %struct.anon { i32 5,
+    i32 add (i32 trunc (i64 sub (i64 ptrtoint (i32** @extgotequiv to i64),
+                                 i64 ptrtoint (i32* getelementptr inbounds ([4 x %struct.data]* @table, i32 0, i64 3, i32 1, i32 1) to i64))
+                                 to i32), i32 24)}
+  }
+], align 16
+
+; Test multiple uses of GOT equivalents.
+
+; CHECK-LABEL: _delta
+; CHECK: Ltmp3:
+; CHECK-NEXT:  .long _extfoo@GOT-Ltmp3
+@delta = global i32 trunc (i64 sub (i64 ptrtoint (i32** @extgotequiv to i64),
+                                    i64 ptrtoint (i32* @delta to i64))
+                           to i32)
+
+; CHECK-LABEL: _deltaplus:
+; CHECK: .long  (l_localgotequiv-_deltaplus)+55
+@deltaplus = global i32 add (i32 trunc (i64 sub (i64 ptrtoint (i32** @localgotequiv to i64),
+                                        i64 ptrtoint (i32* @deltaplus to i64))
+                                        to i32), i32 55)
+
+define i32 @t0(i32 %a) {
+  %x = add i32 trunc (i64 sub (i64 ptrtoint (i32** @bargotequiv to i64),
+                               i64 ptrtoint (i32 (i32)* @t0 to i64))
+                           to i32), %a
+  ret i32 %x
+}
+
+; Check that these got equivalent symbols are emitted on ARM64. Since ARM64
+; does not support encoding an extra offset with @GOT, we still need to emit the
+; equivalents for use by such IR constructs. Check them at the end of the test
+; since they will start out as GOT equivalent candidates, but they are actually
+; needed and are therefore emitted at the end.
+
+; CHECK-LABEL: l_localgotequiv:
+; CHECK-NEXT: .quad   _localfoo
+
+; CHECK-LABEL: l_extgotequiv:
+; CHECK-NEXT: .quad   _extfoo