[PPC64] Fix PR19893 - improve code generation for local function addresses

Rafael opened http://llvm.org/bugs/show_bug.cgi?id=19893 to track non-optimal
code generation for forming a function address that is local to the compile
unit.  The existing code was treating both local and non-local functions
identically.

This patch fixes the problem by properly identifying local functions and
generating the proper addis/addi code.  I also noticed that Rafael's earlier
changes to correct the surrounding code in PPCISelLowering.cpp were also
needed for fast instruction selection in PPCFastISel.cpp, so this patch
fixes that code as well.

The existing test/CodeGen/PowerPC/func-addr.ll is modified to test the new
code generation.  I've added a -O0 run line to test the fast-isel code as
well.

Tested on powerpc64[le]-unknown-linux-gnu with no regressions.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@211056 91177308-0d34-0410-b5e6-96231b3b80d8

Author: wschmidt-ibm

lib/Target/PowerPC/PPCAsmPrinter.cpp

@@ -365,8 +365,8 @@ void PPCAsmPrinter::EmitInstruction(const MachineInstr *MI) {
     // Transform %Xd = ADDIStocHA %X2, <ga:@sym>
     LowerPPCMachineInstrToMCInst(MI, TmpInst, *this, Subtarget.isDarwin());
 
-    // Change the opcode to ADDIS8.  If the global address is external,
-    // has common linkage, is a function address, or is a jump table
+    // Change the opcode to ADDIS8.  If the global address is external, has
+    // common linkage, is a non-local function address, or is a jump table
     // address, then generate a TOC entry and reference that.  Otherwise
     // reference the symbol directly.
     TmpInst.setOpcode(PPC::ADDIS8);
@@ -375,7 +375,7 @@ void PPCAsmPrinter::EmitInstruction(const MachineInstr *MI) {
            "Invalid operand for ADDIStocHA!");
     MCSymbol *MOSymbol = nullptr;
     bool IsExternal = false;
-    bool IsFunction = false;
+    bool IsNonLocalFunction = false;
     bool IsCommon = false;
     bool IsAvailExt = false;
 
@@ -384,15 +384,16 @@ void PPCAsmPrinter::EmitInstruction(const MachineInstr *MI) {
       MOSymbol = getSymbol(GV);
       IsExternal = GV->isDeclaration();
       IsCommon = GV->hasCommonLinkage();
-      IsFunction = GV->getType()->getElementType()->isFunctionTy();
+      IsNonLocalFunction = GV->getType()->getElementType()->isFunctionTy() &&
+        (GV->isDeclaration() || GV->isWeakForLinker());
       IsAvailExt = GV->hasAvailableExternallyLinkage();
     } else if (MO.isCPI())
       MOSymbol = GetCPISymbol(MO.getIndex());
     else if (MO.isJTI())
       MOSymbol = GetJTISymbol(MO.getIndex());
 
-    if (IsExternal || IsFunction || IsCommon || IsAvailExt || MO.isJTI() ||
-        TM.getCodeModel() == CodeModel::Large)
+    if (IsExternal || IsNonLocalFunction || IsCommon || IsAvailExt ||
+        MO.isJTI() || TM.getCodeModel() == CodeModel::Large)
       MOSymbol = lookUpOrCreateTOCEntry(MOSymbol);
 
     const MCExpr *Exp =
@@ -451,17 +452,19 @@ void PPCAsmPrinter::EmitInstruction(const MachineInstr *MI) {
     assert((MO.isGlobal() || MO.isCPI()) && "Invalid operand for ADDItocL");
     MCSymbol *MOSymbol = nullptr;
     bool IsExternal = false;
-    bool IsFunction = false;
+    bool IsNonLocalFunction = false;
 
     if (MO.isGlobal()) {
       const GlobalValue *GV = MO.getGlobal();
       MOSymbol = getSymbol(GV);
       IsExternal = GV->isDeclaration();
-      IsFunction = GV->getType()->getElementType()->isFunctionTy();
+      IsNonLocalFunction = GV->getType()->getElementType()->isFunctionTy() &&
+        (GV->isDeclaration() || GV->isWeakForLinker());
     } else if (MO.isCPI())
       MOSymbol = GetCPISymbol(MO.getIndex());
 
-    if (IsFunction || IsExternal || TM.getCodeModel() == CodeModel::Large)
+    if (IsNonLocalFunction || IsExternal ||
+        TM.getCodeModel() == CodeModel::Large)
       MOSymbol = lookUpOrCreateTOCEntry(MOSymbol);
 
     const MCExpr *Exp =

lib/Target/PowerPC/PPCFastISel.cpp

@@ -1858,7 +1858,6 @@ unsigned PPCFastISel::PPCMaterializeGV(const GlobalValue *GV, MVT VT) {
   // FIXME: Jump tables are not yet required because fast-isel doesn't
   // handle switches; if that changes, we need them as well.  For now,
   // what follows assumes everything's a generic (or TLS) global address.
-  const GlobalVariable *GVar = dyn_cast<GlobalVariable>(GV);
 
   // FIXME: We don't yet handle the complexity of TLS.
   if (GV->isThreadLocal())
@@ -1871,8 +1870,8 @@ unsigned PPCFastISel::PPCMaterializeGV(const GlobalValue *GV, MVT VT) {
         .addGlobalAddress(GV)
         .addReg(PPC::X2);
   else {
-    // If the address is an externally defined symbol, a symbol with
-    // common or externally available linkage, a function address, or a
+    // If the address is an externally defined symbol, a symbol with common
+    // or externally available linkage, a non-local function address, or a
     // jump table address (not yet needed), or if we are generating code
     // for large code model, we generate:
     //       LDtocL(GV, ADDIStocHA(%X2, GV))
@@ -1883,12 +1882,13 @@ unsigned PPCFastISel::PPCMaterializeGV(const GlobalValue *GV, MVT VT) {
     BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(PPC::ADDIStocHA),
             HighPartReg).addReg(PPC::X2).addGlobalAddress(GV);
 
-    // !GVar implies a function address.  An external variable is one
-    // without an initializer.
     // If/when switches are implemented, jump tables should be handled
     // on the "if" path here.
-    if (CModel == CodeModel::Large || !GVar || !GVar->hasInitializer() ||
-        GVar->hasCommonLinkage() || GVar->hasAvailableExternallyLinkage())
+    if (CModel == CodeModel::Large ||
+        (GV->getType()->getElementType()->isFunctionTy() &&
+         (GV->isDeclaration() || GV->isWeakForLinker())) ||
+        GV->isDeclaration() || GV->hasCommonLinkage() ||
+        GV->hasAvailableExternallyLinkage())
       BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc, TII.get(PPC::LDtocL),
               DestReg).addGlobalAddress(GV).addReg(HighPartReg);
     else

lib/Target/PowerPC/PPCISelDAGToDAG.cpp

@@ -1454,10 +1454,10 @@ SDNode *PPCDAGToDAGISel::Select(SDNode *N) {
     if (CModel != CodeModel::Medium && CModel != CodeModel::Large)
       break;
 
-    // The first source operand is a TargetGlobalAddress or a
-    // TargetJumpTable.  If it is an externally defined symbol, a symbol
-    // with common linkage, a function address, or a jump table address,
-    // or if we are generating code for large code model, we generate:
+    // The first source operand is a TargetGlobalAddress or a TargetJumpTable.
+    // If it is an externally defined symbol, a symbol with common linkage,
+    // a non-local function address, or a jump table address, or if we are
+    // generating code for large code model, we generate:
     //   LDtocL(<ga:@sym>, ADDIStocHA(%X2, <ga:@sym>))
     // Otherwise we generate:
     //   ADDItocL(ADDIStocHA(%X2, <ga:@sym>), <ga:@sym>)
@@ -1472,7 +1472,8 @@ SDNode *PPCDAGToDAGISel::Select(SDNode *N) {
 
     if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(GA)) {
       const GlobalValue *GValue = G->getGlobal();
-      if (GValue->getType()->getElementType()->isFunctionTy() ||
+      if ((GValue->getType()->getElementType()->isFunctionTy() &&
+           (GValue->isDeclaration() || GValue->isWeakForLinker())) ||
           GValue->isDeclaration() || GValue->hasCommonLinkage() ||
           GValue->hasAvailableExternallyLinkage())
         return CurDAG->getMachineNode(PPC::LDtocL, dl, MVT::i64, GA,

test/CodeGen/PowerPC/func-addr.ll

@@ -1,20 +1,16 @@
 ; RUN: llc -mtriple powerpc64-linux < %s | FileCheck %s
+; RUN: llc -O0 -mtriple powerpc64-linux < %s | FileCheck %s
 
 define void @foo()  {
   ret void
 }
 declare i32 @bar(i8*)
 
 ; CHECK-LABEL: {{^}}zed:
-; CHECK:        addis 3, 2, .LC1@toc@ha
-; CHECK-NEXT:   ld 3, .LC1@toc@l(3)
+; CHECK:        addis 3, 2, foo@toc@ha
+; CHECK-NEXT:   addi 3, 3, foo@toc@l
 ; CHECK-NEXT:   bl bar
 
-
-; CHECK-LABEL: .section        .toc,"aw",@progbits
-; CHECK:       .LC1:
-; CHECK-NEXT:  .tc foo[TC],foo
-
 define  void @zed() {
   call i32 @bar(i8* bitcast (void ()* @foo to i8*))
   ret void