[RISCV] Replace @plt/@gotpcrel in data directives with %pltpcrel %gotpcrel

clang -fexperimental-relative-c++-abi-vtables might generate `@plt` and
`@gotpcrel` specifiers in data directives. The syntax is not used in
humand-written assembly code, and is not supported by GNU assembler.
Note: the `@plt` in `.word foo@plt` is different from
the legacy `call func@plt` (where `@plt` is simply ignored).

The `@plt` syntax was selected was simply due to a quirk of AsmParser:
the syntax was supported by all targets until I updated it
to be an opt-in feature in a0671758eb6e52a758bd1b096a9b421eec60204c

RISC-V favors the `%specifier(expr)` syntax following MIPS and Sparc,
and we should follow this convention.

This PR adds support for `.word %pltpcrel(foo+offset)` and
`.word %gotpcrel(foo)`, and drops `@plt` and `@gotpcrel`.

* MCValue::SymA can no longer have a SymbolVariant. Add an assert
  similar to that of AArch64ELFObjectWriter.cpp before
  https://reviews.llvm.org/D81446 (see my analysis at
  https://maskray.me/blog/2025-03-16-relocation-generation-in-assemblers
  if intrigued)
* `jump foo@plt, x31` now has a different diagnostic.

Pull Request: https://github.com/llvm/llvm-project/pull/132569

2 files changed, 90 insertions, 53 deletions

diff --git a/llvm/lib/CodeGen/AsmPrinter/AsmPrinter.cpp b/llvm/lib/CodeGen/AsmPrinter/AsmPrinter.cpp
index 40abe28..c626202 100644
--- a/llvm/lib/CodeGen/AsmPrinter/AsmPrinter.cpp
+++ b/llvm/lib/CodeGen/AsmPrinter/AsmPrinter.cpp
@@ -3363,7 +3363,9 @@ void AsmPrinter::emitAlignment(Align Alignment, const GlobalObject *GV,
 // Constant emission.
 //===----------------------------------------------------------------------===//
 
-const MCExpr *AsmPrinter::lowerConstant(const Constant *CV) {
+const MCExpr *AsmPrinter::lowerConstant(const Constant *CV,
+                                        const Constant *BaseCV,
+                                        uint64_t Offset) {
   MCContext &Ctx = OutContext;
 
   if (CV->isNullValue() || isa<UndefValue>(CV))
@@ -3382,7 +3384,8 @@ const MCExpr *AsmPrinter::lowerConstant(const Constant *CV) {
     return lowerBlockAddressConstant(*BA);
 
   if (const auto *Equiv = dyn_cast<DSOLocalEquivalent>(CV))
-    return getObjFileLowering().lowerDSOLocalEquivalent(Equiv, TM);
+    return getObjFileLowering().lowerDSOLocalEquivalent(
+        getSymbol(Equiv->getGlobalValue()), nullptr, 0, std::nullopt, TM);
 
   if (const NoCFIValue *NC = dyn_cast<NoCFIValue>(CV))
     return MCSymbolRefExpr::create(getSymbol(NC->getGlobalValue()), Ctx);
@@ -3428,7 +3431,7 @@ const MCExpr *AsmPrinter::lowerConstant(const Constant *CV) {
     // is reasonable to treat their delta as a 32-bit value.
     [[fallthrough]];
   case Instruction::BitCast:
-    return lowerConstant(CE->getOperand(0));
+    return lowerConstant(CE->getOperand(0), BaseCV, Offset);
 
   case Instruction::IntToPtr: {
     const DataLayout &DL = getDataLayout();
@@ -3467,33 +3470,42 @@ const MCExpr *AsmPrinter::lowerConstant(const Constant *CV) {
   }
 
   case Instruction::Sub: {
-    GlobalValue *LHSGV;
-    APInt LHSOffset;
+    GlobalValue *LHSGV, *RHSGV;
+    APInt LHSOffset, RHSOffset;
     DSOLocalEquivalent *DSOEquiv;
     if (IsConstantOffsetFromGlobal(CE->getOperand(0), LHSGV, LHSOffset,
-                                   getDataLayout(), &DSOEquiv)) {
-      GlobalValue *RHSGV;
-      APInt RHSOffset;
-      if (IsConstantOffsetFromGlobal(CE->getOperand(1), RHSGV, RHSOffset,
-                                     getDataLayout())) {
-        const MCExpr *RelocExpr =
-            getObjFileLowering().lowerRelativeReference(LHSGV, RHSGV, TM);
-        if (!RelocExpr) {
-          const MCExpr *LHSExpr =
-              MCSymbolRefExpr::create(getSymbol(LHSGV), Ctx);
-          if (DSOEquiv &&
-              getObjFileLowering().supportDSOLocalEquivalentLowering())
-            LHSExpr =
-                getObjFileLowering().lowerDSOLocalEquivalent(DSOEquiv, TM);
-          RelocExpr = MCBinaryExpr::createSub(
-              LHSExpr, MCSymbolRefExpr::create(getSymbol(RHSGV), Ctx), Ctx);
-        }
-        int64_t Addend = (LHSOffset - RHSOffset).getSExtValue();
+                                   getDataLayout(), &DSOEquiv) &&
+        IsConstantOffsetFromGlobal(CE->getOperand(1), RHSGV, RHSOffset,
+                                   getDataLayout())) {
+      auto *LHSSym = getSymbol(LHSGV);
+      auto *RHSSym = getSymbol(RHSGV);
+      int64_t Addend = (LHSOffset - RHSOffset).getSExtValue();
+      std::optional<int64_t> PCRelativeOffset;
+      if (getObjFileLowering().hasPLTPCRelative() && RHSGV == BaseCV)
+        PCRelativeOffset = Offset;
+
+      // Try the generic symbol difference first.
+      const MCExpr *Res = getObjFileLowering().lowerRelativeReference(
+          LHSGV, RHSGV, Addend, PCRelativeOffset, TM);
+
+      // (ELF-specific) If the generic symbol difference does not apply, and
+      // LHS is a dso_local_equivalent of a dso_preemptable function,
+      // reference the PLT entry instead.
+      if (DSOEquiv && TM.getTargetTriple().isOSBinFormatELF() &&
+          !(LHSGV->isDSOLocal() || LHSGV->isImplicitDSOLocal()))
+        Res = getObjFileLowering().lowerDSOLocalEquivalent(
+            LHSSym, RHSSym, Addend, PCRelativeOffset, TM);
+
+      // Otherwise, return LHS-RHS+Addend.
+      if (!Res) {
+        Res =
+            MCBinaryExpr::createSub(MCSymbolRefExpr::create(LHSSym, Ctx),
+                                    MCSymbolRefExpr::create(RHSSym, Ctx), Ctx);
         if (Addend != 0)
-          RelocExpr = MCBinaryExpr::createAdd(
-              RelocExpr, MCConstantExpr::create(Addend, Ctx), Ctx);
-        return RelocExpr;
+          Res = MCBinaryExpr::createAdd(
+              Res, MCConstantExpr::create(Addend, Ctx), Ctx);
       }
+      return Res;
     }
 
     const MCExpr *LHS = lowerConstant(CE->getOperand(0));
@@ -4023,7 +4035,7 @@ static void emitGlobalConstantImpl(const DataLayout &DL, const Constant *CV,
 
   // Otherwise, it must be a ConstantExpr.  Lower it to an MCExpr, then emit it
   // thread the streamer with EmitValue.
-  const MCExpr *ME = AP.lowerConstant(CV);
+  const MCExpr *ME = AP.lowerConstant(CV, BaseCV, Offset);
 
   // Since lowerConstant already folded and got rid of all IR pointer and
   // integer casts, detect GOT equivalent accesses by looking into the MCExpr
diff --git a/llvm/lib/CodeGen/TargetLoweringObjectFileImpl.cpp b/llvm/lib/CodeGen/TargetLoweringObjectFileImpl.cpp
index 0e44acd..dd6d85e 100644
--- a/llvm/lib/CodeGen/TargetLoweringObjectFileImpl.cpp
+++ b/llvm/lib/CodeGen/TargetLoweringObjectFileImpl.cpp
@@ -115,10 +115,6 @@ static void GetObjCImageInfo(Module &M, unsigned &Version, unsigned &Flags,
 //                                  ELF
 //===----------------------------------------------------------------------===//
 
-TargetLoweringObjectFileELF::TargetLoweringObjectFileELF() {
-  SupportDSOLocalEquivalentLowering = true;
-}
-
 void TargetLoweringObjectFileELF::Initialize(MCContext &Ctx,
                                              const TargetMachine &TgtM) {
   TargetLoweringObjectFile::Initialize(Ctx, TgtM);
@@ -1174,9 +1170,37 @@ MCSection *TargetLoweringObjectFileELF::getStaticDtorSection(
                                   KeySym);
 }
 
+const MCExpr *TargetLoweringObjectFileELF::lowerSymbolDifference(
+    const MCSymbol *LHS, const MCSymbol *RHS, int64_t Addend,
+    std::optional<int64_t> PCRelativeOffset) const {
+  auto &Ctx = getContext();
+  const MCExpr *Res;
+  // Return a relocatable expression with the PLT specifier, %plt(GV) or
+  // %plt(GV-RHS).
+  if (PCRelativeOffset && PLTPCRelativeSpecifier) {
+    Res = MCSymbolRefExpr::create(LHS, Ctx);
+    // The current location is RHS plus *PCRelativeOffset. Compensate for it.
+    Addend += *PCRelativeOffset;
+    if (Addend)
+      Res = MCBinaryExpr::createAdd(Res, MCConstantExpr::create(Addend, Ctx),
+                                    Ctx);
+    return createTargetMCExpr(Res, PLTPCRelativeSpecifier);
+  }
+
+  if (!PLTRelativeSpecifier)
+    return nullptr;
+  Res = MCBinaryExpr::createSub(
+      MCSymbolRefExpr::create(LHS, PLTRelativeSpecifier, Ctx),
+      MCSymbolRefExpr::create(RHS, Ctx), Ctx);
+  if (Addend)
+    Res =
+        MCBinaryExpr::createAdd(Res, MCConstantExpr::create(Addend, Ctx), Ctx);
+  return Res;
+}
+
 const MCExpr *TargetLoweringObjectFileELF::lowerRelativeReference(
-    const GlobalValue *LHS, const GlobalValue *RHS,
-    const TargetMachine &TM) const {
+    const GlobalValue *LHS, const GlobalValue *RHS, int64_t Addend,
+    std::optional<int64_t> PCRelativeOffset, const TargetMachine &TM) const {
   // We may only use a PLT-relative relocation to refer to unnamed_addr
   // functions.
   if (!LHS->hasGlobalUnnamedAddr() || !LHS->getValueType()->isFunctionTy())
@@ -1188,24 +1212,22 @@ const MCExpr *TargetLoweringObjectFileELF::lowerRelativeReference(
       RHS->isThreadLocal())
     return nullptr;
 
-  return MCBinaryExpr::createSub(
-      MCSymbolRefExpr::create(TM.getSymbol(LHS), PLTRelativeSpecifier,
-                              getContext()),
-      MCSymbolRefExpr::create(TM.getSymbol(RHS), getContext()), getContext());
+  return lowerSymbolDifference(TM.getSymbol(LHS), TM.getSymbol(RHS), Addend,
+                               PCRelativeOffset);
 }
 
+// Reference the PLT entry of a function, optionally with a subtrahend (`RHS`).
 const MCExpr *TargetLoweringObjectFileELF::lowerDSOLocalEquivalent(
-    const DSOLocalEquivalent *Equiv, const TargetMachine &TM) const {
-  assert(supportDSOLocalEquivalentLowering());
-
-  const auto *GV = Equiv->getGlobalValue();
-
-  // A PLT entry is not needed for dso_local globals.
-  if (GV->isDSOLocal() || GV->isImplicitDSOLocal())
-    return MCSymbolRefExpr::create(TM.getSymbol(GV), getContext());
-
-  return MCSymbolRefExpr::create(TM.getSymbol(GV), PLTRelativeSpecifier,
-                                 getContext());
+    const MCSymbol *LHS, const MCSymbol *RHS, int64_t Addend,
+    std::optional<int64_t> PCRelativeOffset, const TargetMachine &TM) const {
+  if (RHS)
+    return lowerSymbolDifference(LHS, RHS, Addend, PCRelativeOffset);
+
+  // Only the legacy MCSymbolRefExpr::VariantKind approach is implemented.
+  // Reference LHS@plt or LHS@plt - RHS.
+  if (PLTRelativeSpecifier)
+    return MCSymbolRefExpr::create(LHS, PLTRelativeSpecifier, getContext());
+  return nullptr;
 }
 
 MCSection *TargetLoweringObjectFileELF::getSectionForCommandLines() const {
@@ -2044,8 +2066,8 @@ MCSection *TargetLoweringObjectFileCOFF::getStaticDtorSection(
 }
 
 const MCExpr *TargetLoweringObjectFileCOFF::lowerRelativeReference(
-    const GlobalValue *LHS, const GlobalValue *RHS,
-    const TargetMachine &TM) const {
+    const GlobalValue *LHS, const GlobalValue *RHS, int64_t Addend,
+    std::optional<int64_t> PCRelativeOffset, const TargetMachine &TM) const {
   const Triple &T = TM.getTargetTriple();
   if (T.isOSCygMing())
     return nullptr;
@@ -2069,9 +2091,12 @@ const MCExpr *TargetLoweringObjectFileCOFF::lowerRelativeReference(
       cast<GlobalVariable>(RHS)->hasInitializer() || RHS->hasSection())
     return nullptr;
 
-  return MCSymbolRefExpr::create(TM.getSymbol(LHS),
-                                 MCSymbolRefExpr::VK_COFF_IMGREL32,
-                                 getContext());
+  const MCExpr *Res = MCSymbolRefExpr::create(
+      TM.getSymbol(LHS), MCSymbolRefExpr::VK_COFF_IMGREL32, getContext());
+  if (Addend != 0)
+    Res = MCBinaryExpr::createAdd(
+        Res, MCConstantExpr::create(Addend, getContext()), getContext());
+  return Res;
 }
 
 static std::string APIntToHexString(const APInt &AI) {