Skip to content

[CIR][NFC] Upstream LValueBaseInfo handling #134928

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Apr 9, 2025

Conversation

andykaylor
Copy link
Contributor

Previous implementations that used the cir::LValue class omitted hanling of the LValueBaseInfo class, which tracks information about the basis for the LValue's alignment. As more code was upstreamed from the incubator, we were accumulating technical debt by adding more places where this wasn't handled correctly. This change puts the interfaces in place to track this information.

The information being tracked isn't used yet, so no functional change is intended. The tracking is being added now because it will become more difficult to add it as more features are implemented.

@llvmbot llvmbot added clang Clang issues not falling into any other category ClangIR Anything related to the ClangIR project labels Apr 8, 2025
@llvmbot
Copy link
Member

llvmbot commented Apr 8, 2025

@llvm/pr-subscribers-clangir

@llvm/pr-subscribers-clang

Author: Andy Kaylor (andykaylor)

Changes

Previous implementations that used the cir::LValue class omitted hanling of the LValueBaseInfo class, which tracks information about the basis for the LValue's alignment. As more code was upstreamed from the incubator, we were accumulating technical debt by adding more places where this wasn't handled correctly. This change puts the interfaces in place to track this information.

The information being tracked isn't used yet, so no functional change is intended. The tracking is being added now because it will become more difficult to add it as more features are implemented.


Full diff: https://github.com/llvm/llvm-project/pull/134928.diff

8 Files Affected:

  • (modified) clang/include/clang/CIR/MissingFeatures.h (+1-1)
  • (modified) clang/lib/CIR/CodeGen/CIRGenDecl.cpp (+2-2)
  • (modified) clang/lib/CIR/CodeGen/CIRGenExpr.cpp (+8-6)
  • (modified) clang/lib/CIR/CodeGen/CIRGenExprAggregate.cpp (+3-3)
  • (modified) clang/lib/CIR/CodeGen/CIRGenFunction.h (+15-3)
  • (modified) clang/lib/CIR/CodeGen/CIRGenModule.cpp (+11-7)
  • (modified) clang/lib/CIR/CodeGen/CIRGenModule.h (+3-1)
  • (modified) clang/lib/CIR/CodeGen/CIRGenValue.h (+46-7)
diff --git a/clang/include/clang/CIR/MissingFeatures.h b/clang/include/clang/CIR/MissingFeatures.h
index ed952620e723e..189e31ce6ffd2 100644
--- a/clang/include/clang/CIR/MissingFeatures.h
+++ b/clang/include/clang/CIR/MissingFeatures.h
@@ -107,9 +107,9 @@ struct MissingFeatures {
   static bool cgFPOptionsRAII() { return false; }
   static bool metaDataNode() { return false; }
   static bool fastMathFlags() { return false; }
-  static bool lvalueBaseInfo() { return false; }
   static bool alignCXXRecordDecl() { return false; }
   static bool setNonGC() { return false; }
+  static bool objCGC() { return false; }
 
   // Missing types
   static bool dataMemberType() { return false; }
diff --git a/clang/lib/CIR/CodeGen/CIRGenDecl.cpp b/clang/lib/CIR/CodeGen/CIRGenDecl.cpp
index d0eb648683e8c..b8e72f299acb6 100644
--- a/clang/lib/CIR/CodeGen/CIRGenDecl.cpp
+++ b/clang/lib/CIR/CodeGen/CIRGenDecl.cpp
@@ -146,7 +146,7 @@ void CIRGenFunction::emitAutoVarInit(
   // its removal/optimization to the CIR lowering.
   if (!constant || isa<CXXTemporaryObjectExpr>(init)) {
     initializeWhatIsTechnicallyUninitialized(addr);
-    LValue lv = LValue::makeAddr(addr, type);
+    LValue lv = makeAddrLValue(addr, type, AlignmentSource::Decl);
     emitExprAsInit(init, &d, lv);
     // In case lv has uses it means we indeed initialized something
     // out of it while trying to build the expression, mark it as such.
@@ -165,7 +165,7 @@ void CIRGenFunction::emitAutoVarInit(
   assert(typedConstant && "expected typed attribute");
   if (!emission.IsConstantAggregate) {
     // For simple scalar/complex initialization, store the value directly.
-    LValue lv = LValue::makeAddr(addr, type);
+    LValue lv = makeAddrLValue(addr, type);
     assert(init && "expected initializer");
     mlir::Location initLoc = getLoc(init->getSourceRange());
     // lv.setNonGC(true);
diff --git a/clang/lib/CIR/CodeGen/CIRGenExpr.cpp b/clang/lib/CIR/CodeGen/CIRGenExpr.cpp
index fe3f2f5de249c..b8e2c73bf6080 100644
--- a/clang/lib/CIR/CodeGen/CIRGenExpr.cpp
+++ b/clang/lib/CIR/CodeGen/CIRGenExpr.cpp
@@ -27,7 +27,8 @@ using namespace cir;
 
 /// Given an expression of pointer type, try to
 /// derive a more accurate bound on the alignment of the pointer.
-Address CIRGenFunction::emitPointerWithAlignment(const Expr *expr) {
+Address CIRGenFunction::emitPointerWithAlignment(const Expr *expr,
+                                                 LValueBaseInfo *baseInfo) {
   // We allow this with ObjC object pointers because of fragile ABIs.
   assert(expr->getType()->isPointerType() ||
          expr->getType()->isObjCObjectPointerType());
@@ -163,7 +164,8 @@ Address CIRGenFunction::emitPointerWithAlignment(const Expr *expr) {
 
   // Otherwise, use the alignment of the type.
   return makeNaturalAddressForPointer(
-      emitScalarExpr(expr), expr->getType()->getPointeeType(), CharUnits());
+      emitScalarExpr(expr), expr->getType()->getPointeeType(), CharUnits(),
+      /*forPointeeType=*/true, baseInfo);
 }
 
 void CIRGenFunction::emitStoreThroughLValue(RValue src, LValue dst,
@@ -299,7 +301,7 @@ LValue CIRGenFunction::emitDeclRefLValue(const DeclRefExpr *e) {
       cgm.errorNYI(vd->getSourceRange(), "emitDeclRefLValue: static local");
     }
 
-    return LValue::makeAddr(addr, ty);
+    return makeAddrLValue(addr, ty, AlignmentSource::Type);
   }
 
   cgm.errorNYI(e->getSourceRange(), "emitDeclRefLValue: unhandled decl type");
@@ -337,9 +339,9 @@ LValue CIRGenFunction::emitUnaryOpLValue(const UnaryOperator *e) {
     QualType t = e->getSubExpr()->getType()->getPointeeType();
     assert(!t.isNull() && "CodeGenFunction::EmitUnaryOpLValue: Illegal type");
 
-    assert(!cir::MissingFeatures::lvalueBaseInfo());
     assert(!cir::MissingFeatures::opTBAA());
-    Address addr = emitPointerWithAlignment(e->getSubExpr());
+    LValueBaseInfo baseInfo;
+    Address addr = emitPointerWithAlignment(e->getSubExpr(), &baseInfo);
 
     // Tag 'load' with deref attribute.
     // FIXME: This misses some derefence cases and has problematic interactions
@@ -349,7 +351,7 @@ LValue CIRGenFunction::emitUnaryOpLValue(const UnaryOperator *e) {
       loadOp.setIsDerefAttr(mlir::UnitAttr::get(&getMLIRContext()));
     }
 
-    LValue lv = LValue::makeAddr(addr, t);
+    LValue lv = makeAddrLValue(addr, t, baseInfo);
     assert(!cir::MissingFeatures::addressSpace());
     assert(!cir::MissingFeatures::setNonGC());
     return lv;
diff --git a/clang/lib/CIR/CodeGen/CIRGenExprAggregate.cpp b/clang/lib/CIR/CodeGen/CIRGenExprAggregate.cpp
index 36da63d5f7d76..368a6cb27c0fd 100644
--- a/clang/lib/CIR/CodeGen/CIRGenExprAggregate.cpp
+++ b/clang/lib/CIR/CodeGen/CIRGenExprAggregate.cpp
@@ -130,7 +130,7 @@ void AggExprEmitter::emitArrayInit(Address destPtr, cir::ArrayType arrayTy,
     }
 
     const Address address = Address(element, cirElementType, elementAlign);
-    const LValue elementLV = LValue::makeAddr(address, elementType);
+    const LValue elementLV = cgf.makeAddrLValue(address, elementType);
     emitInitializationToLValue(args[i], elementLV);
   }
 
@@ -157,7 +157,7 @@ void AggExprEmitter::emitArrayInit(Address destPtr, cir::ArrayType arrayTy,
     const Address tmpAddr = cgf.createTempAlloca(
         cirElementPtrType, cgf.getPointerAlign(), loc, "arrayinit.temp",
         /*insertIntoFnEntryBlock=*/false);
-    LValue tmpLV = LValue::makeAddr(tmpAddr, elementPtrType);
+    LValue tmpLV = cgf.makeAddrLValue(tmpAddr, elementPtrType);
     cgf.emitStoreThroughLValue(RValue::get(element), tmpLV);
 
     // TODO(CIR): Replace this part later with cir::DoWhileOp
@@ -166,7 +166,7 @@ void AggExprEmitter::emitArrayInit(Address destPtr, cir::ArrayType arrayTy,
           builder.createLoad(loc, tmpAddr.getPointer());
 
       // Emit the actual filler expression.
-      const LValue elementLV = LValue::makeAddr(
+      const LValue elementLV = cgf.makeAddrLValue(
           Address(currentElement, cirElementType, elementAlign), elementType);
 
       if (arrayFiller)
diff --git a/clang/lib/CIR/CodeGen/CIRGenFunction.h b/clang/lib/CIR/CodeGen/CIRGenFunction.h
index dde665a75ae57..f0c146315a673 100644
--- a/clang/lib/CIR/CodeGen/CIRGenFunction.h
+++ b/clang/lib/CIR/CodeGen/CIRGenFunction.h
@@ -227,12 +227,23 @@ class CIRGenFunction : public CIRGenTypeCache {
   /// been signed, and the returned Address will have the pointer authentication
   /// information needed to authenticate the signed pointer.
   Address makeNaturalAddressForPointer(mlir::Value ptr, QualType t,
-                                       CharUnits alignment) {
+                                       CharUnits alignment,
+                                       bool forPointeeType = false,
+                                       LValueBaseInfo *baseInfo = nullptr) {
     if (alignment.isZero())
-      alignment = cgm.getNaturalTypeAlignment(t);
+      alignment = cgm.getNaturalTypeAlignment(t, baseInfo);
     return Address(ptr, convertTypeForMem(t), alignment);
   }
 
+  LValue makeAddrLValue(Address addr, QualType ty,
+                        AlignmentSource source = AlignmentSource::Type) {
+    return makeAddrLValue(addr, ty, LValueBaseInfo(source));
+  }
+
+  LValue makeAddrLValue(Address addr, QualType ty, LValueBaseInfo baseInfo) {
+    return LValue::makeAddr(addr, ty, baseInfo);
+  }
+
   cir::FuncOp generateCode(clang::GlobalDecl gd, cir::FuncOp fn,
                            cir::FuncType funcType);
 
@@ -489,7 +500,8 @@ class CIRGenFunction : public CIRGenTypeCache {
   /// into the address of a local variable.  In such a case, it's quite
   /// reasonable to just ignore the returned alignment when it isn't from an
   /// explicit source.
-  Address emitPointerWithAlignment(const clang::Expr *expr);
+  Address emitPointerWithAlignment(const clang::Expr *expr,
+                                   LValueBaseInfo *baseInfo);
 
   mlir::LogicalResult emitReturnStmt(const clang::ReturnStmt &s);
 
diff --git a/clang/lib/CIR/CodeGen/CIRGenModule.cpp b/clang/lib/CIR/CodeGen/CIRGenModule.cpp
index 78d995b4d9648..d2259a9c41d22 100644
--- a/clang/lib/CIR/CodeGen/CIRGenModule.cpp
+++ b/clang/lib/CIR/CodeGen/CIRGenModule.cpp
@@ -74,18 +74,20 @@ CIRGenModule::CIRGenModule(mlir::MLIRContext &mlirContext,
                      builder.getStringAttr(getTriple().str()));
 }
 
-CharUnits CIRGenModule::getNaturalTypeAlignment(QualType t) {
+CharUnits CIRGenModule::getNaturalTypeAlignment(QualType t,
+                                                LValueBaseInfo *baseInfo) {
   assert(!cir::MissingFeatures::opTBAA());
 
-  // FIXME: This duplicates logic in ASTContext::getTypeAlignIfKnown. But
-  // that doesn't return the information we need to compute BaseInfo.
+  // FIXME: This duplicates logic in ASTContext::getTypeAlignIfKnown, but
+  // that doesn't return the information we need to compute baseInfo.
 
   // Honor alignment typedef attributes even on incomplete types.
   // We also honor them straight for C++ class types, even as pointees;
   // there's an expressivity gap here.
   if (const auto *tt = t->getAs<TypedefType>()) {
     if (unsigned align = tt->getDecl()->getMaxAlignment()) {
-      assert(!cir::MissingFeatures::lvalueBaseInfo());
+      if (baseInfo)
+        *baseInfo = LValueBaseInfo(AlignmentSource::AttributedType);
       return astContext.toCharUnitsFromBits(align);
     }
   }
@@ -99,13 +101,15 @@ CharUnits CIRGenModule::getNaturalTypeAlignment(QualType t) {
     // ASTContext::getTypeAlignIfKnown, but nothing uses the alignment if the
     // type is incomplete, so it's impossible to test. We could try to reuse
     // getTypeAlignIfKnown, but that doesn't return the information we need
-    // to set BaseInfo.  So just ignore the possibility that the alignment is
+    // to set baseInfo.  So just ignore the possibility that the alignment is
     // greater than one.
-    assert(!cir::MissingFeatures::lvalueBaseInfo());
+    if (baseInfo)
+      *baseInfo = LValueBaseInfo(AlignmentSource::Type);
     return CharUnits::One();
   }
 
-  assert(!cir::MissingFeatures::lvalueBaseInfo());
+  if (baseInfo)
+    *baseInfo = LValueBaseInfo(AlignmentSource::Type);
 
   CharUnits alignment;
   if (t.getQualifiers().hasUnaligned()) {
diff --git a/clang/lib/CIR/CodeGen/CIRGenModule.h b/clang/lib/CIR/CodeGen/CIRGenModule.h
index 66e9faa5c48e5..fe39244fc7834 100644
--- a/clang/lib/CIR/CodeGen/CIRGenModule.h
+++ b/clang/lib/CIR/CodeGen/CIRGenModule.h
@@ -16,6 +16,7 @@
 #include "CIRGenBuilder.h"
 #include "CIRGenTypeCache.h"
 #include "CIRGenTypes.h"
+#include "CIRGenValue.h"
 
 #include "clang/AST/CharUnits.h"
 #include "clang/CIR/Dialect/IR/CIRDialect.h"
@@ -91,7 +92,8 @@ class CIRGenModule : public CIRGenTypeCache {
 
   /// FIXME: this could likely be a common helper and not necessarily related
   /// with codegen.
-  clang::CharUnits getNaturalTypeAlignment(clang::QualType t);
+  clang::CharUnits getNaturalTypeAlignment(clang::QualType t,
+                                           LValueBaseInfo *baseInfo);
 
   void emitTopLevelDecl(clang::Decl *decl);
 
diff --git a/clang/lib/CIR/CodeGen/CIRGenValue.h b/clang/lib/CIR/CodeGen/CIRGenValue.h
index 68aecc6ee4a10..1b702daae4b4c 100644
--- a/clang/lib/CIR/CodeGen/CIRGenValue.h
+++ b/clang/lib/CIR/CodeGen/CIRGenValue.h
@@ -75,6 +75,28 @@ enum class AlignmentSource {
   Type
 };
 
+/// Given that the base address has the given alignment source, what's
+/// our confidence in the alignment of the field?
+static inline AlignmentSource getFieldAlignmentSource(AlignmentSource source) {
+  // For now, we don't distinguish fields of opaque pointers from
+  // top-level declarations, but maybe we should.
+  return AlignmentSource::Decl;
+}
+
+class LValueBaseInfo {
+  AlignmentSource alignSource;
+
+public:
+  explicit LValueBaseInfo(AlignmentSource source = AlignmentSource::Type)
+      : alignSource(source) {}
+  AlignmentSource getAlignmentSource() const { return alignSource; }
+  void setAlignmentSource(AlignmentSource source) { alignSource = source; }
+
+  void mergeForCast(const LValueBaseInfo &info) {
+    setAlignmentSource(info.getAlignmentSource());
+  }
+};
+
 class LValue {
   enum {
     Simple,       // This is a normal l-value, use getAddress().
@@ -87,13 +109,26 @@ class LValue {
   clang::QualType type;
   clang::Qualifiers quals;
 
+  // The alignment to use when accessing this lvalue. (For vector elements,
+  // this is the alignment of the whole vector)
+  unsigned alignment;
   mlir::Value v;
   mlir::Type elementType;
+  LValueBaseInfo baseInfo;
 
-  void initialize(clang::QualType type, clang::Qualifiers quals) {
-    assert(!cir::MissingFeatures::lvalueBaseInfo());
+  void initialize(clang::QualType type, clang::Qualifiers quals,
+                  clang::CharUnits alignment, LValueBaseInfo baseInfo) {
+    assert((!alignment.isZero() || type->isIncompleteType()) &&
+           "initializing l-value with zero alignment!");
     this->type = type;
     this->quals = quals;
+    const unsigned maxAlign = 1U << 31;
+    this->alignment = alignment.getQuantity() <= maxAlign
+                          ? alignment.getQuantity()
+                          : maxAlign;
+    assert(this->alignment == alignment.getQuantity() &&
+           "Alignment exceeds allowed max!");
+    this->baseInfo = baseInfo;
   }
 
 public:
@@ -108,9 +143,9 @@ class LValue {
   mlir::Value getPointer() const { return v; }
 
   clang::CharUnits getAlignment() const {
-    // TODO: Handle alignment
-    return clang::CharUnits::One();
+    return clang::CharUnits::fromQuantity(alignment);
   }
+  void setAlignment(clang::CharUnits a) { alignment = a.getQuantity(); }
 
   Address getAddress() const {
     return Address(getPointer(), elementType, getAlignment());
@@ -118,13 +153,17 @@ class LValue {
 
   const clang::Qualifiers &getQuals() const { return quals; }
 
-  static LValue makeAddr(Address address, clang::QualType t) {
+  static LValue makeAddr(Address address, clang::QualType t,
+                         LValueBaseInfo baseInfo) {
+    // Classic codegen sets the objc gc qualifier here. That requires an
+    // ASTContext, which is passed in from CIRGenFunction::makeAddrLValue.
+    assert(!cir::MissingFeatures::objCGC());
+
     LValue r;
     r.lvType = Simple;
     r.v = address.getPointer();
     r.elementType = address.getElementType();
-    r.initialize(t, t.getQualifiers());
-    assert(!cir::MissingFeatures::lvalueBaseInfo());
+    r.initialize(t, t.getQualifiers(), address.getAlignment(), baseInfo);
     return r;
   }
 };

Copy link
Collaborator

@erichkeane erichkeane left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I don't have any concerns with this, but would like the rest of the group to review/approve this.

@@ -27,7 +27,8 @@ using namespace cir;

/// Given an expression of pointer type, try to
/// derive a more accurate bound on the alignment of the pointer.
Address CIRGenFunction::emitPointerWithAlignment(const Expr *expr) {
Address CIRGenFunction::emitPointerWithAlignment(const Expr *expr,
LValueBaseInfo *baseInfo) {
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We have a new warning from the previous PR related to unused variable, should we quickly handle it here?

if (auto const *ece = dyn_cast<ExplicitCastExpr>(ce)) { <----- ece
   cgm.errorNYI(expr->getSourceRange(),
                   "emitPointerWithAlignment: explicit cast");
   return Address::invalid();
}

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

No. It's best to keep changes like that separate to avoid mingling unrelated changes in the commit history. We wouldn't want a warning to be reintroduced if the commit for this PR were reverted, for example. Thanks for brining this to my attention though, I'll take care of it.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, that make sense

@andykaylor
Copy link
Contributor Author

@mmha Still having to add you as a reviewer by comment

Copy link
Member

@bcardosolopes bcardosolopes left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ship it!

Previous implementations that used the cir::LValue class omitted hanling
of the LValueBaseInfo class, which tracks information about the basis for
the LValue's alignment. As more code was upstreamed from the incubator,
we were accumulating technical debt by adding more places where this wasn't
handled correctly. This change puts the interfaces in place to track this
information.

The information being tracked isn't used yet, so no functional change is
intended. The tracking is being added now because it will become more
difficult to add it as more features are implemented.
@andykaylor andykaylor merged commit 58b91d1 into llvm:main Apr 9, 2025
9 of 10 checks passed
AllinLeeYL pushed a commit to AllinLeeYL/llvm-project that referenced this pull request Apr 10, 2025
Previous implementations that used the cir::LValue class omitted hanling
of the LValueBaseInfo class, which tracks information about the basis
for the LValue's alignment. As more code was upstreamed from the
incubator, we were accumulating technical debt by adding more places
where this wasn't handled correctly. This change puts the interfaces in
place to track this information.

The information being tracked isn't used yet, so no functional change is
intended. The tracking is being added now because it will become more
difficult to add it as more features are implemented.
@andykaylor andykaylor deleted the cir-aligninfo branch April 10, 2025 21:16
var-const pushed a commit to ldionne/llvm-project that referenced this pull request Apr 17, 2025
Previous implementations that used the cir::LValue class omitted hanling
of the LValueBaseInfo class, which tracks information about the basis
for the LValue's alignment. As more code was upstreamed from the
incubator, we were accumulating technical debt by adding more places
where this wasn't handled correctly. This change puts the interfaces in
place to track this information.

The information being tracked isn't used yet, so no functional change is
intended. The tracking is being added now because it will become more
difficult to add it as more features are implemented.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
clang Clang issues not falling into any other category ClangIR Anything related to the ClangIR project
Projects
None yet
Development

Successfully merging this pull request may close these issues.

5 participants