[CIR] Add support for SourceLocExpr (__builtin_LINE, etc.)

This patch implements SourceLocExpr code generation, which enables
support for the following builtin functions:
- __builtin_LINE
- __builtin_FILE
- __builtin_FUNCTION
- __builtin_COLUMN

These builtins are evaluated at compile-time and emit constant values
or global string literals. The implementation follows the traditional
CodeGen approach of using ConstantEmitter to emit abstract constants
from evaluated APValues.

A key challenge was handling synthetic StringLiterals created during
constant evaluation (for __builtin_FILE and __builtin_FUNCTION).
These synthetic literals don't have valid source locations, which
caused assertion failures when creating global string constants.
The fix adds a check in getGlobalForStringLiteral to use
UnknownLoc when the StringLiteral has an invalid source range.

This feature unblocks:
- std::source_location (C++20)
- Logging and debugging macros that use source location builtins
- Assertion/diagnostic frameworks

Test coverage includes:
- Basic __builtin_LINE, __builtin_FILE, __builtin_FUNCTION, __builtin_COLUMN
- Usage in global variable initializers
- Default function arguments
- Lambda expressions
- Combined usage of multiple source location builtins


ghstack-source-id: 083424c
Pull-Request: #1988

Author: lanza

clang/lib/CIR/CodeGen/CIRGenExprScalar.cpp

@@ -11,6 +11,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "Address.h"
+#include "CIRGenCstEmitter.h"
 #include "CIRGenFunction.h"
 #include "CIRGenModule.h"
 #include "CIRGenOpenMPRuntime.h"
@@ -705,7 +706,17 @@ class ScalarExprEmitter : public StmtVisitor<ScalarExprEmitter, mlir::Value> {
   mlir::Value VisitMaterializeTemporaryExpr(const MaterializeTemporaryExpr *E) {
     return emitLoadOfLValue(E);
   }
-  mlir::Value VisitSourceLocExpr(SourceLocExpr *E) { llvm_unreachable("NYI"); }
+  mlir::Value VisitSourceLocExpr(SourceLocExpr *E) {
+    auto &Ctx = CGF.getContext();
+    APValue Evaluated =
+        E->EvaluateInContext(Ctx, CGF.CurSourceLocExprScope.getDefaultExpr());
+    auto C = ConstantEmitter(CGF).emitAbstract(E->getLocation(), Evaluated,
+                                               E->getType());
+    mlir::TypedAttr typedAttr = mlir::dyn_cast_or_null<mlir::TypedAttr>(C);
+    assert(typedAttr && "SourceLocExpr must produce a typed constant");
+    return cir::ConstantOp::create(Builder, CGF.getLoc(E->getExprLoc()),
+                                   typedAttr);
+  }
   mlir::Value VisitCXXDefaultArgExpr(CXXDefaultArgExpr *DAE) {
     CIRGenFunction::CXXDefaultArgExprScope Scope(CGF, DAE);
     return Visit(DAE->getExpr());

clang/lib/CIR/CodeGen/CIRGenModule.cpp

@@ -1855,7 +1855,10 @@ cir::GlobalOp CIRGenModule::getGlobalForStringLiteral(const StringLiteral *s,
     // Unlike LLVM IR, CIR doesn't automatically unique names for globals, so
     // we need to do that explicitly.
     std::string uniqueName = getUniqueGlobalName(globalVariableName.str());
-    auto loc = getLoc(s->getSourceRange());
+    // Synthetic string literals (e.g., from SourceLocExpr) may not have valid
+    // source locations. Use unknown location in those cases.
+    auto loc = s->getBeginLoc().isValid() ? getLoc(s->getSourceRange())
+                                          : builder.getUnknownLoc();
     auto typedC = llvm::dyn_cast<mlir::TypedAttr>(c);
     if (!typedC)
       llvm_unreachable("this should never be untyped at this point");

clang/test/CIR/CodeGen/source-loc-expr.cpp

@@ -0,0 +1,73 @@
+// RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -std=c++20 -fclangir -emit-cir %s -o %t.cir
+// RUN: FileCheck --input-file=%t.cir %s
+
+// Test __builtin_LINE
+int test_builtin_LINE() {
+  // CHECK-LABEL: cir.func {{.*}}@{{.*}}test_builtin_LINE
+  // CHECK: %{{.*}} = cir.const #cir.int<8> : !u32i
+  return __builtin_LINE();
+}
+
+// Test __builtin_FILE
+const char* test_builtin_FILE() {
+  // CHECK-LABEL: cir.func {{.*}}@{{.*}}test_builtin_FILE
+  // CHECK: %{{.*}} = cir.const #cir.global_view<@".str{{.*}}"> : !cir.ptr<!s8i>
+  return __builtin_FILE();
+}
+
+// Test __builtin_FUNCTION
+const char* test_builtin_FUNCTION() {
+  // CHECK-LABEL: cir.func {{.*}}@{{.*}}test_builtin_FUNCTION
+  // CHECK: %{{.*}} = cir.const #cir.global_view<@".str{{.*}}"> : !cir.ptr<!s8i>
+  return __builtin_FUNCTION();
+}
+
+// Test __builtin_COLUMN
+int test_builtin_COLUMN() {
+  // CHECK-LABEL: cir.func {{.*}}@{{.*}}test_builtin_COLUMN
+  // The column number is the position of '__builtin_COLUMN'
+  // CHECK: %{{.*}} = cir.const #cir.int<10> : !u32i
+  return __builtin_COLUMN();
+}
+
+// Test in global context
+#line 100 "test_file.cpp"
+int global_line = __builtin_LINE();
+// CHECK: cir.global external @global_line = #cir.int<100> : !s32i
+
+// Test default argument
+int get_line(int l = __builtin_LINE()) {
+  return l;
+}
+
+void test_default_arg() {
+  // CHECK-LABEL: cir.func {{.*}}@{{.*}}test_default_arg
+  // The LINE should be from the call site, not the default argument definition
+  #line 111
+  int x = get_line();
+  // CHECK: %{{.*}} = cir.const #cir.int<111> : !u32i
+  // CHECK: %{{.*}} = cir.call @{{.*}}get_line{{.*}}({{.*}}) :
+}
+
+#line 200 "lambda-test.cpp"
+// Test in lambda (this tests that source location correctly captures context)
+void test_in_lambda() {
+  // CHECK-LABEL: cir.func {{.*}}@{{.*}}test_in_lambda
+  auto lambda = []() {
+    return __builtin_LINE();
+  };
+  int x = lambda();
+}
+
+#line 214 "combined-test.cpp"
+// Test multiple builtins in one expression
+void test_combined() {
+  // CHECK-LABEL: cir.func {{.*}}@{{.*}}test_combined
+  const char* file = __builtin_FILE();
+  int line = __builtin_LINE();
+  const char* func = __builtin_FUNCTION();
+  // All should produce constants
+  // CHECK: cir.const
+  // CHECK: cir.const
+  // CHECK: cir.const
+}