[CIR] Handle overlapping values in constant init expressions

[andykaylor]

This adds handling for generating constant initializers in the case where a value we are emitting overlaps with a previous constant, such as can happen when initializing a structure with bitfields.

[llvmbot]

@llvm/pr-subscribers-clang

@llvm/pr-subscribers-clangir

Author: Andy Kaylor (andykaylor)

Changes

This adds handling for generating constant initializers in the case where a value we are emitting overlaps with a previous constant, such as can happen when initializing a structure with bitfields.

---

Full diff: https://github.com/llvm/llvm-project/pull/164508.diff

2 Files Affected:

• (modified) clang/lib/CIR/CodeGen/CIRGenExprConstant.cpp (+17-2)
• (modified) clang/test/CIR/CodeGen/struct-init.cpp (+15)

diff --git a/clang/lib/CIR/CodeGen/CIRGenExprConstant.cpp b/clang/lib/CIR/CodeGen/CIRGenExprConstant.cpp
index 65e6a3915f241..13bcd54ece291 100644
--- a/clang/lib/CIR/CodeGen/CIRGenExprConstant.cpp
+++ b/clang/lib/CIR/CodeGen/CIRGenExprConstant.cpp
@@ -179,8 +179,23 @@ bool ConstantAggregateBuilder::add(mlir::TypedAttr typedAttr, CharUnits offset,
   }
 
   // Uncommon case: constant overlaps what we've already created.
-  cgm.errorNYI("overlapping constants");
-  return false;
+  std::optional<size_t> firstElemToReplace = splitAt(offset);
+  if (!firstElemToReplace)
+    return false;
+
+  CharUnits cSize = getSize(typedAttr);
+  std::optional<size_t> lastElemToReplace = splitAt(offset + cSize);
+  if (!lastElemToReplace)
+    return false;
+
+  assert((firstElemToReplace == lastElemToReplace || allowOverwrite) &&
+         "unexpectedly overwriting field");
+
+  Element newElt(typedAttr, offset);
+  replace(elements, *firstElemToReplace, *lastElemToReplace, {newElt});
+  size = std::max(size, offset + cSize);
+  naturalLayout = false;
+  return true;
 }
 
 bool ConstantAggregateBuilder::addBits(llvm::APInt bits, uint64_t offsetInBits,
diff --git a/clang/test/CIR/CodeGen/struct-init.cpp b/clang/test/CIR/CodeGen/struct-init.cpp
index 2887e6f404ffc..ff1b06308e74e 100644
--- a/clang/test/CIR/CodeGen/struct-init.cpp
+++ b/clang/test/CIR/CodeGen/struct-init.cpp
@@ -5,6 +5,21 @@
 // RUN: %clang_cc1 -triple x86_64-unknown-linux-gnu -emit-llvm %s -o %t.ll
 // RUN: FileCheck --check-prefix=OGCG --input-file=%t.ll %s
 
+struct O {
+  unsigned int a:4;
+  unsigned int b:14;
+  unsigned int c:14;
+};
+
+O overlapping_init = { 3, 2, 1 };
+
+// This is unintuitive. The bitfields are initialized using a struct of constants
+// that maps to the bitfields but splits the value into bytes.
+
+// CIR: cir.global external @overlapping_init = #cir.const_record<{#cir.int<35> : !u8i, #cir.int<0> : !u8i, #cir.int<4> : !u8i, #cir.int<0> : !u8i}> : !rec_anon_struct
+// LLVM: @overlapping_init = global { i8, i8, i8, i8 } { i8 35, i8 0, i8 4, i8 0 }
+// OGCG: @overlapping_init = global { i8, i8, i8, i8 } { i8 35, i8 0, i8 4, i8 0 }
+
 struct S {
   int a, b, c;
 };

[AmrDeveloper]

LGTM

[erichkeane]

Would love for this to have a better name :) Just a few characters even. Otherwise O overlapping_init looks a LOT like 1 word/was jarring.

[andykaylor]

Yeah, I didn't like that either and I'm the one who wrote it! I'm not sure why I left it that way.