[clang] [clang] Lower _BitInt(129+) to a different type in LLVM IR (PR #91364)

[John McCall via cfe-commits]

================
@@ -1774,6 +1784,22 @@ llvm::Constant *ConstantEmitter::emitForMemory(CodeGenModule &CGM,
     return Res;
   }
 
+  if (destType->isBitIntType()) {
+    if (!CGM.getTypes().LLVMTypeLayoutMatchesAST(destType, C->getType())) {
+      // Long _BitInt has array of bytes as in-memory type.
+      // So, split constant into individual bytes.
+      ConstantAggregateBuilder Builder(CGM);
+      llvm::Type *DesiredTy = CGM.getTypes().ConvertTypeForMem(destType);
+      // LLVM type doesn't match AST type only for big enough _BitInts, these
+      // types don't appear in constant expressions involving ptrtoint, so it
+      // is safe to expect a constant int here.
+      auto *CI = cast<llvm::ConstantInt>(C);
+      llvm::APInt Value = CI->getValue();
+      Builder.addBits(Value, /*OffsetInBits=*/0, /*AllowOverwrite=*/false);
+      return Builder.build(DesiredTy, /*AllowOversized*/ false);
----------------
rjmccall wrote:

`C` is of the scalar type, not the load/store type, right?  So tail-padding with `zeroinitializer` is still wrong:
- Given a negative value, we need to fill the "padding" with 1s, not 0s.
- On a big-endian target, the padding needs to come before the value, not after it.
Both of these are fixed by sign/zero-extending the constant to the load/store type.

https://github.com/llvm/llvm-project/pull/91364