[clang] [CIR] Upstream get_bitfield operation to load bit-field members from structs (PR #145971)

[via cfe-commits]

================
@@ -2244,6 +2245,54 @@ mlir::LogicalResult CIRToLLVMComplexImagOpLowering::matchAndRewrite(
   return mlir::success();
 }
 
+mlir::LogicalResult CIRToLLVMGetBitfieldOpLowering::matchAndRewrite(
+    cir::GetBitfieldOp op, OpAdaptor adaptor,
+    mlir::ConversionPatternRewriter &rewriter) const {
+
+  mlir::OpBuilder::InsertionGuard guard(rewriter);
+  rewriter.setInsertionPoint(op);
+
+  cir::BitfieldInfoAttr info = op.getBitfieldInfo();
+  uint64_t size = info.getSize();
+  uint64_t offset = info.getOffset();
+  mlir::Type storageType = info.getStorageType();
+  mlir::MLIRContext *context = storageType.getContext();
+  unsigned storageSize = 0;
+
+  if (auto arTy = mlir::dyn_cast<cir::ArrayType>(storageType))
----------------
Andres-Salamanca wrote:

Yes, we can still produce arrays when clipping. Here's an example: https://godbolt.org/z/sGae8vcze. The CodeGen explanation for this case is here:
https://github.com/llvm/llvm-project/blob/fb24b4d46a0a8278031f42c1cba6c030eb6c6010/clang/lib/CodeGen/CGRecordLayoutBuilder.cpp#L44C1-L49C27

Also, clipping can occur when the target does not support unaligned bitfield access, when `!astContext.getTargetInfo().hasCheapUnalignedBitFieldAccess()` is true. However, for our case, that's currently NYI.

https://github.com/llvm/llvm-project/pull/145971