Fix tbaa.struct metadata for bitfields using big endian. (llvm#87753)

When generating tbaa.struct metadata we treat multiple adjacent
bitfields as a single "field", with one corresponding entry in the
metadata. At the moment this is achieved by adding an entry for the
first bitfield in the run using its StorageSize and skipping the
remaining bitfields. The problem is that "first" is determined by
checking that the Offset of the field in the run is 0, which breaks for
big endian.

PR: llvm#87753

clang/lib/CodeGen/CodeGenTBAA.cpp

@@ -22,6 +22,7 @@
 #include "clang/AST/Mangle.h"
 #include "clang/AST/RecordLayout.h"
 #include "clang/Basic/CodeGenOptions.h"
+#include "clang/Basic/TargetInfo.h"
 #include "llvm/ADT/SmallSet.h"
 #include "llvm/IR/Constants.h"
 #include "llvm/IR/LLVMContext.h"
@@ -319,7 +320,13 @@ CodeGenTBAA::CollectFields(uint64_t BaseOffset,
       // base type.
       if ((*i)->isBitField()) {
         const CGBitFieldInfo &Info = CGRL.getBitFieldInfo(*i);
-        if (Info.Offset != 0)
+        // For big endian targets the first bitfield in the consecutive run is
+        // at the most-significant end; see CGRecordLowering::setBitFieldInfo
+        // for more information.
+        bool IsBE = Context.getTargetInfo().isBigEndian();
+        bool IsFirst = IsBE ? Info.StorageSize - (Info.Offset + Info.Size) == 0
+                            : Info.Offset == 0;
+        if (!IsFirst)
           continue;
         unsigned CurrentBitFieldSize = Info.StorageSize;
         uint64_t Size =

clang/test/CodeGen/tbaa-struct-bitfield-endianness.cpp

@@ -1,13 +1,10 @@
 // RUN: %clang_cc1 -triple aarch64_be-apple-darwin -emit-llvm -o - -O1 %s | \
-// RUN:     FileCheck -check-prefixes=CHECK,CHECK-BE %s
+// RUN:     FileCheck -check-prefixes=CHECK %s
 // RUN: %clang_cc1 -triple aarch64-apple-darwin -emit-llvm -o - -O1 %s | \
-// RUN:     FileCheck -check-prefixes=CHECK,CHECK-LE %s
+// RUN:     FileCheck -check-prefixes=CHECK %s
 //
 // Check that TBAA metadata for structs containing bitfields is
 // consistent between big and little endian layouts.
-//
-// FIXME: The metadata below is invalid for the big endian layout: the
-// start offset of 2 is incorrect.
 
 struct NamedBitfields {
   int f1 : 8;
@@ -28,8 +25,7 @@ void copy(NamedBitfields *a1, NamedBitfields *a2) {
   *a1 = *a2;
 }
 
-// CHECK-BE: [[TBAA_STRUCT2]] = !{i64 2, i64 4, [[META3:![0-9]+]], i64 4, i64 4, [[META6:![0-9]+]], i64 8, i64 8, [[META8:![0-9]+]]}
-// CHECK-LE: [[TBAA_STRUCT2]] = !{i64 0, i64 4, [[META3:![0-9]+]], i64 4, i64 4, [[META6:![0-9]+]], i64 8, i64 8, [[META8:![0-9]+]]}
+// CHECK: [[TBAA_STRUCT2]] = !{i64 0, i64 4, [[META3:![0-9]+]], i64 4, i64 4, [[META6:![0-9]+]], i64 8, i64 8, [[META8:![0-9]+]]}
 // CHECK: [[META3]] = !{[[META4:![0-9]+]], [[META4]], i64 0}
 // CHECK: [[META4]] = !{!"omnipotent char", [[META5:![0-9]+]], i64 0}
 // CHECK: [[META5]] = !{!"Simple C++ TBAA"}