[clang] Fix tbaa.struct metadata for bitfields using big endian. (PR #87753)

[Julian Nagele via cfe-commits]

https://github.com/juliannagele created https://github.com/llvm/llvm-project/pull/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.

>From cd73a85e55d683f9aa18e1099302202d663e5e7d Mon Sep 17 00:00:00 2001
From: Julian Nagele <j.nagele at apple.com>
Date: Thu, 4 Apr 2024 11:48:19 +0100
Subject: [PATCH] Fix tbaa.struct metadata for bitfields using big endian.

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. This breaks for big endian.
---
 clang/lib/CodeGen/CodeGenTBAA.cpp                      |  6 +++++-
 clang/test/CodeGen/tbaa-struct-bitfield-endianness.cpp | 10 +++-------
 2 files changed, 8 insertions(+), 8 deletions(-)

diff --git a/clang/lib/CodeGen/CodeGenTBAA.cpp b/clang/lib/CodeGen/CodeGenTBAA.cpp
index a1e14c5f0a8c78..0ddefc4751b08c 100644
--- a/clang/lib/CodeGen/CodeGenTBAA.cpp
+++ b/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 +320,10 @@ CodeGenTBAA::CollectFields(uint64_t BaseOffset,
       // base type.
       if ((*i)->isBitField()) {
         const CGBitFieldInfo &Info = CGRL.getBitFieldInfo(*i);
-        if (Info.Offset != 0)
+        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 =
diff --git a/clang/test/CodeGen/tbaa-struct-bitfield-endianness.cpp b/clang/test/CodeGen/tbaa-struct-bitfield-endianness.cpp
index 80884b49ddc669..e8bb46982537bb 100644
--- a/clang/test/CodeGen/tbaa-struct-bitfield-endianness.cpp
+++ b/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"}