[llvm] [bolt] Simplify rodata/literal load for X86_64 & AArch64 (PR #179474)

[YongKang Zhu via llvm-commits]

================
@@ -1261,28 +1262,49 @@ bool SimplifyRODataLoads::simplifyRODataLoads(BinaryFunction &BF) {
       }
 
       // Get the contents of the section containing the target address of the
-      // memory operand. We are only interested in read-only sections.
+      // memory operand. We are only interested in read-only sections for X86,
+      // for aarch64 the sections can be read-only or executable.
       ErrorOr<BinarySection &> DataSection =
           BC.getSectionForAddress(TargetAddress);
       if (!DataSection || DataSection->isWritable())
         continue;
 
+      if (DataSection->isText()) {
+        // If data is not part of a function, check if it is part of a global CI
+        // Do not proceed if there aren't data markers for CIs
+        BinaryFunction *TargetBF =
+            BC.getBinaryFunctionContainingAddress(TargetAddress,
+                                                  /*CheckPastEnd*/ false,
+                                                  /*UseMaxSize*/ true);
+        const bool IsInsideFunc =
+            TargetBF && TargetBF->isInConstantIsland(TargetAddress);
+
+        auto CIEndIter = BC.AddressToConstantIslandMap.end();
+        auto CIIter = BC.AddressToConstantIslandMap.find(TargetAddress);
+        if (!IsInsideFunc && CIIter == CIEndIter)
----------------
yozhu wrote:

Your code is correct. I got it wrong ..

https://github.com/llvm/llvm-project/pull/179474