diff --git a/bolt/include/bolt/Core/MCPlusBuilder.h b/bolt/include/bolt/Core/MCPlusBuilder.h
index 2e167ffb29091cec2eace1c3302ef041d7ec49fc..7f4f8fcc090dd91f468e1a0f9df0ac401e678a2e 100644
--- a/bolt/include/bolt/Core/MCPlusBuilder.h
+++ b/bolt/include/bolt/Core/MCPlusBuilder.h
@@ -1978,7 +1978,8 @@ public:
   }
 
   /// Print each annotation attached to \p Inst.
-  void printAnnotations(const MCInst &Inst, raw_ostream &OS) const;
+  void printAnnotations(const MCInst &Inst, raw_ostream &OS,
+                        bool PrintMemData = false) const;
 
   /// Remove annotation with a given \p Index.
   ///
diff --git a/bolt/lib/Core/BinaryContext.cpp b/bolt/lib/Core/BinaryContext.cpp
index 7c3b4a87cfac8b63f4ca98e0dcf76bcdeecf7351..446d12b8312beecaff18b6794fb7e29a59a9e7e3 100644
--- a/bolt/lib/Core/BinaryContext.cpp
+++ b/bolt/lib/Core/BinaryContext.cpp
@@ -1996,7 +1996,7 @@ void BinaryContext::printInstruction(raw_ostream &OS, const MCInst &Instruction,
   if (MCSymbol *Label = MIB->getInstLabel(Instruction))
     OS << " # Label: " << *Label;
 
-  MIB->printAnnotations(Instruction, OS);
+  MIB->printAnnotations(Instruction, OS, PrintMemData || opts::PrintMemData);
 
   if (opts::PrintDebugInfo)
     printDebugInfo(OS, Instruction, Function, DwCtx.get());
diff --git a/bolt/lib/Core/MCPlusBuilder.cpp b/bolt/lib/Core/MCPlusBuilder.cpp
index 778528d251b6b83980654929c0c456f4c848be46..7eb6c5ff390bcc4044dee1949b1ffa61564205d5 100644
--- a/bolt/lib/Core/MCPlusBuilder.cpp
+++ b/bolt/lib/Core/MCPlusBuilder.cpp
@@ -379,8 +379,8 @@ void MCPlusBuilder::stripAnnotations(MCInst &Inst, bool KeepTC) const {
     setTailCall(Inst);
 }
 
-void MCPlusBuilder::printAnnotations(const MCInst &Inst,
-                                     raw_ostream &OS) const {
+void MCPlusBuilder::printAnnotations(const MCInst &Inst, raw_ostream &OS,
+                                     bool PrintMemData) const {
   std::optional<unsigned> FirstAnnotationOp = getFirstAnnotationOpIndex(Inst);
   if (!FirstAnnotationOp)
     return;
@@ -391,7 +391,11 @@ void MCPlusBuilder::printAnnotations(const MCInst &Inst,
     const int64_t Value = extractAnnotationValue(Imm);
     const auto *Annotation = reinterpret_cast<const MCAnnotation *>(Value);
     if (Index >= MCAnnotation::kGeneric) {
-      OS << " # " << AnnotationNames[Index - MCAnnotation::kGeneric] << ": ";
+      std::string AnnotationName =
+          AnnotationNames[Index - MCAnnotation::kGeneric];
+      if (!PrintMemData && AnnotationName == "MemoryAccessProfile")
+        continue;
+      OS << " # " << AnnotationName << ": ";
       Annotation->print(OS);
     }
   }
diff --git a/bolt/lib/Passes/BinaryPasses.cpp b/bolt/lib/Passes/BinaryPasses.cpp
index fa95ad7324ac1cda944c386df94deabe72dbb007..5b7f7d40244c075ab56dac960908fde8166068ca 100644
--- a/bolt/lib/Passes/BinaryPasses.cpp
+++ b/bolt/lib/Passes/BinaryPasses.cpp
@@ -59,7 +59,12 @@ static cl::opt<DynoStatsSortOrder> DynoStatsSortOrderOpt(
     "print-sorted-by-order",
     cl::desc("use ascending or descending order when printing functions "
              "ordered by dyno stats"),
-    cl::init(DynoStatsSortOrder::Descending), cl::cat(BoltOptCategory));
+    cl::init(DynoStatsSortOrder::Descending),
+    cl::values(clEnumValN(DynoStatsSortOrder::Ascending, "ascending",
+                          "Ascending order"),
+               clEnumValN(DynoStatsSortOrder::Descending, "descending",
+                          "Descending order")),
+    cl::cat(BoltOptCategory));
 
 cl::list<std::string>
 HotTextMoveSections("hot-text-move-sections",
diff --git a/bolt/test/AArch64/print-mem-data.test b/bolt/test/AArch64/print-mem-data.test
new file mode 100644
index 0000000000000000000000000000000000000000..09d4f4640a45413cd4e0dce066267bdb4691b004
--- /dev/null
+++ b/bolt/test/AArch64/print-mem-data.test
@@ -0,0 +1,40 @@
+# Check that --print-mem-data option works properly in llvm-bolt
+
+# RUN: split-file %s %t
+# RUN: %clang %cflags -fPIC -pie %t/main.s -o %t.exe -nostdlib -Wl,-q
+# RUN: llvm-bolt %t.exe -o %t.bolt --print-mem-data=true --print-cfg \
+# RUN:   --data %t/fdata | FileCheck %s -check-prefix=CHECK-PRINT
+# RUN: llvm-bolt %t.exe -o %t.bolt --print-cfg \
+# RUN:   --data %t/fdata | FileCheck %s -check-prefix=CHECK-DEFAULT
+
+# CHECK-PRINT: ldr	w2, [x1], #0x4 # MemoryAccessProfile: 7 total counts :
+# CHECK-PRINT-NEXT: { 0x123: 1 },
+# CHECK-PRINT-NEXT: { 0x456: 2 },
+# CHECK-PRINT-NEXT: { 0xabc: 4 }
+# CHECK-DEFAULT-NOT: MemoryAccessProfile
+
+#--- main.s
+  .text
+  .align 4
+  .global main
+  .type	main, %function
+main:
+  sub sp, sp, #48
+  add x1, sp, 8
+  add x3, sp, 48
+  mov w0, 0
+.L2:
+  ldr w2, [x1], 4
+  add w0, w0, w2
+  cmp x1, x3
+  bne .L2
+  add sp, sp, 48
+  ret
+  .size main, .-main
+
+# The three memory access data generated by the load at
+# offset 0x10 in the main.
+#--- fdata
+4 main 10 4 otherSym 123 1
+4 main 10 4 otherSym 456 2
+4 main 10 4 otherSym abc 4
diff --git a/bolt/test/AArch64/print-sorted-by-order.s b/bolt/test/AArch64/print-sorted-by-order.s
new file mode 100644
index 0000000000000000000000000000000000000000..30b5297ec5cd2e43fa126de1958f15fdad2918e3
--- /dev/null
+++ b/bolt/test/AArch64/print-sorted-by-order.s
@@ -0,0 +1,31 @@
+# Check that --print-sorted-by-order=<ascending/descending> option works properly in llvm-bolt
+#
+# RUN: llvm-mc -filetype=obj -triple aarch64-unknown-unknown %s -o %t.o
+# RUN: %clang %cflags -fPIC -pie %t.o -o %t.exe -nostdlib -Wl,-q
+# RUN: link_fdata %s %t.o %t.fdata
+# RUN: llvm-bolt %t.exe -o %t.bolt --print-sorted-by=all --print-sorted-by-order=ascending \
+# RUN:   --data %t.fdata | FileCheck %s -check-prefix=CHECK-ASCEND
+# RUN: llvm-bolt %t.exe -o %t.bolt --print-sorted-by=all --print-sorted-by-order=descending \
+# RUN:   --data %t.fdata | FileCheck %s -check-prefix=CHECK-DESCEND
+
+# CHECK-ASCEND: BOLT-INFO: top functions sorted by dyno stats are:
+# CHECK-ASCEND-NEXT: bar
+# CHECK-ASCEND-NEXT: foo
+# CHECK-DESCEND: BOLT-INFO: top functions sorted by dyno stats are:
+# CHECK-DESCEND-NEXT: foo
+# CHECK-DESCEND-NEXT: bar
+
+  .text
+  .align 4
+  .global bar
+  .type bar, %function
+bar:
+  mov w0, wzr
+  ret
+
+  .global foo
+  .type foo, %function
+foo:
+# FDATA: 1 foo 0 1 bar 0 0 1
+  bl bar
+  ret