AArch64: lower "fence singlethread" to a pure compiler barrier.

Single-threaded fences aren't required to provide any synchronization with
other processing elements so there's no need for a DMB. They should still be a
barrier for compiler optimizations though.

llvm-svn: 300905

4 files changed, 14 insertions, 0 deletions

diff --git a/llvm/lib/Target/AArch64/AArch64ExpandPseudoInsts.cpp b/llvm/lib/Target/AArch64/AArch64ExpandPseudoInsts.cpp
index d0c0956b87c..629ad5c61b7 100644
--- a/llvm/lib/Target/AArch64/AArch64ExpandPseudoInsts.cpp
+++ b/llvm/lib/Target/AArch64/AArch64ExpandPseudoInsts.cpp
@@ -942,6 +942,7 @@ bool AArch64ExpandPseudo::expandMI(MachineBasicBlock &MBB,
                           AArch64::XZR, NextMBBI);
   case AArch64::CMP_SWAP_128:
     return expandCMP_SWAP_128(MBB, MBBI, NextMBBI);
+
   }
   return false;
 }
diff --git a/llvm/lib/Target/AArch64/AArch64InstrAtomics.td b/llvm/lib/Target/AArch64/AArch64InstrAtomics.td
index 867074c3c37..71826bec6b1 100644
--- a/llvm/lib/Target/AArch64/AArch64InstrAtomics.td
+++ b/llvm/lib/Target/AArch64/AArch64InstrAtomics.td
@@ -14,6 +14,9 @@
 //===----------------------------------
 // Atomic fences
 //===----------------------------------
+let AddedComplexity = 15, Size = 0 in
+def CompilerBarrier : Pseudo<(outs), (ins i32imm:$ordering),
+                             [(atomic_fence imm:$ordering, 0)]>, Sched<[]>;
 def : Pat<(atomic_fence (i64 4), (imm)), (DMB (i32 0x9))>;
 def : Pat<(atomic_fence (imm), (imm)), (DMB (i32 0xb))>;
 
diff --git a/llvm/lib/Target/AArch64/InstPrinter/AArch64InstPrinter.cpp b/llvm/lib/Target/AArch64/InstPrinter/AArch64InstPrinter.cpp
index 41ae70f85e5..fc89657bffd 100644
--- a/llvm/lib/Target/AArch64/InstPrinter/AArch64InstPrinter.cpp
+++ b/llvm/lib/Target/AArch64/InstPrinter/AArch64InstPrinter.cpp
@@ -17,6 +17,7 @@
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/ADT/StringExtras.h"
 #include "llvm/ADT/StringRef.h"
+#include "llvm/MC/MCAsmInfo.h"
 #include "llvm/MC/MCExpr.h"
 #include "llvm/MC/MCInst.h"
 #include "llvm/MC/MCRegisterInfo.h"
@@ -275,6 +276,12 @@ void AArch64InstPrinter::printInst(const MCInst *MI, raw_ostream &O,
     }
   }
 
+  if (Opcode == AArch64::CompilerBarrier) {
+    O << '\t' << MAI.getCommentString() << " COMPILER BARRIER";
+    printAnnotation(O, Annot);
+    return;
+  }
+
   if (!printAliasInstr(MI, STI, O))
     printInstruction(MI, STI, O);
 
diff --git a/llvm/lib/Target/AArch64/MCTargetDesc/AArch64MCCodeEmitter.cpp b/llvm/lib/Target/AArch64/MCTargetDesc/AArch64MCCodeEmitter.cpp
index 62dfa59483e..33698d2b8c3 100644
--- a/llvm/lib/Target/AArch64/MCTargetDesc/AArch64MCCodeEmitter.cpp
+++ b/llvm/lib/Target/AArch64/MCTargetDesc/AArch64MCCodeEmitter.cpp
@@ -565,6 +565,9 @@ void AArch64MCCodeEmitter::encodeInstruction(const MCInst &MI, raw_ostream &OS,
     MCFixupKind Fixup = MCFixupKind(AArch64::fixup_aarch64_tlsdesc_call);
     Fixups.push_back(MCFixup::create(0, MI.getOperand(0).getExpr(), Fixup));
     return;
+  } else if (MI.getOpcode() == AArch64::CompilerBarrier) {
+    // This just prevents the compiler from reordering accesses, no actual code.
+    return;
   }
 
   uint64_t Binary = getBinaryCodeForInstr(MI, Fixups, STI);