3 files changed, 55 insertions, 19 deletions

diff --git a/llvm/lib/Target/ARM/ARM.td b/llvm/lib/Target/ARM/ARM.td
index 742b3551889..2e62a079041 100644
--- a/llvm/lib/Target/ARM/ARM.td
+++ b/llvm/lib/Target/ARM/ARM.td
@@ -141,6 +141,10 @@ def FeatureFPAO           : SubtargetFeature<"fpao", "HasFPAO", "true",
 def FeatureFuseAES        : SubtargetFeature<"fuse-aes", "HasFuseAES", "true",
                                              "CPU fuses AES crypto operations">;
 
+// Fast execution of bottom and top halves of literal generation
+def FeatureFuseLiterals   : SubtargetFeature<"fuse-literals", "HasFuseLiterals", "true",
+                                             "CPU fuses literal generation operations">;
+
 // The way of reading thread pointer                                             
 def FeatureReadTp :  SubtargetFeature<"read-tp-hard", "ReadTPHard", "true",
                                       "Reading thread pointer from register">;
diff --git a/llvm/lib/Target/ARM/ARMMacroFusion.cpp b/llvm/lib/Target/ARM/ARMMacroFusion.cpp
index f2dc650a6f3..d11fe9d5c50 100644
--- a/llvm/lib/Target/ARM/ARMMacroFusion.cpp
+++ b/llvm/lib/Target/ARM/ARMMacroFusion.cpp
@@ -19,6 +19,47 @@
 
 namespace llvm {
 
+// Fuse AES crypto encoding or decoding.
+static bool isAESPair(const MachineInstr *FirstMI,
+                      const MachineInstr &SecondMI) {
+  // Assume the 1st instr to be a wildcard if it is unspecified.
+  unsigned FirstOpcode =
+      FirstMI ? FirstMI->getOpcode()
+              : static_cast<unsigned>(ARM::INSTRUCTION_LIST_END);
+  unsigned SecondOpcode = SecondMI.getOpcode();
+
+  switch(SecondOpcode) {
+  // AES encode.
+  case ARM::AESMC :
+    return FirstOpcode == ARM::AESE ||
+           FirstOpcode == ARM::INSTRUCTION_LIST_END;
+  // AES decode.
+  case ARM::AESIMC:
+    return FirstOpcode == ARM::AESD ||
+           FirstOpcode == ARM::INSTRUCTION_LIST_END;
+  }
+
+  return false;
+}
+
+// Fuse literal generation.
+static bool isLiteralsPair(const MachineInstr *FirstMI,
+                           const MachineInstr &SecondMI) {
+  // Assume the 1st instr to be a wildcard if it is unspecified.
+  unsigned FirstOpcode =
+      FirstMI ? FirstMI->getOpcode()
+              : static_cast<unsigned>(ARM::INSTRUCTION_LIST_END);
+  unsigned SecondOpcode = SecondMI.getOpcode();
+
+  // 32 bit immediate.
+  if ((FirstOpcode == ARM::INSTRUCTION_LIST_END ||
+       FirstOpcode == ARM::MOVi16) &&
+      SecondOpcode == ARM::MOVTi16)
+    return true;
+
+  return false;
+}
+
 /// Check if the instr pair, FirstMI and SecondMI, should be fused
 /// together. Given SecondMI, when FirstMI is unspecified, then check if
 /// SecondMI may be part of a fused pair at all.
@@ -28,24 +69,10 @@ static bool shouldScheduleAdjacent(const TargetInstrInfo &TII,
                                    const MachineInstr &SecondMI) {
   const ARMSubtarget &ST = static_cast<const ARMSubtarget&>(TSI);
 
-  // Assume wildcards for unspecified instrs.
-  unsigned FirstOpcode =
-    FirstMI ? FirstMI->getOpcode()
-            : static_cast<unsigned>(ARM::INSTRUCTION_LIST_END);
-  unsigned SecondOpcode = SecondMI.getOpcode();
-
-  if (ST.hasFuseAES())
-    // Fuse AES crypto operations.
-    switch(SecondOpcode) {
-    // AES encode.
-    case ARM::AESMC :
-      return FirstOpcode == ARM::AESE ||
-             FirstOpcode == ARM::INSTRUCTION_LIST_END;
-    // AES decode.
-    case ARM::AESIMC:
-      return FirstOpcode == ARM::AESD ||
-             FirstOpcode == ARM::INSTRUCTION_LIST_END;
-    }
+  if (ST.hasFuseAES() && isAESPair(FirstMI, SecondMI))
+    return true;
+  if (ST.hasFuseLiterals() && isLiteralsPair(FirstMI, SecondMI))
+    return true;
 
   return false;
 }
diff --git a/llvm/lib/Target/ARM/ARMSubtarget.h b/llvm/lib/Target/ARM/ARMSubtarget.h
index 165077926c8..74aee9a8ed3 100644
--- a/llvm/lib/Target/ARM/ARMSubtarget.h
+++ b/llvm/lib/Target/ARM/ARMSubtarget.h
@@ -327,6 +327,10 @@ protected:
   /// pairs faster.
   bool HasFuseAES = false;
 
+  /// HasFuseLiterals - if true, processor executes back to back
+  /// bottom and top halves of literal generation faster.
+  bool HasFuseLiterals = false;
+
   /// If true, if conversion may decide to leave some instructions unpredicated.
   bool IsProfitableToUnpredicate = false;
 
@@ -616,8 +620,9 @@ public:
   bool hasFullFP16() const { return HasFullFP16; }
 
   bool hasFuseAES() const { return HasFuseAES; }
+  bool hasFuseLiterals() const { return HasFuseLiterals; }
   /// Return true if the CPU supports any kind of instruction fusion.
-  bool hasFusion() const { return hasFuseAES(); }
+  bool hasFusion() const { return hasFuseAES() || hasFuseLiterals(); }
 
   const Triple &getTargetTriple() const { return TargetTriple; }