[AArch64] Add new target feature to fuse arithmetic and logic operations

[evandro]

Summary

This feature enables the fusion of some arithmetic and logic instructions together.

[fhahn]

Would it be possible to have a MIR test here, which just runs the scheduler?

llvm/lib/Target/AArch64/AArch64Subtarget.h

324

Not related to this patch directly, but is there a reason hasFuseCryptoEOR() is missing here?

[evandro]

In D56572#1354333, @fhahn wrote:

Would it be possible to have a MIR test here, which just runs the scheduler?

I tried to keep the existing pattern, but I could add one too. Should it be done now or later?

llvm/lib/Target/AArch64/AArch64Subtarget.h

324

Since Cyclone is the only target using it, I guess that it doesn't care about running the MachineScheduler. At least not in the public repository. Though in the associated test case it is enabled in the command line.

[fhahn]

In D56572#1354350, @evandro wrote:

In D56572#1354333, @fhahn wrote:

Would it be possible to have a MIR test here, which just runs the scheduler?

I tried to keep the existing pattern, but I could add one too. Should it be done now or later?

I think it would be good to get a MIR test in from the start as it is easier to see what's going, similar to https://github.com/llvm-mirror/llvm/blob/master/test/CodeGen/AArch64/misched-fusion-crypto-eor.mir .

[evandro]

I'm afraid that I'll need more time to craft the MIR file. Then, I might as well do the same for the other fusion tests.

[evandro]

Replaced the test case with a MIR file.

[fhahn]

LGTM, thanks! It would be great if you could extend the test cases a bit more before committing, covering cases with shifted regs and cases where some SUs have multiple successors and not all of them should be fused.

llvm/lib/Target/AArch64/AArch64MacroFusion.cpp

342

This indent is off by one level I think.

llvm/lib/Target/AArch64/AArch64.td

 def FeatureFuseAddress : SubtargetFeature<
     "fuse-address", "HasFuseAddress", "true",
     "CPU fuses address generation and memory operations">;
 
 def FeatureFuseAES : SubtargetFeature<
     "fuse-aes", "HasFuseAES", "true",
     "CPU fuses AES crypto operations">;
 
-def FeatureFuseCryptoEOR : SubtargetFeature<
-    "fuse-crypto-eor", "HasFuseCryptoEOR", "true",
-    "CPU fuses AES/PMULL and EOR operations">;
+def FeatureFuseArithmeticLogic : SubtargetFeature<
+    "fuse-arith-logic", "HasFuseArithmeticLogic", "true",
+    "CPU fuses arithmetic and logic operations">;
 
 def FeatureFuseCCSelect : SubtargetFeature<
     "fuse-csel", "HasFuseCCSelect", "true",
     "CPU fuses conditional select operations">;
 
+def FeatureFuseCryptoEOR : SubtargetFeature<
+    "fuse-crypto-eor", "HasFuseCryptoEOR", "true",
+    "CPU fuses AES/PMULL and EOR operations">;
+
 def FeatureFuseLiterals : SubtargetFeature<
     "fuse-literals", "HasFuseLiterals", "true",
     "CPU fuses literal generation operations">;
 
 def FeatureDisableLatencySchedHeuristic : SubtargetFeature<
     "disable-latency-sched-heuristic", "DisableLatencySchedHeuristic", "true",
     "Disable latency scheduling heuristic">;
 

llvm/lib/Target/AArch64/AArch64MacroFusion.cpp

if (FirstMI->definesRegister(AArch64::XZR))
       case AArch64::SUBSXri:
         return true;
       }
   }
 
   return false;
 }
 
-/// Check if the instr pair, FirstMI and SecondMI, should be fused
+// Arithmetic and logic.
+static bool isArithmeticLogicPair(const MachineInstr *FirstMI,
+                                  const MachineInstr &SecondMI) {
+  if (AArch64InstrInfo::hasShiftedReg(SecondMI))
+    return false;
+
+  switch (SecondMI.getOpcode()) {
+  // Arithmetic
+  case AArch64::ADDWrr:
+  case AArch64::ADDXrr:
+  case AArch64::SUBWrr:
+  case AArch64::SUBXrr:
+  case AArch64::ADDWrs:
+  case AArch64::ADDXrs:
+  case AArch64::SUBWrs:
+  case AArch64::SUBXrs:
+  // Logic
+  case AArch64::ANDWrr:
+  case AArch64::ANDXrr:
+  case AArch64::BICWrr:
+  case AArch64::BICXrr:
+  case AArch64::EONWrr:
+  case AArch64::EONXrr:
+  case AArch64::EORWrr:
+  case AArch64::EORXrr:
+  case AArch64::ORNWrr:
+  case AArch64::ORNXrr:
+  case AArch64::ORRWrr:
+  case AArch64::ORRXrr:
+  case AArch64::ANDWrs:
+  case AArch64::ANDXrs:
+  case AArch64::BICWrs:
+  case AArch64::BICXrs:
+  case AArch64::EONWrs:
+  case AArch64::EONXrs:
+  case AArch64::EORWrs:
+  case AArch64::EORXrs:
+  case AArch64::ORNWrs:
+  case AArch64::ORNXrs:
+  case AArch64::ORRWrs:
+  case AArch64::ORRXrs:
+    // Assume the 1st instr to be a wildcard if it is unspecified.
+    if (FirstMI == nullptr)
+      return true;
+
+    // Arithmetic
+    switch (FirstMI->getOpcode()) {
+    case AArch64::ADDWrr:
+    case AArch64::ADDXrr:
+    case AArch64::ADDSWrr:
+    case AArch64::ADDSXrr:
+    case AArch64::SUBWrr:
+    case AArch64::SUBXrr:
+    case AArch64::SUBSWrr:
+    case AArch64::SUBSXrr:
+      return true;
+    case AArch64::ADDWrs:
+    case AArch64::ADDXrs:
+    case AArch64::ADDSWrs:
+    case AArch64::ADDSXrs:
+    case AArch64::SUBWrs:
+    case AArch64::SUBXrs:
+    case AArch64::SUBSWrs:
+    case AArch64::SUBSXrs:
+      return !AArch64InstrInfo::hasShiftedReg(*FirstMI);
+    }
+    break;
+
+  // Arithmetic setting flags.
+    case AArch64::ADDSWrr:
+    case AArch64::ADDSXrr:
+    case AArch64::SUBSWrr:
+    case AArch64::SUBSXrr:
+    case AArch64::ADDSWrs:
+    case AArch64::ADDSXrs:
+    case AArch64::SUBSWrs:
+    case AArch64::SUBSXrs:
+    // Assume the 1st instr to be a wildcard if it is unspecified.
+    if (FirstMI == nullptr)
+      return true;
+
+    // Arithmetic
+    switch (FirstMI->getOpcode()) {
+    case AArch64::ADDWrr:
+    case AArch64::ADDXrr:
+    case AArch64::SUBWrr:
+    case AArch64::SUBXrr:
+      return true;
+    case AArch64::ADDWrs:
+    case AArch64::ADDXrs:
+    case AArch64::SUBWrs:
+    case AArch64::SUBXrs:
+      return !AArch64InstrInfo::hasShiftedReg(*FirstMI);
+    }
+    break;
+  }
+
+  return false;
+}
+
+/// \brief 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.
 static bool shouldScheduleAdjacent(const TargetInstrInfo &TII,
                                    const TargetSubtargetInfo &TSI,
                                    const MachineInstr *FirstMI,
                                    const MachineInstr &SecondMI) {
   const AArch64Subtarget &ST = static_cast<const AArch64Subtarget&>(TSI);
 
   // All checking functions assume that the 1st instr is a wildcard if it is
   // unspecified.
   if (ST.hasArithmeticBccFusion() && isArithmeticBccPair(FirstMI, SecondMI))
     return true;
   if (ST.hasArithmeticCbzFusion() && isArithmeticCbzPair(FirstMI, SecondMI))
     return true;
   if (ST.hasFuseAES() && isAESPair(FirstMI, SecondMI))
     return true;
   if (ST.hasFuseCryptoEOR() && isCryptoEORPair(FirstMI, SecondMI))
     return true;
   if (ST.hasFuseLiterals() && isLiteralsPair(FirstMI, SecondMI))
     return true;
   if (ST.hasFuseAddress() && isAddressLdStPair(FirstMI, SecondMI))
     return true;
   if (ST.hasFuseCCSelect() && isCCSelectPair(FirstMI, SecondMI))
     return true;
+  if (ST.hasFuseArithmeticLogic() && isArithmeticLogicPair(FirstMI, SecondMI))
+    return true;
 
   return false;
 }
 
 } // end namespace
 
 
 namespace llvm {
 
 std::unique_ptr<ScheduleDAGMutation> createAArch64MacroFusionDAGMutation () {
   return createMacroFusionDAGMutation(shouldScheduleAdjacent);
 }
 
 } // end namespace llvm

llvm/lib/Target/AArch64/AArch64Subtarget.h

protected:
   bool Misaligned128StoreIsSlow = false;
   bool Paired128IsSlow = false;
   bool STRQroIsSlow = false;
   bool UseAlternateSExtLoadCVTF32Pattern = false;
   bool HasArithmeticBccFusion = false;
   bool HasArithmeticCbzFusion = false;
   bool HasFuseAddress = false;
   bool HasFuseAES = false;
-  bool HasFuseCryptoEOR = false;
+  bool HasFuseArithmeticLogic = false;
   bool HasFuseCCSelect = false;
+  bool HasFuseCryptoEOR = false;
   bool HasFuseLiterals = false;
   bool DisableLatencySchedHeuristic = false;
   bool UseRSqrt = false;
   bool Force32BitJumpTables = false;
   uint8_t MaxInterleaveFactor = 2;
   uint8_t VectorInsertExtractBaseCost = 3;
   uint16_t CacheLineSize = 0;
   uint16_t PrefetchDistance = 0;
public:
   bool isSTRQroSlow() const { return STRQroIsSlow; }
   bool useAlternateSExtLoadCVTF32Pattern() const {
     return UseAlternateSExtLoadCVTF32Pattern;
   }
   bool hasArithmeticBccFusion() const { return HasArithmeticBccFusion; }
   bool hasArithmeticCbzFusion() const { return HasArithmeticCbzFusion; }
   bool hasFuseAddress() const { return HasFuseAddress; }
   bool hasFuseAES() const { return HasFuseAES; }
-  bool hasFuseCryptoEOR() const { return HasFuseCryptoEOR; }
+  bool hasFuseArithmeticLogic() const { return HasFuseArithmeticLogic; }
   bool hasFuseCCSelect() const { return HasFuseCCSelect; }
+  bool hasFuseCryptoEOR() const { return HasFuseCryptoEOR; }
   bool hasFuseLiterals() const { return HasFuseLiterals; }
 
   /// Return true if the CPU supports any kind of instruction fusion.
   bool hasFusion() const {
     return hasArithmeticBccFusion() || hasArithmeticCbzFusion() ||
-           hasFuseAES() || hasFuseCCSelect() || hasFuseLiterals();
+           hasFuseAES() || hasFuseArithmeticLogic() ||
+           hasFuseCCSelect() || hasFuseLiterals();
   }
 
   bool useRSqrt() const { return UseRSqrt; }
   bool force32BitJumpTables() const { return Force32BitJumpTables; }
   unsigned getMaxInterleaveFactor() const { return MaxInterleaveFactor; }
   unsigned getVectorInsertExtractBaseCost() const {
     return VectorInsertExtractBaseCost;
   }

llvm/test/CodeGen/AArch64/misched-fusion-arith-logic.mir

+# RUN: llc -o - %s -mtriple aarch64-unknown -mattr=fuse-arith-logic -run-pass=machine-scheduler -misched-print-dags |& FileCheck %s
+# REQUIRES: asserts
+
+---
+name: arith
+body: |
+  bb.0.entry:
+    %0:gpr32 = SUBWrr undef $w0, undef $w1
+    %1:gpr32 = ADDWrr undef $w1, undef $w2
+    %2:gpr32 = SUBWrr %0, undef $w2
+    %3:gpr32 = ADDWrr %1, undef $w3
+
+    ; CHECK: SU(0):      %0:gpr32 = SUBWrr undef $w0, undef $w1
+    ; CHECK: Successors:
+    ; CHECK: SU(2): Ord  Latency=0 Cluster
+    ; CHECK: SU(1):      %1:gpr32 = ADDWrr undef $w1, undef $w2
+    ; CHECK: Successors:
+    ; CHECK: SU(3): Ord  Latency=0 Cluster
+    ; CHECK: SU(2): dead %2:gpr32 = SUBWrr %0:gpr32, undef $w2
+    ; CHECK: Predecessors:
+    ; CHECK: SU(0): Ord  Latency=0 Cluster
+    ; CHECK: SU(3): dead %3:gpr32 = ADDWrr %1:gpr32, undef $w3
+    ; CHECK: Predecessors:
+    ; CHECK: SU(1): Ord  Latency=0 Cluster
+...
+---
+name: compare
+body: |
+  bb.0.entry:
+    %0:gpr64 = ADDXrr undef $x0, undef $x1
+    %1:gpr64 = SUBXrr undef $x1, undef $x2
+    %2:gpr64 = ADDSXrr %0, undef $x3, implicit-def $nzcv
+    %3:gpr64 = SUBSXrr %1, undef $x4, implicit-def $nzcv
+
+    ; CHECK: SU(0):      %0:gpr64 = ADDXrr undef $x0, undef $x1
+    ; CHECK: Successors:
+    ; CHECK: SU(2): Ord  Latency=0 Cluster
+    ; CHECK: SU(1):      %1:gpr64 = SUBXrr undef $x1, undef $x2
+    ; CHECK: Successors:
+    ; CHECK: SU(3): Ord  Latency=0 Cluster
+    ; CHECK: SU(2): dead %2:gpr64 = ADDSXrr %0:gpr64, undef $x3, implicit-def $nzcv
+    ; CHECK: Predecessors:
+    ; CHECK: SU(0): Ord  Latency=0 Cluster
+    ; CHECK: SU(3): dead %3:gpr64 = SUBSXrr %1:gpr64, undef $x4, implicit-def $nzcv
+    ; CHECK: Predecessors:
+    ; CHECK: SU(1): Ord  Latency=0 Cluster
+...
+---
+name: logic
+body: |
+  bb.0.entry:
+    %0:gpr32 = ADDWrr undef $w0, undef $w1
+    %1:gpr64 = SUBXrr undef $x1, undef $x2
+    %3:gpr32 = ANDWrr %0, undef $w3
+    %4:gpr64 = ORRXrr %1, undef $x4
+
+    ; CHECK: SU(0):      %0:gpr32 = ADDWrr undef $w0, undef $w1
+    ; CHECK: Successors:
+    ; CHECK: SU(2): Ord  Latency=0 Cluster
+    ; CHECK: SU(1):      %1:gpr64 = SUBXrr undef $x1, undef $x2
+    ; CHECK: Successors:
+    ; CHECK: SU(3): Ord  Latency=0 Cluster
+    ; CHECK: SU(2): dead %2:gpr32 = ANDWrr %0:gpr32, undef $w3
+    ; CHECK: Predecessors:
+    ; CHECK: SU(0): Ord  Latency=0 Cluster
+    ; CHECK: SU(3): dead %3:gpr64 = ORRXrr %1:gpr64, undef $x4
+    ; CHECK: Predecessors:
+    ; CHECK: SU(1): Ord  Latency=0 Cluster
+...