[LoongArch] Override hooks to enable sink-and-fold support in MachineSink

[zhaoqi5]

No description provided.

[llvmbot]

@llvm/pr-subscribers-backend-loongarch

Author: ZhaoQi (zhaoqi5)

Changes

---

Full diff: https://github.com/llvm/llvm-project/pull/163721.diff

5 Files Affected:

• (modified) llvm/lib/Target/LoongArch/LoongArchInstrInfo.cpp (+77)
• (modified) llvm/lib/Target/LoongArch/LoongArchInstrInfo.h (+6)
• (modified) llvm/lib/Target/LoongArch/LoongArchTargetMachine.cpp (+8-1)
• (modified) llvm/test/CodeGen/LoongArch/ldptr.ll (+4-6)
• (modified) llvm/test/CodeGen/LoongArch/stptr.ll (+3-5)

diff --git a/llvm/lib/Target/LoongArch/LoongArchInstrInfo.cpp b/llvm/lib/Target/LoongArch/LoongArchInstrInfo.cpp
index c89212dae72d9..f2349c730b1ec 100644
--- a/llvm/lib/Target/LoongArch/LoongArchInstrInfo.cpp
+++ b/llvm/lib/Target/LoongArch/LoongArchInstrInfo.cpp
@@ -756,6 +756,83 @@ LoongArchInstrInfo::getSerializableBitmaskMachineOperandTargetFlags() const {
   return ArrayRef(TargetFlags);
 }
 
+bool LoongArchInstrInfo::canFoldIntoAddrMode(const MachineInstr &MemI,
+                                             Register Reg,
+                                             const MachineInstr &AddrI,
+                                             ExtAddrMode &AM) const {
+  unsigned OffsetWidth = 0;
+  switch (MemI.getOpcode()) {
+  default:
+    return false;
+  case LoongArch::LD_B:
+  case LoongArch::LD_H:
+  case LoongArch::LD_W:
+  case LoongArch::LD_D:
+  case LoongArch::LD_BU:
+  case LoongArch::LD_HU:
+  case LoongArch::LD_WU:
+  case LoongArch::ST_B:
+  case LoongArch::ST_H:
+  case LoongArch::ST_W:
+  case LoongArch::ST_D:
+  case LoongArch::FLD_S:
+  case LoongArch::FLD_D:
+  case LoongArch::FST_S:
+  case LoongArch::FST_D:
+    OffsetWidth = 12;
+    break;
+  case LoongArch::LDPTR_W:
+  case LoongArch::LDPTR_D:
+  case LoongArch::STPTR_W:
+  case LoongArch::STPTR_D:
+    OffsetWidth = 14;
+    break;
+  }
+
+  if (MemI.getOperand(0).getReg() == Reg)
+    return false;
+
+  if ((AddrI.getOpcode() != LoongArch::ADDI_W &&
+       AddrI.getOpcode() != LoongArch::ADDI_D) ||
+      !AddrI.getOperand(1).isReg() || !AddrI.getOperand(2).isImm())
+    return false;
+
+  int64_t OldOffset = MemI.getOperand(2).getImm();
+  int64_t Disp = AddrI.getOperand(2).getImm();
+  int64_t NewOffset = OldOffset + Disp;
+  if (!STI.is64Bit())
+    NewOffset = SignExtend64<32>(NewOffset);
+
+  if (!(OffsetWidth == 12 && isInt<12>(NewOffset)) &&
+      !(OffsetWidth == 14 && isShiftedInt<14, 2>(NewOffset) && STI.hasUAL()))
+    return false;
+
+  AM.BaseReg = AddrI.getOperand(1).getReg();
+  AM.ScaledReg = 0;
+  AM.Scale = 0;
+  AM.Displacement = NewOffset;
+  AM.Form = ExtAddrMode::Formula::Basic;
+  return true;
+}
+
+MachineInstr *
+LoongArchInstrInfo::emitLdStWithAddr(MachineInstr &MemI,
+                                     const ExtAddrMode &AM) const {
+  const DebugLoc &DL = MemI.getDebugLoc();
+  MachineBasicBlock &MBB = *MemI.getParent();
+
+  assert(AM.ScaledReg == 0 && AM.Scale == 0 &&
+         "Addressing mode not supported for folding");
+
+  return BuildMI(MBB, MemI, DL, get(MemI.getOpcode()))
+      .addReg(MemI.getOperand(0).getReg(),
+              MemI.mayLoad() ? RegState::Define : 0)
+      .addReg(AM.BaseReg)
+      .addImm(AM.Displacement)
+      .setMemRefs(MemI.memoperands())
+      .setMIFlags(MemI.getFlags());
+}
+
 // Returns true if this is the sext.w pattern, addi.w rd, rs, 0.
 bool LoongArch::isSEXT_W(const MachineInstr &MI) {
   return MI.getOpcode() == LoongArch::ADDI_W && MI.getOperand(1).isReg() &&
diff --git a/llvm/lib/Target/LoongArch/LoongArchInstrInfo.h b/llvm/lib/Target/LoongArch/LoongArchInstrInfo.h
index f25958a32bec4..f69a558bdeca9 100644
--- a/llvm/lib/Target/LoongArch/LoongArchInstrInfo.h
+++ b/llvm/lib/Target/LoongArch/LoongArchInstrInfo.h
@@ -93,6 +93,12 @@ class LoongArchInstrInfo : public LoongArchGenInstrInfo {
   ArrayRef<std::pair<unsigned, const char *>>
   getSerializableBitmaskMachineOperandTargetFlags() const override;
 
+  bool canFoldIntoAddrMode(const MachineInstr &MemI, Register Reg,
+                           const MachineInstr &AddrI,
+                           ExtAddrMode &AM) const override;
+  MachineInstr *emitLdStWithAddr(MachineInstr &MemI,
+                                 const ExtAddrMode &AM) const override;
+
 protected:
   const LoongArchSubtarget &STI;
 };
diff --git a/llvm/lib/Target/LoongArch/LoongArchTargetMachine.cpp b/llvm/lib/Target/LoongArch/LoongArchTargetMachine.cpp
index c5e26c106b5df..a7a9de7170a09 100644
--- a/llvm/lib/Target/LoongArch/LoongArchTargetMachine.cpp
+++ b/llvm/lib/Target/LoongArch/LoongArchTargetMachine.cpp
@@ -62,6 +62,11 @@ static cl::opt<bool>
                           cl::desc("Enable the merge base offset pass"),
                           cl::init(true), cl::Hidden);
 
+static cl::opt<bool>
+    EnableSinkFold("loongarch-enable-sink-fold",
+                   cl::desc("Enable sinking and folding of instruction copies"),
+                   cl::init(true), cl::Hidden);
+
 static Reloc::Model getEffectiveRelocModel(const Triple &TT,
                                            std::optional<Reloc::Model> RM) {
   return RM.value_or(Reloc::Static);
@@ -147,7 +152,9 @@ namespace {
 class LoongArchPassConfig : public TargetPassConfig {
 public:
   LoongArchPassConfig(LoongArchTargetMachine &TM, PassManagerBase &PM)
-      : TargetPassConfig(TM, PM) {}
+      : TargetPassConfig(TM, PM) {
+    setEnableSinkAndFold(EnableSinkFold);
+  }
 
   LoongArchTargetMachine &getLoongArchTargetMachine() const {
     return getTM<LoongArchTargetMachine>();
diff --git a/llvm/test/CodeGen/LoongArch/ldptr.ll b/llvm/test/CodeGen/LoongArch/ldptr.ll
index c3656a6bdafba..9bafa10c47e3f 100644
--- a/llvm/test/CodeGen/LoongArch/ldptr.ll
+++ b/llvm/test/CodeGen/LoongArch/ldptr.ll
@@ -24,8 +24,7 @@ define signext i32 @ldptr_w(ptr %p) nounwind {
 ; LA32-LABEL: ldptr_w:
 ; LA32:       # %bb.0: # %entry
 ; LA32-NEXT:    addi.w $a0, $a0, 2047
-; LA32-NEXT:    addi.w $a0, $a0, 1
-; LA32-NEXT:    ld.w $a0, $a0, 0
+; LA32-NEXT:    ld.w $a0, $a0, 1
 ; LA32-NEXT:    ret
 ;
 ; LA64-LABEL: ldptr_w:
@@ -81,10 +80,9 @@ entry:
 define i64 @ldptr_d(ptr %p) nounwind {
 ; LA32-LABEL: ldptr_d:
 ; LA32:       # %bb.0: # %entry
-; LA32-NEXT:    addi.w $a0, $a0, 2047
-; LA32-NEXT:    addi.w $a1, $a0, 1
-; LA32-NEXT:    ld.w $a0, $a1, 0
-; LA32-NEXT:    ld.w $a1, $a1, 4
+; LA32-NEXT:    addi.w $a1, $a0, 2047
+; LA32-NEXT:    ld.w $a0, $a1, 1
+; LA32-NEXT:    ld.w $a1, $a1, 5
 ; LA32-NEXT:    ret
 ;
 ; LA64-LABEL: ldptr_d:
diff --git a/llvm/test/CodeGen/LoongArch/stptr.ll b/llvm/test/CodeGen/LoongArch/stptr.ll
index d70f9f4ba1603..23b433aa15856 100644
--- a/llvm/test/CodeGen/LoongArch/stptr.ll
+++ b/llvm/test/CodeGen/LoongArch/stptr.ll
@@ -23,8 +23,7 @@ define void @stptr_w(ptr %p, i32 signext %val) nounwind {
 ; LA32-LABEL: stptr_w:
 ; LA32:       # %bb.0:
 ; LA32-NEXT:    addi.w $a0, $a0, 2047
-; LA32-NEXT:    addi.w $a0, $a0, 1
-; LA32-NEXT:    st.w $a1, $a0, 0
+; LA32-NEXT:    st.w $a1, $a0, 1
 ; LA32-NEXT:    ret
 ;
 ; LA64-LABEL: stptr_w:
@@ -77,9 +76,8 @@ define void @stptr_d(ptr %p, i64 %val) nounwind {
 ; LA32-LABEL: stptr_d:
 ; LA32:       # %bb.0:
 ; LA32-NEXT:    addi.w $a0, $a0, 2047
-; LA32-NEXT:    addi.w $a0, $a0, 1
-; LA32-NEXT:    st.w $a2, $a0, 4
-; LA32-NEXT:    st.w $a1, $a0, 0
+; LA32-NEXT:    st.w $a2, $a0, 5
+; LA32-NEXT:    st.w $a1, $a0, 1
 ; LA32-NEXT:    ret
 ;
 ; LA64-LABEL: stptr_d: