[RISCV] Increase default tail duplication threshold to 6 at -O3

This is just like AArch64.

Changing the threshold to 6 will increase the code size, but will
also decrease direct branches. CPUs with wide fetch/issue units
can benefit from it.

The value 6 may be debatable, we can set it to `SchedModel.IssueWidth`.

Author: wangpc-pp

llvm/lib/Target/RISCV/RISCVInstrInfo.cpp

@@ -3763,6 +3763,11 @@ RISCVInstrInfo::getSerializableMachineMemOperandTargetFlags() const {
   return ArrayRef(TargetFlags);
 }
 
+unsigned int
+RISCVInstrInfo::getTailDuplicateSize(CodeGenOptLevel OptLevel) const {
+  return OptLevel >= CodeGenOptLevel::Aggressive ? 6 : 2;
+}
+
 // Returns true if this is the sext.w pattern, addiw rd, rs1, 0.
 bool RISCV::isSEXT_W(const MachineInstr &MI) {
   return MI.getOpcode() == RISCV::ADDIW && MI.getOperand(1).isReg() &&

llvm/lib/Target/RISCV/RISCVInstrInfo.h

@@ -288,6 +288,8 @@ class RISCVInstrInfo : public RISCVGenInstrInfo {
   ArrayRef<std::pair<MachineMemOperand::Flags, const char *>>
   getSerializableMachineMemOperandTargetFlags() const override;
 
+  unsigned int getTailDuplicateSize(CodeGenOptLevel OptLevel) const override;
+
   unsigned getUndefInitOpcode(unsigned RegClassID) const override {
     switch (RegClassID) {
     case RISCV::VRRegClassID:

llvm/test/CodeGen/RISCV/riscv-tail-dup-size.ll

@@ -4,7 +4,7 @@
 
 ; RUN: llc -mtriple=riscv64 -mattr=+m -tail-dup-size=4 < %s | FileCheck %s --check-prefix=CHECK-O2
 ; RUN: llc -mtriple=riscv64 -mattr=+m -tail-dup-placement-threshold=4 < %s | FileCheck %s --check-prefix=CHECK-O2
-; RUN: llc -mtriple=riscv64 -mattr=+m -tail-dup-placement-threshold=4 < %s | FileCheck %s --check-prefix=CHECK-O3
+; RUN: llc -mtriple=riscv64 -mattr=+m -tail-dup-placement-threshold=6 < %s | FileCheck %s --check-prefix=CHECK-O3
 
 @a = external dso_local local_unnamed_addr global i32
 @b = external dso_local local_unnamed_addr global i32
@@ -41,12 +41,15 @@ define dso_local i32 @test(i32 %n) {
 ; CHECK-O3-NEXT:    lui a1, %hi(a)
 ; CHECK-O3-NEXT:    lw a1, %lo(a)(a1)
 ; CHECK-O3-NEXT:    mul a0, a1, a0
-; CHECK-O3-NEXT:    j .LBB0_3
+; CHECK-O3-NEXT:    lui a1, %hi(c)
+; CHECK-O3-NEXT:    lw a1, %lo(c)(a1)
+; CHECK-O3-NEXT:    addi a0, a0, -1
+; CHECK-O3-NEXT:    mulw a0, a0, a1
+; CHECK-O3-NEXT:    tail foo
 ; CHECK-O3-NEXT:  .LBB0_2: # %if.else
 ; CHECK-O3-NEXT:    lui a1, %hi(b)
 ; CHECK-O3-NEXT:    lw a1, %lo(b)(a1)
 ; CHECK-O3-NEXT:    divw a0, a1, a0
-; CHECK-O3-NEXT:  .LBB0_3: # %if.end
 ; CHECK-O3-NEXT:    lui a1, %hi(c)
 ; CHECK-O3-NEXT:    lw a1, %lo(c)(a1)
 ; CHECK-O3-NEXT:    addi a0, a0, -1