[RISCV] Add late optimization pass for RISC-V to optimize branch instructions

This is an alternative to llvm#117060, and is stacked on llvm#131684. Marking
@mikhailramalho as Co-Author here because I got the idea of a late peephole pass
and the test case from llvm#117060.

I use a late pass because we introduce the optimizable branches so late in the
pipeline.

Co-authored-by: Mikhail R. Gadelha <[email protected]>

Author: michaelmaitland

llvm/lib/Target/RISCV/CMakeLists.txt

@@ -35,6 +35,7 @@ add_llvm_target(RISCVCodeGen
   RISCVConstantPoolValue.cpp
   RISCVDeadRegisterDefinitions.cpp
   RISCVMakeCompressible.cpp
+  RISCVLatePeephole.cpp
   RISCVExpandAtomicPseudoInsts.cpp
   RISCVExpandPseudoInsts.cpp
   RISCVFoldMemOffset.cpp

llvm/lib/Target/RISCV/RISCV.h

@@ -43,6 +43,9 @@ FunctionPass *createRISCVISelDag(RISCVTargetMachine &TM,
 FunctionPass *createRISCVMakeCompressibleOptPass();
 void initializeRISCVMakeCompressibleOptPass(PassRegistry &);
 
+FunctionPass *createRISCVLatePeepholeOptPass();
+void initializeRISCVLatePeepholeOptPass(PassRegistry &);
+
 FunctionPass *createRISCVGatherScatterLoweringPass();
 void initializeRISCVGatherScatterLoweringPass(PassRegistry &);
 

llvm/lib/Target/RISCV/RISCVInstrInfo.h

@@ -306,19 +306,6 @@ class RISCVInstrInfo : public RISCVGenInstrInfo {
 
   static bool isLdStSafeToPair(const MachineInstr &LdSt,
                                const TargetRegisterInfo *TRI);
-
-protected:
-  const RISCVSubtarget &STI;
-
-private:
-  unsigned getInstBundleLength(const MachineInstr &MI) const;
-
-  bool isVectorAssociativeAndCommutative(const MachineInstr &MI,
-                                         bool Invert = false) const;
-  bool areRVVInstsReassociable(const MachineInstr &MI1,
-                               const MachineInstr &MI2) const;
-  bool hasReassociableVectorSibling(const MachineInstr &Inst,
-                                    bool &Commuted) const;
   /// Return true if the branch represented by the conditional branch with
   /// components TBB, FBB, and CurCond was folded into an unconditional branch.
   ///
@@ -339,6 +326,19 @@ class RISCVInstrInfo : public RISCVGenInstrInfo {
   bool trySimplifyCondBr(MachineBasicBlock &MBB, MachineBasicBlock *TBB,
                          MachineBasicBlock *FBB,
                          SmallVectorImpl<MachineOperand> &Cond) const;
+
+protected:
+  const RISCVSubtarget &STI;
+
+private:
+  unsigned getInstBundleLength(const MachineInstr &MI) const;
+
+  bool isVectorAssociativeAndCommutative(const MachineInstr &MI,
+                                         bool Invert = false) const;
+  bool areRVVInstsReassociable(const MachineInstr &MI1,
+                               const MachineInstr &MI2) const;
+  bool hasReassociableVectorSibling(const MachineInstr &Inst,
+                                    bool &Commuted) const;
 };
 
 namespace RISCV {

llvm/lib/Target/RISCV/RISCVLatePeephole.cpp

@@ -0,0 +1,85 @@
+//===-- RISCVLatePeephole.cpp - Late stage peephole optimization ----------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+///
+/// This file provides RISC-V late peephole optimizations
+///
+//===----------------------------------------------------------------------===//
+
+#include "MCTargetDesc/RISCVMCTargetDesc.h"
+#include "RISCV.h"
+#include "RISCVInstrInfo.h"
+#include "RISCVSubtarget.h"
+#include "llvm/CodeGen/MachineBasicBlock.h"
+#include "llvm/CodeGen/MachineBranchProbabilityInfo.h"
+#include "llvm/CodeGen/MachineDominators.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/Passes.h"
+#include "llvm/CodeGen/RegisterScavenging.h"
+#include "llvm/MC/TargetRegistry.h"
+#include "llvm/Support/Debug.h"
+
+using namespace llvm;
+
+#define DEBUG_TYPE "riscv-late-peephole"
+#define RISCV_LATE_PEEPHOLE_NAME "RISC-V Late Stage Peephole"
+
+namespace {
+
+struct RISCVLatePeepholeOpt : public MachineFunctionPass {
+  static char ID;
+
+  RISCVLatePeepholeOpt() : MachineFunctionPass(ID) {}
+
+  StringRef getPassName() const override { return RISCV_LATE_PEEPHOLE_NAME; }
+
+  void getAnalysisUsage(AnalysisUsage &AU) const override {
+    MachineFunctionPass::getAnalysisUsage(AU);
+  }
+
+  bool runOnMachineFunction(MachineFunction &Fn) override;
+
+private:
+  bool optimizeBlock(MachineBasicBlock &MBB);
+
+  const RISCVInstrInfo *TII = nullptr;
+};
+} // namespace
+
+char RISCVLatePeepholeOpt::ID = 0;
+INITIALIZE_PASS(RISCVLatePeepholeOpt, "riscv-late-peephole",
+                RISCV_LATE_PEEPHOLE_NAME, false, false)
+
+bool RISCVLatePeepholeOpt::optimizeBlock(MachineBasicBlock &MBB) {
+
+  // Use trySimplifyCondBr directly to know whether the optimization occured.
+  MachineBasicBlock *TBB, *FBB;
+  SmallVector<MachineOperand, 4> Cond;
+  if (!TII->analyzeBranch(MBB, TBB, FBB, Cond, false))
+    return TII->trySimplifyCondBr(MBB, TBB, FBB, Cond);
+
+  return false;
+}
+
+bool RISCVLatePeepholeOpt::runOnMachineFunction(MachineFunction &MF) {
+  if (skipFunction(MF.getFunction()))
+    return false;
+
+  TII = MF.getSubtarget<RISCVSubtarget>().getInstrInfo();
+
+  bool MadeChange = false;
+
+  for (MachineBasicBlock &MBB : MF)
+    MadeChange |= optimizeBlock(MBB);
+
+  return MadeChange;
+}
+
+/// Returns an instance of the Make Compressible Optimization pass.
+FunctionPass *llvm::createRISCVLatePeepholeOptPass() {
+  return new RISCVLatePeepholeOpt();
+}

llvm/lib/Target/RISCV/RISCVTargetMachine.cpp

@@ -128,6 +128,7 @@ extern "C" LLVM_EXTERNAL_VISIBILITY void LLVMInitializeRISCVTarget() {
   initializeKCFIPass(*PR);
   initializeRISCVDeadRegisterDefinitionsPass(*PR);
   initializeRISCVMakeCompressibleOptPass(*PR);
+  initializeRISCVLatePeepholeOptPass(*PR);
   initializeRISCVGatherScatterLoweringPass(*PR);
   initializeRISCVCodeGenPreparePass(*PR);
   initializeRISCVPostRAExpandPseudoPass(*PR);
@@ -567,6 +568,7 @@ void RISCVPassConfig::addPreEmitPass() {
     addPass(createMachineCopyPropagationPass(true));
   addPass(&BranchRelaxationPassID);
   addPass(createRISCVMakeCompressibleOptPass());
+  addPass(createRISCVLatePeepholeOptPass());
 }
 
 void RISCVPassConfig::addPreEmitPass2() {

llvm/test/CodeGen/RISCV/O0-pipeline.ll

@@ -64,6 +64,7 @@
 ; CHECK-NEXT:       Implement the 'patchable-function' attribute
 ; CHECK-NEXT:       Branch relaxation pass
 ; CHECK-NEXT:       RISC-V Make Compressible
+; CHECK-NEXT:       RISC-V Late Stage Peephole
 ; CHECK-NEXT:       Contiguously Lay Out Funclets
 ; CHECK-NEXT:       Remove Loads Into Fake Uses
 ; CHECK-NEXT:       StackMap Liveness Analysis

llvm/test/CodeGen/RISCV/O3-pipeline.ll

@@ -196,6 +196,7 @@
 ; CHECK-NEXT:       Machine Copy Propagation Pass
 ; CHECK-NEXT:       Branch relaxation pass
 ; CHECK-NEXT:       RISC-V Make Compressible
+; CHECK-NEXT:       RISC-V Late Stage Peephole
 ; CHECK-NEXT:       Contiguously Lay Out Funclets
 ; CHECK-NEXT:       Remove Loads Into Fake Uses
 ; CHECK-NEXT:       StackMap Liveness Analysis

llvm/test/CodeGen/RISCV/simplify-condbr.ll

@@ -128,7 +128,7 @@ define ptr @Perl_pp_refassign(ptr %PL_stack_sp, i1 %tobool.not, i1 %tobool3.not,
 ; CHECK-NEXT:    andi a2, a2, 1
 ; CHECK-NEXT:    beqz a2, .LBB1_2
 ; CHECK-NEXT:  .LBB1_4:
-; CHECK-NEXT:    beqz zero, .LBB1_6
+; CHECK-NEXT:    j .LBB1_6
 ; CHECK-NEXT:  .LBB1_5: # %sw.bb85
 ; CHECK-NEXT:    addi sp, sp, -16
 ; CHECK-NEXT:    sd ra, 8(sp) # 8-byte Folded Spill