[X86][NewPM] Port X86 FP Stackifier Pass to NewPM

[boomanaiden154]

No description provided.

[llvmbot]

@llvm/pr-subscribers-backend-x86
@llvm/pr-subscribers-tools-llvm-exegesis

@llvm/pr-subscribers-debuginfo

Author: Aiden Grossman (boomanaiden154)

Changes

---

Full diff: https://github.com/llvm/llvm-project/pull/167911.diff

7 Files Affected:

• (modified) llvm/lib/Target/X86/X86.h (+8-2)
• (modified) llvm/lib/Target/X86/X86FloatingPoint.cpp (+77-38)
• (modified) llvm/lib/Target/X86/X86PassRegistry.def (+1-1)
• (modified) llvm/lib/Target/X86/X86TargetMachine.cpp (+2-2)
• (modified) llvm/test/CodeGen/X86/x87-stack-pop.mir (+2-1)
• (modified) llvm/test/DebugInfo/MIR/InstrRef/x86-fp-stackifier-drop-locations.mir (+2-1)
• (modified) llvm/tools/llvm-exegesis/lib/X86/Target.cpp (+1-1)

diff --git a/llvm/lib/Target/X86/X86.h b/llvm/lib/Target/X86/X86.h
index 200ca80adb232..03706aaaab237 100644
--- a/llvm/lib/Target/X86/X86.h
+++ b/llvm/lib/Target/X86/X86.h
@@ -44,7 +44,13 @@ FunctionPass *createCleanupLocalDynamicTLSPass();
 /// This function returns a pass which converts floating-point register
 /// references and pseudo instructions into floating-point stack references and
 /// physical instructions.
-FunctionPass *createX86FloatingPointStackifierPass();
+class X86FPStackifierPass : public PassInfoMixin<X86FPStackifierPass> {
+public:
+  PreservedAnalyses run(MachineFunction &MF,
+                        MachineFunctionAnalysisManager &MFAM);
+};
+
+FunctionPass *createX86FPStackifierLegacyPass();
 
 /// This pass inserts AVX vzeroupper instructions before each call to avoid
 /// transition penalty between functions encoded with AVX and SSE.
@@ -229,7 +235,6 @@ FunctionPass *createX86ArgumentStackSlotPass();
 FunctionPass *createX86SuppressAPXForRelocationPass();
 
 void initializeCompressEVEXPassPass(PassRegistry &);
-void initializeFPSPass(PassRegistry &);
 void initializeFixupBWInstPassPass(PassRegistry &);
 void initializeFixupLEAPassPass(PassRegistry &);
 void initializeX86ArgumentStackSlotPassPass(PassRegistry &);
@@ -246,6 +251,7 @@ void initializeX86DomainReassignmentPass(PassRegistry &);
 void initializeX86DynAllocaExpanderLegacyPass(PassRegistry &);
 void initializeX86ExecutionDomainFixPass(PassRegistry &);
 void initializeX86ExpandPseudoPass(PassRegistry &);
+void initializeX86FPStackifierLegacyPass(PassRegistry &);
 void initializeX86FastPreTileConfigPass(PassRegistry &);
 void initializeX86FastTileConfigPass(PassRegistry &);
 void initializeX86FixupSetCCPassPass(PassRegistry &);
diff --git a/llvm/lib/Target/X86/X86FloatingPoint.cpp b/llvm/lib/Target/X86/X86FloatingPoint.cpp
index bc9d354cd8e1a..6af2050d8b9d0 100644
--- a/llvm/lib/Target/X86/X86FloatingPoint.cpp
+++ b/llvm/lib/Target/X86/X86FloatingPoint.cpp
@@ -31,6 +31,8 @@
 #include "llvm/ADT/Statistic.h"
 #include "llvm/CodeGen/EdgeBundles.h"
 #include "llvm/CodeGen/LiveRegUnits.h"
+#include "llvm/CodeGen/MachineFunction.h"
+#include "llvm/CodeGen/MachineFunctionAnalysisManager.h"
 #include "llvm/CodeGen/MachineFunctionPass.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/MachineRegisterInfo.h"
@@ -38,6 +40,7 @@
 #include "llvm/CodeGen/TargetInstrInfo.h"
 #include "llvm/CodeGen/TargetSubtargetInfo.h"
 #include "llvm/Config/llvm-config.h"
+#include "llvm/IR/Analysis.h"
 #include "llvm/IR/InlineAsm.h"
 #include "llvm/InitializePasses.h"
 #include "llvm/Support/Debug.h"
@@ -48,7 +51,7 @@
 #include <bitset>
 using namespace llvm;
 
-#define DEBUG_TYPE "x86-codegen"
+#define DEBUG_TYPE "x86-fp-stackifier"
 
 STATISTIC(NumFXCH, "Number of fxch instructions inserted");
 STATISTIC(NumFP, "Number of floating point instructions");
@@ -56,25 +59,10 @@ STATISTIC(NumFP, "Number of floating point instructions");
 namespace {
 const unsigned ScratchFPReg = 7;
 
-struct FPS : public MachineFunctionPass {
-  static char ID;
-  FPS() : MachineFunctionPass(ID) {}
-
-  void getAnalysisUsage(AnalysisUsage &AU) const override {
-    AU.setPreservesCFG();
-    AU.addRequired<EdgeBundlesWrapperLegacy>();
-    AU.addPreservedID(MachineLoopInfoID);
-    AU.addPreservedID(MachineDominatorsID);
-    MachineFunctionPass::getAnalysisUsage(AU);
-  }
-
-  bool runOnMachineFunction(MachineFunction &MF) override;
-
-  MachineFunctionProperties getRequiredProperties() const override {
-    return MachineFunctionProperties().setNoVRegs();
-  }
-
-  StringRef getPassName() const override { return "X86 FP Stackifier"; }
+class FPS {
+public:
+  bool shouldRun(MachineFunction &MF);
+  bool run(MachineFunction &MF, EdgeBundles *EdgeBundles);
 
 private:
   const TargetInstrInfo *TII = nullptr; // Machine instruction info.
@@ -292,15 +280,43 @@ struct FPS : public MachineFunctionPass {
 
   void setKillFlags(MachineBasicBlock &MBB) const;
 };
+
+class X86FPStackifierLegacy : public MachineFunctionPass {
+public:
+  X86FPStackifierLegacy() : MachineFunctionPass(ID) {}
+
+  static char ID;
+
+private:
+  void getAnalysisUsage(AnalysisUsage &AU) const override {
+    AU.setPreservesCFG();
+    AU.addRequired<EdgeBundlesWrapperLegacy>();
+    AU.addPreservedID(MachineLoopInfoID);
+    AU.addPreservedID(MachineDominatorsID);
+    MachineFunctionPass::getAnalysisUsage(AU);
+  }
+
+  bool runOnMachineFunction(MachineFunction &MF) override;
+
+  MachineFunctionProperties getRequiredProperties() const override {
+    return MachineFunctionProperties().setNoVRegs();
+  }
+
+  StringRef getPassName() const override { return "X86 FP Stackifier"; }
+};
 } // namespace
 
-char FPS::ID = 0;
+char X86FPStackifierLegacy::ID = 0;
 
-INITIALIZE_PASS_BEGIN(FPS, DEBUG_TYPE, "X86 FP Stackifier", false, false)
+INITIALIZE_PASS_BEGIN(X86FPStackifierLegacy, DEBUG_TYPE, "X86 FP Stackifier",
+                      false, false)
 INITIALIZE_PASS_DEPENDENCY(EdgeBundlesWrapperLegacy)
-INITIALIZE_PASS_END(FPS, DEBUG_TYPE, "X86 FP Stackifier", false, false)
+INITIALIZE_PASS_END(X86FPStackifierLegacy, DEBUG_TYPE, "X86 FP Stackifier",
+                    false, false)
 
-FunctionPass *llvm::createX86FloatingPointStackifierPass() { return new FPS(); }
+FunctionPass *llvm::createX86FPStackifierLegacyPass() {
+  return new X86FPStackifierLegacy();
+}
 
 /// getFPReg - Return the X86::FPx register number for the specified operand.
 /// For example, this returns 3 for X86::FP3.
@@ -311,28 +327,25 @@ static unsigned getFPReg(const MachineOperand &MO) {
   return Reg - X86::FP0;
 }
 
-/// runOnMachineFunction - Loop over all of the basic blocks, transforming FP
-/// register references into FP stack references.
-///
-bool FPS::runOnMachineFunction(MachineFunction &MF) {
+bool FPS::shouldRun(MachineFunction &MF) {
   // We only need to run this pass if there are any FP registers used in this
   // function.  If it is all integer, there is nothing for us to do!
-  bool FPIsUsed = false;
-
   static_assert(X86::FP6 == X86::FP0 + 6,
                 "Register enums aren't sorted right!");
   const MachineRegisterInfo &MRI = MF.getRegInfo();
-  for (unsigned i = 0; i <= 6; ++i)
-    if (!MRI.reg_nodbg_empty(X86::FP0 + i)) {
-      FPIsUsed = true;
-      break;
+  for (unsigned I = 0; I <= 6; ++I)
+    if (!MRI.reg_nodbg_empty(X86::FP0 + I)) {
+      return true;
     }
 
-  // Early exit.
-  if (!FPIsUsed)
-    return false;
+  return false;
+}
 
-  Bundles = &getAnalysis<EdgeBundlesWrapperLegacy>().getEdgeBundles();
+/// runOnMachineFunction - Loop over all of the basic blocks, transforming FP
+/// register references into FP stack references.
+///
+bool FPS::run(MachineFunction &MF, EdgeBundles *FunctionBundles) {
+  Bundles = FunctionBundles;
   TII = MF.getSubtarget().getInstrInfo();
 
   // Prepare cross-MBB liveness.
@@ -1812,3 +1825,29 @@ void FPS::setKillFlags(MachineBasicBlock &MBB) const {
     LPR.stepBackward(MI);
   }
 }
+
+bool X86FPStackifierLegacy::runOnMachineFunction(MachineFunction &MF) {
+  FPS Impl;
+  if (!Impl.shouldRun(MF))
+    return false;
+
+  EdgeBundles *Bundles =
+      &getAnalysis<EdgeBundlesWrapperLegacy>().getEdgeBundles();
+  return FPS().run(MF, Bundles);
+}
+
+PreservedAnalyses
+X86FPStackifierPass::run(MachineFunction &MF,
+                         MachineFunctionAnalysisManager &MFAM) {
+  FPS Impl;
+  if (!Impl.shouldRun(MF))
+    return PreservedAnalyses::all();
+
+  EdgeBundles *Bundles = &MFAM.getResult<EdgeBundlesAnalysis>(MF);
+  bool Changed = Impl.run(MF, Bundles);
+  if (!Changed)
+    return PreservedAnalyses::all();
+  PreservedAnalyses PA = PreservedAnalyses::none();
+  PA.preserveSet<CFGAnalyses>();
+  return PA;
+}
diff --git a/llvm/lib/Target/X86/X86PassRegistry.def b/llvm/lib/Target/X86/X86PassRegistry.def
index 0d7095b18daa8..b80ad38c146df 100644
--- a/llvm/lib/Target/X86/X86PassRegistry.def
+++ b/llvm/lib/Target/X86/X86PassRegistry.def
@@ -31,6 +31,7 @@ DUMMY_FUNCTION_PASS("x86-winehstate", WinEHStatePass())
 #endif
 MACHINE_FUNCTION_PASS("x86-avoid-trailing-call", X86AvoidTrailingCallPass())
 MACHINE_FUNCTION_PASS("x86-dyn-alloca-expander", X86DynAllocaExpanderPass())
+MACHINE_FUNCTION_PASS("x86-fp-stackifier", X86FPStackifierPass())
 MACHINE_FUNCTION_PASS("x86-isel", X86ISelDAGToDAGPass(*this))
 #undef MACHINE_FUNCTION_PASS
 
@@ -40,7 +41,6 @@ MACHINE_FUNCTION_PASS("x86-isel", X86ISelDAGToDAGPass(*this))
 DUMMY_MACHINE_FUNCTION_PASS("x86-avoid-SFB", X86AvoidSFBPass())
 DUMMY_MACHINE_FUNCTION_PASS("x86-cf-opt", X86CallFrameOptimization())
 DUMMY_MACHINE_FUNCTION_PASS("x86-cmov-conversion", X86CmovConverterPass())
-DUMMY_MACHINE_FUNCTION_PASS("x86-codege", FPS())
 DUMMY_MACHINE_FUNCTION_PASS("x86-compress-evex", CompressEVEXPass())
 DUMMY_MACHINE_FUNCTION_PASS("x86-domain-reassignment", X86DomainReassignment())
 DUMMY_MACHINE_FUNCTION_PASS("x86-execution-domain-fix", X86ExecutionDomainFix())
diff --git a/llvm/lib/Target/X86/X86TargetMachine.cpp b/llvm/lib/Target/X86/X86TargetMachine.cpp
index c1214149dfa1d..543220b2fd3b9 100644
--- a/llvm/lib/Target/X86/X86TargetMachine.cpp
+++ b/llvm/lib/Target/X86/X86TargetMachine.cpp
@@ -76,7 +76,7 @@ extern "C" LLVM_C_ABI void LLVMInitializeX86Target() {
   initializeFixupBWInstPassPass(PR);
   initializeCompressEVEXPassPass(PR);
   initializeFixupLEAPassPass(PR);
-  initializeFPSPass(PR);
+  initializeX86FPStackifierLegacyPass(PR);
   initializeX86FixupSetCCPassPass(PR);
   initializeX86CallFrameOptimizationPass(PR);
   initializeX86CmovConverterPassPass(PR);
@@ -531,7 +531,7 @@ void X86PassConfig::addMachineSSAOptimization() {
 
 void X86PassConfig::addPostRegAlloc() {
   addPass(createX86LowerTileCopyPass());
-  addPass(createX86FloatingPointStackifierPass());
+  addPass(createX86FPStackifierLegacyPass());
   // When -O0 is enabled, the Load Value Injection Hardening pass will fall back
   // to using the Speculative Execution Side Effect Suppression pass for
   // mitigation. This is to prevent slow downs due to
diff --git a/llvm/test/CodeGen/X86/x87-stack-pop.mir b/llvm/test/CodeGen/X86/x87-stack-pop.mir
index 1c4ffa54b150f..73144fd484521 100644
--- a/llvm/test/CodeGen/X86/x87-stack-pop.mir
+++ b/llvm/test/CodeGen/X86/x87-stack-pop.mir
@@ -1,5 +1,6 @@
 # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
-# RUN: llc -mtriple=i686-- -run-pass x86-codegen -O2 -o - %s | FileCheck %s
+# RUN: llc -mtriple=i686-- -run-pass=x86-fp-stackifier -O2 -o - %s | FileCheck %s
+# RUN: llc -mtriple=i686-- -passes=x86-fp-stackifier -O2 -o - %s | FileCheck %s
 
 ---
 name: func_fxam
diff --git a/llvm/test/DebugInfo/MIR/InstrRef/x86-fp-stackifier-drop-locations.mir b/llvm/test/DebugInfo/MIR/InstrRef/x86-fp-stackifier-drop-locations.mir
index be12082c45b9a..ed7a41cc7fc1d 100644
--- a/llvm/test/DebugInfo/MIR/InstrRef/x86-fp-stackifier-drop-locations.mir
+++ b/llvm/test/DebugInfo/MIR/InstrRef/x86-fp-stackifier-drop-locations.mir
@@ -1,4 +1,5 @@
-# RUN: llc %s -run-pass=x86-codegen -o - -experimental-debug-variable-locations | FileCheck %s --implicit-check-not=debug-instr-number
+# RUN: llc %s -run-pass=x86-fp-stackifier -o - -experimental-debug-variable-locations | FileCheck %s --implicit-check-not=debug-instr-number
+# RUN: llc %s -passes=x86-fp-stackifier -o - -experimental-debug-variable-locations | FileCheck %s --implicit-check-not=debug-instr-number
 #
 # The x87 FP instructions below have debug instr numbers attached -- but the
 # operands get rewritten when it's converted to stack-form. Rather than trying
diff --git a/llvm/tools/llvm-exegesis/lib/X86/Target.cpp b/llvm/tools/llvm-exegesis/lib/X86/Target.cpp
index 77b2f491f6a72..3f2b9112e5c76 100644
--- a/llvm/tools/llvm-exegesis/lib/X86/Target.cpp
+++ b/llvm/tools/llvm-exegesis/lib/X86/Target.cpp
@@ -869,7 +869,7 @@ constexpr MCPhysReg kDefaultLoopCounterReg = X86::R8;
 
 void ExegesisX86Target::addTargetSpecificPasses(PassManagerBase &PM) const {
   // Lowers FP pseudo-instructions, e.g. ABS_Fp32 -> ABS_F.
-  PM.add(createX86FloatingPointStackifierPass());
+  PM.add(createX86FPStackifierLegacyPass());
 }
 
 MCRegister ExegesisX86Target::getScratchMemoryRegister(const Triple &TT) const {

[RKSimon]

LGTM