Initial implementation of PPCTargetTransformInfo

This provides a place to add customized operation cost information and
control some other target-specific IR-level transformations.

The only non-trivial logic in this checkin assigns a higher cost to
unaligned loads and stores (covered by the included test case).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@173520 91177308-0d34-0410-b5e6-96231b3b80d8

7 files changed, 277 insertions, 0 deletions

diff --git a/lib/Target/PowerPC/CMakeLists.txt b/lib/Target/PowerPC/CMakeLists.txt
index 192d18d664..6036428fad 100644
--- a/lib/Target/PowerPC/CMakeLists.txt
+++ b/lib/Target/PowerPC/CMakeLists.txt
@@ -26,6 +26,7 @@ add_llvm_target(PowerPCCodeGen
   PPCRegisterInfo.cpp
   PPCSubtarget.cpp
   PPCTargetMachine.cpp
+  PPCTargetTransformInfo.cpp
   PPCSelectionDAGInfo.cpp
   )
 
diff --git a/lib/Target/PowerPC/PPC.h b/lib/Target/PowerPC/PPC.h
index e6d38ebf21..72e0ce3568 100644
--- a/lib/Target/PowerPC/PPC.h
+++ b/lib/Target/PowerPC/PPC.h
@@ -25,6 +25,7 @@
 namespace llvm {
   class PPCTargetMachine;
   class FunctionPass;
+  class ImmutablePass;
   class JITCodeEmitter;
   class MachineInstr;
   class AsmPrinter;
@@ -37,6 +38,9 @@ namespace llvm {
                                             JITCodeEmitter &MCE);
   void LowerPPCMachineInstrToMCInst(const MachineInstr *MI, MCInst &OutMI,
                                     AsmPrinter &AP, bool isDarwin);
+
+  /// \brief Creates an PPC-specific Target Transformation Info pass.
+  ImmutablePass *createPPCTargetTransformInfoPass(const PPCTargetMachine *TM);
   
   namespace PPCII {
     
diff --git a/lib/Target/PowerPC/PPCTargetMachine.cpp b/lib/Target/PowerPC/PPCTargetMachine.cpp
index b8b7882ac0..fe851c1b6f 100644
--- a/lib/Target/PowerPC/PPCTargetMachine.cpp
+++ b/lib/Target/PowerPC/PPCTargetMachine.cpp
@@ -126,3 +126,12 @@ bool PPCTargetMachine::addCodeEmitter(PassManagerBase &PM,
 
   return false;
 }
+
+void PPCTargetMachine::addAnalysisPasses(PassManagerBase &PM) {
+  // Add first the target-independent BasicTTI pass, then our PPC pass. This
+  // allows the PPC pass to delegate to the target independent layer when
+  // appropriate.
+  PM.add(createBasicTargetTransformInfoPass(getTargetLowering()));
+  PM.add(createPPCTargetTransformInfoPass(this));
+}
+
diff --git a/lib/Target/PowerPC/PPCTargetMachine.h b/lib/Target/PowerPC/PPCTargetMachine.h
index d917d99ded..606ccb3141 100644
--- a/lib/Target/PowerPC/PPCTargetMachine.h
+++ b/lib/Target/PowerPC/PPCTargetMachine.h
@@ -68,6 +68,9 @@ public:
   virtual TargetPassConfig *createPassConfig(PassManagerBase &PM);
   virtual bool addCodeEmitter(PassManagerBase &PM,
                               JITCodeEmitter &JCE);
+
+  /// \brief Register PPC analysis passes with a pass manager.
+  virtual void addAnalysisPasses(PassManagerBase &PM);
 };
 
 /// PPC32TargetMachine - PowerPC 32-bit target machine.
diff --git a/lib/Target/PowerPC/PPCTargetTransformInfo.cpp b/lib/Target/PowerPC/PPCTargetTransformInfo.cpp
new file mode 100644
index 0000000000..1afef33df0
--- /dev/null
+++ b/lib/Target/PowerPC/PPCTargetTransformInfo.cpp
@@ -0,0 +1,220 @@
+//===-- PPCTargetTransformInfo.cpp - PPC specific TTI pass ----------------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+/// \file
+/// This file implements a TargetTransformInfo analysis pass specific to the
+/// PPC target machine. It uses the target's detailed information to provide
+/// more precise answers to certain TTI queries, while letting the target
+/// independent and default TTI implementations handle the rest.
+///
+//===----------------------------------------------------------------------===//
+
+#define DEBUG_TYPE "ppctti"
+#include "PPC.h"
+#include "PPCTargetMachine.h"
+#include "llvm/Analysis/TargetTransformInfo.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Target/TargetLowering.h"
+#include "llvm/Target/CostTable.h"
+using namespace llvm;
+
+// Declare the pass initialization routine locally as target-specific passes
+// don't havve a target-wide initialization entry point, and so we rely on the
+// pass constructor initialization.
+namespace llvm {
+void initializePPCTTIPass(PassRegistry &);
+}
+
+namespace {
+
+class PPCTTI : public ImmutablePass, public TargetTransformInfo {
+  const PPCTargetMachine *TM;
+  const PPCSubtarget *ST;
+  const PPCTargetLowering *TLI;
+
+  /// Estimate the overhead of scalarizing an instruction. Insert and Extract
+  /// are set if the result needs to be inserted and/or extracted from vectors.
+  unsigned getScalarizationOverhead(Type *Ty, bool Insert, bool Extract) const;
+
+public:
+  PPCTTI() : ImmutablePass(ID), TM(0), ST(0), TLI(0) {
+    llvm_unreachable("This pass cannot be directly constructed");
+  }
+
+  PPCTTI(const PPCTargetMachine *TM)
+      : ImmutablePass(ID), TM(TM), ST(TM->getSubtargetImpl()),
+        TLI(TM->getTargetLowering()) {
+    initializePPCTTIPass(*PassRegistry::getPassRegistry());
+  }
+
+  virtual void initializePass() {
+    pushTTIStack(this);
+  }
+
+  virtual void finalizePass() {
+    popTTIStack();
+  }
+
+  virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+    TargetTransformInfo::getAnalysisUsage(AU);
+  }
+
+  /// Pass identification.
+  static char ID;
+
+  /// Provide necessary pointer adjustments for the two base classes.
+  virtual void *getAdjustedAnalysisPointer(const void *ID) {
+    if (ID == &TargetTransformInfo::ID)
+      return (TargetTransformInfo*)this;
+    return this;
+  }
+
+  /// \name Scalar TTI Implementations
+  /// @{
+  virtual PopcntSupportKind getPopcntSupport(unsigned TyWidth) const;
+
+  /// @}
+
+  /// \name Vector TTI Implementations
+  /// @{
+
+  virtual unsigned getNumberOfRegisters(bool Vector) const;
+  virtual unsigned getRegisterBitWidth(bool Vector) const;
+  virtual unsigned getMaximumUnrollFactor() const;
+  virtual unsigned getArithmeticInstrCost(unsigned Opcode, Type *Ty) const;
+  virtual unsigned getShuffleCost(ShuffleKind Kind, Type *Tp,
+                                  int Index, Type *SubTp) const;
+  virtual unsigned getCastInstrCost(unsigned Opcode, Type *Dst,
+                                    Type *Src) const;
+  virtual unsigned getCmpSelInstrCost(unsigned Opcode, Type *ValTy,
+                                      Type *CondTy) const;
+  virtual unsigned getVectorInstrCost(unsigned Opcode, Type *Val,
+                                      unsigned Index) const;
+  virtual unsigned getMemoryOpCost(unsigned Opcode, Type *Src,
+                                   unsigned Alignment,
+                                   unsigned AddressSpace) const;
+
+  /// @}
+};
+
+} // end anonymous namespace
+
+INITIALIZE_AG_PASS(PPCTTI, TargetTransformInfo, "ppctti",
+                   "PPC Target Transform Info", true, true, false)
+char PPCTTI::ID = 0;
+
+ImmutablePass *
+llvm::createPPCTargetTransformInfoPass(const PPCTargetMachine *TM) {
+  return new PPCTTI(TM);
+}
+
+
+//===----------------------------------------------------------------------===//
+//
+// PPC cost model.
+//
+//===----------------------------------------------------------------------===//
+
+PPCTTI::PopcntSupportKind PPCTTI::getPopcntSupport(unsigned TyWidth) const {
+  assert(isPowerOf2_32(TyWidth) && "Ty width must be power of 2");
+  // FIXME: PPC currently does not have custom popcnt lowering even though
+  // there is hardware support. Once this is fixed, update this function
+  // to reflect the real capabilities of the hardware.
+  return PSK_Software;
+}
+
+unsigned PPCTTI::getNumberOfRegisters(bool Vector) const {
+  if (Vector && !ST->hasAltivec())
+    return 0;
+  return 32;
+}
+
+unsigned PPCTTI::getRegisterBitWidth(bool Vector) const {
+  if (Vector) {
+    if (ST->hasAltivec()) return 128;
+    return 0;
+  }
+
+  if (ST->isPPC64())
+    return 64;
+  return 32;
+
+}
+
+unsigned PPCTTI::getMaximumUnrollFactor() const {
+  unsigned Directive = ST->getDarwinDirective();
+  // The 440 has no SIMD support, but floating-point instructions
+  // have a 5-cycle latency, so unroll by 5x for latency hiding.
+  if (Directive == PPC::DIR_440)
+    return 5;
+
+  // The A2 has no SIMD support, but floating-point instructions
+  // have a 6-cycle latency, so unroll by 6x for latency hiding.
+  if (Directive == PPC::DIR_A2)
+    return 6;
+
+  // FIXME: For lack of any better information, do no harm...
+  if (Directive == PPC::DIR_E500mc || Directive == PPC::DIR_E5500)
+    return 1;
+
+  // For most things, modern systems have two execution units (and
+  // out-of-order execution).
+  return 2;
+}
+
+unsigned PPCTTI::getArithmeticInstrCost(unsigned Opcode, Type *Ty) const {
+  int ISD = TLI->InstructionOpcodeToISD(Opcode);
+  assert(ISD && "Invalid opcode");
+
+  // Fallback to the default implementation.
+  return TargetTransformInfo::getArithmeticInstrCost(Opcode, Ty);
+}
+
+unsigned PPCTTI::getShuffleCost(ShuffleKind Kind, Type *Tp, int Index,
+                                Type *SubTp) const {
+  return TargetTransformInfo::getShuffleCost(Kind, Tp, Index, SubTp);
+}
+
+unsigned PPCTTI::getCastInstrCost(unsigned Opcode, Type *Dst, Type *Src) const {
+  int ISD = TLI->InstructionOpcodeToISD(Opcode);
+  assert(ISD && "Invalid opcode");
+
+  return TargetTransformInfo::getCastInstrCost(Opcode, Dst, Src);
+}
+
+unsigned PPCTTI::getCmpSelInstrCost(unsigned Opcode, Type *ValTy,
+                                    Type *CondTy) const {
+  return TargetTransformInfo::getCmpSelInstrCost(Opcode, ValTy, CondTy);
+}
+
+unsigned PPCTTI::getVectorInstrCost(unsigned Opcode, Type *Val,
+                                    unsigned Index) const {
+  assert(Val->isVectorTy() && "This must be a vector type");
+
+  return TargetTransformInfo::getVectorInstrCost(Opcode, Val, Index);
+}
+
+unsigned PPCTTI::getMemoryOpCost(unsigned Opcode, Type *Src, unsigned Alignment,
+                                 unsigned AddressSpace) const {
+  // Legalize the type.
+  std::pair<unsigned, MVT> LT = TLI->getTypeLegalizationCost(Src);
+  assert((Opcode == Instruction::Load || Opcode == Instruction::Store) &&
+         "Invalid Opcode");
+
+  // Each load/store unit costs 1.
+  unsigned Cost = LT.first * 1;
+
+  // PPC in general does not support unaligned loads and stores. They'll need
+  // to be decomposed based on the alignment factor.
+  unsigned SrcBytes = LT.second.getStoreSize();
+  if (SrcBytes && Alignment && Alignment < SrcBytes)
+    Cost *= (SrcBytes/Alignment);
+
+  return Cost;
+}
+
diff --git a/test/Analysis/CostModel/PowerPC/lit.local.cfg b/test/Analysis/CostModel/PowerPC/lit.local.cfg
new file mode 100644
index 0000000000..4019eca0bb
--- /dev/null
+++ b/test/Analysis/CostModel/PowerPC/lit.local.cfg
@@ -0,0 +1,6 @@
+config.suffixes = ['.ll', '.c', '.cpp']
+
+targets = set(config.root.targets_to_build.split())
+if not 'PowerPC' in targets:
+    config.unsupported = True
+
diff --git a/test/Analysis/CostModel/PowerPC/load_store.ll b/test/Analysis/CostModel/PowerPC/load_store.ll
new file mode 100644
index 0000000000..c77cce955a
--- /dev/null
+++ b/test/Analysis/CostModel/PowerPC/load_store.ll
@@ -0,0 +1,34 @@
+; RUN: opt < %s  -cost-model -analyze -mtriple=powerpc64-unknown-linux-gnu -mcpu=pwr7 | FileCheck %s
+target datalayout = "E-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-f128:128:128-v128:128:128-n32:64"
+target triple = "powerpc64-unknown-linux-gnu"
+
+define i32 @stores(i32 %arg) {
+
+  ; CHECK: cost of 1 {{.*}} store
+  store i8 undef, i8* undef, align 4
+  ; CHECK: cost of 1 {{.*}} store
+  store i16 undef, i16* undef, align 4
+  ; CHECK: cost of 1 {{.*}} store
+  store i32 undef, i32* undef, align 4
+  ; CHECK: cost of 2 {{.*}} store
+  store i64 undef, i64* undef, align 4
+  ; CHECK: cost of 4 {{.*}} store
+  store i128 undef, i128* undef, align 4
+
+  ret i32 undef
+}
+define i32 @loads(i32 %arg) {
+  ; CHECK: cost of 1 {{.*}} load
+  load i8* undef, align 4
+  ; CHECK: cost of 1 {{.*}} load
+  load i16* undef, align 4
+  ; CHECK: cost of 1 {{.*}} load
+  load i32* undef, align 4
+  ; CHECK: cost of 2 {{.*}} load
+  load i64* undef, align 4
+  ; CHECK: cost of 4 {{.*}} load
+  load i128* undef, align 4
+
+  ret i32 undef
+}
+