From: Preston Gurd <preston.gurd@intel.com>
Date: Mon, 23 Apr 2012 21:39:35 +0000 (+0000)
Subject: This patch fixes a problem which arose when using the Post-RA scheduler
X-Git-Url: http://demsky.eecs.uci.edu/git/?a=commitdiff_plain;h=6a8c7bf8e72338e55f0f9583e1828f62da165d4a;p=oota-llvm.git

This patch fixes a problem which arose when using the Post-RA scheduler
on X86 Atom. Some of our tests failed because the tail merging part of
the BranchFolding pass was creating new basic blocks which did not
contain live-in information. When the anti-dependency code in the Post-RA
scheduler ran, it would sometimes rename the register containing
the function return value because the fact that the return value was
live-in to the subsequent block had been lost. To fix this, it is necessary
to run the RegisterScavenging code in the BranchFolding pass.

This patch makes sure that the register scavenging code is invoked
in the X86 subtarget only when post-RA scheduling is being done.
Post RA scheduling in the X86 subtarget is only done for Atom.

This patch adds a new function to the TargetRegisterClass to control
whether or not live-ins should be preserved during branch folding.
This is necessary in order for the anti-dependency optimizations done
during the PostRASchedulerList pass to work properly when doing
Post-RA scheduling for the X86 in general and for the Intel Atom in particular.

The patch adds and invokes the new function trackLivenessAfterRegAlloc()
instead of using the existing requiresRegisterScavenging().
It changes BranchFolding.cpp to call trackLivenessAfterRegAlloc() instead of
requiresRegisterScavenging(). It changes the all the targets that
implemented requiresRegisterScavenging() to also implement
trackLivenessAfterRegAlloc().

It adds an assertion in the Post RA scheduler to make sure that post RA
liveness information is available when it is needed.

It changes the X86 break-anti-dependencies test to use –mcpu=atom, in order
to avoid running into the added assertion.

Finally, this patch restores the use of anti-dependency checking
(which was turned off temporarily for the 3.1 release) for
Intel Atom in the Post RA scheduler.

Patch by Andy Zhang!

Thanks to Jakob and Anton for their reviews.




git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@155395 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/include/llvm/Target/TargetRegisterInfo.h b/include/llvm/Target/TargetRegisterInfo.h
index 471d9aeb473..f265460972c 100644
--- a/include/llvm/Target/TargetRegisterInfo.h
+++ b/include/llvm/Target/TargetRegisterInfo.h
@@ -612,6 +612,12 @@ public:
     return false;
   }
 
+  /// trackLivenessAfterRegAlloc - returns true if the live-ins should be tracked
+  /// after register allocation.
+  virtual bool trackLivenessAfterRegAlloc(const MachineFunction &MF) const {
+    return false;
+  }
+
   /// needsStackRealignment - true if storage within the function requires the
   /// stack pointer to be aligned more than the normal calling convention calls
   /// for.
diff --git a/lib/CodeGen/BranchFolding.cpp b/lib/CodeGen/BranchFolding.cpp
index ef1d2baed9c..c7f2d64a36b 100644
--- a/lib/CodeGen/BranchFolding.cpp
+++ b/lib/CodeGen/BranchFolding.cpp
@@ -188,7 +188,7 @@ bool BranchFolder::OptimizeFunction(MachineFunction &MF,
 
   // Use a RegScavenger to help update liveness when required.
   MachineRegisterInfo &MRI = MF.getRegInfo();
-  if (MRI.tracksLiveness() && TRI->requiresRegisterScavenging(MF))
+  if (MRI.tracksLiveness() && TRI->trackLivenessAfterRegAlloc(MF))
     RS = new RegScavenger();
   else
     MRI.invalidateLiveness();
diff --git a/lib/CodeGen/PostRASchedulerList.cpp b/lib/CodeGen/PostRASchedulerList.cpp
index 24d3e5ab0c9..13feaec7d8d 100644
--- a/lib/CodeGen/PostRASchedulerList.cpp
+++ b/lib/CodeGen/PostRASchedulerList.cpp
@@ -206,6 +206,10 @@ SchedulePostRATDList::SchedulePostRATDList(
   const InstrItineraryData *InstrItins = TM.getInstrItineraryData();
   HazardRec =
     TM.getInstrInfo()->CreateTargetPostRAHazardRecognizer(InstrItins, this);
+
+  assert((AntiDepMode == TargetSubtargetInfo::ANTIDEP_NONE ||
+          MRI.tracksLiveness()) &&
+         "Live-ins must be accurate for anti-dependency breaking");
   AntiDepBreak =
     ((AntiDepMode == TargetSubtargetInfo::ANTIDEP_ALL) ?
      (AntiDepBreaker *)new AggressiveAntiDepBreaker(MF, RCI, CriticalPathRCs) :
diff --git a/lib/Target/ARM/ARMBaseRegisterInfo.cpp b/lib/Target/ARM/ARMBaseRegisterInfo.cpp
index 2a09b20b9c6..9442e4bb712 100644
--- a/lib/Target/ARM/ARMBaseRegisterInfo.cpp
+++ b/lib/Target/ARM/ARMBaseRegisterInfo.cpp
@@ -711,6 +711,11 @@ requiresRegisterScavenging(const MachineFunction &MF) const {
   return true;
 }
 
+bool ARMBaseRegisterInfo::
+trackLivenessAfterRegAlloc(const MachineFunction &MF) const {
+  return true;
+}
+
 bool ARMBaseRegisterInfo::
 requiresFrameIndexScavenging(const MachineFunction &MF) const {
   return true;
diff --git a/lib/Target/ARM/ARMBaseRegisterInfo.h b/lib/Target/ARM/ARMBaseRegisterInfo.h
index af7935147e4..4eeee70cac7 100644
--- a/lib/Target/ARM/ARMBaseRegisterInfo.h
+++ b/lib/Target/ARM/ARMBaseRegisterInfo.h
@@ -173,6 +173,8 @@ public:
 
   virtual bool requiresRegisterScavenging(const MachineFunction &MF) const;
 
+  virtual bool trackLivenessAfterRegAlloc(const MachineFunction &MF) const;
+
   virtual bool requiresFrameIndexScavenging(const MachineFunction &MF) const;
 
   virtual bool requiresVirtualBaseRegisters(const MachineFunction &MF) const;
diff --git a/lib/Target/CellSPU/SPURegisterInfo.h b/lib/Target/CellSPU/SPURegisterInfo.h
index e5ab2242250..9fae8f2947e 100644
--- a/lib/Target/CellSPU/SPURegisterInfo.h
+++ b/lib/Target/CellSPU/SPURegisterInfo.h
@@ -63,6 +63,11 @@ namespace llvm {
     virtual bool requiresRegisterScavenging(const MachineFunction &MF) const
     { return true; }
 
+    //! Enable tracking of liveness after register allocation, since register
+    // scavenging is enabled.
+    virtual bool trackLivenessAfterRegAlloc(const MachineFunction &MF) const
+    { return true; }
+
     //! Return the reserved registers
     BitVector getReservedRegs(const MachineFunction &MF) const;
 
diff --git a/lib/Target/Hexagon/HexagonRegisterInfo.h b/lib/Target/Hexagon/HexagonRegisterInfo.h
index 6cf727bc027..85355ae7beb 100644
--- a/lib/Target/Hexagon/HexagonRegisterInfo.h
+++ b/lib/Target/Hexagon/HexagonRegisterInfo.h
@@ -73,6 +73,10 @@ struct HexagonRegisterInfo : public HexagonGenRegisterInfo {
     return true;
   }
 
+  bool trackLivenessAfterRegAlloc(const MachineFunction &MF) const {
+    return true;
+  }
+
   // Debug information queries.
   unsigned getRARegister() const;
   unsigned getFrameRegister(const MachineFunction &MF) const;
diff --git a/lib/Target/Mips/MipsRegisterInfo.cpp b/lib/Target/Mips/MipsRegisterInfo.cpp
index 67f2834ab68..952666301ac 100644
--- a/lib/Target/Mips/MipsRegisterInfo.cpp
+++ b/lib/Target/Mips/MipsRegisterInfo.cpp
@@ -136,6 +136,11 @@ MipsRegisterInfo::requiresRegisterScavenging(const MachineFunction &MF) const {
   return true;
 }
 
+bool
+MipsRegisterInfo::trackLivenessAfterRegAlloc(const MachineFunction &MF) const {
+  return true;
+}
+
 // This function eliminate ADJCALLSTACKDOWN,
 // ADJCALLSTACKUP pseudo instructions
 void MipsRegisterInfo::
diff --git a/lib/Target/Mips/MipsRegisterInfo.h b/lib/Target/Mips/MipsRegisterInfo.h
index 0716d29b2f3..6d3f83f5061 100644
--- a/lib/Target/Mips/MipsRegisterInfo.h
+++ b/lib/Target/Mips/MipsRegisterInfo.h
@@ -49,6 +49,8 @@ struct MipsRegisterInfo : public MipsGenRegisterInfo {
 
   virtual bool requiresRegisterScavenging(const MachineFunction &MF) const;
 
+  virtual bool trackLivenessAfterRegAlloc(const MachineFunction &MF) const;
+
   void eliminateCallFramePseudoInstr(MachineFunction &MF,
                                      MachineBasicBlock &MBB,
                                      MachineBasicBlock::iterator I) const;
diff --git a/lib/Target/PowerPC/PPCRegisterInfo.cpp b/lib/Target/PowerPC/PPCRegisterInfo.cpp
index ef1357137de..41e724ffa9d 100644
--- a/lib/Target/PowerPC/PPCRegisterInfo.cpp
+++ b/lib/Target/PowerPC/PPCRegisterInfo.cpp
@@ -89,6 +89,12 @@ PPCRegisterInfo::PPCRegisterInfo(const PPCSubtarget &ST,
   ImmToIdxMap[PPC::ADDI8] = PPC::ADD8; ImmToIdxMap[PPC::STD_32] = PPC::STDX_32;
 }
 
+bool
+PPCRegisterInfo::trackLivenessAfterRegAlloc(const MachineFunction &MF) const {
+  return requiresRegisterScavenging(MF);
+}
+
+
 /// getPointerRegClass - Return the register class to use to hold pointers.
 /// This is used for addressing modes.
 const TargetRegisterClass *
diff --git a/lib/Target/PowerPC/PPCRegisterInfo.h b/lib/Target/PowerPC/PPCRegisterInfo.h
index b1e6a7218ee..775c3f13115 100644
--- a/lib/Target/PowerPC/PPCRegisterInfo.h
+++ b/lib/Target/PowerPC/PPCRegisterInfo.h
@@ -50,6 +50,8 @@ public:
   /// FIXME (64-bit): Should be inlined.
   bool requiresRegisterScavenging(const MachineFunction &MF) const;
 
+  bool trackLivenessAfterRegAlloc(const MachineFunction &MF) const;
+
   void eliminateCallFramePseudoInstr(MachineFunction &MF,
                                      MachineBasicBlock &MBB,
                                      MachineBasicBlock::iterator I) const;
diff --git a/lib/Target/X86/X86RegisterInfo.cpp b/lib/Target/X86/X86RegisterInfo.cpp
index e612f496da5..6e00a552b60 100644
--- a/lib/Target/X86/X86RegisterInfo.cpp
+++ b/lib/Target/X86/X86RegisterInfo.cpp
@@ -90,6 +90,12 @@ int X86RegisterInfo::getCompactUnwindRegNum(unsigned RegNum, bool isEH) const {
   return -1;
 }
 
+bool
+X86RegisterInfo::trackLivenessAfterRegAlloc(const MachineFunction &MF) const {
+  // Only enable when post-RA scheduling is enabled and this is needed.
+  return TM.getSubtargetImpl()->postRAScheduler();
+}
+
 int
 X86RegisterInfo::getSEHRegNum(unsigned i) const {
   int reg = X86_MC::getX86RegNum(i);
diff --git a/lib/Target/X86/X86RegisterInfo.h b/lib/Target/X86/X86RegisterInfo.h
index bee03936f1f..4809fd5c110 100644
--- a/lib/Target/X86/X86RegisterInfo.h
+++ b/lib/Target/X86/X86RegisterInfo.h
@@ -65,7 +65,8 @@ public:
   int getCompactUnwindRegNum(unsigned RegNum, bool isEH) const;
 
   /// Code Generation virtual methods...
-  /// 
+  ///
+  virtual bool trackLivenessAfterRegAlloc(const MachineFunction &MF) const;
 
   /// getMatchingSuperRegClass - Return a subclass of the specified register
   /// class A so that each register in it has a sub-register of the
diff --git a/lib/Target/X86/X86Subtarget.cpp b/lib/Target/X86/X86Subtarget.cpp
index ed1a40965aa..452dd7eba32 100644
--- a/lib/Target/X86/X86Subtarget.cpp
+++ b/lib/Target/X86/X86Subtarget.cpp
@@ -424,9 +424,7 @@ bool X86Subtarget::enablePostRAScheduler(
            CodeGenOpt::Level OptLevel,
            TargetSubtargetInfo::AntiDepBreakMode& Mode,
            RegClassVector& CriticalPathRCs) const {
-  //TODO: change back to ANTIDEP_CRITICAL when the
-  // X86 subtarget properly sets up post RA liveness.
-  Mode = TargetSubtargetInfo::ANTIDEP_NONE;
+  Mode = TargetSubtargetInfo::ANTIDEP_CRITICAL;
   CriticalPathRCs.clear();
   return PostRAScheduler && OptLevel >= CodeGenOpt::Default;
 }
diff --git a/lib/Target/X86/X86Subtarget.h b/lib/Target/X86/X86Subtarget.h
index 7fd832bf067..766f2daccbd 100644
--- a/lib/Target/X86/X86Subtarget.h
+++ b/lib/Target/X86/X86Subtarget.h
@@ -307,6 +307,8 @@ public:
                              TargetSubtargetInfo::AntiDepBreakMode& Mode,
                              RegClassVector& CriticalPathRCs) const;
 
+  bool postRAScheduler() const { return PostRAScheduler; }
+
   /// getInstrItins = Return the instruction itineraries based on the
   /// subtarget selection.
   const InstrItineraryData &getInstrItineraryData() const { return InstrItins; }
diff --git a/lib/Target/XCore/XCoreRegisterInfo.cpp b/lib/Target/XCore/XCoreRegisterInfo.cpp
index 6812f8f56da..cdd0a0893b9 100644
--- a/lib/Target/XCore/XCoreRegisterInfo.cpp
+++ b/lib/Target/XCore/XCoreRegisterInfo.cpp
@@ -91,6 +91,11 @@ XCoreRegisterInfo::requiresRegisterScavenging(const MachineFunction &MF) const {
   return TFI->hasFP(MF);
 }
 
+bool
+XCoreRegisterInfo::trackLivenessAfterRegAlloc(const MachineFunction &MF) const {
+  return requiresRegisterScavenging(MF);
+}
+
 bool
 XCoreRegisterInfo::useFPForScavengingIndex(const MachineFunction &MF) const {
   return false;
diff --git a/lib/Target/XCore/XCoreRegisterInfo.h b/lib/Target/XCore/XCoreRegisterInfo.h
index 7391cfdf073..c4dcb6b533c 100644
--- a/lib/Target/XCore/XCoreRegisterInfo.h
+++ b/lib/Target/XCore/XCoreRegisterInfo.h
@@ -50,6 +50,8 @@ public:
   
   bool requiresRegisterScavenging(const MachineFunction &MF) const;
 
+  bool trackLivenessAfterRegAlloc(const MachineFunction &MF) const;
+
   bool useFPForScavengingIndex(const MachineFunction &MF) const;
 
   void eliminateCallFramePseudoInstr(MachineFunction &MF,
diff --git a/test/CodeGen/X86/atom-sched.ll b/test/CodeGen/X86/atom-sched.ll
index 4dd9a9e3481..0d97e853582 100644
--- a/test/CodeGen/X86/atom-sched.ll
+++ b/test/CodeGen/X86/atom-sched.ll
@@ -1,9 +1,6 @@
-; XFAIL: *
 ; RUN: llc <%s -O2 -mcpu=atom -march=x86 -relocation-model=static | FileCheck -check-prefix=atom %s
 ; RUN: llc <%s -O2 -mcpu=core2 -march=x86 -relocation-model=static | FileCheck %s
 ;
-; FIXME: Atom's scheduler is temporarily disabled.
-; XFAIL: *
 
 @a = common global i32 0, align 4
 @b = common global i32 0, align 4
diff --git a/test/CodeGen/X86/break-anti-dependencies.ll b/test/CodeGen/X86/break-anti-dependencies.ll
index 93b20437e1e..cf774591d8b 100644
--- a/test/CodeGen/X86/break-anti-dependencies.ll
+++ b/test/CodeGen/X86/break-anti-dependencies.ll
@@ -1,8 +1,10 @@
 ; Without list-burr scheduling we may not see the difference in codegen here.
-; RUN: llc < %s -march=x86-64 -post-RA-scheduler -pre-RA-sched=list-burr -break-anti-dependencies=none > %t
+; Use a subtarget that has post-RA scheduling enabled because the anti-dependency
+; breaker requires liveness information to be kept.
+; RUN: llc < %s -march=x86-64 -mcpu=atom -post-RA-scheduler -pre-RA-sched=list-burr -break-anti-dependencies=none > %t
 ; RUN:   grep {%xmm0} %t | count 14
 ; RUN:   not grep {%xmm1} %t
-; RUN: llc < %s -march=x86-64 -post-RA-scheduler -break-anti-dependencies=critical > %t
+; RUN: llc < %s -march=x86-64 -mcpu=atom -post-RA-scheduler -break-anti-dependencies=critical > %t
 ; RUN:   grep {%xmm0} %t | count 7
 ; RUN:   grep {%xmm1} %t | count 7