Pad Short Functions for Intel Atom

The current Intel Atom microarchitecture has a feature whereby
when a function returns early then it is slightly faster to execute
a sequence of NOP instructions to wait until the return address is ready,
as opposed to simply stalling on the ret instruction until
the return address is ready.

When compiling for X86 Atom only, this patch will run a pass,
called "X86PadShortFunction" which will add NOP instructions where less
than four cycles elapse between function entry and return.

It includes tests.

This patch has been updated to address Nadav's review comments
- Optimize only at >= O1 and don't do optimization if -Os is set
- Stores MachineBasicBlock* instead of BBNum
- Uses DenseMap instead of std::map
- Fixes placement of braces

Patch by Andy Zhang.




git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@171879 91177308-0d34-0410-b5e6-96231b3b80d8

lib/Target/X86/CMakeLists.txt

@@ -25,6 +25,7 @@ set(sources
   X86JITInfo.cpp
   X86MCInstLower.cpp
   X86MachineFunctionInfo.cpp
+  X86PadShortFunction.cpp
   X86RegisterInfo.cpp
   X86SelectionDAGInfo.cpp
   X86Subtarget.cpp

lib/Target/X86/X86.h

@@ -66,6 +66,10 @@ FunctionPass *createEmitX86CodeToMemory();
 /// \brief Creates an X86-specific Target Transformation Info pass.
 ImmutablePass *createX86TargetTransformInfoPass(const X86TargetMachine *TM);
 
+/// createX86PadShortFunctions - Return a pass that pads short functions
+/// with NOOPs. This will prevent a stall when returning on the Atom.
+FunctionPass *createX86PadShortFunctions();
+
 } // End llvm namespace
 
 #endif

lib/Target/X86/X86.td

@@ -123,8 +123,11 @@ def FeatureRTM     : SubtargetFeature<"rtm", "HasRTM", "true",
 def FeatureLeaForSP : SubtargetFeature<"lea-sp", "UseLeaForSP", "true",
                                      "Use LEA for adjusting the stack pointer">;
 def FeatureSlowDivide : SubtargetFeature<"idiv-to-divb",
-                          "HasSlowDivide", "true",
-                          "Use small divide for positive values less than 256">;
+                                     "HasSlowDivide", "true",
+                                     "Use small divide for positive values less than 256">;
+def FeaturePadShortFunctions : SubtargetFeature<"pad-short-functions",
+                                     "PadShortFunctions", "true",
+                                     "Pad short functions">;
 
 //===----------------------------------------------------------------------===//
 // X86 processors supported.
@@ -167,7 +170,7 @@ def : Proc<"penryn",          [FeatureSSE41, FeatureCMPXCHG16B,
                                FeatureSlowBTMem]>;
 def : AtomProc<"atom",        [ProcIntelAtom, FeatureSSSE3, FeatureCMPXCHG16B,
                                FeatureMOVBE, FeatureSlowBTMem, FeatureLeaForSP,
-                               FeatureSlowDivide]>;
+                               FeatureSlowDivide, FeaturePadShortFunctions]>;
 // "Arrandale" along with corei3 and corei5
 def : Proc<"corei7",          [FeatureSSE42, FeatureCMPXCHG16B,
                                FeatureSlowBTMem, FeatureFastUAMem,

lib/Target/X86/X86PadShortFunction.cpp

@@ -0,0 +1,171 @@
+//===-------- X86PadShortFunction.cpp - pad short functions -----------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file defines the pass which will pad short functions to prevent
+// a stall if a function returns before the return address is ready. This
+// is needed for some Intel Atom processors.
+//
+//===----------------------------------------------------------------------===//
+
+#include <algorithm>
+
+#define DEBUG_TYPE "x86-pad-short-functions"
+#include "X86.h"
+#include "X86InstrInfo.h"
+#include "llvm/ADT/Statistic.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineRegisterInfo.h"
+#include "llvm/CodeGen/Passes.h"
+#include "llvm/IR/Function.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/TargetInstrInfo.h"
+
+using namespace llvm;
+
+STATISTIC(NumBBsPadded, "Number of basic blocks padded");
+
+namespace {
+  struct PadShortFunc : public MachineFunctionPass {
+    static char ID;
+    PadShortFunc() : MachineFunctionPass(ID)
+                   , Threshold(4), TM(0), TII(0) {}
+
+    virtual bool runOnMachineFunction(MachineFunction &MF);
+
+    virtual const char *getPassName() const {
+      return "X86 Atom pad short functions";
+    }
+
+  private:
+    void findReturns(MachineBasicBlock *MBB,
+                     unsigned int Cycles = 0);
+
+    bool cyclesUntilReturn(MachineBasicBlock *MBB,
+                           unsigned int &Cycles,
+                           MachineBasicBlock::iterator *Location = 0);
+
+    void addPadding(MachineBasicBlock *MBB,
+                    MachineBasicBlock::iterator &MBBI,
+                    unsigned int NOOPsToAdd);
+
+    const unsigned int Threshold;
+    DenseMap<MachineBasicBlock*, unsigned int> ReturnBBs;
+
+    const TargetMachine *TM;
+    const TargetInstrInfo *TII;
+  };
+
+  char PadShortFunc::ID = 0;
+}
+
+FunctionPass *llvm::createX86PadShortFunctions() {
+  return new PadShortFunc();
+}
+
+/// runOnMachineFunction - Loop over all of the basic blocks, inserting
+/// NOOP instructions before early exits.
+bool PadShortFunc::runOnMachineFunction(MachineFunction &MF) {
+  bool OptForSize = MF.getFunction()->getAttributes().
+    hasAttribute(AttributeSet::FunctionIndex, Attribute::OptimizeForSize);
+
+  if (OptForSize)
+    return false;
+
+  TM = &MF.getTarget();
+  TII = TM->getInstrInfo();
+
+  // Search through basic blocks and mark the ones that have early returns
+  ReturnBBs.clear();
+  findReturns(MF.begin());
+
+  bool MadeChange = false;
+
+  MachineBasicBlock::iterator ReturnLoc;
+  MachineBasicBlock *MBB;
+  unsigned int Cycles = 0;
+  unsigned int BBCycles;
+
+  // Pad the identified basic blocks with NOOPs
+  for (DenseMap<MachineBasicBlock*, unsigned int>::iterator I = ReturnBBs.begin();
+       I != ReturnBBs.end(); ++I) {
+    MBB = I->first;
+    Cycles = I->second;
+
+    if (Cycles < Threshold) {
+      if (!cyclesUntilReturn(MBB, BBCycles, &ReturnLoc))
+        continue;
+
+      addPadding(MBB, ReturnLoc, Threshold - Cycles);
+      NumBBsPadded++;
+      MadeChange = true;
+    }
+  }
+
+  return MadeChange;
+}
+
+/// findReturn - Starting at MBB, follow control flow and add all
+/// basic blocks that contain a return to ReturnBBs.
+void PadShortFunc::findReturns(MachineBasicBlock *MBB, unsigned int Cycles) {
+  // If this BB has a return, note how many cycles it takes to get there.
+  bool hasReturn = cyclesUntilReturn(MBB, Cycles);
+  if (Cycles >= Threshold)
+    return;
+
+  if (hasReturn) {
+    ReturnBBs[MBB] = std::max(ReturnBBs[MBB], Cycles);
+    return;
+  }
+
+  // Follow branches in BB and look for returns
+  for (MachineBasicBlock::succ_iterator I = MBB->succ_begin();
+      I != MBB->succ_end(); ++I) {
+    findReturns(*I, Cycles);
+  }
+}
+
+/// cyclesUntilReturn - if the MBB has a return instruction, set Location
+/// to the instruction and return true. Return false otherwise.
+/// Cycles will be incremented by the number of cycles taken to reach the
+/// return or the end of the BB, whichever occurs first.
+bool PadShortFunc::cyclesUntilReturn(MachineBasicBlock *MBB,
+                                     unsigned int &Cycles,
+                                     MachineBasicBlock::iterator *Location) {
+  for (MachineBasicBlock::iterator MBBI = MBB->begin();
+        MBBI != MBB->end(); ++MBBI) {
+    MachineInstr *MI = MBBI;
+    // Mark basic blocks with a return instruction. Calls to other
+    // functions do not count because the called function will be padded,
+    // if necessary.
+    if (MI->isReturn() && !MI->isCall()) {
+      if (Location)
+        *Location = MBBI;
+      return true;
+    }
+
+    Cycles += TII->getInstrLatency(TM->getInstrItineraryData(), MI);
+  }
+
+  return false;
+}
+
+/// addPadding - Add the given number of NOOP instructions to the function
+/// just prior to the return at MBBI
+void PadShortFunc::addPadding(MachineBasicBlock *MBB,
+                              MachineBasicBlock::iterator &MBBI,
+                              unsigned int NOOPsToAdd) {
+  DebugLoc DL = MBBI->getDebugLoc();
+
+  while (NOOPsToAdd-- > 0) {
+    BuildMI(*MBB, MBBI, DL, TII->get(X86::NOOP));
+    BuildMI(*MBB, MBBI, DL, TII->get(X86::NOOP));
+  }
+}

lib/Target/X86/X86Subtarget.cpp

@@ -350,6 +350,7 @@ X86Subtarget::X86Subtarget(const std::string &TT, const std::string &CPU,
   , UseLeaForSP(false)
   , HasSlowDivide(false)
   , PostRAScheduler(false)
+  , PadShortFunctions(false)
   , stackAlignment(4)
   // FIXME: this is a known good value for Yonah. How about others?
   , MaxInlineSizeThreshold(128)

lib/Target/X86/X86Subtarget.h

@@ -146,6 +146,10 @@ class X86Subtarget : public X86GenSubtargetInfo {
   /// PostRAScheduler - True if using post-register-allocation scheduler.
   bool PostRAScheduler;
 
+  /// PadShortFunctions - True if the short functions should be padded to prevent
+  /// a stall when returning too early.
+  bool PadShortFunctions;
+
   /// stackAlignment - The minimum alignment known to hold of the stack frame on
   /// entry to the function and which must be maintained by every function.
   unsigned stackAlignment;
@@ -231,6 +235,7 @@ class X86Subtarget : public X86GenSubtargetInfo {
   bool hasCmpxchg16b() const { return HasCmpxchg16b; }
   bool useLeaForSP() const { return UseLeaForSP; }
   bool hasSlowDivide() const { return HasSlowDivide; }
+  bool padShortFunctions() const { return PadShortFunctions; }
 
   bool isAtom() const { return X86ProcFamily == IntelAtom; }
 

lib/Target/X86/X86TargetMachine.cpp

@@ -202,6 +202,12 @@ bool X86PassConfig::addPreEmitPass() {
     ShouldPrint = true;
   }
 
+  if (getOptLevel() != CodeGenOpt::None &&
+      getX86Subtarget().padShortFunctions()) {
+    addPass(createX86PadShortFunctions());
+    ShouldPrint = true;
+  }
+
   return ShouldPrint;
 }
 

test/CodeGen/X86/atom-pad-short-functions.ll

@@ -0,0 +1,78 @@
+; RUN: llc < %s -O1 -mcpu=atom -mtriple=i686-linux  | FileCheck %s
+
+declare void @external_function(...)
+
+define i32 @test_return_val(i32 %a) nounwind {
+; CHECK: test_return_val
+; CHECK: movl
+; CHECK: nop
+; CHECK: nop
+; CHECK: nop
+; CHECK: nop
+; CHECK: nop
+; CHECK: nop
+; CHECK: ret
+  ret i32 %a
+}
+
+define i32 @test_optsize(i32 %a) nounwind optsize {
+; CHECK: test_optsize
+; CHECK: movl
+; CHECK-NEXT: ret
+  ret i32 %a
+}
+
+define i32 @test_add(i32 %a, i32 %b) nounwind {
+; CHECK: test_add
+; CHECK: addl
+; CHECK: nop
+; CHECK: nop
+; CHECK: nop
+; CHECK: nop
+; CHECK: ret
+  %result = add i32 %a, %b
+  ret i32 %result
+}
+
+define i32 @test_multiple_ret(i32 %a, i32 %b, i1 %c) nounwind {
+; CHECK: @test_multiple_ret
+; CHECK: je
+
+; CHECK: nop
+; CHECK: nop
+; CHECK: ret
+
+; CHECK: nop
+; CHECK: nop
+; CHECK: ret
+
+  br i1 %c, label %bb1, label %bb2
+
+bb1:
+  ret i32 %a
+
+bb2:
+  ret i32 %b
+}
+
+define void @test_call_others(i32 %x) nounwind
+{
+; CHECK: test_call_others
+; CHECK: je
+  %tobool = icmp eq i32 %x, 0
+  br i1 %tobool, label %if.end, label %true.case
+
+; CHECK: jmp external_function
+true.case:
+  tail call void bitcast (void (...)* @external_function to void ()*)() nounwind
+  br label %if.end
+
+; CHECK: nop
+; CHECK: nop
+; CHECK: nop
+; CHECK: nop
+; CHECK: ret
+if.end:
+  ret void
+
+}

test/CodeGen/X86/fast-isel-x86-64.ll

@@ -1,5 +1,5 @@
-; RUN: llc < %s -mattr=-avx -fast-isel -O0 -regalloc=fast -asm-verbose=0 -fast-isel-abort | FileCheck %s
-; RUN: llc < %s -mattr=+avx -fast-isel -O0 -regalloc=fast -asm-verbose=0 -fast-isel-abort | FileCheck %s --check-prefix=AVX
+; RUN: llc < %s -mattr=-avx -fast-isel -mcpu=core2 -O0 -regalloc=fast -asm-verbose=0 -fast-isel-abort | FileCheck %s
+; RUN: llc < %s -mattr=+avx -fast-isel -mcpu=core2 -O0 -regalloc=fast -asm-verbose=0 -fast-isel-abort | FileCheck %s --check-prefix=AVX
 
 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-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64"
 target triple = "x86_64-apple-darwin10.0.0"

test/CodeGen/X86/ret-mmx.ll

@@ -1,4 +1,4 @@
-; RUN: llc < %s -mtriple=x86_64-apple-darwin11 -mattr=+mmx,+sse2 | FileCheck %s
+; RUN: llc < %s -mtriple=x86_64-apple-darwin11 -mcpu=core2 -mattr=+mmx,+sse2 | FileCheck %s
 ; rdar://6602459
 
 @g_v1di = external global <1 x i64>

test/CodeGen/X86/select.ll

@@ -282,7 +282,7 @@ define i32 @test13(i32 %a, i32 %b) nounwind {
 ; ATOM: test13:
 ; ATOM: cmpl
 ; ATOM-NEXT: sbbl
-; ATOM-NEXT: ret
+; ATOM: ret
 }
 
 define i32 @test14(i32 %a, i32 %b) nounwind {
@@ -299,7 +299,7 @@ define i32 @test14(i32 %a, i32 %b) nounwind {
 ; ATOM: cmpl
 ; ATOM-NEXT: sbbl
 ; ATOM-NEXT: notl
-; ATOM-NEXT: ret
+; ATOM: ret
 }
 
 ; rdar://10961709