[ARM] Promote small global constants to constant pools
If a constant is unamed_addr and is only used within one function, we can save
on the code size and runtime cost of an indirection by changing the global's storage
to inside the constant pool. For example, instead of:
ldr r0, .CPI0
bl printf
bx lr
.CPI0: &format_string
format_string: .asciz "hello, world!\n"
We can emit:
adr r0, .CPI0
bl printf
bx lr
.CPI0: .asciz "hello, world!\n"
This can cause significant code size savings when many small strings are used in one
function (4 bytes per string).
This recommit contains fixes for a nasty bug related to fast-isel fallback - because
fast-isel doesn't know about this optimization, if it runs and emits references to
a string that we inline (because fast-isel fell back to SDAG) we will end up
with an inlined string and also an out-of-line string, and we won't emit the
out-of-line string, causing backend failures.
It also contains fixes for emitting .text relocations which made the sanitizer
bots unhappy.
llvm-svn: 281715
diff --git a/llvm/lib/Target/ARM/ARMAsmPrinter.cpp b/llvm/lib/Target/ARM/ARMAsmPrinter.cpp
index 616b71c..0ffa5b6 100644
--- a/llvm/lib/Target/ARM/ARMAsmPrinter.cpp
+++ b/llvm/lib/Target/ARM/ARMAsmPrinter.cpp
@@ -97,6 +97,13 @@
OutStreamer->EmitValue(E, Size);
}
+void ARMAsmPrinter::EmitGlobalVariable(const GlobalVariable *GV) {
+ if (PromotedGlobals.count(GV))
+ // The global was promoted into a constant pool. It should not be emitted.
+ return;
+ AsmPrinter::EmitGlobalVariable(GV);
+}
+
/// runOnMachineFunction - This uses the EmitInstruction()
/// method to print assembly for each instruction.
///
@@ -109,6 +116,12 @@
const Function* F = MF.getFunction();
const TargetMachine& TM = MF.getTarget();
+ // Collect all globals that had their storage promoted to a constant pool.
+ // Functions are emitted before variables, so this accumulates promoted
+ // globals from all functions in PromotedGlobals.
+ for (auto *GV : AFI->getGlobalsPromotedToConstantPool())
+ PromotedGlobals.insert(GV);
+
// Calculate this function's optimization goal.
unsigned OptimizationGoal;
if (F->hasFnAttribute(Attribute::OptimizeNone))
diff --git a/llvm/lib/Target/ARM/ARMAsmPrinter.h b/llvm/lib/Target/ARM/ARMAsmPrinter.h
index 97f5ca0..30abdbc 100644
--- a/llvm/lib/Target/ARM/ARMAsmPrinter.h
+++ b/llvm/lib/Target/ARM/ARMAsmPrinter.h
@@ -56,6 +56,12 @@
/// -1 if uninitialized, 0 if conflicting goals
int OptimizationGoals;
+ /// List of globals that have had their storage promoted to a constant
+ /// pool. This lives between calls to runOnMachineFunction and collects
+ /// data from every MachineFunction. It is used during doFinalization
+ /// when all non-function globals are emitted.
+ SmallPtrSet<const GlobalVariable*,2> PromotedGlobals;
+
public:
explicit ARMAsmPrinter(TargetMachine &TM,
std::unique_ptr<MCStreamer> Streamer);
@@ -90,7 +96,8 @@
void EmitStartOfAsmFile(Module &M) override;
void EmitEndOfAsmFile(Module &M) override;
void EmitXXStructor(const DataLayout &DL, const Constant *CV) override;
-
+ void EmitGlobalVariable(const GlobalVariable *GV) override;
+
// lowerOperand - Convert a MachineOperand into the equivalent MCOperand.
bool lowerOperand(const MachineOperand &MO, MCOperand &MCOp);
diff --git a/llvm/lib/Target/ARM/ARMISelLowering.cpp b/llvm/lib/Target/ARM/ARMISelLowering.cpp
index 434b454..8a864c2 100644
--- a/llvm/lib/Target/ARM/ARMISelLowering.cpp
+++ b/llvm/lib/Target/ARM/ARMISelLowering.cpp
@@ -59,12 +59,28 @@
STATISTIC(NumTailCalls, "Number of tail calls");
STATISTIC(NumMovwMovt, "Number of GAs materialized with movw + movt");
STATISTIC(NumLoopByVals, "Number of loops generated for byval arguments");
+STATISTIC(NumConstpoolPromoted,
+ "Number of constants with their storage promoted into constant pools");
static cl::opt<bool>
ARMInterworking("arm-interworking", cl::Hidden,
cl::desc("Enable / disable ARM interworking (for debugging only)"),
cl::init(true));
+static cl::opt<bool> EnableConstpoolPromotion(
+ "arm-promote-constant", cl::Hidden,
+ cl::desc("Enable / disable promotion of unnamed_addr constants into "
+ "constant pools"),
+ cl::init(true));
+static cl::opt<unsigned> ConstpoolPromotionMaxSize(
+ "arm-promote-constant-max-size", cl::Hidden,
+ cl::desc("Maximum size of constant to promote into a constant pool"),
+ cl::init(64));
+static cl::opt<unsigned> ConstpoolPromotionMaxTotal(
+ "arm-promote-constant-max-total", cl::Hidden,
+ cl::desc("Maximum size of ALL constants to promote into a constant pool"),
+ cl::init(128));
+
namespace {
class ARMCCState : public CCState {
public:
@@ -2963,6 +2979,153 @@
llvm_unreachable("bogus TLS model");
}
+/// Return true if all users of V are within function F, looking through
+/// ConstantExprs.
+static bool allUsersAreInFunction(const Value *V, const Function *F) {
+ SmallVector<const User*,4> Worklist;
+ for (auto *U : V->users())
+ Worklist.push_back(U);
+ while (!Worklist.empty()) {
+ auto *U = Worklist.pop_back_val();
+ if (isa<ConstantExpr>(U)) {
+ for (auto *UU : U->users())
+ Worklist.push_back(UU);
+ continue;
+ }
+
+ auto *I = dyn_cast<Instruction>(U);
+ if (!I || I->getParent()->getParent() != F)
+ return false;
+ }
+ return true;
+}
+
+/// Return true if all users of V are within some (any) function, looking through
+/// ConstantExprs. In other words, are there any global constant users?
+static bool allUsersAreInFunctions(const Value *V) {
+ SmallVector<const User*,4> Worklist;
+ for (auto *U : V->users())
+ Worklist.push_back(U);
+ while (!Worklist.empty()) {
+ auto *U = Worklist.pop_back_val();
+ if (isa<ConstantExpr>(U)) {
+ for (auto *UU : U->users())
+ Worklist.push_back(UU);
+ continue;
+ }
+
+ if (!isa<Instruction>(U))
+ return false;
+ }
+ return true;
+}
+
+// Return true if T is an integer, float or an array/vector of either.
+static bool isSimpleType(Type *T) {
+ if (T->isIntegerTy() || T->isFloatingPointTy())
+ return true;
+ Type *SubT = nullptr;
+ if (T->isArrayTy())
+ SubT = T->getArrayElementType();
+ else if (T->isVectorTy())
+ SubT = T->getVectorElementType();
+ else
+ return false;
+ return SubT->isIntegerTy() || SubT->isFloatingPointTy();
+}
+
+static SDValue promoteToConstantPool(const GlobalValue *GV, SelectionDAG &DAG,
+ EVT PtrVT, SDLoc dl) {
+ // If we're creating a pool entry for a constant global with unnamed address,
+ // and the global is small enough, we can emit it inline into the constant pool
+ // to save ourselves an indirection.
+ //
+ // This is a win if the constant is only used in one function (so it doesn't
+ // need to be duplicated) or duplicating the constant wouldn't increase code
+ // size (implying the constant is no larger than 4 bytes).
+ const Function *F = DAG.getMachineFunction().getFunction();
+
+ // We rely on this decision to inline being idemopotent and unrelated to the
+ // use-site. We know that if we inline a variable at one use site, we'll
+ // inline it elsewhere too (and reuse the constant pool entry). Fast-isel
+ // doesn't know about this optimization, so bail out if it's enabled else
+ // we could decide to inline here (and thus never emit the GV) but require
+ // the GV from fast-isel generated code.
+ if (!EnableConstpoolPromotion ||
+ DAG.getMachineFunction().getTarget().Options.EnableFastISel)
+ return SDValue();
+
+ auto *GVar = dyn_cast<GlobalVariable>(GV);
+ if (!GVar || !GVar->hasInitializer() ||
+ !GVar->isConstant() || !GVar->hasGlobalUnnamedAddr() ||
+ !GVar->hasLocalLinkage())
+ return SDValue();
+
+ // Ensure that we don't try and inline any type that contains pointers. If
+ // we inline a value that contains relocations, we move the relocations from
+ // .data to .text which is not ideal.
+ auto *Init = GVar->getInitializer();
+ if (!isSimpleType(Init->getType()))
+ return SDValue();
+
+ // The constant islands pass can only really deal with alignment requests
+ // <= 4 bytes and cannot pad constants itself. Therefore we cannot promote
+ // any type wanting greater alignment requirements than 4 bytes. We also
+ // can only promote constants that are multiples of 4 bytes in size or
+ // are paddable to a multiple of 4. Currently we only try and pad constants
+ // that are strings for simplicity.
+ auto *CDAInit = dyn_cast<ConstantDataArray>(Init);
+ unsigned Size = DAG.getDataLayout().getTypeAllocSize(Init->getType());
+ unsigned Align = DAG.getDataLayout().getABITypeAlignment(Init->getType());
+ unsigned RequiredPadding = 4 - (Size % 4);
+ bool PaddingPossible =
+ RequiredPadding == 4 || (CDAInit && CDAInit->isString());
+ if (!PaddingPossible || Align > 4 || Size > ConstpoolPromotionMaxSize)
+ return SDValue();
+
+ unsigned PaddedSize = Size + ((RequiredPadding == 4) ? 0 : RequiredPadding);
+ MachineFunction &MF = DAG.getMachineFunction();
+ ARMFunctionInfo *AFI = MF.getInfo<ARMFunctionInfo>();
+
+ // We can't bloat the constant pool too much, else the ConstantIslands pass
+ // may fail to converge. If we haven't promoted this global yet (it may have
+ // multiple uses), and promoting it would increase the constant pool size (Sz
+ // > 4), ensure we have space to do so up to MaxTotal.
+ if (!AFI->getGlobalsPromotedToConstantPool().count(GVar) && Size > 4)
+ if (AFI->getPromotedConstpoolIncrease() + PaddedSize - 4 >=
+ ConstpoolPromotionMaxTotal)
+ return SDValue();
+
+ // This is only valid if all users are in a single function OR it has users
+ // in multiple functions but it no larger than a pointer. We also check if
+ // GVar has constant (non-ConstantExpr) users. If so, it essentially has its
+ // address taken.
+ if (!allUsersAreInFunction(GVar, F) &&
+ !(Size <= 4 && allUsersAreInFunctions(GVar)))
+ return SDValue();
+
+ // We're going to inline this global. Pad it out if needed.
+ if (RequiredPadding != 4) {
+ StringRef S = CDAInit->getAsString();
+
+ SmallVector<uint8_t,16> V(S.size());
+ std::copy(S.bytes_begin(), S.bytes_end(), V.begin());
+ while (RequiredPadding--)
+ V.push_back(0);
+ Init = ConstantDataArray::get(*DAG.getContext(), V);
+ }
+
+ SDValue CPAddr =
+ DAG.getTargetConstantPool(Init, PtrVT, /*Align=*/4);
+ if (!AFI->getGlobalsPromotedToConstantPool().count(GVar)) {
+ AFI->markGlobalAsPromotedToConstantPool(GVar);
+ AFI->setPromotedConstpoolIncrease(AFI->getPromotedConstpoolIncrease() +
+ PaddedSize - 4);
+ }
+ ++NumConstpoolPromoted;
+ return DAG.getNode(ARMISD::Wrapper, dl, MVT::i32, CPAddr);
+}
+
SDValue ARMTargetLowering::LowerGlobalAddressELF(SDValue Op,
SelectionDAG &DAG) const {
EVT PtrVT = getPointerTy(DAG.getDataLayout());
@@ -2974,6 +3137,11 @@
bool IsRO =
(isa<GlobalVariable>(GV) && cast<GlobalVariable>(GV)->isConstant()) ||
isa<Function>(GV);
+
+ if (TM.shouldAssumeDSOLocal(*GV->getParent(), GV))
+ if (SDValue V = promoteToConstantPool(GV, DAG, PtrVT, dl))
+ return V;
+
if (isPositionIndependent()) {
bool UseGOT_PREL = !TM.shouldAssumeDSOLocal(*GV->getParent(), GV);
diff --git a/llvm/lib/Target/ARM/ARMMachineFunctionInfo.cpp b/llvm/lib/Target/ARM/ARMMachineFunctionInfo.cpp
index b6dee9f..72e37a3 100644
--- a/llvm/lib/Target/ARM/ARMMachineFunctionInfo.cpp
+++ b/llvm/lib/Target/ARM/ARMMachineFunctionInfo.cpp
@@ -21,4 +21,4 @@
FramePtrSpillOffset(0), GPRCS1Offset(0), GPRCS2Offset(0), DPRCSOffset(0),
GPRCS1Size(0), GPRCS2Size(0), DPRCSSize(0),
PICLabelUId(0), VarArgsFrameIndex(0), HasITBlocks(false),
- ArgumentStackSize(0), IsSplitCSR(false) {}
+ ArgumentStackSize(0), IsSplitCSR(false), PromotedGlobalsIncrease(0) {}
diff --git a/llvm/lib/Target/ARM/ARMMachineFunctionInfo.h b/llvm/lib/Target/ARM/ARMMachineFunctionInfo.h
index f714972..8c485e8 100644
--- a/llvm/lib/Target/ARM/ARMMachineFunctionInfo.h
+++ b/llvm/lib/Target/ARM/ARMMachineFunctionInfo.h
@@ -121,6 +121,12 @@
/// copies.
bool IsSplitCSR;
+ /// Globals that have had their storage promoted into the constant pool.
+ SmallPtrSet<const GlobalVariable*,2> PromotedGlobals;
+
+ /// The amount the literal pool has been increasedby due to promoted globals.
+ int PromotedGlobalsIncrease;
+
public:
ARMFunctionInfo() :
isThumb(false),
@@ -131,7 +137,8 @@
FramePtrSpillOffset(0), GPRCS1Offset(0), GPRCS2Offset(0), DPRCSOffset(0),
GPRCS1Size(0), GPRCS2Size(0), DPRCSAlignGapSize(0), DPRCSSize(0),
NumAlignedDPRCS2Regs(0), PICLabelUId(0),
- VarArgsFrameIndex(0), HasITBlocks(false), IsSplitCSR(false) {}
+ VarArgsFrameIndex(0), HasITBlocks(false), IsSplitCSR(false),
+ PromotedGlobalsIncrease(0) {}
explicit ARMFunctionInfo(MachineFunction &MF);
@@ -226,6 +233,22 @@
}
return It;
}
+
+ /// Indicate to the backend that \c GV has had its storage changed to inside
+ /// a constant pool. This means it no longer needs to be emitted as a
+ /// global variable.
+ void markGlobalAsPromotedToConstantPool(const GlobalVariable *GV) {
+ PromotedGlobals.insert(GV);
+ }
+ SmallPtrSet<const GlobalVariable*, 2>& getGlobalsPromotedToConstantPool() {
+ return PromotedGlobals;
+ }
+ int getPromotedConstpoolIncrease() const {
+ return PromotedGlobalsIncrease;
+ }
+ void setPromotedConstpoolIncrease(int Sz) {
+ PromotedGlobalsIncrease = Sz;
+ }
};
} // End llvm namespace