aboutsummaryrefslogtreecommitdiffstats
path: root/lib/Transforms/InstCombine
diff options
context:
space:
mode:
Diffstat (limited to 'lib/Transforms/InstCombine')
-rw-r--r--lib/Transforms/InstCombine/InstCombineCalls.cpp12
-rw-r--r--lib/Transforms/InstCombine/InstCombineCompares.cpp41
-rw-r--r--lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp172
-rw-r--r--lib/Transforms/InstCombine/InstCombineSelect.cpp15
-rw-r--r--lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp29
5 files changed, 260 insertions, 9 deletions
diff --git a/lib/Transforms/InstCombine/InstCombineCalls.cpp b/lib/Transforms/InstCombine/InstCombineCalls.cpp
index c1d9d01..cbe1ca4 100644
--- a/lib/Transforms/InstCombine/InstCombineCalls.cpp
+++ b/lib/Transforms/InstCombine/InstCombineCalls.cpp
@@ -51,8 +51,8 @@ Instruction *InstCombiner::SimplifyMemTransfer(MemIntrinsic *MI) {
// if the size is something we can handle with a single primitive load/store.
// A single load+store correctly handles overlapping memory in the memmove
// case.
- unsigned Size = MemOpLength->getZExtValue();
- if (Size == 0) return MI; // Delete this mem transfer.
+ uint64_t Size = MemOpLength->getLimitedValue();
+ assert(Size && "0-sized memory transfering should be removed already.");
if (Size > 8 || (Size&(Size-1)))
return 0; // If not 1/2/4/8 bytes, exit.
@@ -133,11 +133,9 @@ Instruction *InstCombiner::SimplifyMemSet(MemSetInst *MI) {
ConstantInt *FillC = dyn_cast<ConstantInt>(MI->getValue());
if (!LenC || !FillC || !FillC->getType()->isIntegerTy(8))
return 0;
- uint64_t Len = LenC->getZExtValue();
+ uint64_t Len = LenC->getLimitedValue();
Alignment = MI->getAlignment();
-
- // If the length is zero, this is a no-op
- if (Len == 0) return MI; // memset(d,c,0,a) -> noop
+ assert(Len && "0-sized memory setting should be removed already.");
// memset(s,c,n) -> store s, c (for n=1,2,4,8)
if (Len <= 8 && isPowerOf2_32((uint32_t)Len)) {
@@ -795,7 +793,7 @@ Instruction *InstCombiner::tryOptimizeCall(CallInst *CI, const TargetData *TD) {
if (CI->getCalledFunction() == 0) return 0;
InstCombineFortifiedLibCalls Simplifier(this);
- Simplifier.fold(CI, TD);
+ Simplifier.fold(CI, TD, TLI);
return Simplifier.NewInstruction;
}
diff --git a/lib/Transforms/InstCombine/InstCombineCompares.cpp b/lib/Transforms/InstCombine/InstCombineCompares.cpp
index 7076d88..c3fc18c 100644
--- a/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -17,6 +17,7 @@
#include "llvm/Analysis/InstructionSimplify.h"
#include "llvm/Analysis/MemoryBuiltins.h"
#include "llvm/Target/TargetData.h"
+#include "llvm/Target/TargetLibraryInfo.h"
#include "llvm/Support/ConstantRange.h"
#include "llvm/Support/GetElementPtrTypeIterator.h"
#include "llvm/Support/PatternMatch.h"
@@ -2824,7 +2825,7 @@ Instruction *InstCombiner::FoldFCmp_IntToFP_Cst(FCmpInst &I,
case ICmpInst::ICMP_UGE:
// (float)int >= -4.4 --> true
// (float)int >= 4.4 --> int > 4
- if (!RHS.isNegative())
+ if (RHS.isNegative())
return ReplaceInstUsesWith(I, ConstantInt::getTrue(I.getContext()));
Pred = ICmpInst::ICMP_UGT;
break;
@@ -2985,6 +2986,44 @@ Instruction *InstCombiner::visitFCmpInst(FCmpInst &I) {
return Res;
}
break;
+ case Instruction::Call: {
+ CallInst *CI = cast<CallInst>(LHSI);
+ LibFunc::Func Func;
+ // Various optimization for fabs compared with zero.
+ if (RHSC->isNullValue() && CI->getCalledFunction() &&
+ TLI->getLibFunc(CI->getCalledFunction()->getName(), Func) &&
+ TLI->has(Func)) {
+ if (Func == LibFunc::fabs || Func == LibFunc::fabsf ||
+ Func == LibFunc::fabsl) {
+ switch (I.getPredicate()) {
+ default: break;
+ // fabs(x) < 0 --> false
+ case FCmpInst::FCMP_OLT:
+ return ReplaceInstUsesWith(I, Builder->getFalse());
+ // fabs(x) > 0 --> x != 0
+ case FCmpInst::FCMP_OGT:
+ return new FCmpInst(FCmpInst::FCMP_ONE, CI->getArgOperand(0),
+ RHSC);
+ // fabs(x) <= 0 --> x == 0
+ case FCmpInst::FCMP_OLE:
+ return new FCmpInst(FCmpInst::FCMP_OEQ, CI->getArgOperand(0),
+ RHSC);
+ // fabs(x) >= 0 --> !isnan(x)
+ case FCmpInst::FCMP_OGE:
+ return new FCmpInst(FCmpInst::FCMP_ORD, CI->getArgOperand(0),
+ RHSC);
+ // fabs(x) == 0 --> x == 0
+ // fabs(x) != 0 --> x != 0
+ case FCmpInst::FCMP_OEQ:
+ case FCmpInst::FCMP_UEQ:
+ case FCmpInst::FCMP_ONE:
+ case FCmpInst::FCMP_UNE:
+ return new FCmpInst(I.getPredicate(), CI->getArgOperand(0),
+ RHSC);
+ }
+ }
+ }
+ }
}
}
diff --git a/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp b/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp
index c485844..6ecb4c5 100644
--- a/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp
+++ b/lib/Transforms/InstCombine/InstCombineLoadStoreAlloca.cpp
@@ -20,7 +20,154 @@
#include "llvm/ADT/Statistic.h"
using namespace llvm;
-STATISTIC(NumDeadStore, "Number of dead stores eliminated");
+STATISTIC(NumDeadStore, "Number of dead stores eliminated");
+STATISTIC(NumGlobalCopies, "Number of allocas copied from constant global");
+
+/// pointsToConstantGlobal - Return true if V (possibly indirectly) points to
+/// some part of a constant global variable. This intentionally only accepts
+/// constant expressions because we can't rewrite arbitrary instructions.
+static bool pointsToConstantGlobal(Value *V) {
+ if (GlobalVariable *GV = dyn_cast<GlobalVariable>(V))
+ return GV->isConstant();
+ if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V))
+ if (CE->getOpcode() == Instruction::BitCast ||
+ CE->getOpcode() == Instruction::GetElementPtr)
+ return pointsToConstantGlobal(CE->getOperand(0));
+ return false;
+}
+
+/// isOnlyCopiedFromConstantGlobal - Recursively walk the uses of a (derived)
+/// pointer to an alloca. Ignore any reads of the pointer, return false if we
+/// see any stores or other unknown uses. If we see pointer arithmetic, keep
+/// track of whether it moves the pointer (with IsOffset) but otherwise traverse
+/// the uses. If we see a memcpy/memmove that targets an unoffseted pointer to
+/// the alloca, and if the source pointer is a pointer to a constant global, we
+/// can optimize this.
+static bool
+isOnlyCopiedFromConstantGlobal(Value *V, MemTransferInst *&TheCopy,
+ SmallVectorImpl<Instruction *> &ToDelete,
+ bool IsOffset = false) {
+ // We track lifetime intrinsics as we encounter them. If we decide to go
+ // ahead and replace the value with the global, this lets the caller quickly
+ // eliminate the markers.
+
+ for (Value::use_iterator UI = V->use_begin(), E = V->use_end(); UI!=E; ++UI) {
+ User *U = cast<Instruction>(*UI);
+
+ if (LoadInst *LI = dyn_cast<LoadInst>(U)) {
+ // Ignore non-volatile loads, they are always ok.
+ if (!LI->isSimple()) return false;
+ continue;
+ }
+
+ if (BitCastInst *BCI = dyn_cast<BitCastInst>(U)) {
+ // If uses of the bitcast are ok, we are ok.
+ if (!isOnlyCopiedFromConstantGlobal(BCI, TheCopy, ToDelete, IsOffset))
+ return false;
+ continue;
+ }
+ if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(U)) {
+ // If the GEP has all zero indices, it doesn't offset the pointer. If it
+ // doesn't, it does.
+ if (!isOnlyCopiedFromConstantGlobal(GEP, TheCopy, ToDelete,
+ IsOffset || !GEP->hasAllZeroIndices()))
+ return false;
+ continue;
+ }
+
+ if (CallSite CS = U) {
+ // If this is the function being called then we treat it like a load and
+ // ignore it.
+ if (CS.isCallee(UI))
+ continue;
+
+ // If this is a readonly/readnone call site, then we know it is just a
+ // load (but one that potentially returns the value itself), so we can
+ // ignore it if we know that the value isn't captured.
+ unsigned ArgNo = CS.getArgumentNo(UI);
+ if (CS.onlyReadsMemory() &&
+ (CS.getInstruction()->use_empty() || CS.doesNotCapture(ArgNo)))
+ continue;
+
+ // If this is being passed as a byval argument, the caller is making a
+ // copy, so it is only a read of the alloca.
+ if (CS.isByValArgument(ArgNo))
+ continue;
+ }
+
+ // Lifetime intrinsics can be handled by the caller.
+ if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(U)) {
+ if (II->getIntrinsicID() == Intrinsic::lifetime_start ||
+ II->getIntrinsicID() == Intrinsic::lifetime_end) {
+ assert(II->use_empty() && "Lifetime markers have no result to use!");
+ ToDelete.push_back(II);
+ continue;
+ }
+ }
+
+ // If this is isn't our memcpy/memmove, reject it as something we can't
+ // handle.
+ MemTransferInst *MI = dyn_cast<MemTransferInst>(U);
+ if (MI == 0)
+ return false;
+
+ // If the transfer is using the alloca as a source of the transfer, then
+ // ignore it since it is a load (unless the transfer is volatile).
+ if (UI.getOperandNo() == 1) {
+ if (MI->isVolatile()) return false;
+ continue;
+ }
+
+ // If we already have seen a copy, reject the second one.
+ if (TheCopy) return false;
+
+ // If the pointer has been offset from the start of the alloca, we can't
+ // safely handle this.
+ if (IsOffset) return false;
+
+ // If the memintrinsic isn't using the alloca as the dest, reject it.
+ if (UI.getOperandNo() != 0) return false;
+
+ // If the source of the memcpy/move is not a constant global, reject it.
+ if (!pointsToConstantGlobal(MI->getSource()))
+ return false;
+
+ // Otherwise, the transform is safe. Remember the copy instruction.
+ TheCopy = MI;
+ }
+ return true;
+}
+
+/// isOnlyCopiedFromConstantGlobal - Return true if the specified alloca is only
+/// modified by a copy from a constant global. If we can prove this, we can
+/// replace any uses of the alloca with uses of the global directly.
+static MemTransferInst *
+isOnlyCopiedFromConstantGlobal(AllocaInst *AI,
+ SmallVectorImpl<Instruction *> &ToDelete) {
+ MemTransferInst *TheCopy = 0;
+ if (isOnlyCopiedFromConstantGlobal(AI, TheCopy, ToDelete))
+ return TheCopy;
+ return 0;
+}
+
+/// getPointeeAlignment - Compute the minimum alignment of the value pointed
+/// to by the given pointer.
+static unsigned getPointeeAlignment(Value *V, const TargetData &TD) {
+ if (ConstantExpr *CE = dyn_cast<ConstantExpr>(V))
+ if (CE->getOpcode() == Instruction::BitCast ||
+ (CE->getOpcode() == Instruction::GetElementPtr &&
+ cast<GEPOperator>(CE)->hasAllZeroIndices()))
+ return getPointeeAlignment(CE->getOperand(0), TD);
+
+ if (GlobalVariable *GV = dyn_cast<GlobalVariable>(V))
+ if (!GV->isDeclaration())
+ return TD.getPreferredAlignment(GV);
+
+ if (PointerType *PT = dyn_cast<PointerType>(V->getType()))
+ return TD.getABITypeAlignment(PT->getElementType());
+
+ return 0;
+}
Instruction *InstCombiner::visitAllocaInst(AllocaInst &AI) {
// Ensure that the alloca array size argument has type intptr_t, so that
@@ -113,6 +260,29 @@ Instruction *InstCombiner::visitAllocaInst(AllocaInst &AI) {
}
}
+ // Check to see if this allocation is only modified by a memcpy/memmove from
+ // a constant global whose alignment is equal to or exceeds that of the
+ // allocation. If this is the case, we can change all users to use
+ // the constant global instead. This is commonly produced by the CFE by
+ // constructs like "void foo() { int A[] = {1,2,3,4,5,6,7,8,9...}; }" if 'A'
+ // is only subsequently read.
+ SmallVector<Instruction *, 4> ToDelete;
+ if (MemTransferInst *Copy = isOnlyCopiedFromConstantGlobal(&AI, ToDelete)) {
+ if (AI.getAlignment() <= getPointeeAlignment(Copy->getSource(), *TD)) {
+ DEBUG(dbgs() << "Found alloca equal to global: " << AI << '\n');
+ DEBUG(dbgs() << " memcpy = " << *Copy << '\n');
+ for (unsigned i = 0, e = ToDelete.size(); i != e; ++i)
+ EraseInstFromFunction(*ToDelete[i]);
+ Constant *TheSrc = cast<Constant>(Copy->getSource());
+ Instruction *NewI
+ = ReplaceInstUsesWith(AI, ConstantExpr::getBitCast(TheSrc,
+ AI.getType()));
+ EraseInstFromFunction(*Copy);
+ ++NumGlobalCopies;
+ return NewI;
+ }
+ }
+
// At last, use the generic allocation site handler to aggressively remove
// unused allocas.
return visitAllocSite(AI);
diff --git a/lib/Transforms/InstCombine/InstCombineSelect.cpp b/lib/Transforms/InstCombine/InstCombineSelect.cpp
index eb9945b..291e800 100644
--- a/lib/Transforms/InstCombine/InstCombineSelect.cpp
+++ b/lib/Transforms/InstCombine/InstCombineSelect.cpp
@@ -881,12 +881,16 @@ Instruction *InstCombiner::visitSelectInst(SelectInst &SI) {
if (SelectInst *TrueSI = dyn_cast<SelectInst>(TrueVal)) {
if (TrueSI->getCondition() == CondVal) {
+ if (SI.getTrueValue() == TrueSI->getTrueValue())
+ return 0;
SI.setOperand(1, TrueSI->getTrueValue());
return &SI;
}
}
if (SelectInst *FalseSI = dyn_cast<SelectInst>(FalseVal)) {
if (FalseSI->getCondition() == CondVal) {
+ if (SI.getFalseValue() == FalseSI->getFalseValue())
+ return 0;
SI.setOperand(2, FalseSI->getFalseValue());
return &SI;
}
@@ -899,5 +903,16 @@ Instruction *InstCombiner::visitSelectInst(SelectInst &SI) {
return &SI;
}
+ if (VectorType* VecTy = dyn_cast<VectorType>(SI.getType())) {
+ unsigned VWidth = VecTy->getNumElements();
+ APInt UndefElts(VWidth, 0);
+ APInt AllOnesEltMask(APInt::getAllOnesValue(VWidth));
+ if (Value *V = SimplifyDemandedVectorElts(&SI, AllOnesEltMask, UndefElts)) {
+ if (V != &SI)
+ return ReplaceInstUsesWith(SI, V);
+ return &SI;
+ }
+ }
+
return 0;
}
diff --git a/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp b/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp
index 125c74a..54be8ed 100644
--- a/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp
+++ b/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp
@@ -989,6 +989,29 @@ Value *InstCombiner::SimplifyDemandedVectorElts(Value *V, APInt DemandedElts,
}
break;
}
+ case Instruction::Select: {
+ APInt LeftDemanded(DemandedElts), RightDemanded(DemandedElts);
+ if (ConstantVector* CV = dyn_cast<ConstantVector>(I->getOperand(0))) {
+ for (unsigned i = 0; i < VWidth; i++) {
+ if (CV->getAggregateElement(i)->isNullValue())
+ LeftDemanded.clearBit(i);
+ else
+ RightDemanded.clearBit(i);
+ }
+ }
+
+ TmpV = SimplifyDemandedVectorElts(I->getOperand(1), LeftDemanded,
+ UndefElts, Depth+1);
+ if (TmpV) { I->setOperand(1, TmpV); MadeChange = true; }
+
+ TmpV = SimplifyDemandedVectorElts(I->getOperand(2), RightDemanded,
+ UndefElts2, Depth+1);
+ if (TmpV) { I->setOperand(2, TmpV); MadeChange = true; }
+
+ // Output elements are undefined if both are undefined.
+ UndefElts &= UndefElts2;
+ break;
+ }
case Instruction::BitCast: {
// Vector->vector casts only.
VectorType *VTy = dyn_cast<VectorType>(I->getOperand(0)->getType());
@@ -1074,6 +1097,12 @@ Value *InstCombiner::SimplifyDemandedVectorElts(Value *V, APInt DemandedElts,
// like undef&0. The result is known zero, not undef.
UndefElts &= UndefElts2;
break;
+ case Instruction::FPTrunc:
+ case Instruction::FPExt:
+ TmpV = SimplifyDemandedVectorElts(I->getOperand(0), DemandedElts,
+ UndefElts, Depth+1);
+ if (TmpV) { I->setOperand(0, TmpV); MadeChange = true; }
+ break;
case Instruction::Call: {
IntrinsicInst *II = dyn_cast<IntrinsicInst>(I);
generated by cgit v1.1 at 2025-11-10 19:01:29 +0000