RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-11-24 03:33:20 +01:00
Code Issues Projects Releases Wiki Activity

This reverts commit r265913 and r265912

Browse Source 
See PR27315

r265913: "[IndVars] Eliminate op.with.overflow when possible"

r265912: "[SCEV] See through op.with.overflow intrinsics"
llvm-svn: 265950
This commit is contained in:
Sanjoy Das 2016-04-11 15:26:18 +00:00
parent a8eb1f20c0
commit 23c7d74ce2
7 changed files with 6 additions and 670 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
include/llvm/Analysis/ValueTracking.h
View File

@ -18,7 +18,6 @@
#include "llvm/ADT/ArrayRef.h"
#include "llvm/IR/ConstantRange.h"
#include "llvm/IR/Instruction.h"
#include "llvm/IR/IntrinsicInst.h"
#include "llvm/Support/DataTypes.h"
namespace llvm {
@ -326,11 +325,6 @@ namespace llvm {
const Instruction *CxtI = nullptr,
const DominatorTree *DT = nullptr);
/// Returns true if the arithmetic part of the \p II 's result is
/// used only along the paths control dependent on the computation
/// not overflowing, \p II being an <op>.with.overflow intrinsic.
bool isOverflowIntrinsicNoWrap(IntrinsicInst *II, DominatorTree &DT);
/// Return true if this function can prove that the instruction I will
/// always transfer execution to one of its successors (including the next
/// instruction that follows within a basic block). E.g. this is not

54
lib/Analysis/ScalarEvolution.cpp
View File

@ -3831,7 +3831,7 @@ struct BinaryOp {
/// Try to map \p V into a BinaryOp, and return \c None on failure.
static Optional<BinaryOp> MatchBinaryOp(Value *V, DominatorTree &DT) {
static Optional<BinaryOp> MatchBinaryOp(Value *V) {
auto *Op = dyn_cast<Operator>(V);
if (!Op)
return None;
@ -3877,50 +3877,6 @@ static Optional<BinaryOp> MatchBinaryOp(Value *V, DominatorTree &DT) {
}
return BinaryOp(Op);
case Instruction::ExtractValue: {
auto *EVI = cast<ExtractValueInst>(Op);
if (EVI->getNumIndices() != 1 || EVI->getIndices()[0] != 0)
break;
auto *CI = dyn_cast<CallInst>(EVI->getAggregateOperand());
if (!CI)
break;
if (auto *F = CI->getCalledFunction())
switch (F->getIntrinsicID()) {
case Intrinsic::sadd_with_overflow:
case Intrinsic::uadd_with_overflow: {
if (!isOverflowIntrinsicNoWrap(cast<IntrinsicInst>(CI), DT))
return BinaryOp(Instruction::Add, CI->getArgOperand(0),
CI->getArgOperand(1));
// Now that we know that all uses of the arithmetic-result component of
// CI are guarded by the overflow check, we can go ahead and pretend
// that the arithmetic is non-overflowing.
if (F->getIntrinsicID() == Intrinsic::sadd_with_overflow)
return BinaryOp(Instruction::Add, CI->getArgOperand(0),
CI->getArgOperand(1), /* IsNSW = */ true,
/* IsNUW = */ false);
else
return BinaryOp(Instruction::Add, CI->getArgOperand(0),
CI->getArgOperand(1), /* IsNSW = */ false,
/* IsNUW*/ true);
}
case Intrinsic::ssub_with_overflow:
case Intrinsic::usub_with_overflow:
return BinaryOp(Instruction::Sub, CI->getArgOperand(0),
CI->getArgOperand(1));
case Intrinsic::smul_with_overflow:
case Intrinsic::umul_with_overflow:
return BinaryOp(Instruction::Mul, CI->getArgOperand(0),
CI->getArgOperand(1));
default:
break;
}
}
default:
break;
}
@ -3997,7 +3953,7 @@ const SCEV *ScalarEvolution::createAddRecFromPHI(PHINode *PN) {
// If the increment doesn't overflow, then neither the addrec nor
// the post-increment will overflow.
if (auto BO = MatchBinaryOp(BEValueV, DT)) {
if (auto BO = MatchBinaryOp(BEValueV)) {
if (BO->Opcode == Instruction::Add && BO->LHS == PN) {
if (BO->IsNUW)
Flags = setFlags(Flags, SCEV::FlagNUW);
@ -4877,7 +4833,7 @@ const SCEV *ScalarEvolution::createSCEV(Value *V) {
return getUnknown(V);
Operator *U = cast<Operator>(V);
if (auto BO = MatchBinaryOp(U, DT)) {
if (auto BO = MatchBinaryOp(U)) {
switch (BO->Opcode) {
case Instruction::Add: {
// The simple thing to do would be to just call getSCEV on both operands
@ -4918,7 +4874,7 @@ const SCEV *ScalarEvolution::createSCEV(Value *V) {
else
AddOps.push_back(getSCEV(BO->RHS));
auto NewBO = MatchBinaryOp(BO->LHS, DT);
auto NewBO = MatchBinaryOp(BO->LHS);
if (!NewBO || (NewBO->Opcode != Instruction::Add &&
NewBO->Opcode != Instruction::Sub)) {
AddOps.push_back(getSCEV(BO->LHS));
@ -4948,7 +4904,7 @@ const SCEV *ScalarEvolution::createSCEV(Value *V) {
}
MulOps.push_back(getSCEV(BO->RHS));
auto NewBO = MatchBinaryOp(BO->LHS, DT);
auto NewBO = MatchBinaryOp(BO->LHS);
if (!NewBO || NewBO->Opcode != Instruction::Mul) {
MulOps.push_back(getSCEV(BO->LHS));
break;

61
lib/Analysis/ValueTracking.cpp
View File

@ -3253,67 +3253,6 @@ static OverflowResult computeOverflowForSignedAdd(
return OverflowResult::MayOverflow;
}
bool llvm::isOverflowIntrinsicNoWrap(IntrinsicInst *II, DominatorTree &DT) {
#ifndef NDEBUG
auto IID = II->getIntrinsicID();
assert((IID == Intrinsic::sadd_with_overflow ||
IID == Intrinsic::uadd_with_overflow ||
IID == Intrinsic::ssub_with_overflow ||
IID == Intrinsic::usub_with_overflow ||
IID == Intrinsic::smul_with_overflow ||
IID == Intrinsic::umul_with_overflow) &&
"Not an overflow intrinsic!");
#endif
SmallVector<BranchInst *, 2> GuardingBranches;
SmallVector<ExtractValueInst *, 2> Results;
for (User *U : II->users()) {
if (auto *EVI = dyn_cast<ExtractValueInst>(U)) {
assert(EVI->getNumIndices() == 1 && "Obvious from CI's type");
if (EVI->getIndices()[0] == 0)
Results.push_back(EVI);
else {
assert(EVI->getIndices()[0] == 1 && "Obvious from CI's type");
for (auto *U : EVI->users())
if (auto *B = dyn_cast<BranchInst>(U)) {
assert(B->isConditional() && "How else is it using an i1?");
GuardingBranches.push_back(B);
}
}
} else {
// We are using the aggregate directly in a way we don't want to analyze
// here (storing it to a global, say).
return false;
}
}
auto AllUsesGuardedByBranch = [&](BranchInst *BI) {
BasicBlockEdge NoWrapEdge(BI->getParent(), BI->getSuccessor(1));
if (!NoWrapEdge.isSingleEdge())
return false;
// Check if all users of the add are provably no-wrap.
for (auto *Result : Results) {
// If the extractvalue itself is not executed on overflow, the we don't
// need to check each use separately, since domination is transitive.
if (DT.dominates(NoWrapEdge, Result->getParent()))
continue;
for (auto &RU : Result->uses())
if (!DT.dominates(NoWrapEdge, RU))
return false;
}
return true;
};
return any_of(GuardingBranches, AllUsesGuardedByBranch);
}
OverflowResult llvm::computeOverflowForSignedAdd(AddOperator *Add,
const DataLayout &DL,
AssumptionCache *AC,

107
lib/Transforms/Utils/SimplifyIndVar.cpp
View File

@ -71,7 +71,6 @@ namespace {
bool eliminateIdentitySCEV(Instruction *UseInst, Instruction *IVOperand);
bool eliminateOverflowIntrinsic(CallInst *CI);
bool eliminateIVUser(Instruction *UseInst, Instruction *IVOperand);
void eliminateIVComparison(ICmpInst *ICmp, Value *IVOperand);
void eliminateIVRemainder(BinaryOperator *Rem, Value *IVOperand,
@ -319,108 +318,6 @@ void SimplifyIndvar::eliminateIVRemainder(BinaryOperator *Rem,
DeadInsts.emplace_back(Rem);
}
bool SimplifyIndvar::eliminateOverflowIntrinsic(CallInst *CI) {
auto *F = CI->getCalledFunction();
if (!F)
return false;
typedef const SCEV *(ScalarEvolution::*OperationFunctionTy)(
const SCEV *, const SCEV *, SCEV::NoWrapFlags);
typedef const SCEV *(ScalarEvolution::*ExtensionFunctionTy)(
const SCEV *, Type *);
OperationFunctionTy Operation;
ExtensionFunctionTy Extension;
Instruction::BinaryOps RawOp;
// We always have exactly one of nsw or nuw. If NoSignedOverflow is false, we
// have nuw.
bool NoSignedOverflow;
switch (F->getIntrinsicID()) {
default:
return false;
case Intrinsic::sadd_with_overflow:
Operation = &ScalarEvolution::getAddExpr;
Extension = &ScalarEvolution::getSignExtendExpr;
RawOp = Instruction::Add;
NoSignedOverflow = true;
break;
case Intrinsic::uadd_with_overflow:
Operation = &ScalarEvolution::getAddExpr;
Extension = &ScalarEvolution::getZeroExtendExpr;
RawOp = Instruction::Add;
NoSignedOverflow = false;
break;
case Intrinsic::ssub_with_overflow:
Operation = &ScalarEvolution::getMinusSCEV;
Extension = &ScalarEvolution::getSignExtendExpr;
RawOp = Instruction::Sub;
NoSignedOverflow = true;
break;
case Intrinsic::usub_with_overflow:
Operation = &ScalarEvolution::getMinusSCEV;
Extension = &ScalarEvolution::getZeroExtendExpr;
RawOp = Instruction::Sub;
NoSignedOverflow = false;
break;
}
const SCEV *LHS = SE->getSCEV(CI->getArgOperand(0));
const SCEV *RHS = SE->getSCEV(CI->getArgOperand(1));
auto *NarrowTy = cast<IntegerType>(LHS->getType());
auto *WideTy =
IntegerType::get(NarrowTy->getContext(), NarrowTy->getBitWidth() * 2);
const SCEV *A =
(SE->*Extension)((SE->*Operation)(LHS, RHS, SCEV::FlagAnyWrap), WideTy);
const SCEV *B =
(SE->*Operation)((SE->*Extension)(LHS, WideTy),
(SE->*Extension)(RHS, WideTy), SCEV::FlagAnyWrap);
if (A != B)
return false;
// Proved no overflow, nuke the overflow check and, if possible, the overflow
// intrinsic as well.
BinaryOperator *NewResult = BinaryOperator::Create(
RawOp, CI->getArgOperand(0), CI->getArgOperand(1), "", CI);
if (NoSignedOverflow)
NewResult->setHasNoSignedWrap(true);
else
NewResult->setHasNoUnsignedWrap(true);
SmallVector<ExtractValueInst *, 4> ToDelete;
for (auto *U : CI->users()) {
if (auto *EVI = dyn_cast<ExtractValueInst>(U)) {
if (EVI->getIndices()[0] == 1)
EVI->replaceAllUsesWith(ConstantInt::getFalse(CI->getContext()));
else {
assert(EVI->getIndices()[0] == 0 && "Only two possibilities!");
EVI->replaceAllUsesWith(NewResult);
}
ToDelete.push_back(EVI);
}
}
for (auto *EVI : ToDelete)
EVI->eraseFromParent();
if (CI->use_empty())
CI->eraseFromParent();
return true;
}
/// Eliminate an operation that consumes a simple IV and has no observable
/// side-effect given the range of IV values. IVOperand is guaranteed SCEVable,
/// but UseInst may not be.
@ -438,10 +335,6 @@ bool SimplifyIndvar::eliminateIVUser(Instruction *UseInst,
}
}
if (auto *CI = dyn_cast<CallInst>(UseInst))
if (eliminateOverflowIntrinsic(CI))
return true;
if (eliminateIdentitySCEV(UseInst, IVOperand))
return true;

309
test/Analysis/ScalarEvolution/overflow-intrinsics.ll
View File

@ -1,309 +0,0 @@
; RUN: opt -analyze -scalar-evolution < %s | FileCheck %s
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-linux-gnu"
define void @f_sadd_0(i8* %a) {
; CHECK-LABEL: Classifying expressions for: @f_sadd_0
entry:
br label %for.body
for.cond.cleanup: ; preds = %cont
ret void
for.body: ; preds = %entry, %cont
; CHECK: %i.04 = phi i32 [ 0, %entry ], [ %tmp2, %cont ]
; CHECK-NEXT: --> {0,+,1}<nuw><nsw><%for.body> U: [0,16) S: [0,16)
%i.04 = phi i32 [ 0, %entry ], [ %tmp2, %cont ]
%idxprom = sext i32 %i.04 to i64
%arrayidx = getelementptr inbounds i8, i8* %a, i64 %idxprom
store i8 0, i8* %arrayidx, align 1
%tmp0 = tail call { i32, i1 } @llvm.sadd.with.overflow.i32(i32 %i.04, i32 1)
%tmp1 = extractvalue { i32, i1 } %tmp0, 1
br i1 %tmp1, label %trap, label %cont, !nosanitize !{}
trap: ; preds = %for.body
tail call void @llvm.trap() #2, !nosanitize !{}
unreachable, !nosanitize !{}
cont: ; preds = %for.body
%tmp2 = extractvalue { i32, i1 } %tmp0, 0
%cmp = icmp slt i32 %tmp2, 16
br i1 %cmp, label %for.body, label %for.cond.cleanup
; CHECK: Loop %for.body: max backedge-taken count is 15
}
define void @f_sadd_1(i8* %a) {
; CHECK-LABEL: Classifying expressions for: @f_sadd_1
entry:
br label %for.body
for.cond.cleanup: ; preds = %cont
ret void
for.body: ; preds = %entry, %cont
; CHECK: %i.04 = phi i32 [ 0, %entry ], [ %tmp2, %cont ]
; CHECK-NEXT: --> {0,+,1}<%for.body> U: [0,16) S: [0,16)
; SCEV can prove <nsw> for the above induction variable; but it does
; not bother so before it sees the sext below since it is not a 100%
; obvious.
%i.04 = phi i32 [ 0, %entry ], [ %tmp2, %cont ]
%idxprom = sext i32 %i.04 to i64
%arrayidx = getelementptr inbounds i8, i8* %a, i64 %idxprom
store i8 0, i8* %arrayidx, align 1
%tmp0 = tail call { i32, i1 } @llvm.sadd.with.overflow.i32(i32 %i.04, i32 1)
%tmp1 = extractvalue { i32, i1 } %tmp0, 1
br i1 %tmp1, label %trap, label %cont, !nosanitize !{}
trap: ; preds = %for.body
br label %cont
cont: ; preds = %for.body
%tmp2 = extractvalue { i32, i1 } %tmp0, 0
%cmp = icmp slt i32 %tmp2, 16
br i1 %cmp, label %for.body, label %for.cond.cleanup
; CHECK: Loop %for.body: max backedge-taken count is 15
}
define void @f_sadd_2(i8* %a, i1* %c) {
; CHECK-LABEL: Classifying expressions for: @f_sadd_2
entry:
br label %for.body
for.cond.cleanup: ; preds = %cont
ret void
for.body: ; preds = %entry, %cont
; CHECK: %i.04 = phi i32 [ 0, %entry ], [ %tmp2, %cont ]
; CHECK-NEXT: --> {0,+,1}<%for.body>
%i.04 = phi i32 [ 0, %entry ], [ %tmp2, %cont ]
%idxprom = sext i32 %i.04 to i64
%arrayidx = getelementptr inbounds i8, i8* %a, i64 %idxprom
store i8 0, i8* %arrayidx, align 1
%tmp0 = tail call { i32, i1 } @llvm.sadd.with.overflow.i32(i32 %i.04, i32 1)
%tmp1 = extractvalue { i32, i1 } %tmp0, 1
br i1 %tmp1, label %trap, label %cont, !nosanitize !{}
trap: ; preds = %for.body
br label %cont
cont: ; preds = %for.body
%tmp2 = extractvalue { i32, i1 } %tmp0, 0
%cond = load volatile i1, i1* %c
br i1 %cond, label %for.body, label %for.cond.cleanup
}
define void @f_sadd_3(i8* %a, i1* %c) {
; CHECK-LABEL: Classifying expressions for: @f_sadd_3
entry:
br label %for.body
for.cond.cleanup: ; preds = %cont
ret void
for.body: ; preds = %entry, %cont
; CHECK: %i.04 = phi i32 [ 0, %entry ], [ %tmp2, %for.body ]
; CHECK-NEXT: --> {0,+,1}<nuw><nsw><%for.body>
%i.04 = phi i32 [ 0, %entry ], [ %tmp2, %for.body ]
%idxprom = sext i32 %i.04 to i64
%arrayidx = getelementptr inbounds i8, i8* %a, i64 %idxprom
store i8 0, i8* %arrayidx, align 1
%tmp0 = tail call { i32, i1 } @llvm.sadd.with.overflow.i32(i32 %i.04, i32 1)
%tmp1 = extractvalue { i32, i1 } %tmp0, 1
%tmp2 = extractvalue { i32, i1 } %tmp0, 0
br i1 %tmp1, label %trap, label %for.body, !nosanitize !{}
trap: ; preds = %for.body
tail call void @llvm.trap() #2, !nosanitize !{}
unreachable, !nosanitize !{}
}
define void @f_sadd_4(i8* %a, i1* %c) {
; CHECK-LABEL: Classifying expressions for: @f_sadd_4
entry:
br label %for.body
for.cond.cleanup: ; preds = %cont
ret void
for.body: ; preds = %entry, %cont
; CHECK: %i.04 = phi i32 [ 0, %entry ], [ %tmp2, %merge ]
; CHECK-NEXT: --> {0,+,1}<nuw><nsw><%for.body>
%i.04 = phi i32 [ 0, %entry ], [ %tmp2, %merge ]
%idxprom = sext i32 %i.04 to i64
%arrayidx = getelementptr inbounds i8, i8* %a, i64 %idxprom
store i8 0, i8* %arrayidx, align 1
%tmp0 = tail call { i32, i1 } @llvm.sadd.with.overflow.i32(i32 %i.04, i32 1)
%tmp1 = extractvalue { i32, i1 } %tmp0, 1
%tmp2 = extractvalue { i32, i1 } %tmp0, 0
br i1 %tmp1, label %notrap, label %merge
notrap:
br label %merge
merge:
%tmp3 = extractvalue { i32, i1 } %tmp0, 1
br i1 %tmp3, label %trap, label %for.body, !nosanitize !{}
trap: ; preds = %for.body
tail call void @llvm.trap() #2, !nosanitize !{}
unreachable, !nosanitize !{}
}
define void @f_sadd_may_overflow(i8* %a, i1* %c) {
; CHECK-LABEL: Classifying expressions for: @f_sadd_may_overflow
entry:
br label %for.body
for.cond.cleanup: ; preds = %cont
ret void
for.body: ; preds = %entry, %cont
; CHECK: %i.04 = phi i32 [ 0, %entry ], [ %tmp1, %cont ]
; CHECK-NEXT: --> {0,+,1}<%for.body> U: full-set S: full-set
%i.04 = phi i32 [ 0, %entry ], [ %tmp1, %cont ]
%idxprom = sext i32 %i.04 to i64
%arrayidx = getelementptr inbounds i8, i8* %a, i64 %idxprom
store i8 0, i8* %arrayidx, align 1
%tmp0 = tail call { i32, i1 } @llvm.sadd.with.overflow.i32(i32 %i.04, i32 1)
%cond1 = load volatile i1, i1* %c
br i1 %cond1, label %trap, label %cont, !nosanitize !{}
trap: ; preds = %for.body
tail call void @llvm.trap() #2, !nosanitize !{}
unreachable, !nosanitize !{}
cont: ; preds = %for.body
%tmp1 = extractvalue { i32, i1 } %tmp0, 0
%cond = load volatile i1, i1* %c
br i1 %cond, label %for.body, label %for.cond.cleanup
}
define void @f_uadd(i8* %a) {
; CHECK-LABEL: Classifying expressions for: @f_uadd
entry:
br label %for.body
for.cond.cleanup: ; preds = %cont
ret void
for.body: ; preds = %entry, %cont
; CHECK: %i.04 = phi i32 [ 0, %entry ], [ %tmp2, %cont ]
; CHECK-NEXT: --> {0,+,1}<nuw><%for.body> U: [0,16) S: [0,16)
%i.04 = phi i32 [ 0, %entry ], [ %tmp2, %cont ]
%idxprom = sext i32 %i.04 to i64
%arrayidx = getelementptr inbounds i8, i8* %a, i64 %idxprom
store i8 0, i8* %arrayidx, align 1
%tmp0 = tail call { i32, i1 } @llvm.uadd.with.overflow.i32(i32 %i.04, i32 1)
%tmp1 = extractvalue { i32, i1 } %tmp0, 1
br i1 %tmp1, label %trap, label %cont, !nosanitize !{}
trap: ; preds = %for.body
tail call void @llvm.trap(), !nosanitize !{}
unreachable, !nosanitize !{}
cont: ; preds = %for.body
%tmp2 = extractvalue { i32, i1 } %tmp0, 0
%cmp = icmp slt i32 %tmp2, 16
br i1 %cmp, label %for.body, label %for.cond.cleanup
; CHECK: Loop %for.body: max backedge-taken count is 15
}
define void @f_ssub(i8* nocapture %a) {
; CHECK-LABEL: Classifying expressions for: @f_ssub
entry:
br label %for.body
for.cond.cleanup: ; preds = %cont
ret void
for.body: ; preds = %entry, %cont
; CHECK: %i.04 = phi i32 [ 15, %entry ], [ %tmp2, %cont ]
; CHECK-NEXT: --> {15,+,-1}<%for.body> U: [0,16) S: [0,16)
%i.04 = phi i32 [ 15, %entry ], [ %tmp2, %cont ]
%idxprom = sext i32 %i.04 to i64
%arrayidx = getelementptr inbounds i8, i8* %a, i64 %idxprom
store i8 0, i8* %arrayidx, align 1
%tmp0 = tail call { i32, i1 } @llvm.ssub.with.overflow.i32(i32 %i.04, i32 1)
%tmp1 = extractvalue { i32, i1 } %tmp0, 1
br i1 %tmp1, label %trap, label %cont, !nosanitize !{}
trap: ; preds = %for.body
tail call void @llvm.trap(), !nosanitize !{}
unreachable, !nosanitize !{}
cont: ; preds = %for.body
%tmp2 = extractvalue { i32, i1 } %tmp0, 0
%cmp = icmp sgt i32 %tmp2, -1
br i1 %cmp, label %for.body, label %for.cond.cleanup
; CHECK: Loop %for.body: max backedge-taken count is 15
}
define void @f_usub(i8* nocapture %a) {
; CHECK-LABEL: Classifying expressions for: @f_usub
entry:
br label %for.body
for.cond.cleanup: ; preds = %cont
ret void
for.body: ; preds = %entry, %cont
; CHECK: %i.04 = phi i32 [ 15, %entry ], [ %tmp2, %cont ]
; CHECK-NEXT: --> {15,+,-1}<%for.body> U: [0,16) S: [0,16)
%i.04 = phi i32 [ 15, %entry ], [ %tmp2, %cont ]
%idxprom = sext i32 %i.04 to i64
%arrayidx = getelementptr inbounds i8, i8* %a, i64 %idxprom
store i8 0, i8* %arrayidx, align 1
%tmp0 = tail call { i32, i1 } @llvm.usub.with.overflow.i32(i32 %i.04, i32 1)
%tmp1 = extractvalue { i32, i1 } %tmp0, 1
br i1 %tmp1, label %trap, label %cont, !nosanitize !{}
trap: ; preds = %for.body
tail call void @llvm.trap(), !nosanitize !{}
unreachable, !nosanitize !{}
cont: ; preds = %for.body
%tmp2 = extractvalue { i32, i1 } %tmp0, 0
%cmp = icmp sgt i32 %tmp2, -1
br i1 %cmp, label %for.body, label %for.cond.cleanup
; CHECK: Loop %for.body: max backedge-taken count is 15
}
define i32 @f_smul(i32 %val_a, i32 %val_b) {
; CHECK-LABEL: Classifying expressions for: @f_smul
%agg = tail call { i32, i1 } @llvm.smul.with.overflow.i32(i32 %val_a, i32 %val_b)
; CHECK: %mul = extractvalue { i32, i1 } %agg, 0
; CHECK-NEXT: --> (%val_a * %val_b) U: full-set S: full-set
%mul = extractvalue { i32, i1 } %agg, 0
ret i32 %mul
}
define i32 @f_umul(i32 %val_a, i32 %val_b) {
; CHECK-LABEL: Classifying expressions for: @f_umul
%agg = tail call { i32, i1 } @llvm.umul.with.overflow.i32(i32 %val_a, i32 %val_b)
; CHECK: %mul = extractvalue { i32, i1 } %agg, 0
; CHECK-NEXT: --> (%val_a * %val_b) U: full-set S: full-set
%mul = extractvalue { i32, i1 } %agg, 0
ret i32 %mul
}
declare { i32, i1 } @llvm.sadd.with.overflow.i32(i32, i32) nounwind readnone
declare { i32, i1 } @llvm.uadd.with.overflow.i32(i32, i32) nounwind readnone
declare { i32, i1 } @llvm.ssub.with.overflow.i32(i32, i32) nounwind readnone
declare { i32, i1 } @llvm.usub.with.overflow.i32(i32, i32) nounwind readnone
declare { i32, i1 } @llvm.smul.with.overflow.i32(i32, i32) nounwind readnone
declare { i32, i1 } @llvm.umul.with.overflow.i32(i32, i32) nounwind readnone
declare void @llvm.trap() #2

137
test/Transforms/IndVarSimplify/overflow-intrinsics.ll
View File

@ -1,137 +0,0 @@
; RUN: opt -S -indvars < %s | FileCheck %s
target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
target triple = "x86_64-unknown-linux-gnu"
define void @f_sadd(i8* %a) {
; CHECK-LABEL: @f_sadd(
entry:
br label %for.body
for.cond.cleanup: ; preds = %cont
ret void
for.body: ; preds = %entry, %cont
%i.04 = phi i32 [ 0, %entry ], [ %2, %cont ]
%idxprom = sext i32 %i.04 to i64
%arrayidx = getelementptr inbounds i8, i8* %a, i64 %idxprom
store i8 0, i8* %arrayidx, align 1
%0 = tail call { i32, i1 } @llvm.sadd.with.overflow.i32(i32 %i.04, i32 1)
%1 = extractvalue { i32, i1 } %0, 1
; CHECK: for.body:
; CHECK-NOT: @llvm.sadd.with.overflow
; CHECK: br i1 false, label %trap, label %cont, !nosanitize !0
br i1 %1, label %trap, label %cont, !nosanitize !{}
trap: ; preds = %for.body
tail call void @llvm.trap() #2, !nosanitize !{}
unreachable, !nosanitize !{}
cont: ; preds = %for.body
%2 = extractvalue { i32, i1 } %0, 0
%cmp = icmp slt i32 %2, 16
br i1 %cmp, label %for.body, label %for.cond.cleanup
}
define void @f_uadd(i8* %a) {
; CHECK-LABEL: @f_uadd(
entry:
br label %for.body
for.cond.cleanup: ; preds = %cont
ret void
for.body: ; preds = %entry, %cont
%i.04 = phi i32 [ 0, %entry ], [ %2, %cont ]
%idxprom = sext i32 %i.04 to i64
%arrayidx = getelementptr inbounds i8, i8* %a, i64 %idxprom
store i8 0, i8* %arrayidx, align 1
%0 = tail call { i32, i1 } @llvm.uadd.with.overflow.i32(i32 %i.04, i32 1)
%1 = extractvalue { i32, i1 } %0, 1
; CHECK: for.body:
; CHECK-NOT: @llvm.uadd.with.overflow
; CHECK: br i1 false, label %trap, label %cont, !nosanitize !0
br i1 %1, label %trap, label %cont, !nosanitize !{}
trap: ; preds = %for.body
tail call void @llvm.trap(), !nosanitize !{}
unreachable, !nosanitize !{}
cont: ; preds = %for.body
%2 = extractvalue { i32, i1 } %0, 0
%cmp = icmp slt i32 %2, 16
br i1 %cmp, label %for.body, label %for.cond.cleanup
}
define void @f_ssub(i8* nocapture %a) {
; CHECK-LABEL: @f_ssub(
entry:
br label %for.body
for.cond.cleanup: ; preds = %cont
ret void
for.body: ; preds = %entry, %cont
%i.04 = phi i32 [ 15, %entry ], [ %2, %cont ]
%idxprom = sext i32 %i.04 to i64
%arrayidx = getelementptr inbounds i8, i8* %a, i64 %idxprom
store i8 0, i8* %arrayidx, align 1
%0 = tail call { i32, i1 } @llvm.ssub.with.overflow.i32(i32 %i.04, i32 1)
%1 = extractvalue { i32, i1 } %0, 1
; CHECK: for.body:
; CHECK-NOT: @llvm.ssub.with.overflow.i32
; CHECK: br i1 false, label %trap, label %cont, !nosanitize !0
br i1 %1, label %trap, label %cont, !nosanitize !{}
trap: ; preds = %for.body
tail call void @llvm.trap(), !nosanitize !{}
unreachable, !nosanitize !{}
cont: ; preds = %for.body
%2 = extractvalue { i32, i1 } %0, 0
%cmp = icmp sgt i32 %2, -1
br i1 %cmp, label %for.body, label %for.cond.cleanup
}
define void @f_usub(i8* nocapture %a) {
; CHECK-LABEL: @f_usub(
entry:
br label %for.body
for.cond.cleanup: ; preds = %cont
ret void
for.body: ; preds = %entry, %cont
%i.04 = phi i32 [ 15, %entry ], [ %2, %cont ]
%idxprom = sext i32 %i.04 to i64
%arrayidx = getelementptr inbounds i8, i8* %a, i64 %idxprom
store i8 0, i8* %arrayidx, align 1
%0 = tail call { i32, i1 } @llvm.usub.with.overflow.i32(i32 %i.04, i32 1)
%1 = extractvalue { i32, i1 } %0, 1
; It is theoretically possible to prove this, but SCEV cannot
; represent non-unsigned-wrapping subtraction operations.
; CHECK: for.body:
; CHECK: [[COND:%[^ ]+]] = extractvalue { i32, i1 } %1, 1
; CHECK-NEXT: br i1 [[COND]], label %trap, label %cont, !nosanitize !0
br i1 %1, label %trap, label %cont, !nosanitize !{}
trap: ; preds = %for.body
tail call void @llvm.trap(), !nosanitize !{}
unreachable, !nosanitize !{}
cont: ; preds = %for.body
%2 = extractvalue { i32, i1 } %0, 0
%cmp = icmp sgt i32 %2, -1
br i1 %cmp, label %for.body, label %for.cond.cleanup
}
declare { i32, i1 } @llvm.sadd.with.overflow.i32(i32, i32) nounwind readnone
declare { i32, i1 } @llvm.uadd.with.overflow.i32(i32, i32) nounwind readnone
declare { i32, i1 } @llvm.ssub.with.overflow.i32(i32, i32) nounwind readnone
declare { i32, i1 } @llvm.usub.with.overflow.i32(i32, i32) nounwind readnone
declare { i32, i1 } @llvm.smul.with.overflow.i32(i32, i32) nounwind readnone
declare { i32, i1 } @llvm.umul.with.overflow.i32(i32, i32) nounwind readnone
declare void @llvm.trap() #2

2
test/Transforms/IndVarSimplify/overflowcheck.ll
View File

@ -10,7 +10,7 @@ target triple = "x86_64-apple-macosx"
; CHECK-LABEL: loop2:
; CHECK-NOT: extractvalue
; CHECK: add nuw
; CHECK-NOT: @llvm.sadd.with.overflow
; CHECK: @llvm.sadd.with.overflow
; CHECK-LABEL: loop3:
; CHECK-NOT: extractvalue
; CHECK: ret