[DAG] Expose NoSignedWrap, NoUnsignedWrap and Exact flags to SelectionDAG.

This patch modifies SelectionDAGBuilder to construct SDNodes with associated
NoSignedWrap, NoUnsignedWrap and Exact flags coming from IR BinaryOperator
instructions.

Added a new SDNode type called 'BinaryWithFlagsSDNode' to allow accessing
nsw/nuw/exact flags during codegen.

Patch by Marcello Maggioni.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@210467 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Andrea Di Biagio 2014-06-09 12:32:53 +00:00
parent a2bc6951a0
commit bc72c8f0d8
7 changed files with 192 additions and 26 deletions

View File

@ -612,14 +612,14 @@ public:
/// ///
SDValue getNode(unsigned Opcode, SDLoc DL, EVT VT); SDValue getNode(unsigned Opcode, SDLoc DL, EVT VT);
SDValue getNode(unsigned Opcode, SDLoc DL, EVT VT, SDValue N); SDValue getNode(unsigned Opcode, SDLoc DL, EVT VT, SDValue N);
SDValue getNode(unsigned Opcode, SDLoc DL, EVT VT, SDValue N1, SDValue N2); SDValue getNode(unsigned Opcode, SDLoc DL, EVT VT, SDValue N1, SDValue N2,
SDValue getNode(unsigned Opcode, SDLoc DL, EVT VT, bool nuw = false, bool nsw = false, bool exact = false);
SDValue N1, SDValue N2, SDValue N3); SDValue getNode(unsigned Opcode, SDLoc DL, EVT VT, SDValue N1, SDValue N2,
SDValue getNode(unsigned Opcode, SDLoc DL, EVT VT, SDValue N3);
SDValue N1, SDValue N2, SDValue N3, SDValue N4); SDValue getNode(unsigned Opcode, SDLoc DL, EVT VT, SDValue N1, SDValue N2,
SDValue getNode(unsigned Opcode, SDLoc DL, EVT VT, SDValue N3, SDValue N4);
SDValue N1, SDValue N2, SDValue N3, SDValue N4, SDValue getNode(unsigned Opcode, SDLoc DL, EVT VT, SDValue N1, SDValue N2,
SDValue N5); SDValue N3, SDValue N4, SDValue N5);
SDValue getNode(unsigned Opcode, SDLoc DL, EVT VT, ArrayRef<SDUse> Ops); SDValue getNode(unsigned Opcode, SDLoc DL, EVT VT, ArrayRef<SDUse> Ops);
SDValue getNode(unsigned Opcode, SDLoc DL, EVT VT, SDValue getNode(unsigned Opcode, SDLoc DL, EVT VT,
ArrayRef<SDValue> Ops); ArrayRef<SDValue> Ops);
@ -927,7 +927,9 @@ public:
/// getNodeIfExists - Get the specified node if it's already available, or /// getNodeIfExists - Get the specified node if it's already available, or
/// else return NULL. /// else return NULL.
SDNode *getNodeIfExists(unsigned Opcode, SDVTList VTs, ArrayRef<SDValue> Ops); SDNode *getNodeIfExists(unsigned Opcode, SDVTList VTs, ArrayRef<SDValue> Ops,
bool nuw = false, bool nsw = false,
bool exact = false);
/// getDbgValue - Creates a SDDbgValue node. /// getDbgValue - Creates a SDDbgValue node.
/// ///
@ -1184,6 +1186,10 @@ private:
void allnodes_clear(); void allnodes_clear();
BinarySDNode *GetBinarySDNode(unsigned Opcode, SDLoc DL, SDVTList VTs,
SDValue N1, SDValue N2, bool nuw, bool nsw,
bool exact);
/// VTList - List of non-single value types. /// VTList - List of non-single value types.
FoldingSet<SDVTListNode> VTListMap; FoldingSet<SDVTListNode> VTListMap;

View File

@ -49,6 +49,24 @@ template <typename T> struct DenseMapInfo;
template <typename T> struct simplify_type; template <typename T> struct simplify_type;
template <typename T> struct ilist_traits; template <typename T> struct ilist_traits;
/// isBinOpWithFlags - Returns true if the opcode is a binary operation
/// with flags.
static bool isBinOpWithFlags(unsigned Opcode) {
switch (Opcode) {
case ISD::SDIV:
case ISD::UDIV:
case ISD::SRA:
case ISD::SRL:
case ISD::MUL:
case ISD::ADD:
case ISD::SUB:
case ISD::SHL:
return true;
default:
return false;
}
}
void checkForCycles(const SDNode *N, const SelectionDAG *DAG = nullptr, void checkForCycles(const SDNode *N, const SelectionDAG *DAG = nullptr,
bool force = false); bool force = false);
@ -939,6 +957,36 @@ public:
} }
}; };
/// BinaryWithFlagsSDNode - This class is an extension of BinarySDNode
/// used from those opcodes that have associated extra flags.
class BinaryWithFlagsSDNode : public BinarySDNode {
enum { NUW = (1 << 0), NSW = (1 << 1), EXACT = (1 << 2) };
public:
BinaryWithFlagsSDNode(unsigned Opc, unsigned Order, DebugLoc dl, SDVTList VTs,
SDValue X, SDValue Y)
: BinarySDNode(Opc, Order, dl, VTs, X, Y) {}
/// getRawSubclassData - Return the SubclassData value, which contains an
/// encoding of the flags.
/// This function should be used to add subclass data to the NodeID value.
unsigned getRawSubclassData() const { return SubclassData; }
void setHasNoUnsignedWrap(bool b) {
SubclassData = (SubclassData & ~NUW) | (b ? NUW : 0);
}
void setHasNoSignedWrap(bool b) {
SubclassData = (SubclassData & ~NSW) | (b ? NSW : 0);
}
void setIsExact(bool b) {
SubclassData = (SubclassData & ~EXACT) | (b ? EXACT : 0);
}
bool hasNoUnsignedWrap() const { return SubclassData & NUW; }
bool hasNoSignedWrap() const { return SubclassData & NSW; }
bool isExact() const { return SubclassData & EXACT; }
static bool classof(const SDNode *N) {
return isBinOpWithFlags(N->getOpcode());
}
};
/// TernarySDNode - This class is used for three-operand SDNodes. This is solely /// TernarySDNode - This class is used for three-operand SDNodes. This is solely
/// to allow co-allocation of node operands with the node itself. /// to allow co-allocation of node operands with the node itself.
class TernarySDNode : public SDNode { class TernarySDNode : public SDNode {

View File

@ -1325,9 +1325,16 @@ SDValue DAGCombiner::combine(SDNode *N) {
// Constant operands are canonicalized to RHS. // Constant operands are canonicalized to RHS.
if (isa<ConstantSDNode>(N0) || !isa<ConstantSDNode>(N1)) { if (isa<ConstantSDNode>(N0) || !isa<ConstantSDNode>(N1)) {
SDValue Ops[] = { N1, N0 }; SDValue Ops[] = {N1, N0};
SDNode *CSENode = DAG.getNodeIfExists(N->getOpcode(), N->getVTList(), SDNode *CSENode;
Ops); if (const BinaryWithFlagsSDNode *BinNode =
dyn_cast<BinaryWithFlagsSDNode>(N)) {
CSENode = DAG.getNodeIfExists(
N->getOpcode(), N->getVTList(), Ops, BinNode->hasNoUnsignedWrap(),
BinNode->hasNoSignedWrap(), BinNode->isExact());
} else {
CSENode = DAG.getNodeIfExists(N->getOpcode(), N->getVTList(), Ops);
}
if (CSENode) if (CSENode)
return SDValue(CSENode, 0); return SDValue(CSENode, 0);
} }

View File

@ -48,6 +48,7 @@
#include "llvm/Target/TargetSelectionDAGInfo.h" #include "llvm/Target/TargetSelectionDAGInfo.h"
#include <algorithm> #include <algorithm>
#include <cmath> #include <cmath>
using namespace llvm; using namespace llvm;
/// makeVTList - Return an instance of the SDVTList struct initialized with the /// makeVTList - Return an instance of the SDVTList struct initialized with the
@ -381,6 +382,20 @@ static void AddNodeIDOperands(FoldingSetNodeID &ID,
} }
} }
static void AddBinaryNodeIDCustom(FoldingSetNodeID &ID, bool nuw, bool nsw,
bool exact) {
ID.AddBoolean(nuw);
ID.AddBoolean(nsw);
ID.AddBoolean(exact);
}
/// AddBinaryNodeIDCustom - Add BinarySDNodes special infos
static void AddBinaryNodeIDCustom(FoldingSetNodeID &ID, unsigned Opcode,
bool nuw, bool nsw, bool exact) {
if (isBinOpWithFlags(Opcode))
AddBinaryNodeIDCustom(ID, nuw, nsw, exact);
}
static void AddNodeIDNode(FoldingSetNodeID &ID, unsigned short OpC, static void AddNodeIDNode(FoldingSetNodeID &ID, unsigned short OpC,
SDVTList VTList, ArrayRef<SDValue> OpList) { SDVTList VTList, ArrayRef<SDValue> OpList) {
AddNodeIDOpcode(ID, OpC); AddNodeIDOpcode(ID, OpC);
@ -473,6 +488,19 @@ static void AddNodeIDCustom(FoldingSetNodeID &ID, const SDNode *N) {
ID.AddInteger(ST->getPointerInfo().getAddrSpace()); ID.AddInteger(ST->getPointerInfo().getAddrSpace());
break; break;
} }
case ISD::SDIV:
case ISD::UDIV:
case ISD::SRA:
case ISD::SRL:
case ISD::MUL:
case ISD::ADD:
case ISD::SUB:
case ISD::SHL: {
const BinaryWithFlagsSDNode *BinNode = cast<BinaryWithFlagsSDNode>(N);
AddBinaryNodeIDCustom(ID, N->getOpcode(), BinNode->hasNoUnsignedWrap(),
BinNode->hasNoSignedWrap(), BinNode->isExact());
break;
}
case ISD::ATOMIC_CMP_SWAP: case ISD::ATOMIC_CMP_SWAP:
case ISD::ATOMIC_SWAP: case ISD::ATOMIC_SWAP:
case ISD::ATOMIC_LOAD_ADD: case ISD::ATOMIC_LOAD_ADD:
@ -926,6 +954,25 @@ void SelectionDAG::allnodes_clear() {
DeallocateNode(AllNodes.begin()); DeallocateNode(AllNodes.begin());
} }
BinarySDNode *SelectionDAG::GetBinarySDNode(unsigned Opcode, SDLoc DL,
SDVTList VTs, SDValue N1,
SDValue N2, bool nuw, bool nsw,
bool exact) {
if (isBinOpWithFlags(Opcode)) {
BinaryWithFlagsSDNode *FN = new (NodeAllocator) BinaryWithFlagsSDNode(
Opcode, DL.getIROrder(), DL.getDebugLoc(), VTs, N1, N2);
FN->setHasNoUnsignedWrap(nuw);
FN->setHasNoSignedWrap(nsw);
FN->setIsExact(exact);
return FN;
}
BinarySDNode *N = new (NodeAllocator)
BinarySDNode(Opcode, DL.getIROrder(), DL.getDebugLoc(), VTs, N1, N2);
return N;
}
void SelectionDAG::clear() { void SelectionDAG::clear() {
allnodes_clear(); allnodes_clear();
OperandAllocator.Reset(); OperandAllocator.Reset();
@ -2936,7 +2983,7 @@ SDValue SelectionDAG::FoldConstantArithmetic(unsigned Opcode, EVT VT,
} }
SDValue SelectionDAG::getNode(unsigned Opcode, SDLoc DL, EVT VT, SDValue N1, SDValue SelectionDAG::getNode(unsigned Opcode, SDLoc DL, EVT VT, SDValue N1,
SDValue N2) { SDValue N2, bool nuw, bool nsw, bool exact) {
ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1.getNode()); ConstantSDNode *N1C = dyn_cast<ConstantSDNode>(N1.getNode());
ConstantSDNode *N2C = dyn_cast<ConstantSDNode>(N2.getNode()); ConstantSDNode *N2C = dyn_cast<ConstantSDNode>(N2.getNode());
switch (Opcode) { switch (Opcode) {
@ -3376,22 +3423,25 @@ SDValue SelectionDAG::getNode(unsigned Opcode, SDLoc DL, EVT VT, SDValue N1,
} }
// Memoize this node if possible. // Memoize this node if possible.
SDNode *N; BinarySDNode *N;
SDVTList VTs = getVTList(VT); SDVTList VTs = getVTList(VT);
const bool BinOpHasFlags = isBinOpWithFlags(Opcode);
if (VT != MVT::Glue) { if (VT != MVT::Glue) {
SDValue Ops[] = { N1, N2 }; SDValue Ops[] = {N1, N2};
FoldingSetNodeID ID; FoldingSetNodeID ID;
AddNodeIDNode(ID, Opcode, VTs, Ops); AddNodeIDNode(ID, Opcode, VTs, Ops);
if (BinOpHasFlags)
AddBinaryNodeIDCustom(ID, Opcode, nuw, nsw, exact);
void *IP = nullptr; void *IP = nullptr;
if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP)) if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
return SDValue(E, 0); return SDValue(E, 0);
N = new (NodeAllocator) BinarySDNode(Opcode, DL.getIROrder(), N = GetBinarySDNode(Opcode, DL, VTs, N1, N2, nuw, nsw, exact);
DL.getDebugLoc(), VTs, N1, N2);
CSEMap.InsertNode(N, IP); CSEMap.InsertNode(N, IP);
} else { } else {
N = new (NodeAllocator) BinarySDNode(Opcode, DL.getIROrder(),
DL.getDebugLoc(), VTs, N1, N2); N = GetBinarySDNode(Opcode, DL, VTs, N1, N2, nuw, nsw, exact);
} }
AllNodes.push_back(N); AllNodes.push_back(N);
@ -5606,10 +5656,13 @@ SelectionDAG::getTargetInsertSubreg(int SRIdx, SDLoc DL, EVT VT,
/// getNodeIfExists - Get the specified node if it's already available, or /// getNodeIfExists - Get the specified node if it's already available, or
/// else return NULL. /// else return NULL.
SDNode *SelectionDAG::getNodeIfExists(unsigned Opcode, SDVTList VTList, SDNode *SelectionDAG::getNodeIfExists(unsigned Opcode, SDVTList VTList,
ArrayRef<SDValue> Ops) { ArrayRef<SDValue> Ops, bool nuw, bool nsw,
if (VTList.VTs[VTList.NumVTs-1] != MVT::Glue) { bool exact) {
if (VTList.VTs[VTList.NumVTs - 1] != MVT::Glue) {
FoldingSetNodeID ID; FoldingSetNodeID ID;
AddNodeIDNode(ID, Opcode, VTList, Ops); AddNodeIDNode(ID, Opcode, VTList, Ops);
if (isBinOpWithFlags(Opcode))
AddBinaryNodeIDCustom(ID, nuw, nsw, exact);
void *IP = nullptr; void *IP = nullptr;
if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP)) if (SDNode *E = CSEMap.FindNodeOrInsertPos(ID, IP))
return E; return E;

View File

@ -2784,8 +2784,22 @@ void SelectionDAGBuilder::visitFSub(const User &I) {
void SelectionDAGBuilder::visitBinary(const User &I, unsigned OpCode) { void SelectionDAGBuilder::visitBinary(const User &I, unsigned OpCode) {
SDValue Op1 = getValue(I.getOperand(0)); SDValue Op1 = getValue(I.getOperand(0));
SDValue Op2 = getValue(I.getOperand(1)); SDValue Op2 = getValue(I.getOperand(1));
setValue(&I, DAG.getNode(OpCode, getCurSDLoc(),
Op1.getValueType(), Op1, Op2)); bool nuw = false;
bool nsw = false;
bool exact = false;
if (const OverflowingBinaryOperator *OFBinOp =
dyn_cast<const OverflowingBinaryOperator>(&I)) {
nuw = OFBinOp->hasNoUnsignedWrap();
nsw = OFBinOp->hasNoSignedWrap();
}
if (const PossiblyExactOperator *ExactOp =
dyn_cast<const PossiblyExactOperator>(&I))
exact = ExactOp->isExact();
SDValue BinNodeValue = DAG.getNode(OpCode, getCurSDLoc(), Op1.getValueType(),
Op1, Op2, nuw, nsw, exact);
setValue(&I, BinNodeValue);
} }
void SelectionDAGBuilder::visitShift(const User &I, unsigned Opcode) { void SelectionDAGBuilder::visitShift(const User &I, unsigned Opcode) {
@ -2816,8 +2830,25 @@ void SelectionDAGBuilder::visitShift(const User &I, unsigned Opcode) {
Op2 = DAG.getZExtOrTrunc(Op2, DL, MVT::i32); Op2 = DAG.getZExtOrTrunc(Op2, DL, MVT::i32);
} }
setValue(&I, DAG.getNode(Opcode, getCurSDLoc(), bool nuw = false;
Op1.getValueType(), Op1, Op2)); bool nsw = false;
bool exact = false;
if (Opcode == ISD::SRL || Opcode == ISD::SRA || Opcode == ISD::SHL) {
if (const OverflowingBinaryOperator *OFBinOp =
dyn_cast<const OverflowingBinaryOperator>(&I)) {
nuw = OFBinOp->hasNoUnsignedWrap();
nsw = OFBinOp->hasNoSignedWrap();
}
if (const PossiblyExactOperator *ExactOp =
dyn_cast<const PossiblyExactOperator>(&I))
exact = ExactOp->isExact();
}
SDValue Res = DAG.getNode(Opcode, getCurSDLoc(), Op1.getValueType(), Op1, Op2,
nuw, nsw, exact);
setValue(&I, Res);
} }
void SelectionDAGBuilder::visitSDiv(const User &I) { void SelectionDAGBuilder::visitSDiv(const User &I) {

View File

@ -2617,7 +2617,8 @@ SDValue TargetLowering::BuildExactSDIV(SDValue Op1, SDValue Op2, SDLoc dl,
if (ShAmt) { if (ShAmt) {
// TODO: For UDIV use SRL instead of SRA. // TODO: For UDIV use SRL instead of SRA.
SDValue Amt = DAG.getConstant(ShAmt, getShiftAmountTy(Op1.getValueType())); SDValue Amt = DAG.getConstant(ShAmt, getShiftAmountTy(Op1.getValueType()));
Op1 = DAG.getNode(ISD::SRA, dl, Op1.getValueType(), Op1, Amt); Op1 = DAG.getNode(ISD::SRA, dl, Op1.getValueType(), Op1, Amt, false, false,
true);
d = d.ashr(ShAmt); d = d.ashr(ShAmt);
} }

View File

@ -0,0 +1,20 @@
; RUN: llc < %s -march=x86-64 | FileCheck %s
; CHECK: addl
; The two additions are the same , but have different flags.
; In theory this code should never be generated by the frontend, but this
; tries to test that two identical instructions with two different flags
; actually generate two different nodes.
;
; Normally the combiner would see this condition without the flags
; and optimize the result of the sub into a register clear
; (the final result would be 0). With the different flags though the combiner
; needs to keep the add + sub nodes, because the two nodes result as different
; nodes and so cannot assume that the subtraction of the two nodes
; generates 0 as result
define i32 @foo(i32 %a, i32 %b) {
%1 = add i32 %a, %b
%2 = add nsw i32 %a, %b
%3 = sub i32 %1, %2
ret i32 %3
}