aboutsummaryrefslogtreecommitdiff
path: root/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
diff options
context:
space:
mode:
authorSimon Pilgrim <llvm-dev@redking.me.uk>2018-02-15 12:14:15 +0000
committerSimon Pilgrim <llvm-dev@redking.me.uk>2018-02-15 12:14:15 +0000
commit80663ee986aa136a13d6c0af1d0a79a669432f50 (patch)
tree6945a98e22f22f081b66ffa432f85cf284733917 /llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
parent9430c8cd1c6fd35777604a2abc055075bc29d9b8 (diff)
downloadllvm-80663ee986aa136a13d6c0af1d0a79a669432f50.zip
llvm-80663ee986aa136a13d6c0af1d0a79a669432f50.tar.gz
llvm-80663ee986aa136a13d6c0af1d0a79a669432f50.tar.bz2
[SelectionDAG] Add initial implementation of TargetLowering::SimplifyDemandedVectorElts
This is mainly a move of simplifyShuffleOperands from DAGCombiner::visitVECTOR_SHUFFLE to create a more general purpose TargetLowering::SimplifyDemandedVectorElts implementation. Further features can be moved/added in future patches. Differential Revision: https://reviews.llvm.org/D42896 llvm-svn: 325232
Diffstat (limited to 'llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp')
-rw-r--r--llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp203
1 files changed, 203 insertions, 0 deletions
diff --git a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
index 79f7d16..a066217b 100644
--- a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -1279,6 +1279,197 @@ bool TargetLowering::SimplifyDemandedBits(SDValue Op,
return false;
}
+bool TargetLowering::SimplifyDemandedVectorElts(SDValue Op,
+ const APInt &DemandedElts,
+ APInt &KnownUndef,
+ APInt &KnownZero,
+ DAGCombinerInfo &DCI) const {
+ SelectionDAG &DAG = DCI.DAG;
+ TargetLoweringOpt TLO(DAG, !DCI.isBeforeLegalize(),
+ !DCI.isBeforeLegalizeOps());
+
+ bool Simplified =
+ SimplifyDemandedVectorElts(Op, DemandedElts, KnownUndef, KnownZero, TLO);
+ if (Simplified)
+ DCI.CommitTargetLoweringOpt(TLO);
+ return Simplified;
+}
+
+bool TargetLowering::SimplifyDemandedVectorElts(
+ SDValue Op, const APInt &DemandedEltMask, APInt &KnownUndef,
+ APInt &KnownZero, TargetLoweringOpt &TLO, unsigned Depth,
+ bool AssumeSingleUse) const {
+ EVT VT = Op.getValueType();
+ APInt DemandedElts = DemandedEltMask;
+ unsigned NumElts = DemandedElts.getBitWidth();
+ assert(VT.isVector() && "Expected vector op");
+ assert(VT.getVectorNumElements() == NumElts &&
+ "Mask size mismatches value type element count!");
+
+ KnownUndef = KnownZero = APInt::getNullValue(NumElts);
+
+ // Undef operand.
+ if (Op.isUndef()) {
+ KnownUndef.setAllBits();
+ return false;
+ }
+
+ // If Op has other users, assume that all elements are needed.
+ if (!Op.getNode()->hasOneUse() && !AssumeSingleUse)
+ DemandedElts.setAllBits();
+
+ // Not demanding any elements from Op.
+ if (DemandedElts == 0) {
+ KnownUndef.setAllBits();
+ return TLO.CombineTo(Op, TLO.DAG.getUNDEF(VT));
+ }
+
+ // Limit search depth.
+ if (Depth >= 6)
+ return false;
+
+ SDLoc DL(Op);
+ unsigned EltSizeInBits = VT.getScalarSizeInBits();
+
+ switch (Op.getOpcode()) {
+ case ISD::SCALAR_TO_VECTOR: {
+ if (!DemandedElts[0]) {
+ KnownUndef.setAllBits();
+ return TLO.CombineTo(Op, TLO.DAG.getUNDEF(VT));
+ }
+ KnownUndef.setHighBits(NumElts - 1);
+ break;
+ }
+ case ISD::BUILD_VECTOR: {
+ // Check all elements and simplify any unused elements with UNDEF.
+ if (!DemandedElts.isAllOnesValue()) {
+ // Don't simplify BROADCASTS.
+ if (llvm::any_of(Op->op_values(),
+ [&](SDValue Elt) { return Op.getOperand(0) != Elt; })) {
+ SmallVector<SDValue, 32> Ops(Op->op_begin(), Op->op_end());
+ bool Updated = false;
+ for (unsigned i = 0; i != NumElts; ++i) {
+ if (!DemandedElts[i] && !Ops[i].isUndef()) {
+ Ops[i] = TLO.DAG.getUNDEF(Ops[0].getValueType());
+ KnownUndef.setBit(i);
+ Updated = true;
+ }
+ }
+ if (Updated)
+ return TLO.CombineTo(Op, TLO.DAG.getBuildVector(VT, DL, Ops));
+ }
+ }
+ for (unsigned i = 0; i != NumElts; ++i) {
+ SDValue SrcOp = Op.getOperand(i);
+ if (SrcOp.isUndef()) {
+ KnownUndef.setBit(i);
+ } else if (EltSizeInBits == SrcOp.getScalarValueSizeInBits() &&
+ (isNullConstant(SrcOp) || isNullFPConstant(SrcOp))) {
+ KnownZero.setBit(i);
+ }
+ }
+ break;
+ }
+ case ISD::CONCAT_VECTORS: {
+ EVT SubVT = Op.getOperand(0).getValueType();
+ unsigned NumSubVecs = Op.getNumOperands();
+ unsigned NumSubElts = SubVT.getVectorNumElements();
+ for (unsigned i = 0; i != NumSubVecs; ++i) {
+ SDValue SubOp = Op.getOperand(i);
+ APInt SubElts = DemandedElts.extractBits(NumSubElts, i * NumSubElts);
+ APInt SubUndef, SubZero;
+ if (SimplifyDemandedVectorElts(SubOp, SubElts, SubUndef, SubZero, TLO,
+ Depth + 1))
+ return true;
+ KnownUndef.insertBits(SubUndef, i * NumSubElts);
+ KnownZero.insertBits(SubZero, i * NumSubElts);
+ }
+ break;
+ }
+ case ISD::INSERT_SUBVECTOR: {
+ if (!isa<ConstantSDNode>(Op.getOperand(2)))
+ break;
+ SDValue Base = Op.getOperand(0);
+ SDValue Sub = Op.getOperand(1);
+ EVT SubVT = Sub.getValueType();
+ unsigned NumSubElts = SubVT.getVectorNumElements();
+ APInt Idx = cast<ConstantSDNode>(Op.getOperand(2))->getAPIntValue();
+ if (Idx.uge(NumElts - NumSubElts))
+ break;
+ unsigned SubIdx = Idx.getZExtValue();
+ APInt SubElts = DemandedElts.extractBits(NumSubElts, SubIdx);
+ APInt SubUndef, SubZero;
+ if (SimplifyDemandedVectorElts(Sub, SubElts, SubUndef, SubZero, TLO,
+ Depth + 1))
+ return true;
+ APInt BaseElts = DemandedElts;
+ BaseElts.insertBits(APInt::getNullValue(NumSubElts), SubIdx);
+ if (SimplifyDemandedVectorElts(Base, BaseElts, KnownUndef, KnownZero, TLO,
+ Depth + 1))
+ return true;
+ KnownUndef.insertBits(SubUndef, SubIdx);
+ KnownZero.insertBits(SubZero, SubIdx);
+ break;
+ }
+ case ISD::VECTOR_SHUFFLE: {
+ ArrayRef<int> ShuffleMask = cast<ShuffleVectorSDNode>(Op)->getMask();
+
+ // Collect demanded elements from shuffle operands..
+ APInt DemandedLHS(NumElts, 0);
+ APInt DemandedRHS(NumElts, 0);
+ for (unsigned i = 0; i != NumElts; ++i) {
+ int M = ShuffleMask[i];
+ if (M < 0 || !DemandedElts[i])
+ continue;
+ assert(0 <= M && M < (int)(2 * NumElts) && "Shuffle index out of range");
+ if (M < (int)NumElts)
+ DemandedLHS.setBit(M);
+ else
+ DemandedRHS.setBit(M - NumElts);
+ }
+
+ // See if we can simplify either shuffle operand.
+ APInt UndefLHS, ZeroLHS;
+ APInt UndefRHS, ZeroRHS;
+ if (SimplifyDemandedVectorElts(Op.getOperand(0), DemandedLHS, UndefLHS,
+ ZeroLHS, TLO, Depth + 1))
+ return true;
+ if (SimplifyDemandedVectorElts(Op.getOperand(1), DemandedRHS, UndefRHS,
+ ZeroRHS, TLO, Depth + 1))
+ return true;
+
+ // Propagate undef/zero elements from LHS/RHS.
+ for (unsigned i = 0; i != NumElts; ++i) {
+ int M = ShuffleMask[i];
+ if (M < 0) {
+ KnownUndef.setBit(i);
+ } else if (M < (int)NumElts) {
+ if (UndefLHS[M])
+ KnownUndef.setBit(i);
+ if (ZeroLHS[M])
+ KnownZero.setBit(i);
+ } else {
+ if (UndefRHS[M - NumElts])
+ KnownUndef.setBit(i);
+ if (ZeroRHS[M - NumElts])
+ KnownZero.setBit(i);
+ }
+ }
+ break;
+ }
+ default: {
+ if (Op.getOpcode() >= ISD::BUILTIN_OP_END)
+ if (SimplifyDemandedVectorEltsForTargetNode(Op, DemandedElts, KnownUndef,
+ KnownZero, TLO, Depth))
+ return true;
+ break;
+ }
+ }
+
+ assert((KnownUndef & KnownZero) == 0 && "Elements flagged as undef AND zero");
+ return false;
+}
+
/// Determine which of the bits specified in Mask are known to be either zero or
/// one and return them in the Known.
void TargetLowering::computeKnownBitsForTargetNode(const SDValue Op,
@@ -1323,6 +1514,18 @@ unsigned TargetLowering::ComputeNumSignBitsForTargetNode(SDValue Op,
return 1;
}
+bool TargetLowering::SimplifyDemandedVectorEltsForTargetNode(
+ SDValue Op, const APInt &DemandedElts, APInt &KnownUndef, APInt &KnownZero,
+ TargetLoweringOpt &TLO, unsigned Depth) const {
+ assert((Op.getOpcode() >= ISD::BUILTIN_OP_END ||
+ Op.getOpcode() == ISD::INTRINSIC_WO_CHAIN ||
+ Op.getOpcode() == ISD::INTRINSIC_W_CHAIN ||
+ Op.getOpcode() == ISD::INTRINSIC_VOID) &&
+ "Should use SimplifyDemandedVectorElts if you don't know whether Op"
+ " is a target node!");
+ return false;
+}
+
// FIXME: Ideally, this would use ISD::isConstantSplatVector(), but that must
// work with truncating build vectors and vectors with elements of less than
// 8 bits.