summaryrefslogtreecommitdiffstats
path: root/llvm/lib/CodeGen
diff options
context:
space:
mode:
authorDavid Green <david.green@arm.com>2019-11-21 14:56:37 +0000
committerDavid Green <david.green@arm.com>2019-11-26 16:21:01 +0000
commitb5315ae8ffa6fb8befdd558d0dfd04295dbc5523 (patch)
tree410cc867173b5eef25c6b4f0d75b2f5813457304 /llvm/lib/CodeGen
parent549db744bde29c8331411a4b41607a33c363c108 (diff)
downloadbcm5719-llvm-b5315ae8ffa6fb8befdd558d0dfd04295dbc5523.tar.gz
bcm5719-llvm-b5315ae8ffa6fb8befdd558d0dfd04295dbc5523.zip
[Codegen][ARM] Add addressing modes from masked loads and stores
MVE has a basic symmetry between it's normal loads/store operations and the masked variants. This means that masked loads and stores can use pre-inc and post-inc addressing modes, just like the standard loads and stores already do. To enable that, this patch adds all the relevant infrastructure for treating masked loads/stores addressing modes in the same way as normal loads/stores. This involves: - Adding an AddressingMode to MaskedLoadStoreSDNode, along with an extra Offset operand that is added after the PtrBase. - Extending the IndexedModeActions from 8bits to 16bits to store the legality of masked operations as well as normal ones. This array is fairly small, so doubling the size still won't make it very large. Offset masked loads can then be controlled with setIndexedMaskedLoadAction, similar to standard loads. - The same methods that combine to indexed loads, such as CombineToPostIndexedLoadStore, are adjusted to handle masked loads in the same way. - The ARM backend is then adjusted to make use of these indexed masked loads/stores. - The X86 backend is adjusted to hopefully be no functional changes. Differential Revision: https://reviews.llvm.org/D70176
Diffstat (limited to 'llvm/lib/CodeGen')
-rw-r--r--llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp177
-rw-r--r--llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp16
-rw-r--r--llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp36
-rw-r--r--llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp63
-rw-r--r--llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp13
-rw-r--r--llvm/lib/CodeGen/SelectionDAG/SelectionDAGDumper.cpp8
-rw-r--r--llvm/lib/CodeGen/TargetLoweringBase.cpp2
7 files changed, 208 insertions, 107 deletions
diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index 793352c16d3..e6844e556b1 100644
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -8724,6 +8724,10 @@ SDValue DAGCombiner::visitMSTORE(SDNode *N) {
if (ISD::isBuildVectorAllZeros(Mask.getNode()))
return Chain;
+ // Try transforming N to an indexed store.
+ if (CombineToPreIndexedLoadStore(N) || CombineToPostIndexedLoadStore(N))
+ return SDValue(N, 0);
+
return SDValue();
}
@@ -8748,6 +8752,10 @@ SDValue DAGCombiner::visitMLOAD(SDNode *N) {
if (ISD::isBuildVectorAllZeros(Mask.getNode()))
return CombineTo(N, MLD->getPassThru(), MLD->getChain());
+ // Try transforming N to an indexed load.
+ if (CombineToPreIndexedLoadStore(N) || CombineToPostIndexedLoadStore(N))
+ return SDValue(N, 0);
+
return SDValue();
}
@@ -9506,11 +9514,10 @@ static SDValue tryToFoldExtOfMaskedLoad(SelectionDAG &DAG,
SDLoc dl(Ld);
SDValue PassThru = DAG.getNode(ExtOpc, dl, VT, Ld->getPassThru());
- SDValue NewLoad = DAG.getMaskedLoad(VT, dl, Ld->getChain(),
- Ld->getBasePtr(), Ld->getMask(),
- PassThru, Ld->getMemoryVT(),
- Ld->getMemOperand(), ExtLoadType,
- Ld->isExpandingLoad());
+ SDValue NewLoad = DAG.getMaskedLoad(
+ VT, dl, Ld->getChain(), Ld->getBasePtr(), Ld->getOffset(), Ld->getMask(),
+ PassThru, Ld->getMemoryVT(), Ld->getMemOperand(), Ld->getAddressingMode(),
+ ExtLoadType, Ld->isExpandingLoad());
DAG.ReplaceAllUsesOfValueWith(SDValue(Ld, 1), SDValue(NewLoad.getNode(), 1));
return NewLoad;
}
@@ -13612,12 +13619,22 @@ static bool canFoldInAddressingMode(SDNode *N, SDNode *Use,
EVT VT;
unsigned AS;
- if (LoadSDNode *LD = dyn_cast<LoadSDNode>(Use)) {
+ if (LoadSDNode *LD = dyn_cast<LoadSDNode>(Use)) {
if (LD->isIndexed() || LD->getBasePtr().getNode() != N)
return false;
VT = LD->getMemoryVT();
AS = LD->getAddressSpace();
- } else if (StoreSDNode *ST = dyn_cast<StoreSDNode>(Use)) {
+ } else if (StoreSDNode *ST = dyn_cast<StoreSDNode>(Use)) {
+ if (ST->isIndexed() || ST->getBasePtr().getNode() != N)
+ return false;
+ VT = ST->getMemoryVT();
+ AS = ST->getAddressSpace();
+ } else if (MaskedLoadSDNode *LD = dyn_cast<MaskedLoadSDNode>(Use)) {
+ if (LD->isIndexed() || LD->getBasePtr().getNode() != N)
+ return false;
+ VT = LD->getMemoryVT();
+ AS = LD->getAddressSpace();
+ } else if (MaskedStoreSDNode *ST = dyn_cast<MaskedStoreSDNode>(Use)) {
if (ST->isIndexed() || ST->getBasePtr().getNode() != N)
return false;
VT = ST->getMemoryVT();
@@ -13651,38 +13668,64 @@ static bool canFoldInAddressingMode(SDNode *N, SDNode *Use,
VT.getTypeForEVT(*DAG.getContext()), AS);
}
-/// Try turning a load/store into a pre-indexed load/store when the base
-/// pointer is an add or subtract and it has other uses besides the load/store.
-/// After the transformation, the new indexed load/store has effectively folded
-/// the add/subtract in and all of its other uses are redirected to the
-/// new load/store.
-bool DAGCombiner::CombineToPreIndexedLoadStore(SDNode *N) {
- if (Level < AfterLegalizeDAG)
- return false;
-
- bool isLoad = true;
- SDValue Ptr;
- EVT VT;
- if (LoadSDNode *LD = dyn_cast<LoadSDNode>(N)) {
+static bool getCombineLoadStoreParts(SDNode *N, unsigned Inc, unsigned Dec,
+ bool &IsLoad, bool &IsMasked, SDValue &Ptr,
+ const TargetLowering &TLI) {
+ if (LoadSDNode *LD = dyn_cast<LoadSDNode>(N)) {
if (LD->isIndexed())
return false;
- VT = LD->getMemoryVT();
- if (!TLI.isIndexedLoadLegal(ISD::PRE_INC, VT) &&
- !TLI.isIndexedLoadLegal(ISD::PRE_DEC, VT))
+ EVT VT = LD->getMemoryVT();
+ if (!TLI.isIndexedLoadLegal(Inc, VT) && !TLI.isIndexedLoadLegal(Dec, VT))
return false;
Ptr = LD->getBasePtr();
- } else if (StoreSDNode *ST = dyn_cast<StoreSDNode>(N)) {
+ } else if (StoreSDNode *ST = dyn_cast<StoreSDNode>(N)) {
if (ST->isIndexed())
return false;
- VT = ST->getMemoryVT();
- if (!TLI.isIndexedStoreLegal(ISD::PRE_INC, VT) &&
- !TLI.isIndexedStoreLegal(ISD::PRE_DEC, VT))
+ EVT VT = ST->getMemoryVT();
+ if (!TLI.isIndexedStoreLegal(Inc, VT) && !TLI.isIndexedStoreLegal(Dec, VT))
return false;
Ptr = ST->getBasePtr();
- isLoad = false;
+ IsLoad = false;
+ } else if (MaskedLoadSDNode *LD = dyn_cast<MaskedLoadSDNode>(N)) {
+ if (LD->isIndexed())
+ return false;
+ EVT VT = LD->getMemoryVT();
+ if (!TLI.isIndexedMaskedLoadLegal(Inc, VT) &&
+ !TLI.isIndexedMaskedLoadLegal(Dec, VT))
+ return false;
+ Ptr = LD->getBasePtr();
+ IsMasked = true;
+ } else if (MaskedStoreSDNode *ST = dyn_cast<MaskedStoreSDNode>(N)) {
+ if (ST->isIndexed())
+ return false;
+ EVT VT = ST->getMemoryVT();
+ if (!TLI.isIndexedMaskedStoreLegal(Inc, VT) &&
+ !TLI.isIndexedMaskedStoreLegal(Dec, VT))
+ return false;
+ Ptr = ST->getBasePtr();
+ IsLoad = false;
+ IsMasked = true;
} else {
return false;
}
+ return true;
+}
+
+/// Try turning a load/store into a pre-indexed load/store when the base
+/// pointer is an add or subtract and it has other uses besides the load/store.
+/// After the transformation, the new indexed load/store has effectively folded
+/// the add/subtract in and all of its other uses are redirected to the
+/// new load/store.
+bool DAGCombiner::CombineToPreIndexedLoadStore(SDNode *N) {
+ if (Level < AfterLegalizeDAG)
+ return false;
+
+ bool IsLoad = true;
+ bool IsMasked = false;
+ SDValue Ptr;
+ if (!getCombineLoadStoreParts(N, ISD::PRE_INC, ISD::PRE_DEC, IsLoad, IsMasked,
+ Ptr, TLI))
+ return false;
// If the pointer is not an add/sub, or if it doesn't have multiple uses, bail
// out. There is no reason to make this a preinc/predec.
@@ -13724,8 +13767,9 @@ bool DAGCombiner::CombineToPreIndexedLoadStore(SDNode *N) {
return false;
// Check #2.
- if (!isLoad) {
- SDValue Val = cast<StoreSDNode>(N)->getValue();
+ if (!IsLoad) {
+ SDValue Val = IsMasked ? cast<MaskedStoreSDNode>(N)->getValue()
+ : cast<StoreSDNode>(N)->getValue();
// Would require a copy.
if (Val == BasePtr)
@@ -13801,18 +13845,26 @@ bool DAGCombiner::CombineToPreIndexedLoadStore(SDNode *N) {
return false;
SDValue Result;
- if (isLoad)
- Result = DAG.getIndexedLoad(SDValue(N,0), SDLoc(N),
- BasePtr, Offset, AM);
- else
- Result = DAG.getIndexedStore(SDValue(N,0), SDLoc(N),
- BasePtr, Offset, AM);
+ if (!IsMasked) {
+ if (IsLoad)
+ Result = DAG.getIndexedLoad(SDValue(N, 0), SDLoc(N), BasePtr, Offset, AM);
+ else
+ Result =
+ DAG.getIndexedStore(SDValue(N, 0), SDLoc(N), BasePtr, Offset, AM);
+ } else {
+ if (IsLoad)
+ Result = DAG.getIndexedMaskedLoad(SDValue(N, 0), SDLoc(N), BasePtr,
+ Offset, AM);
+ else
+ Result = DAG.getIndexedMaskedStore(SDValue(N, 0), SDLoc(N), BasePtr,
+ Offset, AM);
+ }
++PreIndexedNodes;
++NodesCombined;
LLVM_DEBUG(dbgs() << "\nReplacing.4 "; N->dump(&DAG); dbgs() << "\nWith: ";
Result.getNode()->dump(&DAG); dbgs() << '\n');
WorklistRemover DeadNodes(*this);
- if (isLoad) {
+ if (IsLoad) {
DAG.ReplaceAllUsesOfValueWith(SDValue(N, 0), Result.getValue(0));
DAG.ReplaceAllUsesOfValueWith(SDValue(N, 1), Result.getValue(2));
} else {
@@ -13866,7 +13918,7 @@ bool DAGCombiner::CombineToPreIndexedLoadStore(SDNode *N) {
// We can now generate the new expression.
SDValue NewOp1 = DAG.getConstant(CNV, DL, CN->getValueType(0));
- SDValue NewOp2 = Result.getValue(isLoad ? 1 : 0);
+ SDValue NewOp2 = Result.getValue(IsLoad ? 1 : 0);
SDValue NewUse = DAG.getNode(Opcode,
DL,
@@ -13876,7 +13928,7 @@ bool DAGCombiner::CombineToPreIndexedLoadStore(SDNode *N) {
}
// Replace the uses of Ptr with uses of the updated base value.
- DAG.ReplaceAllUsesOfValueWith(Ptr, Result.getValue(isLoad ? 1 : 0));
+ DAG.ReplaceAllUsesOfValueWith(Ptr, Result.getValue(IsLoad ? 1 : 0));
deleteAndRecombine(Ptr.getNode());
AddToWorklist(Result.getNode());
@@ -13891,29 +13943,12 @@ bool DAGCombiner::CombineToPostIndexedLoadStore(SDNode *N) {
if (Level < AfterLegalizeDAG)
return false;
- bool isLoad = true;
+ bool IsLoad = true;
+ bool IsMasked = false;
SDValue Ptr;
- EVT VT;
- if (LoadSDNode *LD = dyn_cast<LoadSDNode>(N)) {
- if (LD->isIndexed())
- return false;
- VT = LD->getMemoryVT();
- if (!TLI.isIndexedLoadLegal(ISD::POST_INC, VT) &&
- !TLI.isIndexedLoadLegal(ISD::POST_DEC, VT))
- return false;
- Ptr = LD->getBasePtr();
- } else if (StoreSDNode *ST = dyn_cast<StoreSDNode>(N)) {
- if (ST->isIndexed())
- return false;
- VT = ST->getMemoryVT();
- if (!TLI.isIndexedStoreLegal(ISD::POST_INC, VT) &&
- !TLI.isIndexedStoreLegal(ISD::POST_DEC, VT))
- return false;
- Ptr = ST->getBasePtr();
- isLoad = false;
- } else {
+ if (!getCombineLoadStoreParts(N, ISD::POST_INC, ISD::POST_DEC, IsLoad, IsMasked,
+ Ptr, TLI))
return false;
- }
if (Ptr.getNode()->hasOneUse())
return false;
@@ -13949,7 +13984,7 @@ bool DAGCombiner::CombineToPostIndexedLoadStore(SDNode *N) {
// If all the uses are load / store addresses, then don't do the
// transformation.
- if (Use->getOpcode() == ISD::ADD || Use->getOpcode() == ISD::SUB){
+ if (Use->getOpcode() == ISD::ADD || Use->getOpcode() == ISD::SUB) {
bool RealUse = false;
for (SDNode *UseUse : Use->uses()) {
if (!canFoldInAddressingMode(Use, UseUse, DAG, TLI))
@@ -13975,18 +14010,24 @@ bool DAGCombiner::CombineToPostIndexedLoadStore(SDNode *N) {
Worklist.push_back(Op);
if (!SDNode::hasPredecessorHelper(N, Visited, Worklist) &&
!SDNode::hasPredecessorHelper(Op, Visited, Worklist)) {
- SDValue Result = isLoad
- ? DAG.getIndexedLoad(SDValue(N,0), SDLoc(N),
- BasePtr, Offset, AM)
- : DAG.getIndexedStore(SDValue(N,0), SDLoc(N),
- BasePtr, Offset, AM);
+ SDValue Result;
+ if (!IsMasked)
+ Result = IsLoad ? DAG.getIndexedLoad(SDValue(N, 0), SDLoc(N), BasePtr,
+ Offset, AM)
+ : DAG.getIndexedStore(SDValue(N, 0), SDLoc(N),
+ BasePtr, Offset, AM);
+ else
+ Result = IsLoad ? DAG.getIndexedMaskedLoad(SDValue(N, 0), SDLoc(N),
+ BasePtr, Offset, AM)
+ : DAG.getIndexedMaskedStore(SDValue(N, 0), SDLoc(N),
+ BasePtr, Offset, AM);
++PostIndexedNodes;
++NodesCombined;
LLVM_DEBUG(dbgs() << "\nReplacing.5 "; N->dump(&DAG);
dbgs() << "\nWith: "; Result.getNode()->dump(&DAG);
dbgs() << '\n');
WorklistRemover DeadNodes(*this);
- if (isLoad) {
+ if (IsLoad) {
DAG.ReplaceAllUsesOfValueWith(SDValue(N, 0), Result.getValue(0));
DAG.ReplaceAllUsesOfValueWith(SDValue(N, 1), Result.getValue(2));
} else {
@@ -13998,7 +14039,7 @@ bool DAGCombiner::CombineToPostIndexedLoadStore(SDNode *N) {
// Replace the uses of Use with uses of the updated base value.
DAG.ReplaceAllUsesOfValueWith(SDValue(Op, 0),
- Result.getValue(isLoad ? 1 : 0));
+ Result.getValue(IsLoad ? 1 : 0));
deleteAndRecombine(Op);
return true;
}
diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
index 56c13bb0753..9f8da60eb9a 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeIntegerTypes.cpp
@@ -592,8 +592,9 @@ SDValue DAGTypeLegalizer::PromoteIntRes_MLOAD(MaskedLoadSDNode *N) {
SDLoc dl(N);
SDValue Res = DAG.getMaskedLoad(NVT, dl, N->getChain(), N->getBasePtr(),
- N->getMask(), ExtPassThru, N->getMemoryVT(),
- N->getMemOperand(), ISD::EXTLOAD);
+ N->getOffset(), N->getMask(), ExtPassThru,
+ N->getMemoryVT(), N->getMemOperand(),
+ N->getAddressingMode(), ISD::EXTLOAD);
// Legalize the chain result - switch anything that used the old chain to
// use the new one.
ReplaceValueWith(SDValue(N, 1), Res.getValue(1));
@@ -1485,11 +1486,11 @@ SDValue DAGTypeLegalizer::PromoteIntOp_MSTORE(MaskedStoreSDNode *N,
SDLoc dl(N);
bool TruncateStore = false;
- if (OpNo == 3) {
+ if (OpNo == 4) {
Mask = PromoteTargetBoolean(Mask, DataVT);
// Update in place.
SmallVector<SDValue, 4> NewOps(N->op_begin(), N->op_end());
- NewOps[3] = Mask;
+ NewOps[4] = Mask;
return SDValue(DAG.UpdateNodeOperands(N, NewOps), 0);
} else { // Data operand
assert(OpNo == 1 && "Unexpected operand for promotion");
@@ -1497,14 +1498,15 @@ SDValue DAGTypeLegalizer::PromoteIntOp_MSTORE(MaskedStoreSDNode *N,
TruncateStore = true;
}
- return DAG.getMaskedStore(N->getChain(), dl, DataOp, N->getBasePtr(), Mask,
- N->getMemoryVT(), N->getMemOperand(),
+ return DAG.getMaskedStore(N->getChain(), dl, DataOp, N->getBasePtr(),
+ N->getOffset(), Mask, N->getMemoryVT(),
+ N->getMemOperand(), N->getAddressingMode(),
TruncateStore, N->isCompressingStore());
}
SDValue DAGTypeLegalizer::PromoteIntOp_MLOAD(MaskedLoadSDNode *N,
unsigned OpNo) {
- assert(OpNo == 2 && "Only know how to promote the mask!");
+ assert(OpNo == 3 && "Only know how to promote the mask!");
EVT DataVT = N->getValueType(0);
SDValue Mask = PromoteTargetBoolean(N->getOperand(OpNo), DataVT);
SmallVector<SDValue, 4> NewOps(N->op_begin(), N->op_end());
diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
index 7bca3ea888e..9403b344ea7 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
@@ -1541,12 +1541,15 @@ void DAGTypeLegalizer::SplitVecRes_LOAD(LoadSDNode *LD, SDValue &Lo,
void DAGTypeLegalizer::SplitVecRes_MLOAD(MaskedLoadSDNode *MLD,
SDValue &Lo, SDValue &Hi) {
+ assert(MLD->isUnindexed() && "Indexed masked load during type legalization!");
EVT LoVT, HiVT;
SDLoc dl(MLD);
std::tie(LoVT, HiVT) = DAG.GetSplitDestVTs(MLD->getValueType(0));
SDValue Ch = MLD->getChain();
SDValue Ptr = MLD->getBasePtr();
+ SDValue Offset = MLD->getOffset();
+ assert(Offset.isUndef() && "Unexpected indexed masked load offset");
SDValue Mask = MLD->getMask();
SDValue PassThru = MLD->getPassThru();
unsigned Alignment = MLD->getOriginalAlignment();
@@ -1578,8 +1581,9 @@ void DAGTypeLegalizer::SplitVecRes_MLOAD(MaskedLoadSDNode *MLD,
MachineMemOperand::MOLoad, LoMemVT.getStoreSize(),
Alignment, MLD->getAAInfo(), MLD->getRanges());
- Lo = DAG.getMaskedLoad(LoVT, dl, Ch, Ptr, MaskLo, PassThruLo, LoMemVT, MMO,
- ExtType, MLD->isExpandingLoad());
+ Lo = DAG.getMaskedLoad(LoVT, dl, Ch, Ptr, Offset, MaskLo, PassThruLo, LoMemVT,
+ MMO, MLD->getAddressingMode(), ExtType,
+ MLD->isExpandingLoad());
Ptr = TLI.IncrementMemoryAddress(Ptr, MaskLo, dl, LoMemVT, DAG,
MLD->isExpandingLoad());
@@ -1590,8 +1594,9 @@ void DAGTypeLegalizer::SplitVecRes_MLOAD(MaskedLoadSDNode *MLD,
HiMemVT.getStoreSize(), Alignment, MLD->getAAInfo(),
MLD->getRanges());
- Hi = DAG.getMaskedLoad(HiVT, dl, Ch, Ptr, MaskHi, PassThruHi, HiMemVT, MMO,
- ExtType, MLD->isExpandingLoad());
+ Hi = DAG.getMaskedLoad(HiVT, dl, Ch, Ptr, Offset, MaskHi, PassThruHi, HiMemVT,
+ MMO, MLD->getAddressingMode(), ExtType,
+ MLD->isExpandingLoad());
// Build a factor node to remember that this load is independent of the
// other one.
@@ -2326,8 +2331,11 @@ SDValue DAGTypeLegalizer::SplitVecOp_MGATHER(MaskedGatherSDNode *MGT,
SDValue DAGTypeLegalizer::SplitVecOp_MSTORE(MaskedStoreSDNode *N,
unsigned OpNo) {
+ assert(N->isUnindexed() && "Indexed masked store of vector?");
SDValue Ch = N->getChain();
SDValue Ptr = N->getBasePtr();
+ SDValue Offset = N->getOffset();
+ assert(Offset.isUndef() && "Unexpected indexed masked store offset");
SDValue Mask = N->getMask();
SDValue Data = N->getValue();
EVT MemoryVT = N->getMemoryVT();
@@ -2361,8 +2369,8 @@ SDValue DAGTypeLegalizer::SplitVecOp_MSTORE(MaskedStoreSDNode *N,
MachineMemOperand::MOStore, LoMemVT.getStoreSize(),
Alignment, N->getAAInfo(), N->getRanges());
- Lo = DAG.getMaskedStore(Ch, DL, DataLo, Ptr, MaskLo, LoMemVT, MMO,
- N->isTruncatingStore(),
+ Lo = DAG.getMaskedStore(Ch, DL, DataLo, Ptr, Offset, MaskLo, LoMemVT, MMO,
+ N->getAddressingMode(), N->isTruncatingStore(),
N->isCompressingStore());
Ptr = TLI.IncrementMemoryAddress(Ptr, MaskLo, DL, LoMemVT, DAG,
@@ -2374,8 +2382,9 @@ SDValue DAGTypeLegalizer::SplitVecOp_MSTORE(MaskedStoreSDNode *N,
HiMemVT.getStoreSize(), Alignment, N->getAAInfo(),
N->getRanges());
- Hi = DAG.getMaskedStore(Ch, DL, DataHi, Ptr, MaskHi, HiMemVT, MMO,
- N->isTruncatingStore(), N->isCompressingStore());
+ Hi = DAG.getMaskedStore(Ch, DL, DataHi, Ptr, Offset, MaskHi, HiMemVT, MMO,
+ N->getAddressingMode(), N->isTruncatingStore(),
+ N->isCompressingStore());
// Build a factor node to remember that this store is independent of the
// other one.
@@ -3699,10 +3708,10 @@ SDValue DAGTypeLegalizer::WidenVecRes_MLOAD(MaskedLoadSDNode *N) {
WidenVT.getVectorNumElements());
Mask = ModifyToType(Mask, WideMaskVT, true);
- SDValue Res = DAG.getMaskedLoad(WidenVT, dl, N->getChain(), N->getBasePtr(),
- Mask, PassThru, N->getMemoryVT(),
- N->getMemOperand(), ExtType,
- N->isExpandingLoad());
+ SDValue Res = DAG.getMaskedLoad(
+ WidenVT, dl, N->getChain(), N->getBasePtr(), N->getOffset(), Mask,
+ PassThru, N->getMemoryVT(), N->getMemOperand(), N->getAddressingMode(),
+ ExtType, N->isExpandingLoad());
// Legalize the chain result - switch anything that used the old chain to
// use the new one.
ReplaceValueWith(SDValue(N, 1), Res.getValue(1));
@@ -4447,7 +4456,8 @@ SDValue DAGTypeLegalizer::WidenVecOp_MSTORE(SDNode *N, unsigned OpNo) {
StVal.getValueType().getVectorNumElements() &&
"Mask and data vectors should have the same number of elements");
return DAG.getMaskedStore(MST->getChain(), dl, StVal, MST->getBasePtr(),
- Mask, MST->getMemoryVT(), MST->getMemOperand(),
+ MST->getOffset(), Mask, MST->getMemoryVT(),
+ MST->getMemOperand(), MST->getAddressingMode(),
false, MST->isCompressingStore());
}
diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
index f1b88d80f43..a20e43462f7 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
@@ -6975,16 +6975,22 @@ SDValue SelectionDAG::getIndexedStore(SDValue OrigStore, const SDLoc &dl,
}
SDValue SelectionDAG::getMaskedLoad(EVT VT, const SDLoc &dl, SDValue Chain,
- SDValue Ptr, SDValue Mask, SDValue PassThru,
- EVT MemVT, MachineMemOperand *MMO,
+ SDValue Base, SDValue Offset, SDValue Mask,
+ SDValue PassThru, EVT MemVT,
+ MachineMemOperand *MMO,
+ ISD::MemIndexedMode AM,
ISD::LoadExtType ExtTy, bool isExpanding) {
- SDVTList VTs = getVTList(VT, MVT::Other);
- SDValue Ops[] = { Chain, Ptr, Mask, PassThru };
+ bool Indexed = AM != ISD::UNINDEXED;
+ assert((Indexed || Offset.isUndef()) &&
+ "Unindexed masked load with an offset!");
+ SDVTList VTs = Indexed ? getVTList(VT, Base.getValueType(), MVT::Other)
+ : getVTList(VT, MVT::Other);
+ SDValue Ops[] = {Chain, Base, Offset, Mask, PassThru};
FoldingSetNodeID ID;
AddNodeIDNode(ID, ISD::MLOAD, VTs, Ops);
ID.AddInteger(MemVT.getRawBits());
ID.AddInteger(getSyntheticNodeSubclassData<MaskedLoadSDNode>(
- dl.getIROrder(), VTs, ExtTy, isExpanding, MemVT, MMO));
+ dl.getIROrder(), VTs, AM, ExtTy, isExpanding, MemVT, MMO));
ID.AddInteger(MMO->getPointerInfo().getAddrSpace());
void *IP = nullptr;
if (SDNode *E = FindNodeOrInsertPos(ID, dl, IP)) {
@@ -6992,7 +6998,7 @@ SDValue SelectionDAG::getMaskedLoad(EVT VT, const SDLoc &dl, SDValue Chain,
return SDValue(E, 0);
}
auto *N = newSDNode<MaskedLoadSDNode>(dl.getIROrder(), dl.getDebugLoc(), VTs,
- ExtTy, isExpanding, MemVT, MMO);
+ AM, ExtTy, isExpanding, MemVT, MMO);
createOperands(N, Ops);
CSEMap.InsertNode(N, IP);
@@ -7002,27 +7008,45 @@ SDValue SelectionDAG::getMaskedLoad(EVT VT, const SDLoc &dl, SDValue Chain,
return V;
}
+SDValue SelectionDAG::getIndexedMaskedLoad(SDValue OrigLoad, const SDLoc &dl,
+ SDValue Base, SDValue Offset,
+ ISD::MemIndexedMode AM) {
+ MaskedLoadSDNode *LD = cast<MaskedLoadSDNode>(OrigLoad);
+ assert(LD->getOffset().isUndef() && "Masked load is already a indexed load!");
+ return getMaskedLoad(OrigLoad.getValueType(), dl, LD->getChain(), Base,
+ Offset, LD->getMask(), LD->getPassThru(),
+ LD->getMemoryVT(), LD->getMemOperand(), AM,
+ LD->getExtensionType(), LD->isExpandingLoad());
+}
+
SDValue SelectionDAG::getMaskedStore(SDValue Chain, const SDLoc &dl,
- SDValue Val, SDValue Ptr, SDValue Mask,
- EVT MemVT, MachineMemOperand *MMO,
- bool IsTruncating, bool IsCompressing) {
+ SDValue Val, SDValue Base, SDValue Offset,
+ SDValue Mask, EVT MemVT,
+ MachineMemOperand *MMO,
+ ISD::MemIndexedMode AM, bool IsTruncating,
+ bool IsCompressing) {
assert(Chain.getValueType() == MVT::Other &&
"Invalid chain type");
- SDVTList VTs = getVTList(MVT::Other);
- SDValue Ops[] = { Chain, Val, Ptr, Mask };
+ bool Indexed = AM != ISD::UNINDEXED;
+ assert((Indexed || Offset.isUndef()) &&
+ "Unindexed masked store with an offset!");
+ SDVTList VTs = Indexed ? getVTList(Base.getValueType(), MVT::Other)
+ : getVTList(MVT::Other);
+ SDValue Ops[] = {Chain, Val, Base, Offset, Mask};
FoldingSetNodeID ID;
AddNodeIDNode(ID, ISD::MSTORE, VTs, Ops);
ID.AddInteger(MemVT.getRawBits());
ID.AddInteger(getSyntheticNodeSubclassData<MaskedStoreSDNode>(
- dl.getIROrder(), VTs, IsTruncating, IsCompressing, MemVT, MMO));
+ dl.getIROrder(), VTs, AM, IsTruncating, IsCompressing, MemVT, MMO));
ID.AddInteger(MMO->getPointerInfo().getAddrSpace());
void *IP = nullptr;
if (SDNode *E = FindNodeOrInsertPos(ID, dl, IP)) {
cast<MaskedStoreSDNode>(E)->refineAlignment(MMO);
return SDValue(E, 0);
}
- auto *N = newSDNode<MaskedStoreSDNode>(dl.getIROrder(), dl.getDebugLoc(), VTs,
- IsTruncating, IsCompressing, MemVT, MMO);
+ auto *N =
+ newSDNode<MaskedStoreSDNode>(dl.getIROrder(), dl.getDebugLoc(), VTs, AM,
+ IsTruncating, IsCompressing, MemVT, MMO);
createOperands(N, Ops);
CSEMap.InsertNode(N, IP);
@@ -7032,6 +7056,17 @@ SDValue SelectionDAG::getMaskedStore(SDValue Chain, const SDLoc &dl,
return V;
}
+SDValue SelectionDAG::getIndexedMaskedStore(SDValue OrigStore, const SDLoc &dl,
+ SDValue Base, SDValue Offset,
+ ISD::MemIndexedMode AM) {
+ MaskedStoreSDNode *ST = cast<MaskedStoreSDNode>(OrigStore);
+ assert(ST->getOffset().isUndef() &&
+ "Masked store is already a indexed store!");
+ return getMaskedStore(ST->getChain(), dl, ST->getValue(), Base, Offset,
+ ST->getMask(), ST->getMemoryVT(), ST->getMemOperand(),
+ AM, ST->isTruncatingStore(), ST->isCompressingStore());
+}
+
SDValue SelectionDAG::getMaskedGather(SDVTList VTs, EVT VT, const SDLoc &dl,
ArrayRef<SDValue> Ops,
MachineMemOperand *MMO,
diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
index 1ed0dc2c979..0aeb3c14aa3 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp
@@ -4295,6 +4295,7 @@ void SelectionDAGBuilder::visitMaskedStore(const CallInst &I,
SDValue Ptr = getValue(PtrOperand);
SDValue Src0 = getValue(Src0Operand);
SDValue Mask = getValue(MaskOperand);
+ SDValue Offset = DAG.getUNDEF(Ptr.getValueType());
EVT VT = Src0.getValueType();
if (!Alignment)
@@ -4311,9 +4312,9 @@ void SelectionDAGBuilder::visitMaskedStore(const CallInst &I,
// vectors.
VT.getStoreSize().getKnownMinSize(),
Alignment, AAInfo);
- SDValue StoreNode = DAG.getMaskedStore(getRoot(), sdl, Src0, Ptr, Mask, VT,
- MMO, false /* Truncating */,
- IsCompressing);
+ SDValue StoreNode =
+ DAG.getMaskedStore(getRoot(), sdl, Src0, Ptr, Offset, Mask, VT, MMO,
+ ISD::UNINDEXED, false /* Truncating */, IsCompressing);
DAG.setRoot(StoreNode);
setValue(&I, StoreNode);
}
@@ -4461,6 +4462,7 @@ void SelectionDAGBuilder::visitMaskedLoad(const CallInst &I, bool IsExpanding) {
SDValue Ptr = getValue(PtrOperand);
SDValue Src0 = getValue(Src0Operand);
SDValue Mask = getValue(MaskOperand);
+ SDValue Offset = DAG.getUNDEF(Ptr.getValueType());
EVT VT = Src0.getValueType();
if (!Alignment)
@@ -4491,8 +4493,9 @@ void SelectionDAGBuilder::visitMaskedLoad(const CallInst &I, bool IsExpanding) {
VT.getStoreSize().getKnownMinSize(),
Alignment, AAInfo, Ranges);
- SDValue Load = DAG.getMaskedLoad(VT, sdl, InChain, Ptr, Mask, Src0, VT, MMO,
- ISD::NON_EXTLOAD, IsExpanding);
+ SDValue Load =
+ DAG.getMaskedLoad(VT, sdl, InChain, Ptr, Offset, Mask, Src0, VT, MMO,
+ ISD::UNINDEXED, ISD::NON_EXTLOAD, IsExpanding);
if (AddToChain)
PendingLoads.push_back(Load.getValue(1));
setValue(&I, Load);
diff --git a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGDumper.cpp b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGDumper.cpp
index bc10f762123..f863d987648 100644
--- a/llvm/lib/CodeGen/SelectionDAG/SelectionDAGDumper.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/SelectionDAGDumper.cpp
@@ -685,6 +685,10 @@ void SDNode::print_details(raw_ostream &OS, const SelectionDAG *G) const {
if (doExt)
OS << " from " << MLd->getMemoryVT().getEVTString();
+ const char *AM = getIndexedModeName(MLd->getAddressingMode());
+ if (*AM)
+ OS << ", " << AM;
+
if (MLd->isExpandingLoad())
OS << ", expanding";
@@ -696,6 +700,10 @@ void SDNode::print_details(raw_ostream &OS, const SelectionDAG *G) const {
if (MSt->isTruncatingStore())
OS << ", trunc to " << MSt->getMemoryVT().getEVTString();
+ const char *AM = getIndexedModeName(MSt->getAddressingMode());
+ if (*AM)
+ OS << ", " << AM;
+
if (MSt->isCompressingStore())
OS << ", compressing";
diff --git a/llvm/lib/CodeGen/TargetLoweringBase.cpp b/llvm/lib/CodeGen/TargetLoweringBase.cpp
index af7dc432eae..cc436fcc4f6 100644
--- a/llvm/lib/CodeGen/TargetLoweringBase.cpp
+++ b/llvm/lib/CodeGen/TargetLoweringBase.cpp
@@ -633,6 +633,8 @@ void TargetLoweringBase::initActions() {
IM != (unsigned)ISD::LAST_INDEXED_MODE; ++IM) {
setIndexedLoadAction(IM, VT, Expand);
setIndexedStoreAction(IM, VT, Expand);
+ setIndexedMaskedLoadAction(IM, VT, Expand);
+ setIndexedMaskedStoreAction(IM, VT, Expand);
}
// Most backends expect to see the node which just returns the value loaded.
OpenPOWER on IntegriCloud