[SelectionDAG] Allow targets to specify legality of extloads' result type (in addition to the memory type)

[ab]

Summary

The *LoadExt* legalization handling used to only have one type, the memory type. This forced users to assume that as long as the extload for the memory type was declared legal, and the result type was legal, the whole extload was legal.

However, this isn't always the case. For instance, on X86, with AVX, this is legal:

v4i32 load, zext from v4i8

but this isn't:

v4i64 load, zext from v4i8

Whereas v4i64 is (arguably) legal, even without AVX2.

Keep in mind that this is a draft to give an idea, I mechanically changed the *LoadExt* calls in the generic and X86 parts.
I still need to change other in-tree targets, tell out-of-tree backend maintainers to change their code, and run the testsuite.

Note that the same thing was done a while ago for truncstores (r46140), but I assume no one needed it yet for extloads, so here we go.

[arsenm]

Thanks for working on this, this has been bothering me for a long time. This LGTM besides few formatting nits where using a variable would avoid ugly line wrappings.

I have a bunch of R600 tests for this lying around somewhere where I was hacking around this problem for extloads to i64

lib/CodeGen/CodeGenPrepare.cpp

2982–2983 ↗

(On Diff #16936)

Using a variable for TLI->getValueType(I->getType()) above would avoid this ugly wrapping (and would also be usable above)

lib/Target/X86/X86ISelLowering.cpp

888–889 ↗

(On Diff #16936)

Variable for the cast would avoid wrapping

[ab]

lib/CodeGen/CodeGenPrepare.cpp

2982–2983 ↗

(On Diff #16936)

Good point, r223491 for the variables only.

lib/Target/X86/X86ISelLowering.cpp

888–889 ↗

(On Diff #16936)

Right, as I was changing these I added iterator ranges in D6537; reviews welcome!

[hfinkel]

This also looks good to me -- this is the right thing to do -- as you've mentioned, you need to update the other in-tree targets before this lands.

[ab]

FYI: while changing other targets, I decided to flip the order of
MemVT and ValVT in [gs]etLoadExtAction. In this patch, we had:

setLoadExtAction(.., MVT ValVT, MVT MemVT, ..)
setTruncStoreAction(MVT ValVT, MVT MemVT, ..)

I think it's more useful to keep "source" and "destination" types
consistent, that is:

setLoadExtAction(.., MVT MemVT, MVT ValVT, ..)
setTruncStoreAction(MVT ValVT, MVT MemVT, ..)

This avoids having consecutive calls to both, with the same arguments,
but different ordering.
I'll also rename them to SrcVT/DestVT to make that clear.

Anyway, thanks for the reviews,

-Ahmed

llvm/trunk/include/llvm/Target/TargetLowering.h

public:
   bool isOperationLegal(unsigned Op, EVT VT) const {
     return (VT == MVT::Other || isTypeLegal(VT)) &&
            getOperationAction(Op, VT) == Legal;
   }
 
   /// Return how this load with extension should be treated: either it is legal,
   /// needs to be promoted to a larger size, needs to be expanded to some other
   /// code sequence, or the target has a custom expander for it.
-  LegalizeAction getLoadExtAction(unsigned ExtType, EVT VT) const {
-    if (VT.isExtended()) return Expand;
-    unsigned I = (unsigned) VT.getSimpleVT().SimpleTy;
-    assert(ExtType < ISD::LAST_LOADEXT_TYPE && I < MVT::LAST_VALUETYPE &&
-           "Table isn't big enough!");
-    return (LegalizeAction)LoadExtActions[I][ExtType];
+  LegalizeAction getLoadExtAction(unsigned ExtType, EVT ValVT, EVT MemVT) const {
+    if (ValVT.isExtended() || MemVT.isExtended()) return Expand;
+    unsigned ValI = (unsigned) ValVT.getSimpleVT().SimpleTy;
+    unsigned MemI = (unsigned) MemVT.getSimpleVT().SimpleTy;
+    assert(ExtType < ISD::LAST_LOADEXT_TYPE && ValI < MVT::LAST_VALUETYPE &&
+           MemI < MVT::LAST_VALUETYPE && "Table isn't big enough!");
+    return (LegalizeAction)LoadExtActions[ValI][MemI][ExtType];
   }
 
   /// Return true if the specified load with extension is legal on this target.
-  bool isLoadExtLegal(unsigned ExtType, EVT VT) const {
-    return VT.isSimple() &&
-      getLoadExtAction(ExtType, VT.getSimpleVT()) == Legal;
+  bool isLoadExtLegal(unsigned ExtType, EVT ValVT, EVT MemVT) const {
+    return ValVT.isSimple() && MemVT.isSimple() &&
+      getLoadExtAction(ExtType, ValVT, MemVT) == Legal;
   }
 
   /// Return how this store with truncation should be treated: either it is
   /// legal, needs to be promoted to a larger size, needs to be expanded to some
   /// other code sequence, or the target has a custom expander for it.
   LegalizeAction getTruncStoreAction(EVT ValVT, EVT MemVT) const {
     if (ValVT.isExtended() || MemVT.isExtended()) return Expand;
     unsigned ValI = (unsigned) ValVT.getSimpleVT().SimpleTy;
protected:
   void setOperationAction(unsigned Op, MVT VT,
                           LegalizeAction Action) {
     assert(Op < array_lengthof(OpActions[0]) && "Table isn't big enough!");
     OpActions[(unsigned)VT.SimpleTy][Op] = (uint8_t)Action;
   }
 
   /// Indicate that the specified load with extension does not work with the
   /// specified type and indicate what to do about it.
-  void setLoadExtAction(unsigned ExtType, MVT VT,
+  void setLoadExtAction(unsigned ExtType, MVT ValVT, MVT MemVT,
                         LegalizeAction Action) {
-    assert(ExtType < ISD::LAST_LOADEXT_TYPE && VT.isValid() &&
-           "Table isn't big enough!");
-    LoadExtActions[VT.SimpleTy][ExtType] = (uint8_t)Action;
+    assert(ExtType < ISD::LAST_LOADEXT_TYPE && ValVT.isValid() &&
+           MemVT.isValid() && "Table isn't big enough!");
+    LoadExtActions[ValVT.SimpleTy][MemVT.SimpleTy][ExtType] = (uint8_t)Action;
   }
 
   /// Indicate that the specified truncating store does not work with the
   /// specified type and indicate what to do about it.
   void setTruncStoreAction(MVT ValVT, MVT MemVT,
                            LegalizeAction Action) {
     assert(ValVT.isValid() && MemVT.isValid() && "Table isn't big enough!");
     TruncStoreActions[ValVT.SimpleTy][MemVT.SimpleTy] = (uint8_t)Action;
private:
   /// operations are Legal (aka, supported natively by the target), but
   /// operations that are not should be described.  Note that operations on
   /// non-legal value types are not described here.
   uint8_t OpActions[MVT::LAST_VALUETYPE][ISD::BUILTIN_OP_END];
 
   /// For each load extension type and each value type, keep a LegalizeAction
   /// that indicates how instruction selection should deal with a load of a
   /// specific value type and extension type.
-  uint8_t LoadExtActions[MVT::LAST_VALUETYPE][ISD::LAST_LOADEXT_TYPE];
+  uint8_t LoadExtActions[MVT::LAST_VALUETYPE][MVT::LAST_VALUETYPE]
+                        [ISD::LAST_LOADEXT_TYPE];
 
   /// For each value type pair keep a LegalizeAction that indicates whether a
   /// truncating store of a specific value type and truncating type is legal.
   uint8_t TruncStoreActions[MVT::LAST_VALUETYPE][MVT::LAST_VALUETYPE];
 
   /// For each indexed mode and each value type, keep a pair of LegalizeAction
   /// that indicates how instruction selection should deal with the load /
   /// store.

llvm/trunk/lib/CodeGen/BasicTargetTransformInfo.cpp

if (Src->isVectorTy() &&
     // itself. Unless the corresponding extending load or truncating store is
     // legal, then this will scalarize.
     TargetLowering::LegalizeAction LA = TargetLowering::Expand;
     EVT MemVT = getTLI()->getValueType(Src, true);
     if (MemVT.isSimple() && MemVT != MVT::Other) {
       if (Opcode == Instruction::Store)
         LA = getTLI()->getTruncStoreAction(LT.second, MemVT.getSimpleVT());
       else
-        LA = getTLI()->getLoadExtAction(ISD::EXTLOAD, MemVT.getSimpleVT());
+        LA = getTLI()->getLoadExtAction(ISD::EXTLOAD, LT.second, MemVT);
     }
 
     if (LA != TargetLowering::Legal && LA != TargetLowering::Custom) {
       // This is a vector load/store for some illegal type that is scalarized.
       // We must account for the cost of building or decomposing the vector.
       Cost += getScalarizationOverhead(Src, Opcode != Instruction::Store,
                                             Opcode == Instruction::Store);
     }

llvm/trunk/lib/CodeGen/CodeGenPrepare.cpp

bool CodeGenPrepare::MoveExtToFormExtLoad(Instruction *&I) {
   // Check whether the target supports casts folded into loads.
   unsigned LType;
   if (isa<ZExtInst>(I))
     LType = ISD::ZEXTLOAD;
   else {
     assert(isa<SExtInst>(I) && "Unexpected ext type!");
     LType = ISD::SEXTLOAD;
   }
-  if (TLI && !TLI->isLoadExtLegal(LType, LoadVT)) {
+  if (TLI && !TLI->isLoadExtLegal(LType, VT, LoadVT)) {
     I = OldExt;
     TPT.rollback(LastKnownGood);
     return false;
   }
 
   // Move the extend into the same block as the load, so that SelectionDAG
   // can fold it.
   TPT.commit();

llvm/trunk/lib/CodeGen/SelectionDAG/DAGCombiner.cpp

 }
 
 SDValue DAGCombiner::PromoteOperand(SDValue Op, EVT PVT, bool &Replace) {
   Replace = false;
   SDLoc dl(Op);
   if (LoadSDNode *LD = dyn_cast<LoadSDNode>(Op)) {
     EVT MemVT = LD->getMemoryVT();
     ISD::LoadExtType ExtType = ISD::isNON_EXTLoad(LD)
-      ? (TLI.isLoadExtLegal(ISD::ZEXTLOAD, MemVT) ? ISD::ZEXTLOAD
+      ? (TLI.isLoadExtLegal(ISD::ZEXTLOAD, PVT, MemVT) ? ISD::ZEXTLOAD
                                                        : ISD::EXTLOAD)
       : LD->getExtensionType();
     Replace = true;
     return DAG.getExtLoad(ExtType, dl, PVT,
                           LD->getChain(), LD->getBasePtr(),
                           MemVT, LD->getMemOperand());
   }
 
   unsigned Opc = Op.getOpcode();
bool DAGCombiner::PromoteLoad(SDValue Op) {
   if (TLI.IsDesirableToPromoteOp(Op, PVT)) {
     assert(PVT != VT && "Don't know what type to promote to!");
 
     SDLoc dl(Op);
     SDNode *N = Op.getNode();
     LoadSDNode *LD = cast<LoadSDNode>(N);
     EVT MemVT = LD->getMemoryVT();
     ISD::LoadExtType ExtType = ISD::isNON_EXTLoad(LD)
-      ? (TLI.isLoadExtLegal(ISD::ZEXTLOAD, MemVT) ? ISD::ZEXTLOAD
+      ? (TLI.isLoadExtLegal(ISD::ZEXTLOAD, PVT, MemVT) ? ISD::ZEXTLOAD
                                                        : ISD::EXTLOAD)
       : LD->getExtensionType();
     SDValue NewLD = DAG.getExtLoad(ExtType, dl, PVT,
                                    LD->getChain(), LD->getBasePtr(),
                                    MemVT, LD->getMemOperand());
     SDValue Result = DAG.getNode(ISD::TRUNCATE, dl, VT, NewLD);
 
     DEBUG(dbgs() << "\nPromoting ";
           N->dump(&DAG);
if (const ConstantSDNode *C = dyn_cast<ConstantSDNode>(N1)) {
           Constant &= SplatValue.lshr(i*BitWidth).zextOrTrunc(BitWidth);
       }
     }
 
     // If we want to change an EXTLOAD to a ZEXTLOAD, ensure a ZEXTLOAD is
     // actually legal and isn't going to get expanded, else this is a false
     // optimisation.
     bool CanZextLoadProfitably = TLI.isLoadExtLegal(ISD::ZEXTLOAD,
+                                                    Load->getValueType(0),
                                                     Load->getMemoryVT());
 
     // Resize the constant to the same size as the original memory access before
     // extension. If it is still the AllOnesValue then this AND is completely
     // unneeded.
     Constant =
       Constant.zextOrTrunc(Load->getMemoryVT().getScalarType().getSizeInBits());
 
if (ISD::isEXTLoad(N0.getNode()) && ISD::isUNINDEXEDLoad(N0.getNode())) {
     LoadSDNode *LN0 = cast<LoadSDNode>(N0);
     EVT MemVT = LN0->getMemoryVT();
     // If we zero all the possible extended bits, then we can turn this into
     // a zextload if we are running before legalize or the operation is legal.
     unsigned BitWidth = N1.getValueType().getScalarType().getSizeInBits();
     if (DAG.MaskedValueIsZero(N1, APInt::getHighBitsSet(BitWidth,
                            BitWidth - MemVT.getScalarType().getSizeInBits())) &&
         ((!LegalOperations && !LN0->isVolatile()) ||
-         TLI.isLoadExtLegal(ISD::ZEXTLOAD, MemVT))) {
+         TLI.isLoadExtLegal(ISD::ZEXTLOAD, VT, MemVT))) {
       SDValue ExtLoad = DAG.getExtLoad(ISD::ZEXTLOAD, SDLoc(N0), VT,
                                        LN0->getChain(), LN0->getBasePtr(),
                                        MemVT, LN0->getMemOperand());
       AddToWorklist(N);
       CombineTo(N0.getNode(), ExtLoad, ExtLoad.getValue(1));
       return SDValue(N, 0);   // Return N so it doesn't get rechecked!
     }
   }
   // fold (zext_inreg (sextload x)) -> (zextload x) iff load has one use
   if (ISD::isSEXTLoad(N0.getNode()) && ISD::isUNINDEXEDLoad(N0.getNode()) &&
       N0.hasOneUse()) {
     LoadSDNode *LN0 = cast<LoadSDNode>(N0);
     EVT MemVT = LN0->getMemoryVT();
     // If we zero all the possible extended bits, then we can turn this into
     // a zextload if we are running before legalize or the operation is legal.
     unsigned BitWidth = N1.getValueType().getScalarType().getSizeInBits();
     if (DAG.MaskedValueIsZero(N1, APInt::getHighBitsSet(BitWidth,
                            BitWidth - MemVT.getScalarType().getSizeInBits())) &&
         ((!LegalOperations && !LN0->isVolatile()) ||
-         TLI.isLoadExtLegal(ISD::ZEXTLOAD, MemVT))) {
+         TLI.isLoadExtLegal(ISD::ZEXTLOAD, VT, MemVT))) {
       SDValue ExtLoad = DAG.getExtLoad(ISD::ZEXTLOAD, SDLoc(N0), VT,
                                        LN0->getChain(), LN0->getBasePtr(),
                                        MemVT, LN0->getMemOperand());
       AddToWorklist(N);
       CombineTo(N0.getNode(), ExtLoad, ExtLoad.getValue(1));
       return SDValue(N, 0);   // Return N so it doesn't get rechecked!
     }
   }
LoadSDNode *LN0 = HasAnyExt
       ? cast<LoadSDNode>(N0.getOperand(0))
       : cast<LoadSDNode>(N0);
     if (LN0->getExtensionType() != ISD::SEXTLOAD &&
         LN0->isUnindexed() && N0.hasOneUse() && SDValue(LN0, 0).hasOneUse()) {
       uint32_t ActiveBits = N1C->getAPIntValue().getActiveBits();
       if (ActiveBits > 0 && APIntOps::isMask(ActiveBits, N1C->getAPIntValue())){
         EVT ExtVT = EVT::getIntegerVT(*DAG.getContext(), ActiveBits);
         EVT LoadedVT = LN0->getMemoryVT();
+        EVT LoadResultTy = HasAnyExt ? LN0->getValueType(0) : VT;
 
         if (ExtVT == LoadedVT &&
-            (!LegalOperations || TLI.isLoadExtLegal(ISD::ZEXTLOAD, ExtVT))) {
-          EVT LoadResultTy = HasAnyExt ? LN0->getValueType(0) : VT;
+            (!LegalOperations || TLI.isLoadExtLegal(ISD::ZEXTLOAD, LoadResultTy,
+                                                    ExtVT))) {
 
           SDValue NewLoad =
             DAG.getExtLoad(ISD::ZEXTLOAD, SDLoc(LN0), LoadResultTy,
                            LN0->getChain(), LN0->getBasePtr(), ExtVT,
                            LN0->getMemOperand());
           AddToWorklist(N);
           CombineTo(LN0, NewLoad, NewLoad.getValue(1));
           return SDValue(N, 0);   // Return N so it doesn't get rechecked!
         }
 
         // Do not change the width of a volatile load.
         // Do not generate loads of non-round integer types since these can
         // be expensive (and would be wrong if the type is not byte sized).
         if (!LN0->isVolatile() && LoadedVT.bitsGT(ExtVT) && ExtVT.isRound() &&
-            (!LegalOperations || TLI.isLoadExtLegal(ISD::ZEXTLOAD, ExtVT))) {
+            (!LegalOperations || TLI.isLoadExtLegal(ISD::ZEXTLOAD, LoadResultTy,
+                                                    ExtVT))) {
           EVT PtrType = LN0->getOperand(1).getValueType();
 
           unsigned Alignment = LN0->getAlignment();
           SDValue NewPtr = LN0->getBasePtr();
 
           // For big endian targets, we need to add an offset to the pointer
           // to load the correct bytes.  For little endian systems, we merely
           // need to read fewer bytes from the same pointer.
           if (TLI.isBigEndian()) {
             unsigned LVTStoreBytes = LoadedVT.getStoreSize();
             unsigned EVTStoreBytes = ExtVT.getStoreSize();
             unsigned PtrOff = LVTStoreBytes - EVTStoreBytes;
             NewPtr = DAG.getNode(ISD::ADD, SDLoc(LN0), PtrType,
                                  NewPtr, DAG.getConstant(PtrOff, PtrType));
             Alignment = MinAlign(Alignment, PtrOff);
           }
 
           AddToWorklist(NewPtr.getNode());
 
-          EVT LoadResultTy = HasAnyExt ? LN0->getValueType(0) : VT;
           SDValue Load =
             DAG.getExtLoad(ISD::ZEXTLOAD, SDLoc(LN0), LoadResultTy,
                            LN0->getChain(), NewPtr,
                            LN0->getPointerInfo(),
                            ExtVT, LN0->isVolatile(), LN0->isNonTemporal(),
                            LN0->isInvariant(), Alignment, LN0->getAAInfo());
           AddToWorklist(N);
           CombineTo(LN0, Load, Load.getValue(1));
SDValue DAGCombiner::visitSIGN_EXTEND(SDNode *N) {
 
   // fold (sext (load x)) -> (sext (truncate (sextload x)))
   // None of the supported targets knows how to perform load and sign extend
   // on vectors in one instruction.  We only perform this transformation on
   // scalars.
   if (ISD::isNON_EXTLoad(N0.getNode()) && !VT.isVector() &&
       ISD::isUNINDEXEDLoad(N0.getNode()) &&
       ((!LegalOperations && !cast<LoadSDNode>(N0)->isVolatile()) ||
-       TLI.isLoadExtLegal(ISD::SEXTLOAD, N0.getValueType()))) {
+       TLI.isLoadExtLegal(ISD::SEXTLOAD, VT, N0.getValueType()))) {
     bool DoXform = true;
     SmallVector<SDNode*, 4> SetCCs;
     if (!N0.hasOneUse())
       DoXform = ExtendUsesToFormExtLoad(N, N0, ISD::SIGN_EXTEND, SetCCs, TLI);
     if (DoXform) {
       LoadSDNode *LN0 = cast<LoadSDNode>(N0);
       SDValue ExtLoad = DAG.getExtLoad(ISD::SEXTLOAD, SDLoc(N), VT,
                                        LN0->getChain(),
SDValue DAGCombiner::visitSIGN_EXTEND(SDNode *N) {
 
   // fold (sext (sextload x)) -> (sext (truncate (sextload x)))
   // fold (sext ( extload x)) -> (sext (truncate (sextload x)))
   if ((ISD::isSEXTLoad(N0.getNode()) || ISD::isEXTLoad(N0.getNode())) &&
       ISD::isUNINDEXEDLoad(N0.getNode()) && N0.hasOneUse()) {
     LoadSDNode *LN0 = cast<LoadSDNode>(N0);
     EVT MemVT = LN0->getMemoryVT();
     if ((!LegalOperations && !LN0->isVolatile()) ||
-        TLI.isLoadExtLegal(ISD::SEXTLOAD, MemVT)) {
+        TLI.isLoadExtLegal(ISD::SEXTLOAD, VT, MemVT)) {
       SDValue ExtLoad = DAG.getExtLoad(ISD::SEXTLOAD, SDLoc(N), VT,
                                        LN0->getChain(),
                                        LN0->getBasePtr(), MemVT,
                                        LN0->getMemOperand());
       CombineTo(N, ExtLoad);
       CombineTo(N0.getNode(),
                 DAG.getNode(ISD::TRUNCATE, SDLoc(N0),
                             N0.getValueType(), ExtLoad),
                 ExtLoad.getValue(1));
       return SDValue(N, 0);   // Return N so it doesn't get rechecked!
     }
   }
 
   // fold (sext (and/or/xor (load x), cst)) ->
   //      (and/or/xor (sextload x), (sext cst))
   if ((N0.getOpcode() == ISD::AND || N0.getOpcode() == ISD::OR ||
        N0.getOpcode() == ISD::XOR) &&
       isa<LoadSDNode>(N0.getOperand(0)) &&
       N0.getOperand(1).getOpcode() == ISD::Constant &&
-      TLI.isLoadExtLegal(ISD::SEXTLOAD, N0.getValueType()) &&
+      TLI.isLoadExtLegal(ISD::SEXTLOAD, VT, N0.getValueType()) &&
       (!LegalOperations && TLI.isOperationLegal(N0.getOpcode(), VT))) {
     LoadSDNode *LN0 = cast<LoadSDNode>(N0.getOperand(0));
     if (LN0->getExtensionType() != ISD::ZEXTLOAD && LN0->isUnindexed()) {
       bool DoXform = true;
       SmallVector<SDNode*, 4> SetCCs;
       if (!N0.hasOneUse())
         DoXform = ExtendUsesToFormExtLoad(N, N0.getOperand(0), ISD::SIGN_EXTEND,
                                           SetCCs, TLI);
SDValue DAGCombiner::visitZERO_EXTEND(SDNode *N) {
 
   // fold (zext (load x)) -> (zext (truncate (zextload x)))
   // None of the supported targets knows how to perform load and vector_zext
   // on vectors in one instruction.  We only perform this transformation on
   // scalars.
   if (ISD::isNON_EXTLoad(N0.getNode()) && !VT.isVector() &&
       ISD::isUNINDEXEDLoad(N0.getNode()) &&
       ((!LegalOperations && !cast<LoadSDNode>(N0)->isVolatile()) ||
-       TLI.isLoadExtLegal(ISD::ZEXTLOAD, N0.getValueType()))) {
+       TLI.isLoadExtLegal(ISD::ZEXTLOAD, VT, N0.getValueType()))) {
     bool DoXform = true;
     SmallVector<SDNode*, 4> SetCCs;
     if (!N0.hasOneUse())
       DoXform = ExtendUsesToFormExtLoad(N, N0, ISD::ZERO_EXTEND, SetCCs, TLI);
     if (DoXform) {
       LoadSDNode *LN0 = cast<LoadSDNode>(N0);
       SDValue ExtLoad = DAG.getExtLoad(ISD::ZEXTLOAD, SDLoc(N), VT,
                                        LN0->getChain(),
SDValue DAGCombiner::visitZERO_EXTEND(SDNode *N) {
   }
 
   // fold (zext (and/or/xor (load x), cst)) ->
   //      (and/or/xor (zextload x), (zext cst))
   if ((N0.getOpcode() == ISD::AND || N0.getOpcode() == ISD::OR ||
        N0.getOpcode() == ISD::XOR) &&
       isa<LoadSDNode>(N0.getOperand(0)) &&
       N0.getOperand(1).getOpcode() == ISD::Constant &&
-      TLI.isLoadExtLegal(ISD::ZEXTLOAD, N0.getValueType()) &&
+      TLI.isLoadExtLegal(ISD::ZEXTLOAD, VT, N0.getValueType()) &&
       (!LegalOperations && TLI.isOperationLegal(N0.getOpcode(), VT))) {
     LoadSDNode *LN0 = cast<LoadSDNode>(N0.getOperand(0));
     if (LN0->getExtensionType() != ISD::SEXTLOAD && LN0->isUnindexed()) {
       bool DoXform = true;
       SmallVector<SDNode*, 4> SetCCs;
       if (!N0.hasOneUse())
         DoXform = ExtendUsesToFormExtLoad(N, N0.getOperand(0), ISD::ZERO_EXTEND,
                                           SetCCs, TLI);
SDValue DAGCombiner::visitZERO_EXTEND(SDNode *N) {
 
   // fold (zext (zextload x)) -> (zext (truncate (zextload x)))
   // fold (zext ( extload x)) -> (zext (truncate (zextload x)))
   if ((ISD::isZEXTLoad(N0.getNode()) || ISD::isEXTLoad(N0.getNode())) &&
       ISD::isUNINDEXEDLoad(N0.getNode()) && N0.hasOneUse()) {
     LoadSDNode *LN0 = cast<LoadSDNode>(N0);
     EVT MemVT = LN0->getMemoryVT();
     if ((!LegalOperations && !LN0->isVolatile()) ||
-        TLI.isLoadExtLegal(ISD::ZEXTLOAD, MemVT)) {
+        TLI.isLoadExtLegal(ISD::ZEXTLOAD, VT, MemVT)) {
       SDValue ExtLoad = DAG.getExtLoad(ISD::ZEXTLOAD, SDLoc(N), VT,
                                        LN0->getChain(),
                                        LN0->getBasePtr(), MemVT,
                                        LN0->getMemOperand());
       CombineTo(N, ExtLoad);
       CombineTo(N0.getNode(),
                 DAG.getNode(ISD::TRUNCATE, SDLoc(N0), N0.getValueType(),
                             ExtLoad),
SDValue DAGCombiner::visitANY_EXTEND(SDNode *N) {
   }
 
   // fold (aext (load x)) -> (aext (truncate (extload x)))
   // None of the supported targets knows how to perform load and any_ext
   // on vectors in one instruction.  We only perform this transformation on
   // scalars.
   if (ISD::isNON_EXTLoad(N0.getNode()) && !VT.isVector() &&
       ISD::isUNINDEXEDLoad(N0.getNode()) &&
-      TLI.isLoadExtLegal(ISD::EXTLOAD, N0.getValueType())) {
+      TLI.isLoadExtLegal(ISD::EXTLOAD, VT, N0.getValueType())) {
     bool DoXform = true;
     SmallVector<SDNode*, 4> SetCCs;
     if (!N0.hasOneUse())
       DoXform = ExtendUsesToFormExtLoad(N, N0, ISD::ANY_EXTEND, SetCCs, TLI);
     if (DoXform) {
       LoadSDNode *LN0 = cast<LoadSDNode>(N0);
       SDValue ExtLoad = DAG.getExtLoad(ISD::EXTLOAD, SDLoc(N), VT,
                                        LN0->getChain(),
SDValue DAGCombiner::visitANY_EXTEND(SDNode *N) {
   // fold (aext (sextload x)) -> (aext (truncate (sextload x)))
   // fold (aext ( extload x)) -> (aext (truncate (extload  x)))
   if (N0.getOpcode() == ISD::LOAD &&
       !ISD::isNON_EXTLoad(N0.getNode()) && ISD::isUNINDEXEDLoad(N0.getNode()) &&
       N0.hasOneUse()) {
     LoadSDNode *LN0 = cast<LoadSDNode>(N0);
     ISD::LoadExtType ExtType = LN0->getExtensionType();
     EVT MemVT = LN0->getMemoryVT();
-    if (!LegalOperations || TLI.isLoadExtLegal(ExtType, MemVT)) {
+    if (!LegalOperations || TLI.isLoadExtLegal(ExtType, VT, MemVT)) {
       SDValue ExtLoad = DAG.getExtLoad(ExtType, SDLoc(N),
                                        VT, LN0->getChain(), LN0->getBasePtr(),
                                        MemVT, LN0->getMemOperand());
       CombineTo(N, ExtLoad);
       CombineTo(N0.getNode(),
                 DAG.getNode(ISD::TRUNCATE, SDLoc(N0),
                             N0.getValueType(), ExtLoad),
                 ExtLoad.getValue(1));
if (Opc == ISD::SIGN_EXTEND_INREG) {
     // Another special-case: SRL is basically zero-extending a narrower value.
     ExtType = ISD::ZEXTLOAD;
     N0 = SDValue(N, 0);
     ConstantSDNode *N01 = dyn_cast<ConstantSDNode>(N0.getOperand(1));
     if (!N01) return SDValue();
     ExtVT = EVT::getIntegerVT(*DAG.getContext(),
                               VT.getSizeInBits() - N01->getZExtValue());
   }
-  if (LegalOperations && !TLI.isLoadExtLegal(ExtType, ExtVT))
+  if (LegalOperations && !TLI.isLoadExtLegal(ExtType, VT, ExtVT))
     return SDValue();
 
   unsigned EVTBits = ExtVT.getSizeInBits();
 
   // Do not generate loads of non-round integer types since these can
   // be expensive (and would be wrong if the type is not byte sized).
   if (!ExtVT.isRound())
     return SDValue();
if (ConstantSDNode *ShAmt = dyn_cast<ConstantSDNode>(N0.getOperand(1)))
       }
   }
 
   // fold (sext_inreg (extload x)) -> (sextload x)
   if (ISD::isEXTLoad(N0.getNode()) &&
       ISD::isUNINDEXEDLoad(N0.getNode()) &&
       EVT == cast<LoadSDNode>(N0)->getMemoryVT() &&
       ((!LegalOperations && !cast<LoadSDNode>(N0)->isVolatile()) ||
-       TLI.isLoadExtLegal(ISD::SEXTLOAD, EVT))) {
+       TLI.isLoadExtLegal(ISD::SEXTLOAD, VT, EVT))) {
     LoadSDNode *LN0 = cast<LoadSDNode>(N0);
     SDValue ExtLoad = DAG.getExtLoad(ISD::SEXTLOAD, SDLoc(N), VT,
                                      LN0->getChain(),
                                      LN0->getBasePtr(), EVT,
                                      LN0->getMemOperand());
     CombineTo(N, ExtLoad);
     CombineTo(N0.getNode(), ExtLoad, ExtLoad.getValue(1));
     AddToWorklist(ExtLoad.getNode());
     return SDValue(N, 0);   // Return N so it doesn't get rechecked!
   }
   // fold (sext_inreg (zextload x)) -> (sextload x) iff load has one use
   if (ISD::isZEXTLoad(N0.getNode()) && ISD::isUNINDEXEDLoad(N0.getNode()) &&
       N0.hasOneUse() &&
       EVT == cast<LoadSDNode>(N0)->getMemoryVT() &&
       ((!LegalOperations && !cast<LoadSDNode>(N0)->isVolatile()) ||
-       TLI.isLoadExtLegal(ISD::SEXTLOAD, EVT))) {
+       TLI.isLoadExtLegal(ISD::SEXTLOAD, VT, EVT))) {
     LoadSDNode *LN0 = cast<LoadSDNode>(N0);
     SDValue ExtLoad = DAG.getExtLoad(ISD::SEXTLOAD, SDLoc(N), VT,
                                      LN0->getChain(),
                                      LN0->getBasePtr(), EVT,
                                      LN0->getMemOperand());
     CombineTo(N, ExtLoad);
     CombineTo(N0.getNode(), ExtLoad, ExtLoad.getValue(1));
     return SDValue(N, 0);   // Return N so it doesn't get rechecked!
if (N0.getOpcode() == ISD::FP_ROUND
     if (VT.bitsLT(In.getValueType()))
       return DAG.getNode(ISD::FP_ROUND, SDLoc(N), VT,
                          In, N0.getOperand(1));
     return DAG.getNode(ISD::FP_EXTEND, SDLoc(N), VT, In);
   }
 
   // fold (fpext (load x)) -> (fpext (fptrunc (extload x)))
   if (ISD::isNormalLoad(N0.getNode()) && N0.hasOneUse() &&
-       TLI.isLoadExtLegal(ISD::EXTLOAD, N0.getValueType())) {
+       TLI.isLoadExtLegal(ISD::EXTLOAD, VT, N0.getValueType())) {
     LoadSDNode *LN0 = cast<LoadSDNode>(N0);
     SDValue ExtLoad = DAG.getExtLoad(ISD::EXTLOAD, SDLoc(N), VT,
                                      LN0->getChain(),
                                      LN0->getBasePtr(), N0.getValueType(),
                                      LN0->getMemOperand());
     CombineTo(N, ExtLoad);
     CombineTo(N0.getNode(),
               DAG.getNode(ISD::FP_ROUND, SDLoc(N0),
for (unsigned i = 1; i < LoadNodes.size(); ++i) {
     if (TLI.isTypeLegal(StoreTy))
       LastLegalIntegerType = i + 1;
     // Or check whether a truncstore and extload is legal.
     else if (TLI.getTypeAction(*DAG.getContext(), StoreTy) ==
              TargetLowering::TypePromoteInteger) {
       EVT LegalizedStoredValueTy =
         TLI.getTypeToTransformTo(*DAG.getContext(), StoreTy);
       if (TLI.isTruncStoreLegal(LegalizedStoredValueTy, StoreTy) &&
-          TLI.isLoadExtLegal(ISD::ZEXTLOAD, StoreTy) &&
-          TLI.isLoadExtLegal(ISD::SEXTLOAD, StoreTy) &&
-          TLI.isLoadExtLegal(ISD::EXTLOAD, StoreTy))
+          TLI.isLoadExtLegal(ISD::ZEXTLOAD, LegalizedStoredValueTy, StoreTy) &&
+          TLI.isLoadExtLegal(ISD::SEXTLOAD, LegalizedStoredValueTy, StoreTy) &&
+          TLI.isLoadExtLegal(ISD::EXTLOAD, LegalizedStoredValueTy, StoreTy))
         LastLegalIntegerType = i+1;
     }
   }
 
   // Only use vector types if the vector type is larger than the integer type.
   // If they are the same, use integers.
   bool UseVectorTy = LastLegalVectorType > LastLegalIntegerType && !NoVectors;
   unsigned LastLegalType = std::max(LastLegalVectorType, LastLegalIntegerType);
SDValue DAGCombiner::ReplaceExtractVectorEltOfLoadWithNarrowedLoad(
   // Note that this replacement assumes that the extractvalue is the only
   // use of the load; that's okay because we don't want to perform this
   // transformation in other cases anyway.
   SDValue Load;
   SDValue Chain;
   if (ResultVT.bitsGT(VecEltVT)) {
     // If the result type of vextract is wider than the load, then issue an
     // extending load instead.
-    ISD::LoadExtType ExtType = TLI.isLoadExtLegal(ISD::ZEXTLOAD, VecEltVT)
+    ISD::LoadExtType ExtType = TLI.isLoadExtLegal(ISD::ZEXTLOAD, ResultVT,
+                                                  VecEltVT)
                                    ? ISD::ZEXTLOAD
                                    : ISD::EXTLOAD;
     Load = DAG.getExtLoad(
         ExtType, SDLoc(EVE), ResultVT, OriginalLoad->getChain(), NewPtr, MPI,
         VecEltVT, OriginalLoad->isVolatile(), OriginalLoad->isNonTemporal(),
         OriginalLoad->isInvariant(), Align, OriginalLoad->getAAInfo());
     Chain = Load.getValue(1);
   } else {

llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp

SelectionDAGLegalize::ExpandConstantFP(ConstantFPSDNode *CFP, bool UseCP) {
 
   EVT OrigVT = VT;
   EVT SVT = VT;
   while (SVT != MVT::f32 && SVT != MVT::f16) {
     SVT = (MVT::SimpleValueType)(SVT.getSimpleVT().SimpleTy - 1);
     if (ConstantFPSDNode::isValueValidForType(SVT, CFP->getValueAPF()) &&
         // Only do this if the target has a native EXTLOAD instruction from
         // smaller type.
-        TLI.isLoadExtLegal(ISD::EXTLOAD, SVT) &&
+        TLI.isLoadExtLegal(ISD::EXTLOAD, OrigVT, SVT) &&
         TLI.ShouldShrinkFPConstant(OrigVT)) {
       Type *SType = SVT.getTypeForEVT(*DAG.getContext());
       LLVMC = cast<ConstantFP>(ConstantExpr::getFPTrunc(LLVMC, SType));
       VT = SVT;
       Extend = true;
     }
   }
 
if (SrcWidth != SrcVT.getStoreSizeInBits() &&
       // Some targets pretend to have an i1 loading operation, and actually
       // load an i8.  This trick is correct for ZEXTLOAD because the top 7
       // bits are guaranteed to be zero; it helps the optimizers understand
       // that these bits are zero.  It is also useful for EXTLOAD, since it
       // tells the optimizers that those bits are undefined.  It would be
       // nice to have an effective generic way of getting these benefits...
       // Until such a way is found, don't insist on promoting i1 here.
       (SrcVT != MVT::i1 ||
-       TLI.getLoadExtAction(ExtType, MVT::i1) == TargetLowering::Promote)) {
+       TLI.getLoadExtAction(ExtType, Node->getValueType(0), MVT::i1) ==
+         TargetLowering::Promote)) {
     // Promote to a byte-sized load if not loading an integral number of
     // bytes.  For example, promote EXTLOAD:i20 -> EXTLOAD:i24.
     unsigned NewWidth = SrcVT.getStoreSizeInBits();
     EVT NVT = EVT::getIntegerVT(*DAG.getContext(), NewWidth);
     SDValue Ch;
 
     // The extra bits are guaranteed to be zero, since we stored them that
     // way.  A zext load from NVT thus automatically gives zext from SrcVT.
if (TLI.isLittleEndian()) {
 
       // Join the hi and lo parts.
       Value = DAG.getNode(ISD::OR, dl, Node->getValueType(0), Lo, Hi);
     }
 
     Chain = Ch;
   } else {
     bool isCustom = false;
-    switch (TLI.getLoadExtAction(ExtType, SrcVT.getSimpleVT())) {
+    switch (TLI.getLoadExtAction(ExtType, Node->getValueType(0),
+                                 SrcVT.getSimpleVT())) {
     default: llvm_unreachable("This action is not supported yet!");
     case TargetLowering::Custom:
       isCustom = true;
       // FALLTHROUGH
     case TargetLowering::Legal: {
       Value = SDValue(Node, 0);
       Chain = SDValue(Node, 1);
 
case TargetLowering::Legal: {
           if (Align < ABIAlignment){
             ExpandUnalignedLoad(cast<LoadSDNode>(Node), DAG, TLI, Value, Chain);
           }
         }
       }
       break;
     }
     case TargetLowering::Expand:
-      if (!TLI.isLoadExtLegal(ISD::EXTLOAD, SrcVT) && TLI.isTypeLegal(SrcVT)) {
+      if (!TLI.isLoadExtLegal(ISD::EXTLOAD, Node->getValueType(0),
+                              SrcVT) && TLI.isTypeLegal(SrcVT)) {
         SDValue Load = DAG.getLoad(SrcVT, dl, Chain, Ptr, LD->getMemOperand());
         unsigned ExtendOp;
         switch (ExtType) {
         case ISD::EXTLOAD:
           ExtendOp = (SrcVT.isFloatingPoint() ?
                       ISD::FP_EXTEND : ISD::ANY_EXTEND);
           break;
         case ISD::SEXTLOAD: ExtendOp = ISD::SIGN_EXTEND; break;

llvm/trunk/lib/CodeGen/SelectionDAG/LegalizeVectorOps.cpp

for (unsigned i = 0, e = Node->getNumOperands(); i != e; ++i)
     Ops.push_back(LegalizeOp(Node->getOperand(i)));
 
   SDValue Result = SDValue(DAG.UpdateNodeOperands(Op.getNode(), Ops), 0);
 
   if (Op.getOpcode() == ISD::LOAD) {
     LoadSDNode *LD = cast<LoadSDNode>(Op.getNode());
     ISD::LoadExtType ExtType = LD->getExtensionType();
     if (LD->getMemoryVT().isVector() && ExtType != ISD::NON_EXTLOAD)
-      switch (TLI.getLoadExtAction(LD->getExtensionType(), LD->getMemoryVT())) {
+      switch (TLI.getLoadExtAction(LD->getExtensionType(), LD->getValueType(0),
+                                   LD->getMemoryVT())) {
       default: llvm_unreachable("This action is not supported yet!");
       case TargetLowering::Legal:
         return TranslateLegalizeResults(Op, Result);
       case TargetLowering::Custom:
         if (SDValue Lowered = TLI.LowerOperation(Result, DAG)) {
           Changed = true;
           if (Lowered->getNumValues() != Op->getNumValues()) {
             // This expanded to something other than the load. Assume the

llvm/trunk/lib/Target/AArch64/AArch64ISelLowering.cpp

AArch64TargetLowering::AArch64TargetLowering(const TargetMachine &TM)
   // become loads from the constant pool.
   if (Subtarget->isTargetMachO() && TM.getCodeModel() == CodeModel::Large) {
     setOperationAction(ISD::ConstantFP, MVT::f32, Legal);
     setOperationAction(ISD::ConstantFP, MVT::f64, Legal);
   }
 
   // AArch64 does not have floating-point extending loads, i1 sign-extending
   // load, floating-point truncating stores, or v2i32->v2i16 truncating store.
-  setLoadExtAction(ISD::EXTLOAD, MVT::f16, Expand);
-  setLoadExtAction(ISD::EXTLOAD, MVT::f32, Expand);
-  setLoadExtAction(ISD::EXTLOAD, MVT::f64, Expand);
-  setLoadExtAction(ISD::EXTLOAD, MVT::f80, Expand);
-  setLoadExtAction(ISD::SEXTLOAD, MVT::i1, Expand);
+  for (MVT VT : MVT::fp_valuetypes()) {
+    setLoadExtAction(ISD::EXTLOAD, VT, MVT::f16, Expand);
+    setLoadExtAction(ISD::EXTLOAD, VT, MVT::f32, Expand);
+    setLoadExtAction(ISD::EXTLOAD, VT, MVT::f64, Expand);
+    setLoadExtAction(ISD::EXTLOAD, VT, MVT::f80, Expand);
+  }
+  for (MVT VT : MVT::integer_valuetypes())
+    setLoadExtAction(ISD::SEXTLOAD, VT, MVT::i1, Expand);
+
   setTruncStoreAction(MVT::f32, MVT::f16, Expand);
   setTruncStoreAction(MVT::f64, MVT::f32, Expand);
   setTruncStoreAction(MVT::f64, MVT::f16, Expand);
   setTruncStoreAction(MVT::f128, MVT::f80, Expand);
   setTruncStoreAction(MVT::f128, MVT::f64, Expand);
   setTruncStoreAction(MVT::f128, MVT::f32, Expand);
   setTruncStoreAction(MVT::f128, MVT::f16, Expand);
 
for (MVT VT : MVT::vector_valuetypes()) {
 
       setOperationAction(ISD::MULHS, VT, Expand);
       setOperationAction(ISD::SMUL_LOHI, VT, Expand);
       setOperationAction(ISD::MULHU, VT, Expand);
       setOperationAction(ISD::UMUL_LOHI, VT, Expand);
 
       setOperationAction(ISD::BSWAP, VT, Expand);
 
-      for (MVT InnerVT : MVT::vector_valuetypes())
+      for (MVT InnerVT : MVT::vector_valuetypes()) {
         setTruncStoreAction(VT, InnerVT, Expand);
-      setLoadExtAction(ISD::SEXTLOAD, VT, Expand);
-      setLoadExtAction(ISD::ZEXTLOAD, VT, Expand);
-      setLoadExtAction(ISD::EXTLOAD, VT, Expand);
+        setLoadExtAction(ISD::SEXTLOAD, VT, InnerVT, Expand);
+        setLoadExtAction(ISD::ZEXTLOAD, VT, InnerVT, Expand);
+        setLoadExtAction(ISD::EXTLOAD, VT, InnerVT, Expand);
+      }
     }
 
     // AArch64 has implementations of a lot of rounding-like FP operations.
     static MVT RoundingVecTypes[] = {MVT::v2f32, MVT::v4f32, MVT::v2f64 };
     for (unsigned I = 0; I < array_lengthof(RoundingVecTypes); ++I) {
       MVT Ty = RoundingVecTypes[I];
       setOperationAction(ISD::FFLOOR, Ty, Legal);
       setOperationAction(ISD::FNEARBYINT, Ty, Legal);
void AArch64TargetLowering::addTypeForNEON(EVT VT, EVT PromotedBitwiseVT) {
   setOperationAction(ISD::AND, VT.getSimpleVT(), Custom);
   setOperationAction(ISD::OR, VT.getSimpleVT(), Custom);
   setOperationAction(ISD::SETCC, VT.getSimpleVT(), Custom);
   setOperationAction(ISD::CONCAT_VECTORS, VT.getSimpleVT(), Legal);
 
   setOperationAction(ISD::SELECT, VT.getSimpleVT(), Expand);
   setOperationAction(ISD::SELECT_CC, VT.getSimpleVT(), Expand);
   setOperationAction(ISD::VSELECT, VT.getSimpleVT(), Expand);
-  setLoadExtAction(ISD::EXTLOAD, VT.getSimpleVT(), Expand);
+  for (MVT InnerVT : MVT::all_valuetypes())
+    setLoadExtAction(ISD::EXTLOAD, InnerVT, VT.getSimpleVT(), Expand);
 
   // CNT supports only B element sizes.
   if (VT != MVT::v8i8 && VT != MVT::v16i8)
     setOperationAction(ISD::CTPOP, VT.getSimpleVT(), Expand);
 
   setOperationAction(ISD::UDIV, VT.getSimpleVT(), Expand);
   setOperationAction(ISD::SDIV, VT.getSimpleVT(), Expand);
   setOperationAction(ISD::UREM, VT.getSimpleVT(), Expand);

llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp

else
     addRegisterClass(MVT::i32, &ARM::GPRRegClass);
   if (!TM.Options.UseSoftFloat && Subtarget->hasVFP2() &&
       !Subtarget->isThumb1Only()) {
     addRegisterClass(MVT::f32, &ARM::SPRRegClass);
     addRegisterClass(MVT::f64, &ARM::DPRRegClass);
   }
 
   for (MVT VT : MVT::vector_valuetypes()) {
-    for (MVT InnerVT : MVT::vector_valuetypes())
+    for (MVT InnerVT : MVT::vector_valuetypes()) {
       setTruncStoreAction(VT, InnerVT, Expand);
-    setLoadExtAction(ISD::SEXTLOAD, VT, Expand);
-    setLoadExtAction(ISD::ZEXTLOAD, VT, Expand);
-    setLoadExtAction(ISD::EXTLOAD, VT, Expand);
+      setLoadExtAction(ISD::SEXTLOAD, VT, InnerVT, Expand);
+      setLoadExtAction(ISD::ZEXTLOAD, VT, InnerVT, Expand);
+      setLoadExtAction(ISD::EXTLOAD, VT, InnerVT, Expand);
+    }
 
     setOperationAction(ISD::MULHS, VT, Expand);
     setOperationAction(ISD::SMUL_LOHI, VT, Expand);
     setOperationAction(ISD::MULHU, VT, Expand);
     setOperationAction(ISD::UMUL_LOHI, VT, Expand);
 
     setOperationAction(ISD::BSWAP, VT, Expand);
   }
if (Subtarget->hasNEON()) {
     setTargetDAGCombine(ISD::FDIV);
     setTargetDAGCombine(ISD::LOAD);
 
     // It is legal to extload from v4i8 to v4i16 or v4i32.
     MVT Tys[6] = {MVT::v8i8, MVT::v4i8, MVT::v2i8,
                   MVT::v4i16, MVT::v2i16,
                   MVT::v2i32};
     for (unsigned i = 0; i < 6; ++i) {
-      setLoadExtAction(ISD::EXTLOAD, Tys[i], Legal);
-      setLoadExtAction(ISD::ZEXTLOAD, Tys[i], Legal);
-      setLoadExtAction(ISD::SEXTLOAD, Tys[i], Legal);
+      for (MVT VT : MVT::integer_vector_valuetypes()) {
+        setLoadExtAction(ISD::EXTLOAD, VT, Tys[i], Legal);
+        setLoadExtAction(ISD::ZEXTLOAD, VT, Tys[i], Legal);
+        setLoadExtAction(ISD::SEXTLOAD, VT, Tys[i], Legal);
+      }
     }
   }
 
   // ARM and Thumb2 support UMLAL/SMLAL.
   if (!Subtarget->isThumb1Only())
     setTargetDAGCombine(ISD::ADDC);
 
   if (Subtarget->isFPOnlySP()) {
if (Subtarget->isFPOnlySP()) {
     setOperationAction(ISD::FFLOOR,     MVT::f64, Expand);
     setOperationAction(ISD::FP_ROUND,   MVT::f32, Custom);
     setOperationAction(ISD::FP_EXTEND,  MVT::f64, Custom);
   }
 
   computeRegisterProperties();
 
   // ARM does not have floating-point extending loads.
-  setLoadExtAction(ISD::EXTLOAD, MVT::f32, Expand);
-  setLoadExtAction(ISD::EXTLOAD, MVT::f16, Expand);
+  for (MVT VT : MVT::fp_valuetypes()) {
+    setLoadExtAction(ISD::EXTLOAD, VT, MVT::f32, Expand);
+    setLoadExtAction(ISD::EXTLOAD, VT, MVT::f16, Expand);
+  }
 
   // ... or truncating stores
   setTruncStoreAction(MVT::f64, MVT::f32, Expand);
   setTruncStoreAction(MVT::f32, MVT::f16, Expand);
   setTruncStoreAction(MVT::f64, MVT::f16, Expand);
 
   // ARM does not have i1 sign extending load.
-  setLoadExtAction(ISD::SEXTLOAD, MVT::i1, Promote);
+  for (MVT VT : MVT::integer_valuetypes())
+    setLoadExtAction(ISD::SEXTLOAD, VT, MVT::i1, Promote);
 
   // ARM supports all 4 flavors of integer indexed load / store.
   if (!Subtarget->isThumb1Only()) {
     for (unsigned im = (unsigned)ISD::PRE_INC;
          im != (unsigned)ISD::LAST_INDEXED_MODE; ++im) {
       setIndexedLoadAction(im,  MVT::i1,  Legal);
       setIndexedLoadAction(im,  MVT::i8,  Legal);
       setIndexedLoadAction(im,  MVT::i16, Legal);

llvm/trunk/lib/Target/Hexagon/HexagonISelLowering.cpp

HexagonTargetLowering::HexagonTargetLowering(const TargetMachine &targetmachine)
   setIndexedStoreAction(ISD::POST_INC, MVT::i8, Legal);
   setIndexedStoreAction(ISD::POST_INC, MVT::i16, Legal);
   setIndexedStoreAction(ISD::POST_INC, MVT::i32, Legal);
   setIndexedStoreAction(ISD::POST_INC, MVT::i64, Legal);
 
   setOperationAction(ISD::BUILD_PAIR, MVT::i64, Expand);
 
   // Turn FP extload into load/fextend.
-  setLoadExtAction(ISD::EXTLOAD, MVT::f32, Expand);
+  for (MVT VT : MVT::fp_valuetypes())
+    setLoadExtAction(ISD::EXTLOAD, VT, MVT::f32, Expand);
   // Hexagon has a i1 sign extending load.
-  setLoadExtAction(ISD::SEXTLOAD, MVT::i1, Expand);
+  for (MVT VT : MVT::integer_valuetypes())
+    setLoadExtAction(ISD::SEXTLOAD, VT, MVT::i1, Expand);
   // Turn FP truncstore into trunc + store.
   setTruncStoreAction(MVT::f64, MVT::f32, Expand);
 
   // Custom legalize GlobalAddress nodes into CONST32.
   setOperationAction(ISD::GlobalAddress, MVT::i32, Custom);
   setOperationAction(ISD::GlobalAddress, MVT::i8, Custom);
   setOperationAction(ISD::BlockAddress, MVT::i32, Custom);
   // Truncate action?

llvm/trunk/lib/Target/MSP430/MSP430ISelLowering.cpp

MSP430TargetLowering::MSP430TargetLowering(const TargetMachine &TM)
   setStackPointerRegisterToSaveRestore(MSP430::SP);
   setBooleanContents(ZeroOrOneBooleanContent);
   setBooleanVectorContents(ZeroOrOneBooleanContent); // FIXME: Is this correct?
 
   // We have post-incremented loads / stores.
   setIndexedLoadAction(ISD::POST_INC, MVT::i8, Legal);
   setIndexedLoadAction(ISD::POST_INC, MVT::i16, Legal);
 
-  setLoadExtAction(ISD::EXTLOAD,  MVT::i1,  Promote);
-  setLoadExtAction(ISD::SEXTLOAD, MVT::i1,  Promote);
-  setLoadExtAction(ISD::ZEXTLOAD, MVT::i1,  Promote);
-  setLoadExtAction(ISD::SEXTLOAD, MVT::i8,  Expand);
-  setLoadExtAction(ISD::SEXTLOAD, MVT::i16, Expand);
+  for (MVT VT : MVT::integer_valuetypes()) {
+    setLoadExtAction(ISD::EXTLOAD,  VT, MVT::i1,  Promote);
+    setLoadExtAction(ISD::SEXTLOAD, VT, MVT::i1,  Promote);
+    setLoadExtAction(ISD::ZEXTLOAD, VT, MVT::i1,  Promote);
+    setLoadExtAction(ISD::SEXTLOAD, VT, MVT::i8,  Expand);
+    setLoadExtAction(ISD::SEXTLOAD, VT, MVT::i16, Expand);
+  }
 
   // We don't have any truncstores
   setTruncStoreAction(MVT::i16, MVT::i8, Expand);
 
   setOperationAction(ISD::SRA,              MVT::i8,    Custom);
   setOperationAction(ISD::SHL,              MVT::i8,    Custom);
   setOperationAction(ISD::SRL,              MVT::i8,    Custom);
   setOperationAction(ISD::SRA,              MVT::i16,   Custom);

llvm/trunk/lib/Target/Mips/MipsISelLowering.cpp

MipsTargetLowering::MipsTargetLowering(const MipsTargetMachine &TM,
   setBooleanVectorContents(ZeroOrNegativeOneBooleanContent);
   // The cmp.cond.fmt instruction in MIPS32r6/MIPS64r6 uses 0 and -1 like MSA
   // does. Integer booleans still use 0 and 1.
   if (Subtarget.hasMips32r6())
     setBooleanContents(ZeroOrOneBooleanContent,
                        ZeroOrNegativeOneBooleanContent);
 
   // Load extented operations for i1 types must be promoted
-  setLoadExtAction(ISD::EXTLOAD,  MVT::i1,  Promote);
-  setLoadExtAction(ISD::ZEXTLOAD, MVT::i1,  Promote);
-  setLoadExtAction(ISD::SEXTLOAD, MVT::i1,  Promote);
+  for (MVT VT : MVT::integer_valuetypes()) {
+    setLoadExtAction(ISD::EXTLOAD,  VT, MVT::i1,  Promote);
+    setLoadExtAction(ISD::ZEXTLOAD, VT, MVT::i1,  Promote);
+    setLoadExtAction(ISD::SEXTLOAD, VT, MVT::i1,  Promote);
+  }
 
   // MIPS doesn't have extending float->double load/store
-  setLoadExtAction(ISD::EXTLOAD, MVT::f32, Expand);
+  for (MVT VT : MVT::fp_valuetypes())
+    setLoadExtAction(ISD::EXTLOAD, VT, MVT::f32, Expand);
   setTruncStoreAction(MVT::f64, MVT::f32, Expand);
 
   // Used by legalize types to correctly generate the setcc result.
   // Without this, every float setcc comes with a AND/OR with the result,
   // we don't want this, since the fpcmp result goes to a flag register,
   // which is used implicitly by brcond and select operations.
   AddPromotedToType(ISD::SETCC, MVT::i1, MVT::i32);
 
if (!Subtarget.hasMips64())
     setOperationAction(ISD::CTLZ, MVT::i64, Expand);
 
   if (!Subtarget.hasMips32r2())
     setOperationAction(ISD::BSWAP, MVT::i32, Expand);
   if (!Subtarget.hasMips64r2())
     setOperationAction(ISD::BSWAP, MVT::i64, Expand);
 
   if (Subtarget.isGP64bit()) {
-    setLoadExtAction(ISD::SEXTLOAD, MVT::i32, Custom);
-    setLoadExtAction(ISD::ZEXTLOAD, MVT::i32, Custom);
-    setLoadExtAction(ISD::EXTLOAD, MVT::i32, Custom);
+    setLoadExtAction(ISD::SEXTLOAD, MVT::i64, MVT::i32, Custom);
+    setLoadExtAction(ISD::ZEXTLOAD, MVT::i64, MVT::i32, Custom);
+    setLoadExtAction(ISD::EXTLOAD, MVT::i64, MVT::i32, Custom);
     setTruncStoreAction(MVT::i64, MVT::i32, Custom);
   }
 
   setOperationAction(ISD::TRAP, MVT::Other, Legal);
 
   setTargetDAGCombine(ISD::SDIVREM);
   setTargetDAGCombine(ISD::UDIVREM);
   setTargetDAGCombine(ISD::SELECT);

llvm/trunk/lib/Target/Mips/MipsSEISelLowering.cpp

MipsSETargetLowering::MipsSETargetLowering(const MipsTargetMachine &TM,
   addRegisterClass(MVT::i32, &Mips::GPR32RegClass);
 
   if (Subtarget.isGP64bit())
     addRegisterClass(MVT::i64, &Mips::GPR64RegClass);
 
   if (Subtarget.hasDSP() || Subtarget.hasMSA()) {
     // Expand all truncating stores and extending loads.
     for (MVT VT0 : MVT::vector_valuetypes()) {
-      for (MVT VT1 : MVT::vector_valuetypes())
+      for (MVT VT1 : MVT::vector_valuetypes()) {
         setTruncStoreAction(VT0, VT1, Expand);
-
-      setLoadExtAction(ISD::SEXTLOAD, VT0, Expand);
-      setLoadExtAction(ISD::ZEXTLOAD, VT0, Expand);
-      setLoadExtAction(ISD::EXTLOAD, VT0, Expand);
+        setLoadExtAction(ISD::SEXTLOAD, VT0, VT1, Expand);
+        setLoadExtAction(ISD::ZEXTLOAD, VT0, VT1, Expand);
+        setLoadExtAction(ISD::EXTLOAD, VT0, VT1, Expand);
+      }
     }
   }
 
   if (Subtarget.hasDSP()) {
     MVT::SimpleValueType VecTys[2] = {MVT::v2i16, MVT::v4i8};
 
     for (unsigned i = 0; i < array_lengthof(VecTys); ++i) {
       addRegisterClass(VecTys[i], &Mips::DSPRRegClass);

llvm/trunk/lib/Target/NVPTX/NVPTXISelLowering.cpp

NVPTXTargetLowering::NVPTXTargetLowering(const NVPTXTargetMachine &TM)
   setOperationAction(ISD::GlobalAddress, MVT::i32, Custom);
   setOperationAction(ISD::GlobalAddress, MVT::i64, Custom);
 
   // We want to legalize constant related memmove and memcopy
   // intrinsics.
   setOperationAction(ISD::INTRINSIC_W_CHAIN, MVT::Other, Custom);
 
   // Turn FP extload into load/fextend
-  setLoadExtAction(ISD::EXTLOAD, MVT::f16, Expand);
-  setLoadExtAction(ISD::EXTLOAD, MVT::f32, Expand);
+  setLoadExtAction(ISD::EXTLOAD, MVT::f32, MVT::f16, Expand);
+  setLoadExtAction(ISD::EXTLOAD, MVT::f64, MVT::f16, Expand);
+  setLoadExtAction(ISD::EXTLOAD, MVT::f64, MVT::f32, Expand);
   // Turn FP truncstore into trunc + store.
   setTruncStoreAction(MVT::f32, MVT::f16, Expand);
   setTruncStoreAction(MVT::f64, MVT::f16, Expand);
   setTruncStoreAction(MVT::f64, MVT::f32, Expand);
 
   // PTX does not support load / store predicate registers
   setOperationAction(ISD::LOAD, MVT::i1, Custom);
   setOperationAction(ISD::STORE, MVT::i1, Custom);
 
-  setLoadExtAction(ISD::SEXTLOAD, MVT::i1, Promote);
-  setLoadExtAction(ISD::ZEXTLOAD, MVT::i1, Promote);
-  setTruncStoreAction(MVT::i64, MVT::i1, Expand);
-  setTruncStoreAction(MVT::i32, MVT::i1, Expand);
-  setTruncStoreAction(MVT::i16, MVT::i1, Expand);
-  setTruncStoreAction(MVT::i8, MVT::i1, Expand);
+  for (MVT VT : MVT::integer_valuetypes()) {
+    setLoadExtAction(ISD::SEXTLOAD, VT, MVT::i1, Promote);
+    setLoadExtAction(ISD::ZEXTLOAD, VT, MVT::i1, Promote);
+    setTruncStoreAction(VT, MVT::i1, Expand);
+  }
 
   // This is legal in NVPTX
   setOperationAction(ISD::ConstantFP, MVT::f64, Legal);
   setOperationAction(ISD::ConstantFP, MVT::f32, Legal);
 
   // TRAP can be lowered to PTX trap
   setOperationAction(ISD::TRAP, MVT::Other, Legal);
 

llvm/trunk/lib/Target/PowerPC/PPCISelLowering.cpp

PPCTargetLowering::PPCTargetLowering(const PPCTargetMachine &TM)
   setMinStackArgumentAlignment(isPPC64 ? 8:4);
 
   // Set up the register classes.
   addRegisterClass(MVT::i32, &PPC::GPRCRegClass);
   addRegisterClass(MVT::f32, &PPC::F4RCRegClass);
   addRegisterClass(MVT::f64, &PPC::F8RCRegClass);
 
   // PowerPC has an i16 but no i8 (or i1) SEXTLOAD
-  setLoadExtAction(ISD::SEXTLOAD, MVT::i1, Promote);
-  setLoadExtAction(ISD::SEXTLOAD, MVT::i8, Expand);
+  for (MVT VT : MVT::integer_valuetypes()) {
+    setLoadExtAction(ISD::SEXTLOAD, VT, MVT::i1, Promote);
+    setLoadExtAction(ISD::SEXTLOAD, VT, MVT::i8, Expand);
+  }
 
   setTruncStoreAction(MVT::f64, MVT::f32, Expand);
 
   // PowerPC has pre-inc load and store's.
   setIndexedLoadAction(ISD::PRE_INC, MVT::i1, Legal);
   setIndexedLoadAction(ISD::PRE_INC, MVT::i8, Legal);
   setIndexedLoadAction(ISD::PRE_INC, MVT::i16, Legal);
   setIndexedLoadAction(ISD::PRE_INC, MVT::i32, Legal);
if (Subtarget.useCRBits()) {
     // PowerPC does not support direct load / store of condition registers
     setOperationAction(ISD::LOAD, MVT::i1, Custom);
     setOperationAction(ISD::STORE, MVT::i1, Custom);
 
     // FIXME: Remove this once the ANDI glue bug is fixed:
     if (ANDIGlueBug)
       setOperationAction(ISD::TRUNCATE, MVT::i1, Custom);
 
-    setLoadExtAction(ISD::SEXTLOAD, MVT::i1, Promote);
-    setLoadExtAction(ISD::ZEXTLOAD, MVT::i1, Promote);
-    setTruncStoreAction(MVT::i64, MVT::i1, Expand);
-    setTruncStoreAction(MVT::i32, MVT::i1, Expand);
-    setTruncStoreAction(MVT::i16, MVT::i1, Expand);
-    setTruncStoreAction(MVT::i8, MVT::i1, Expand);
+    for (MVT VT : MVT::integer_valuetypes()) {
+      setLoadExtAction(ISD::SEXTLOAD, VT, MVT::i1, Promote);
+      setLoadExtAction(ISD::ZEXTLOAD, VT, MVT::i1, Promote);
+      setTruncStoreAction(VT, MVT::i1, Expand);
+    }
 
     addRegisterClass(MVT::i1, &PPC::CRBITRCRegClass);
   }
 
   // This is used in the ppcf128->int sequence.  Note it has different semantics
   // from FP_ROUND:  that rounds to nearest, this rounds to zero.
   setOperationAction(ISD::FP_ROUND_INREG, MVT::ppcf128, Custom);
 
for (MVT VT : MVT::vector_valuetypes()) {
       setOperationAction(ISD::CTPOP, VT, Expand);
       setOperationAction(ISD::CTLZ, VT, Expand);
       setOperationAction(ISD::CTLZ_ZERO_UNDEF, VT, Expand);
       setOperationAction(ISD::CTTZ, VT, Expand);
       setOperationAction(ISD::CTTZ_ZERO_UNDEF, VT, Expand);
       setOperationAction(ISD::VSELECT, VT, Expand);
       setOperationAction(ISD::SIGN_EXTEND_INREG, VT, Expand);
 
-      for (MVT InnerVT : MVT::vector_valuetypes())
+      for (MVT InnerVT : MVT::vector_valuetypes()) {
         setTruncStoreAction(VT, InnerVT, Expand);
-      setLoadExtAction(ISD::SEXTLOAD, VT, Expand);
-      setLoadExtAction(ISD::ZEXTLOAD, VT, Expand);
-      setLoadExtAction(ISD::EXTLOAD, VT, Expand);
+        setLoadExtAction(ISD::SEXTLOAD, VT, InnerVT, Expand);
+        setLoadExtAction(ISD::ZEXTLOAD, VT, InnerVT, Expand);
+        setLoadExtAction(ISD::EXTLOAD, VT, InnerVT, Expand);
+      }
     }
 
     // We can custom expand all VECTOR_SHUFFLEs to VPERM, others we can handle
     // with merges, splats, etc.
     setOperationAction(ISD::VECTOR_SHUFFLE, MVT::v16i8, Custom);
 
     setOperationAction(ISD::AND   , MVT::v4i32, Legal);
     setOperationAction(ISD::OR    , MVT::v4i32, Legal);

llvm/trunk/lib/Target/R600/AMDGPUISelLowering.cpp

AMDGPUTargetLowering::AMDGPUTargetLowering(TargetMachine &TM) :
   setOperationAction(ISD::CONCAT_VECTORS, MVT::v8f32, Custom);
   setOperationAction(ISD::EXTRACT_SUBVECTOR, MVT::v2f32, Custom);
   setOperationAction(ISD::EXTRACT_SUBVECTOR, MVT::v2i32, Custom);
   setOperationAction(ISD::EXTRACT_SUBVECTOR, MVT::v4f32, Custom);
   setOperationAction(ISD::EXTRACT_SUBVECTOR, MVT::v4i32, Custom);
   setOperationAction(ISD::EXTRACT_SUBVECTOR, MVT::v8f32, Custom);
   setOperationAction(ISD::EXTRACT_SUBVECTOR, MVT::v8i32, Custom);
 
-  setLoadExtAction(ISD::EXTLOAD, MVT::v2i8, Expand);
-  setLoadExtAction(ISD::SEXTLOAD, MVT::v2i8, Expand);
-  setLoadExtAction(ISD::ZEXTLOAD, MVT::v2i8, Expand);
-  setLoadExtAction(ISD::EXTLOAD, MVT::v4i8, Expand);
-  setLoadExtAction(ISD::SEXTLOAD, MVT::v4i8, Expand);
-  setLoadExtAction(ISD::ZEXTLOAD, MVT::v4i8, Expand);
-  setLoadExtAction(ISD::EXTLOAD, MVT::v2i16, Expand);
-  setLoadExtAction(ISD::SEXTLOAD, MVT::v2i16, Expand);
-  setLoadExtAction(ISD::ZEXTLOAD, MVT::v2i16, Expand);
-  setLoadExtAction(ISD::EXTLOAD, MVT::v4i16, Expand);
-  setLoadExtAction(ISD::SEXTLOAD, MVT::v4i16, Expand);
-  setLoadExtAction(ISD::ZEXTLOAD, MVT::v4i16, Expand);
+  for (MVT VT : MVT::integer_vector_valuetypes()) {
+    setLoadExtAction(ISD::EXTLOAD, VT, MVT::v2i8, Expand);
+    setLoadExtAction(ISD::SEXTLOAD, VT, MVT::v2i8, Expand);
+    setLoadExtAction(ISD::ZEXTLOAD, VT, MVT::v2i8, Expand);
+    setLoadExtAction(ISD::EXTLOAD, VT, MVT::v4i8, Expand);
+    setLoadExtAction(ISD::SEXTLOAD, VT, MVT::v4i8, Expand);
+    setLoadExtAction(ISD::ZEXTLOAD, VT, MVT::v4i8, Expand);
+    setLoadExtAction(ISD::EXTLOAD, VT, MVT::v2i16, Expand);
+    setLoadExtAction(ISD::SEXTLOAD, VT, MVT::v2i16, Expand);
+    setLoadExtAction(ISD::ZEXTLOAD, VT, MVT::v2i16, Expand);
+    setLoadExtAction(ISD::EXTLOAD, VT, MVT::v4i16, Expand);
+    setLoadExtAction(ISD::SEXTLOAD, VT, MVT::v4i16, Expand);
+    setLoadExtAction(ISD::ZEXTLOAD, VT, MVT::v4i16, Expand);
+  }
 
   setOperationAction(ISD::BR_CC, MVT::i1, Expand);
 
   if (Subtarget->getGeneration() < AMDGPUSubtarget::SEA_ISLANDS) {
     setOperationAction(ISD::FCEIL, MVT::f64, Custom);
     setOperationAction(ISD::FTRUNC, MVT::f64, Custom);
     setOperationAction(ISD::FRINT, MVT::f64, Custom);
     setOperationAction(ISD::FFLOOR, MVT::f64, Custom);
   }
 
   if (!Subtarget->hasBFI()) {
     // fcopysign can be done in a single instruction with BFI.
     setOperationAction(ISD::FCOPYSIGN, MVT::f32, Expand);
     setOperationAction(ISD::FCOPYSIGN, MVT::f64, Expand);
   }
 
   setOperationAction(ISD::FP16_TO_FP, MVT::f64, Expand);
 
-  setLoadExtAction(ISD::EXTLOAD, MVT::f16, Expand);
+  setLoadExtAction(ISD::EXTLOAD, MVT::f32, MVT::f16, Expand);
+  setLoadExtAction(ISD::EXTLOAD, MVT::f64, MVT::f16, Expand);
   setTruncStoreAction(MVT::f32, MVT::f16, Expand);
   setTruncStoreAction(MVT::f64, MVT::f16, Expand);
 
   const MVT ScalarIntVTs[] = { MVT::i32, MVT::i64 };
   for (MVT VT : ScalarIntVTs) {
     setOperationAction(ISD::SREM, VT, Expand);
     setOperationAction(ISD::SDIV, VT, Expand);
 

llvm/trunk/lib/Target/R600/R600ISelLowering.cpp

R600TargetLowering::R600TargetLowering(TargetMachine &TM) :
 
   // Legalize loads and stores to the private address space.
   setOperationAction(ISD::LOAD, MVT::i32, Custom);
   setOperationAction(ISD::LOAD, MVT::v2i32, Custom);
   setOperationAction(ISD::LOAD, MVT::v4i32, Custom);
 
   // EXTLOAD should be the same as ZEXTLOAD. It is legal for some address
   // spaces, so it is custom lowered to handle those where it isn't.
-  setLoadExtAction(ISD::SEXTLOAD, MVT::i1, Promote);
-  setLoadExtAction(ISD::SEXTLOAD, MVT::i8, Custom);
-  setLoadExtAction(ISD::SEXTLOAD, MVT::i16, Custom);
-
-  setLoadExtAction(ISD::ZEXTLOAD, MVT::i1, Promote);
-  setLoadExtAction(ISD::ZEXTLOAD, MVT::i8, Custom);
-  setLoadExtAction(ISD::ZEXTLOAD, MVT::i16, Custom);
-
-  setLoadExtAction(ISD::EXTLOAD, MVT::i1, Promote);
-  setLoadExtAction(ISD::EXTLOAD, MVT::i8, Custom);
-  setLoadExtAction(ISD::EXTLOAD, MVT::i16, Custom);
+  for (MVT VT : MVT::integer_valuetypes()) {
+    setLoadExtAction(ISD::SEXTLOAD, VT, MVT::i1, Promote);
+    setLoadExtAction(ISD::SEXTLOAD, VT, MVT::i8, Custom);
+    setLoadExtAction(ISD::SEXTLOAD, VT, MVT::i16, Custom);
+
+    setLoadExtAction(ISD::ZEXTLOAD, VT, MVT::i1, Promote);
+    setLoadExtAction(ISD::ZEXTLOAD, VT, MVT::i8, Custom);
+    setLoadExtAction(ISD::ZEXTLOAD, VT, MVT::i16, Custom);
+
+    setLoadExtAction(ISD::EXTLOAD, VT, MVT::i1, Promote);
+    setLoadExtAction(ISD::EXTLOAD, VT, MVT::i8, Custom);
+    setLoadExtAction(ISD::EXTLOAD, VT, MVT::i16, Custom);
+  }
 
   setOperationAction(ISD::STORE, MVT::i8, Custom);
   setOperationAction(ISD::STORE, MVT::i32, Custom);
   setOperationAction(ISD::STORE, MVT::v2i32, Custom);
   setOperationAction(ISD::STORE, MVT::v4i32, Custom);
   setTruncStoreAction(MVT::i32, MVT::i8, Custom);
   setTruncStoreAction(MVT::i32, MVT::i16, Custom);
 

llvm/trunk/lib/Target/R600/SIISelLowering.cpp

SITargetLowering::SITargetLowering(TargetMachine &TM) :
   setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::Other, Custom);
   setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::f32, Custom);
   setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::v16i8, Custom);
   setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::v4f32, Custom);
 
   setOperationAction(ISD::INTRINSIC_VOID, MVT::Other, Custom);
   setOperationAction(ISD::BRCOND, MVT::Other, Custom);
 
-  setLoadExtAction(ISD::SEXTLOAD, MVT::i1, Promote);
-  setLoadExtAction(ISD::SEXTLOAD, MVT::i8, Custom);
-  setLoadExtAction(ISD::SEXTLOAD, MVT::i16, Custom);
-  setLoadExtAction(ISD::SEXTLOAD, MVT::i32, Expand);
-  setLoadExtAction(ISD::SEXTLOAD, MVT::v8i16, Expand);
-  setLoadExtAction(ISD::SEXTLOAD, MVT::v16i16, Expand);
-
-  setLoadExtAction(ISD::ZEXTLOAD, MVT::i1, Promote);
-  setLoadExtAction(ISD::ZEXTLOAD, MVT::i8, Custom);
-  setLoadExtAction(ISD::ZEXTLOAD, MVT::i16, Custom);
-  setLoadExtAction(ISD::ZEXTLOAD, MVT::i32, Expand);
-
-  setLoadExtAction(ISD::EXTLOAD, MVT::i1, Promote);
-  setLoadExtAction(ISD::EXTLOAD, MVT::i8, Custom);
-  setLoadExtAction(ISD::EXTLOAD, MVT::i16, Custom);
-  setLoadExtAction(ISD::EXTLOAD, MVT::i32, Expand);
-  setLoadExtAction(ISD::EXTLOAD, MVT::f32, Expand);
+  for (MVT VT : MVT::integer_valuetypes()) {
+    setLoadExtAction(ISD::SEXTLOAD, VT, MVT::i1, Promote);
+    setLoadExtAction(ISD::SEXTLOAD, VT, MVT::i8, Custom);
+    setLoadExtAction(ISD::SEXTLOAD, VT, MVT::i16, Custom);
+    setLoadExtAction(ISD::SEXTLOAD, VT, MVT::i32, Expand);
+
+    setLoadExtAction(ISD::ZEXTLOAD, VT, MVT::i1, Promote);
+    setLoadExtAction(ISD::ZEXTLOAD, VT, MVT::i8, Custom);
+    setLoadExtAction(ISD::ZEXTLOAD, VT, MVT::i16, Custom);
+    setLoadExtAction(ISD::ZEXTLOAD, VT, MVT::i32, Expand);
+
+    setLoadExtAction(ISD::EXTLOAD, VT, MVT::i1, Promote);
+    setLoadExtAction(ISD::EXTLOAD, VT, MVT::i8, Custom);
+    setLoadExtAction(ISD::EXTLOAD, VT, MVT::i16, Custom);
+    setLoadExtAction(ISD::EXTLOAD, VT, MVT::i32, Expand);
+  }
+
+  for (MVT VT : MVT::integer_vector_valuetypes()) {
+    setLoadExtAction(ISD::SEXTLOAD, VT, MVT::v8i16, Expand);
+    setLoadExtAction(ISD::SEXTLOAD, VT, MVT::v16i16, Expand);
+  }
+
+  for (MVT VT : MVT::fp_valuetypes())
+    setLoadExtAction(ISD::EXTLOAD, VT, MVT::f32, Expand);
 
   setTruncStoreAction(MVT::i32, MVT::i8, Custom);
   setTruncStoreAction(MVT::i32, MVT::i16, Custom);
   setTruncStoreAction(MVT::f64, MVT::f32, Expand);
   setTruncStoreAction(MVT::i64, MVT::i32, Expand);
   setTruncStoreAction(MVT::v8i32, MVT::v8i16, Expand);
   setTruncStoreAction(MVT::v16i32, MVT::v16i16, Expand);
 

llvm/trunk/lib/Target/Sparc/SparcISelLowering.cpp

SparcTargetLowering::SparcTargetLowering(TargetMachine &TM)
   addRegisterClass(MVT::i32, &SP::IntRegsRegClass);
   addRegisterClass(MVT::f32, &SP::FPRegsRegClass);
   addRegisterClass(MVT::f64, &SP::DFPRegsRegClass);
   addRegisterClass(MVT::f128, &SP::QFPRegsRegClass);
   if (Subtarget->is64Bit())
     addRegisterClass(MVT::i64, &SP::I64RegsRegClass);
 
   // Turn FP extload into load/fextend
-  setLoadExtAction(ISD::EXTLOAD, MVT::f32, Expand);
-  setLoadExtAction(ISD::EXTLOAD, MVT::f64, Expand);
+  for (MVT VT : MVT::fp_valuetypes()) {
+    setLoadExtAction(ISD::EXTLOAD, VT, MVT::f32, Expand);
+    setLoadExtAction(ISD::EXTLOAD, VT, MVT::f64, Expand);
+  }
 
   // Sparc doesn't have i1 sign extending load
-  setLoadExtAction(ISD::SEXTLOAD, MVT::i1, Promote);
+  for (MVT VT : MVT::integer_valuetypes())
+    setLoadExtAction(ISD::SEXTLOAD, VT, MVT::i1, Promote);
 
   // Turn FP truncstore into trunc + store.
   setTruncStoreAction(MVT::f64, MVT::f32, Expand);
   setTruncStoreAction(MVT::f128, MVT::f32, Expand);
   setTruncStoreAction(MVT::f128, MVT::f64, Expand);
 
   // Custom legalize GlobalAddress nodes into LO/HI parts.
   setOperationAction(ISD::GlobalAddress, getPointerTy(), Custom);

llvm/trunk/lib/Target/SystemZ/SystemZISelLowering.cpp

SystemZTargetLowering::SystemZTargetLowering(const TargetMachine &tm)
   setOperationAction(ISD::OR, MVT::i64, Custom);
 
   // FIXME: Can we support these natively?
   setOperationAction(ISD::SRL_PARTS, MVT::i64, Expand);
   setOperationAction(ISD::SHL_PARTS, MVT::i64, Expand);
   setOperationAction(ISD::SRA_PARTS, MVT::i64, Expand);
 
   // We have native instructions for i8, i16 and i32 extensions, but not i1.
-  setLoadExtAction(ISD::SEXTLOAD, MVT::i1, Promote);
-  setLoadExtAction(ISD::ZEXTLOAD, MVT::i1, Promote);
-  setLoadExtAction(ISD::EXTLOAD,  MVT::i1, Promote);
   setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i1, Expand);
+  for (MVT VT : MVT::integer_valuetypes()) {
+    setLoadExtAction(ISD::SEXTLOAD, VT, MVT::i1, Promote);
+    setLoadExtAction(ISD::ZEXTLOAD, VT, MVT::i1, Promote);
+    setLoadExtAction(ISD::EXTLOAD,  VT, MVT::i1, Promote);
+  }
 
   // Handle the various types of symbolic address.
   setOperationAction(ISD::ConstantPool,     PtrVT, Custom);
   setOperationAction(ISD::GlobalAddress,    PtrVT, Custom);
   setOperationAction(ISD::GlobalTLSAddress, PtrVT, Custom);
   setOperationAction(ISD::BlockAddress,     PtrVT, Custom);
   setOperationAction(ISD::JumpTable,        PtrVT, Custom);
 
SystemZTargetLowering::SystemZTargetLowering(const TargetMachine &tm)
   // We have fused multiply-addition for f32 and f64 but not f128.
   setOperationAction(ISD::FMA, MVT::f32,  Legal);
   setOperationAction(ISD::FMA, MVT::f64,  Legal);
   setOperationAction(ISD::FMA, MVT::f128, Expand);
 
   // Needed so that we don't try to implement f128 constant loads using
   // a load-and-extend of a f80 constant (in cases where the constant
   // would fit in an f80).
-  setLoadExtAction(ISD::EXTLOAD, MVT::f80, Expand);
+  for (MVT VT : MVT::fp_valuetypes())
+    setLoadExtAction(ISD::EXTLOAD, VT, MVT::f80, Expand);
 
   // Floating-point truncation and stores need to be done separately.
   setTruncStoreAction(MVT::f64,  MVT::f32, Expand);
   setTruncStoreAction(MVT::f128, MVT::f32, Expand);
   setTruncStoreAction(MVT::f128, MVT::f64, Expand);
 
   // We have 64-bit FPR<->GPR moves, but need special handling for
   // 32-bit forms.

llvm/trunk/lib/Target/X86/X86ISelLowering.cpp

void X86TargetLowering::resetOperationActions() {
 
   // Set up the register classes.
   addRegisterClass(MVT::i8, &X86::GR8RegClass);
   addRegisterClass(MVT::i16, &X86::GR16RegClass);
   addRegisterClass(MVT::i32, &X86::GR32RegClass);
   if (Subtarget->is64Bit())
     addRegisterClass(MVT::i64, &X86::GR64RegClass);
 
-  setLoadExtAction(ISD::SEXTLOAD, MVT::i1, Promote);
+  for (MVT VT : MVT::integer_valuetypes())
+    setLoadExtAction(ISD::SEXTLOAD, VT, MVT::i1, Promote);
 
   // We don't accept any truncstore of integer registers.
   setTruncStoreAction(MVT::i64, MVT::i32, Expand);
   setTruncStoreAction(MVT::i64, MVT::i16, Expand);
   setTruncStoreAction(MVT::i64, MVT::i8 , Expand);
   setTruncStoreAction(MVT::i32, MVT::i16, Expand);
   setTruncStoreAction(MVT::i32, MVT::i8 , Expand);
   setTruncStoreAction(MVT::i16, MVT::i8,  Expand);
void X86TargetLowering::resetOperationActions() {
   }
 
   // There's never any support for operations beyond MVT::f32.
   setOperationAction(ISD::FP16_TO_FP, MVT::f64, Expand);
   setOperationAction(ISD::FP16_TO_FP, MVT::f80, Expand);
   setOperationAction(ISD::FP_TO_FP16, MVT::f64, Expand);
   setOperationAction(ISD::FP_TO_FP16, MVT::f80, Expand);
 
-  setLoadExtAction(ISD::EXTLOAD, MVT::f16, Expand);
+  setLoadExtAction(ISD::EXTLOAD, MVT::f32, MVT::f16, Expand);
+  setLoadExtAction(ISD::EXTLOAD, MVT::f64, MVT::f16, Expand);
+  setLoadExtAction(ISD::EXTLOAD, MVT::f80, MVT::f16, Expand);
   setTruncStoreAction(MVT::f32, MVT::f16, Expand);
   setTruncStoreAction(MVT::f64, MVT::f16, Expand);
   setTruncStoreAction(MVT::f80, MVT::f16, Expand);
 
   if (Subtarget->hasPOPCNT()) {
     setOperationAction(ISD::CTPOP          , MVT::i8   , Promote);
   } else {
     setOperationAction(ISD::CTPOP          , MVT::i8   , Expand);
for (MVT VT : MVT::vector_valuetypes()) {
     setOperationAction(ISD::SINT_TO_FP, VT, Expand);
     setOperationAction(ISD::SIGN_EXTEND_INREG, VT,Expand);
     setOperationAction(ISD::TRUNCATE, VT, Expand);
     setOperationAction(ISD::SIGN_EXTEND, VT, Expand);
     setOperationAction(ISD::ZERO_EXTEND, VT, Expand);
     setOperationAction(ISD::ANY_EXTEND, VT, Expand);
     setOperationAction(ISD::VSELECT, VT, Expand);
     setOperationAction(ISD::SELECT_CC, VT, Expand);
-    for (MVT InnerVT : MVT::vector_valuetypes())
-      setTruncStoreAction(VT, InnerVT, Expand);
-    setLoadExtAction(ISD::SEXTLOAD, VT, Expand);
-    setLoadExtAction(ISD::ZEXTLOAD, VT, Expand);
+    for (MVT InnerVT : MVT::vector_valuetypes()) {
+      setTruncStoreAction(InnerVT, VT, Expand);
+
+      setLoadExtAction(ISD::SEXTLOAD, InnerVT, VT, Expand);
+      setLoadExtAction(ISD::ZEXTLOAD, InnerVT, VT, Expand);
 
       // N.b. ISD::EXTLOAD legality is basically ignored except for i1-like types,
       // we have to deal with them whether we ask for Expansion or not. Setting
       // Expand causes its own optimisation problems though, so leave them legal.
       if (VT.getVectorElementType() == MVT::i1)
-      setLoadExtAction(ISD::EXTLOAD, VT, Expand);
+        setLoadExtAction(ISD::EXTLOAD, InnerVT, VT, Expand);
+    }
   }
 
   // FIXME: In order to prevent SSE instructions being expanded to MMX ones
   // with -msoft-float, disable use of MMX as well.
   if (!TM.Options.UseSoftFloat && Subtarget->hasMMX()) {
     addRegisterClass(MVT::x86mmx, &X86::VR64RegClass);
     // No operations on x86mmx supported, everything uses intrinsics.
   }
for (int i = MVT::v16i8; i != MVT::v2i64; ++i) {
       setOperationAction(ISD::VECTOR_SHUFFLE,     VT, Custom);
       setOperationAction(ISD::EXTRACT_VECTOR_ELT, VT, Custom);
     }
 
     // We support custom legalizing of sext and anyext loads for specific
     // memory vector types which we can load as a scalar (or sequence of
     // scalars) and extend in-register to a legal 128-bit vector type. For sext
     // loads these must work with a single scalar load.
-    setLoadExtAction(ISD::SEXTLOAD, MVT::v4i8, Custom);
-    setLoadExtAction(ISD::SEXTLOAD, MVT::v4i16, Custom);
-    setLoadExtAction(ISD::SEXTLOAD, MVT::v8i8, Custom);
-    setLoadExtAction(ISD::EXTLOAD, MVT::v2i8, Custom);
-    setLoadExtAction(ISD::EXTLOAD, MVT::v2i16, Custom);
-    setLoadExtAction(ISD::EXTLOAD, MVT::v2i32, Custom);
-    setLoadExtAction(ISD::EXTLOAD, MVT::v4i8, Custom);
-    setLoadExtAction(ISD::EXTLOAD, MVT::v4i16, Custom);
-    setLoadExtAction(ISD::EXTLOAD, MVT::v8i8, Custom);
+    for (MVT VT : MVT::integer_vector_valuetypes()) {
+      setLoadExtAction(ISD::SEXTLOAD, VT, MVT::v4i8, Custom);
+      setLoadExtAction(ISD::SEXTLOAD, VT, MVT::v4i16, Custom);
+      setLoadExtAction(ISD::SEXTLOAD, VT, MVT::v8i8, Custom);
+      setLoadExtAction(ISD::EXTLOAD, VT, MVT::v2i8, Custom);
+      setLoadExtAction(ISD::EXTLOAD, VT, MVT::v2i16, Custom);
+      setLoadExtAction(ISD::EXTLOAD, VT, MVT::v2i32, Custom);
+      setLoadExtAction(ISD::EXTLOAD, VT, MVT::v4i8, Custom);
+      setLoadExtAction(ISD::EXTLOAD, VT, MVT::v4i16, Custom);
+      setLoadExtAction(ISD::EXTLOAD, VT, MVT::v8i8, Custom);
+    }
 
     setOperationAction(ISD::BUILD_VECTOR,       MVT::v2f64, Custom);
     setOperationAction(ISD::BUILD_VECTOR,       MVT::v2i64, Custom);
     setOperationAction(ISD::VECTOR_SHUFFLE,     MVT::v2f64, Custom);
     setOperationAction(ISD::VECTOR_SHUFFLE,     MVT::v2i64, Custom);
     setOperationAction(ISD::INSERT_VECTOR_ELT,  MVT::v2f64, Custom);
     setOperationAction(ISD::EXTRACT_VECTOR_ELT, MVT::v2f64, Custom);
 
if (!TM.Options.UseSoftFloat && Subtarget->hasSSE2()) {
     // As there is no 64-bit GPR available, we need build a special custom
     // sequence to convert from v2i32 to v2f32.
     if (!Subtarget->is64Bit())
       setOperationAction(ISD::UINT_TO_FP,       MVT::v2f32, Custom);
 
     setOperationAction(ISD::FP_EXTEND,          MVT::v2f32, Custom);
     setOperationAction(ISD::FP_ROUND,           MVT::v2f32, Custom);
 
-    setLoadExtAction(ISD::EXTLOAD,              MVT::v2f32, Legal);
+    for (MVT VT : MVT::fp_vector_valuetypes())
+      setLoadExtAction(ISD::EXTLOAD, VT, MVT::v2f32, Legal);
 
     setOperationAction(ISD::BITCAST,            MVT::v2i32, Custom);
     setOperationAction(ISD::BITCAST,            MVT::v4i16, Custom);
     setOperationAction(ISD::BITCAST,            MVT::v8i8,  Custom);
   }
 
   if (!TM.Options.UseSoftFloat && Subtarget->hasSSE41()) {
     setOperationAction(ISD::FFLOOR,             MVT::f32,   Legal);
if (!TM.Options.UseSoftFloat && Subtarget->hasSSE41()) {
     setOperationAction(ISD::VSELECT,            MVT::v4f32, Custom);
     setOperationAction(ISD::VSELECT,            MVT::v8i16, Custom);
     // There is no BLENDI for byte vectors. We don't need to custom lower
     // some vselects for now.
     setOperationAction(ISD::VSELECT,            MVT::v16i8, Legal);
 
     // SSE41 brings specific instructions for doing vector sign extend even in
     // cases where we don't have SRA.
-    setLoadExtAction(ISD::SEXTLOAD, MVT::v2i8, Custom);
-    setLoadExtAction(ISD::SEXTLOAD, MVT::v2i16, Custom);
-    setLoadExtAction(ISD::SEXTLOAD, MVT::v2i32, Custom);
+    for (MVT VT : MVT::integer_vector_valuetypes()) {
+      setLoadExtAction(ISD::SEXTLOAD, VT, MVT::v2i8, Custom);
+      setLoadExtAction(ISD::SEXTLOAD, VT, MVT::v2i16, Custom);
+      setLoadExtAction(ISD::SEXTLOAD, VT, MVT::v2i32, Custom);
+    }
 
     // i8 and i16 vectors are custom because the source register and source
     // source memory operand types are not the same width.  f32 vectors are
     // custom since the immediate controlling the insert encodes additional
     // information.
     setOperationAction(ISD::INSERT_VECTOR_ELT,  MVT::v16i8, Custom);
     setOperationAction(ISD::INSERT_VECTOR_ELT,  MVT::v8i16, Custom);
     setOperationAction(ISD::INSERT_VECTOR_ELT,  MVT::v4i32, Custom);
if (!TM.Options.UseSoftFloat && Subtarget->hasFp256()) {
 
     setOperationAction(ISD::SINT_TO_FP,         MVT::v8i16, Promote);
     setOperationAction(ISD::SINT_TO_FP,         MVT::v8i32, Legal);
     setOperationAction(ISD::FP_ROUND,           MVT::v4f32, Legal);
 
     setOperationAction(ISD::UINT_TO_FP,         MVT::v8i8,  Custom);
     setOperationAction(ISD::UINT_TO_FP,         MVT::v8i16, Custom);
 
-    setLoadExtAction(ISD::EXTLOAD,              MVT::v4f32, Legal);
+    for (MVT VT : MVT::fp_vector_valuetypes())
+      setLoadExtAction(ISD::EXTLOAD, VT, MVT::v4f32, Legal);
 
     setOperationAction(ISD::SRL,               MVT::v16i16, Custom);
     setOperationAction(ISD::SRL,               MVT::v32i8, Custom);
 
     setOperationAction(ISD::SHL,               MVT::v16i16, Custom);
     setOperationAction(ISD::SHL,               MVT::v32i8, Custom);
 
     setOperationAction(ISD::SRA,               MVT::v16i16, Custom);
if (!TM.Options.UseSoftFloat && Subtarget->hasAVX512()) {
     addRegisterClass(MVT::v16f32, &X86::VR512RegClass);
     addRegisterClass(MVT::v8i64,  &X86::VR512RegClass);
     addRegisterClass(MVT::v8f64,  &X86::VR512RegClass);
 
     addRegisterClass(MVT::i1,     &X86::VK1RegClass);
     addRegisterClass(MVT::v8i1,   &X86::VK8RegClass);
     addRegisterClass(MVT::v16i1,  &X86::VK16RegClass);
 
+    for (MVT VT : MVT::fp_vector_valuetypes())
+      setLoadExtAction(ISD::EXTLOAD, VT, MVT::v8f32, Legal);
+
     setOperationAction(ISD::BR_CC,              MVT::i1,    Expand);
     setOperationAction(ISD::SETCC,              MVT::i1,    Custom);
     setOperationAction(ISD::XOR,                MVT::i1,    Legal);
     setOperationAction(ISD::OR,                 MVT::i1,    Legal);
     setOperationAction(ISD::AND,                MVT::i1,    Legal);
-    setLoadExtAction(ISD::EXTLOAD,              MVT::v8f32, Legal);
     setOperationAction(ISD::LOAD,               MVT::v16f32, Legal);
     setOperationAction(ISD::LOAD,               MVT::v8f64, Legal);
     setOperationAction(ISD::LOAD,               MVT::v8i64, Legal);
     setOperationAction(ISD::LOAD,               MVT::v16i32, Legal);
     setOperationAction(ISD::LOAD,               MVT::v16i1, Legal);
 
     setOperationAction(ISD::FADD,               MVT::v16f32, Legal);
     setOperationAction(ISD::FSUB,               MVT::v16f32, Legal);

llvm/trunk/lib/Target/XCore/XCoreISelLowering.cpp

XCoreTargetLowering::XCoreTargetLowering(const TargetMachine &TM)
 
   setOperationAction(ISD::GlobalAddress, MVT::i32,   Custom);
   setOperationAction(ISD::BlockAddress, MVT::i32 , Custom);
 
   // Conversion of i64 -> double produces constantpool nodes
   setOperationAction(ISD::ConstantPool, MVT::i32,   Custom);
 
   // Loads
-  setLoadExtAction(ISD::EXTLOAD, MVT::i1, Promote);
-  setLoadExtAction(ISD::ZEXTLOAD, MVT::i1, Promote);
-  setLoadExtAction(ISD::SEXTLOAD, MVT::i1, Promote);
+  for (MVT VT : MVT::integer_valuetypes()) {
+    setLoadExtAction(ISD::EXTLOAD, VT, MVT::i1, Promote);
+    setLoadExtAction(ISD::ZEXTLOAD, VT, MVT::i1, Promote);
+    setLoadExtAction(ISD::SEXTLOAD, VT, MVT::i1, Promote);
 
-  setLoadExtAction(ISD::SEXTLOAD, MVT::i8, Expand);
-  setLoadExtAction(ISD::ZEXTLOAD, MVT::i16, Expand);
+    setLoadExtAction(ISD::SEXTLOAD, VT, MVT::i8, Expand);
+    setLoadExtAction(ISD::ZEXTLOAD, VT, MVT::i16, Expand);
+  }
 
   // Custom expand misaligned loads / stores.
   setOperationAction(ISD::LOAD, MVT::i32, Custom);
   setOperationAction(ISD::STORE, MVT::i32, Custom);
 
   // Varargs
   setOperationAction(ISD::VAEND, MVT::Other, Expand);
   setOperationAction(ISD::VACOPY, MVT::Other, Expand);