This commit changes:
1. Legalize now always promotes truncstore of i1 to i8.
2. Remove patterns and gunk related to truncstore i1 from targets.
3. Rename the StoreXAction stuff to TruncStoreAction in TLI.
4. Make the TLI TruncStoreAction table a 2d table to handle from/to conversions.
5. Mark a wide variety of invalid truncstores as such in various targets, e.g.
X86 currently doesn't support truncstore of any of its integer types.
6. Add legalize support for truncstores with invalid value input types.
7. Add a dag combine transform to turn store(truncate) into truncstore when
safe.
The later allows us to compile CodeGen/X86/storetrunc-fp.ll to:
_foo:
fldt 20(%esp)
fldt 4(%esp)
faddp %st(1)
movl 36(%esp), %eax
fstps (%eax)
ret
instead of:
_foo:
subl $4, %esp
fldt 24(%esp)
fldt 8(%esp)
faddp %st(1)
fstps (%esp)
movl 40(%esp), %eax
movss (%esp), %xmm0
movss %xmm0, (%eax)
addl $4, %esp
ret
llvm-svn: 46140
diff --git a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index d7a9229..879a410 100644
--- a/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -1690,8 +1690,7 @@
if (N1C && N0.getOpcode() == ISD::LOAD) {
LoadSDNode *LN0 = cast<LoadSDNode>(N0);
if (LN0->getExtensionType() != ISD::SEXTLOAD &&
- LN0->getAddressingMode() == ISD::UNINDEXED &&
- N0.hasOneUse()) {
+ LN0->isUnindexed() && N0.hasOneUse()) {
MVT::ValueType EVT, LoadedVT;
if (N1C->getValue() == 255)
EVT = MVT::i8;
@@ -3810,7 +3809,7 @@
SDOperand Ptr;
MVT::ValueType VT;
if (LoadSDNode *LD = dyn_cast<LoadSDNode>(N)) {
- if (LD->getAddressingMode() != ISD::UNINDEXED)
+ if (LD->isIndexed())
return false;
VT = LD->getLoadedVT();
if (!TLI.isIndexedLoadLegal(ISD::PRE_INC, VT) &&
@@ -3818,7 +3817,7 @@
return false;
Ptr = LD->getBasePtr();
} else if (StoreSDNode *ST = dyn_cast<StoreSDNode>(N)) {
- if (ST->getAddressingMode() != ISD::UNINDEXED)
+ if (ST->isIndexed())
return false;
VT = ST->getStoredVT();
if (!TLI.isIndexedStoreLegal(ISD::PRE_INC, VT) &&
@@ -3937,7 +3936,7 @@
SDOperand Ptr;
MVT::ValueType VT;
if (LoadSDNode *LD = dyn_cast<LoadSDNode>(N)) {
- if (LD->getAddressingMode() != ISD::UNINDEXED)
+ if (LD->isIndexed())
return false;
VT = LD->getLoadedVT();
if (!TLI.isIndexedLoadLegal(ISD::POST_INC, VT) &&
@@ -3945,7 +3944,7 @@
return false;
Ptr = LD->getBasePtr();
} else if (StoreSDNode *ST = dyn_cast<StoreSDNode>(N)) {
- if (ST->getAddressingMode() != ISD::UNINDEXED)
+ if (ST->isIndexed())
return false;
VT = ST->getStoredVT();
if (!TLI.isIndexedStoreLegal(ISD::POST_INC, VT) &&
@@ -4187,7 +4186,7 @@
// If this is a store of a bit convert, store the input value if the
// resultant store does not need a higher alignment than the original.
if (Value.getOpcode() == ISD::BIT_CONVERT && !ST->isTruncatingStore() &&
- ST->getAddressingMode() == ISD::UNINDEXED) {
+ ST->isUnindexed()) {
unsigned Align = ST->getAlignment();
MVT::ValueType SVT = Value.getOperand(0).getValueType();
unsigned OrigAlign = TLI.getTargetMachine().getTargetData()->
@@ -4285,7 +4284,7 @@
return SDOperand(N, 0);
// FIXME: is there such a thing as a truncating indexed store?
- if (ST->isTruncatingStore() && ST->getAddressingMode() == ISD::UNINDEXED &&
+ if (ST->isTruncatingStore() && ST->isUnindexed() &&
MVT::isInteger(Value.getValueType())) {
// See if we can simplify the input to this truncstore with knowledge that
// only the low bits are being used. For example:
@@ -4308,8 +4307,7 @@
// is dead/noop.
if (LoadSDNode *Ld = dyn_cast<LoadSDNode>(Value)) {
if (Ld->getBasePtr() == Ptr && ST->getStoredVT() == Ld->getLoadedVT() &&
- ST->getAddressingMode() == ISD::UNINDEXED &&
- !ST->isVolatile() &&
+ ST->isUnindexed() && !ST->isVolatile() &&
// There can't be any side effects between the load and store, such as
// a call or store.
Chain.reachesChainWithoutSideEffects(SDOperand(Ld, 1))) {
@@ -4318,6 +4316,18 @@
}
}
+ // If this is an FP_ROUND or TRUNC followed by a store, fold this into a
+ // truncating store. We can do this even if this is already a truncstore.
+ if ((Value.getOpcode() == ISD::FP_ROUND || Value.getOpcode() == ISD::TRUNCATE)
+ && TLI.isTypeLegal(Value.getOperand(0).getValueType()) &&
+ Value.Val->hasOneUse() && ST->isUnindexed() &&
+ TLI.isTruncStoreLegal(Value.getOperand(0).getValueType(),
+ ST->getStoredVT())) {
+ return DAG.getTruncStore(Chain, Value.getOperand(0), Ptr, ST->getSrcValue(),
+ ST->getSrcValueOffset(), ST->getStoredVT(),
+ ST->isVolatile(), ST->getAlignment());
+ }
+
return SDOperand();
}
diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
index 18d3a6d..8d7c85c 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
@@ -2239,15 +2239,24 @@
break;
}
} else {
- // Truncating store
- assert(isTypeLegal(ST->getValue().getValueType()) &&
- "Cannot handle illegal TRUNCSTORE yet!");
- Tmp3 = LegalizeOp(ST->getValue());
+ switch (getTypeAction(ST->getValue().getValueType())) {
+ case Legal:
+ Tmp3 = LegalizeOp(ST->getValue());
+ break;
+ case Promote:
+ // We can promote the value, the truncstore will still take care of it.
+ Tmp3 = PromoteOp(ST->getValue());
+ break;
+ case Expand:
+ // Just store the low part. This may become a non-trunc store, so make
+ // sure to use getTruncStore, not UpdateNodeOperands below.
+ ExpandOp(ST->getValue(), Tmp3, Tmp4);
+ return DAG.getTruncStore(Tmp1, Tmp3, Tmp2, ST->getSrcValue(),
+ SVOffset, MVT::i8, isVolatile, Alignment);
+ }
- // The only promote case we handle is TRUNCSTORE:i1 X into
- // -> TRUNCSTORE:i8 (and X, 1)
- if (ST->getStoredVT() == MVT::i1 &&
- TLI.getStoreXAction(MVT::i1) == TargetLowering::Promote) {
+ // Unconditionally promote TRUNCSTORE:i1 X -> TRUNCSTORE:i8 (and X, 1)
+ if (ST->getStoredVT() == MVT::i1) {
// Promote the bool to a mask then store.
Tmp3 = DAG.getNode(ISD::AND, Tmp3.getValueType(), Tmp3,
DAG.getConstant(1, Tmp3.getValueType()));
@@ -2261,7 +2270,7 @@
}
MVT::ValueType StVT = cast<StoreSDNode>(Result.Val)->getStoredVT();
- switch (TLI.getStoreXAction(StVT)) {
+ switch (TLI.getTruncStoreAction(ST->getValue().getValueType(), StVT)) {
default: assert(0 && "This action is not supported yet!");
case TargetLowering::Legal:
// If this is an unaligned store and the target doesn't support it,
@@ -2275,8 +2284,7 @@
}
break;
case TargetLowering::Custom:
- Tmp1 = TLI.LowerOperation(Result, DAG);
- if (Tmp1.Val) Result = Tmp1;
+ Result = TLI.LowerOperation(Result, DAG);
break;
}
}
diff --git a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
index cb12b7c..8e7e87d 100644
--- a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -158,7 +158,7 @@
// All operations default to being supported.
memset(OpActions, 0, sizeof(OpActions));
memset(LoadXActions, 0, sizeof(LoadXActions));
- memset(&StoreXActions, 0, sizeof(StoreXActions));
+ memset(TruncStoreActions, 0, sizeof(TruncStoreActions));
memset(&IndexedModeActions, 0, sizeof(IndexedModeActions));
memset(&ConvertActions, 0, sizeof(ConvertActions));
diff --git a/llvm/lib/Target/ARM/ARMISelLowering.cpp b/llvm/lib/Target/ARM/ARMISelLowering.cpp
index 4d8a237..bdaf6b2 100644
--- a/llvm/lib/Target/ARM/ARMISelLowering.cpp
+++ b/llvm/lib/Target/ARM/ARMISelLowering.cpp
@@ -125,6 +125,8 @@
if (!UseSoftFloat && Subtarget->hasVFP2() && !Subtarget->isThumb()) {
addRegisterClass(MVT::f32, ARM::SPRRegisterClass);
addRegisterClass(MVT::f64, ARM::DPRRegisterClass);
+
+ setTruncStoreAction(MVT::f64, MVT::f32, Expand);
}
computeRegisterProperties();
diff --git a/llvm/lib/Target/ARM/ARMInstrInfo.td b/llvm/lib/Target/ARM/ARMInstrInfo.td
index 1027b6c..d3bce5d 100644
--- a/llvm/lib/Target/ARM/ARMInstrInfo.td
+++ b/llvm/lib/Target/ARM/ARMInstrInfo.td
@@ -1404,14 +1404,6 @@
def : ARMPat<(extloadi8 addrmode2:$addr), (LDRB addrmode2:$addr)>;
def : ARMPat<(extloadi16 addrmode3:$addr), (LDRH addrmode3:$addr)>;
-// truncstore i1 -> truncstore i8
-def : ARMPat<(truncstorei1 GPR:$src, addrmode2:$dst),
- (STRB GPR:$src, addrmode2:$dst)>;
-def : ARMPat<(pre_truncsti1 GPR:$src, GPR:$base, am2offset:$offset),
- (STRB_PRE GPR:$src, GPR:$base, am2offset:$offset)>;
-def : ARMPat<(post_truncsti1 GPR:$src, GPR:$base, am2offset:$offset),
- (STRB_POST GPR:$src, GPR:$base, am2offset:$offset)>;
-
// smul* and smla*
def : ARMV5TEPat<(mul (sra (shl GPR:$a, 16), 16), (sra (shl GPR:$b, 16), 16)),
(SMULBB GPR:$a, GPR:$b)>;
diff --git a/llvm/lib/Target/ARM/ARMInstrThumb.td b/llvm/lib/Target/ARM/ARMInstrThumb.td
index d6fa76c..7880c23 100644
--- a/llvm/lib/Target/ARM/ARMInstrThumb.td
+++ b/llvm/lib/Target/ARM/ARMInstrThumb.td
@@ -588,10 +588,6 @@
def : ThumbPat<(extloadi8 t_addrmode_s1:$addr), (tLDRB t_addrmode_s1:$addr)>;
def : ThumbPat<(extloadi16 t_addrmode_s2:$addr), (tLDRH t_addrmode_s2:$addr)>;
-// truncstore i1 -> truncstore i8
-def : ThumbPat<(truncstorei1 GPR:$src, t_addrmode_s1:$dst),
- (tSTRB GPR:$src, t_addrmode_s1:$dst)>;
-
// Large immediate handling.
// Two piece imms.
diff --git a/llvm/lib/Target/Alpha/AlphaISelLowering.cpp b/llvm/lib/Target/Alpha/AlphaISelLowering.cpp
index a37e886..c359a5b 100644
--- a/llvm/lib/Target/Alpha/AlphaISelLowering.cpp
+++ b/llvm/lib/Target/Alpha/AlphaISelLowering.cpp
@@ -59,8 +59,6 @@
setLoadXAction(ISD::SEXTLOAD, MVT::i8, Expand);
setLoadXAction(ISD::SEXTLOAD, MVT::i16, Expand);
- setStoreXAction(MVT::i1, Promote);
-
// setOperationAction(ISD::BRIND, MVT::Other, Expand);
setOperationAction(ISD::BR_JT, MVT::Other, Expand);
setOperationAction(ISD::BR_CC, MVT::Other, Expand);
diff --git a/llvm/lib/Target/CellSPU/SPUISelLowering.cpp b/llvm/lib/Target/CellSPU/SPUISelLowering.cpp
index ed6dfe0..e2a1b43 100644
--- a/llvm/lib/Target/CellSPU/SPUISelLowering.cpp
+++ b/llvm/lib/Target/CellSPU/SPUISelLowering.cpp
@@ -129,13 +129,21 @@
setLoadXAction(ISD::EXTLOAD, MVT::i1, Custom);
setLoadXAction(ISD::SEXTLOAD, MVT::i1, Promote);
setLoadXAction(ISD::ZEXTLOAD, MVT::i1, Promote);
- setStoreXAction(MVT::i1, Custom);
+ setTruncStoreAction(MVT::i8, MVT::i1, Custom);
+ setTruncStoreAction(MVT::i16, MVT::i1, Custom);
+ setTruncStoreAction(MVT::i32, MVT::i1, Custom);
+ setTruncStoreAction(MVT::i64, MVT::i1, Custom);
+ setTruncStoreAction(MVT::i128, MVT::i1, Custom);
setLoadXAction(ISD::EXTLOAD, MVT::i8, Custom);
setLoadXAction(ISD::SEXTLOAD, MVT::i8, Custom);
setLoadXAction(ISD::ZEXTLOAD, MVT::i8, Custom);
- setStoreXAction(MVT::i8, Custom);
-
+ setTruncStoreAction(MVT::i8 , MVT::i8, Custom);
+ setTruncStoreAction(MVT::i16 , MVT::i8, Custom);
+ setTruncStoreAction(MVT::i32 , MVT::i8, Custom);
+ setTruncStoreAction(MVT::i64 , MVT::i8, Custom);
+ setTruncStoreAction(MVT::i128, MVT::i8, Custom);
+
setLoadXAction(ISD::EXTLOAD, MVT::i16, Custom);
setLoadXAction(ISD::SEXTLOAD, MVT::i16, Custom);
setLoadXAction(ISD::ZEXTLOAD, MVT::i16, Custom);
diff --git a/llvm/lib/Target/Mips/MipsISelLowering.cpp b/llvm/lib/Target/Mips/MipsISelLowering.cpp
index 56527eb..fa2f72c 100644
--- a/llvm/lib/Target/Mips/MipsISelLowering.cpp
+++ b/llvm/lib/Target/Mips/MipsISelLowering.cpp
@@ -72,9 +72,6 @@
setLoadXAction(ISD::ZEXTLOAD, MVT::i1, Promote);
setLoadXAction(ISD::SEXTLOAD, MVT::i1, Promote);
- // Store operations for i1 types must be promoted
- setStoreXAction(MVT::i1, Promote);
-
// Mips does not have these NodeTypes below.
setOperationAction(ISD::BR_JT, MVT::Other, Expand);
setOperationAction(ISD::BR_CC, MVT::Other, Expand);
diff --git a/llvm/lib/Target/Mips/MipsInstrInfo.td b/llvm/lib/Target/Mips/MipsInstrInfo.td
index 43313e2..9b9e4f7 100644
--- a/llvm/lib/Target/Mips/MipsInstrInfo.td
+++ b/llvm/lib/Target/Mips/MipsInstrInfo.td
@@ -534,8 +534,6 @@
def : Pat<(i32 (extloadi1 addr:$src)), (LBu addr:$src)>;
def : Pat<(i32 (extloadi8 addr:$src)), (LBu addr:$src)>;
def : Pat<(i32 (extloadi16 addr:$src)), (LHu addr:$src)>;
-def : Pat<(truncstorei1 CPURegs:$src, addr:$addr),
- (SB CPURegs:$src, addr:$addr)>;
// some peepholes
def : Pat<(store (i32 0), addr:$dst), (SW ZERO, addr:$dst)>;
diff --git a/llvm/lib/Target/PowerPC/PPCISelLowering.cpp b/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
index af9bb79..6a35baf 100644
--- a/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
+++ b/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
@@ -54,10 +54,9 @@
// PowerPC has an i16 but no i8 (or i1) SEXTLOAD
setLoadXAction(ISD::SEXTLOAD, MVT::i1, Expand);
setLoadXAction(ISD::SEXTLOAD, MVT::i8, Expand);
-
- // PowerPC does not have truncstore for i1.
- setStoreXAction(MVT::i1, Promote);
-
+
+ setTruncStoreAction(MVT::f64, MVT::f32, Expand);
+
// PowerPC has pre-inc load and store's.
setIndexedLoadAction(ISD::PRE_INC, MVT::i1, Legal);
setIndexedLoadAction(ISD::PRE_INC, MVT::i8, Legal);
diff --git a/llvm/lib/Target/Sparc/SparcInstrInfo.td b/llvm/lib/Target/Sparc/SparcInstrInfo.td
index 061cfab..71fa887 100644
--- a/llvm/lib/Target/Sparc/SparcInstrInfo.td
+++ b/llvm/lib/Target/Sparc/SparcInstrInfo.td
@@ -774,9 +774,3 @@
// zextload bool -> zextload byte
def : Pat<(i32 (zextloadi1 ADDRrr:$src)), (LDUBrr ADDRrr:$src)>;
def : Pat<(i32 (zextloadi1 ADDRri:$src)), (LDUBri ADDRri:$src)>;
-
-// truncstore bool -> truncstore byte.
-def : Pat<(truncstorei1 IntRegs:$src, ADDRrr:$addr),
- (STBrr ADDRrr:$addr, IntRegs:$src)>;
-def : Pat<(truncstorei1 IntRegs:$src, ADDRri:$addr),
- (STBri ADDRri:$addr, IntRegs:$src)>;
diff --git a/llvm/lib/Target/TargetSelectionDAG.td b/llvm/lib/Target/TargetSelectionDAG.td
index 696f815..2560d86 100644
--- a/llvm/lib/Target/TargetSelectionDAG.td
+++ b/llvm/lib/Target/TargetSelectionDAG.td
@@ -551,13 +551,6 @@
}]>;
// truncstore fragments.
-def truncstorei1 : PatFrag<(ops node:$val, node:$ptr),
- (st node:$val, node:$ptr), [{
- if (StoreSDNode *ST = dyn_cast<StoreSDNode>(N))
- return ST->isTruncatingStore() && ST->getStoredVT() == MVT::i1 &&
- ST->getAddressingMode() == ISD::UNINDEXED;
- return false;
-}]>;
def truncstorei8 : PatFrag<(ops node:$val, node:$ptr),
(st node:$val, node:$ptr), [{
if (StoreSDNode *ST = dyn_cast<StoreSDNode>(N))
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 2b58826..bd1ee55 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -82,6 +82,14 @@
setLoadXAction(ISD::SEXTLOAD, MVT::i1, Expand);
+ // We don't accept any truncstore of integer registers.
+ setTruncStoreAction(MVT::i64, MVT::i32, Expand);
+ setTruncStoreAction(MVT::i64, MVT::i16, Expand);
+ setTruncStoreAction(MVT::i64, MVT::i8 , Expand);
+ setTruncStoreAction(MVT::i32, MVT::i16, Expand);
+ setTruncStoreAction(MVT::i32, MVT::i8 , Expand);
+ setTruncStoreAction(MVT::i16, MVT::i8, Expand);
+
// Promote all UINT_TO_FP to larger SINT_TO_FP's, as X86 doesn't have this
// operation.
setOperationAction(ISD::UINT_TO_FP , MVT::i1 , Promote);
@@ -638,6 +646,8 @@
AddPromotedToType (ISD::SELECT, (MVT::ValueType)VT, MVT::v2i64);
}
+ setTruncStoreAction(MVT::f64, MVT::f32, Expand);
+
// Custom lower v2i64 and v2f64 selects.
setOperationAction(ISD::LOAD, MVT::v2f64, Legal);
setOperationAction(ISD::LOAD, MVT::v2i64, Legal);
diff --git a/llvm/lib/Target/X86/X86InstrInfo.td b/llvm/lib/Target/X86/X86InstrInfo.td
index de705de..f5a5803 100644
--- a/llvm/lib/Target/X86/X86InstrInfo.td
+++ b/llvm/lib/Target/X86/X86InstrInfo.td
@@ -2598,11 +2598,6 @@
def : Pat<(subc GR32:$src1, i32immSExt8:$src2),
(SUB32ri8 GR32:$src1, i32immSExt8:$src2)>;
-def : Pat<(truncstorei1 (i8 imm:$src), addr:$dst),
- (MOV8mi addr:$dst, imm:$src)>;
-def : Pat<(truncstorei1 GR8:$src, addr:$dst),
- (MOV8mr addr:$dst, GR8:$src)>;
-
// Comparisons.
// TEST R,R is smaller than CMP R,0