[PR27599] [SystemZ] [SelectionDAG] Fix extension of atomic cmpxchg result.
Currently, SelectionDAG assumes 8/16-bit cmpxchg returns either a sign
extended result, or a zero extended result. SystemZ takes a third
option by returning junk in the high bits (rotated contents of the other
bytes in the memory word). In that case, don't use Assert*ext, and
zero-extend the result ourselves if a comparison is needed.
Differential Revision: http://reviews.llvm.org/D19800
llvm-svn: 269075
diff --git a/llvm/include/llvm/Target/TargetLowering.h b/llvm/include/llvm/Target/TargetLowering.h
index c47e3c0..b0002b5 100644
--- a/llvm/include/llvm/Target/TargetLowering.h
+++ b/llvm/include/llvm/Target/TargetLowering.h
@@ -1248,9 +1248,10 @@
return nullptr;
}
- /// Returns true if the platform's atomic operations are sign extended.
- virtual bool hasSignExtendedAtomicOps() const {
- return false;
+ /// Returns how the platform's atomic operations are extended (ZERO_EXTEND,
+ /// SIGN_EXTEND, or ANY_EXTEND).
+ virtual ISD::NodeType getExtendForAtomicOps() const {
+ return ISD::ZERO_EXTEND;
}
/// @}
diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
index a06932b..b8918d2 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
@@ -2837,25 +2837,38 @@
cast<AtomicSDNode>(Node)->getFailureOrdering(),
cast<AtomicSDNode>(Node)->getSynchScope());
+ SDValue ExtRes = Res;
SDValue LHS = Res;
SDValue RHS = Node->getOperand(1);
EVT AtomicType = cast<AtomicSDNode>(Node)->getMemoryVT();
EVT OuterType = Node->getValueType(0);
- if (TLI.hasSignExtendedAtomicOps()) {
+ switch (TLI.getExtendForAtomicOps()) {
+ case ISD::SIGN_EXTEND:
LHS = DAG.getNode(ISD::AssertSext, dl, OuterType, Res,
DAG.getValueType(AtomicType));
RHS = DAG.getNode(ISD::SIGN_EXTEND_INREG, dl, OuterType,
Node->getOperand(2), DAG.getValueType(AtomicType));
- } else {
- LHS = DAG.getNode(ISD::AssertZext, dl, OuterType, Res, DAG.getValueType(AtomicType));
+ ExtRes = LHS;
+ break;
+ case ISD::ZERO_EXTEND:
+ LHS = DAG.getNode(ISD::AssertZext, dl, OuterType, Res,
+ DAG.getValueType(AtomicType));
RHS = DAG.getNode(ISD::ZERO_EXTEND, dl, OuterType, Node->getOperand(2));
+ ExtRes = LHS;
+ break;
+ case ISD::ANY_EXTEND:
+ LHS = DAG.getZeroExtendInReg(Res, dl, AtomicType);
+ RHS = DAG.getNode(ISD::ZERO_EXTEND, dl, OuterType, Node->getOperand(2));
+ break;
+ default:
+ llvm_unreachable("Invalid atomic op extension");
}
SDValue Success =
DAG.getSetCC(dl, Node->getValueType(1), LHS, RHS, ISD::SETEQ);
- Results.push_back(LHS.getValue(0));
+ Results.push_back(ExtRes.getValue(0));
Results.push_back(Success);
Results.push_back(Res.getValue(1));
break;
diff --git a/llvm/lib/Target/Mips/MipsISelLowering.h b/llvm/lib/Target/Mips/MipsISelLowering.h
index 31ae235..a56918c 100644
--- a/llvm/lib/Target/Mips/MipsISelLowering.h
+++ b/llvm/lib/Target/Mips/MipsISelLowering.h
@@ -238,8 +238,8 @@
bool isCheapToSpeculateCttz() const override;
bool isCheapToSpeculateCtlz() const override;
- bool hasSignExtendedAtomicOps() const override {
- return true;
+ ISD::NodeType getExtendForAtomicOps() const override {
+ return ISD::SIGN_EXTEND;
}
void LowerOperationWrapper(SDNode *N,
diff --git a/llvm/lib/Target/SystemZ/SystemZISelLowering.h b/llvm/lib/Target/SystemZ/SystemZISelLowering.h
index f26e1dc..bfd0854 100644
--- a/llvm/lib/Target/SystemZ/SystemZISelLowering.h
+++ b/llvm/lib/Target/SystemZ/SystemZISelLowering.h
@@ -459,6 +459,10 @@
SelectionDAG &DAG) const override;
SDValue PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const override;
+ ISD::NodeType getExtendForAtomicOps() const override {
+ return ISD::ANY_EXTEND;
+ }
+
bool supportSwiftError() const override {
return true;
}
diff --git a/llvm/test/CodeGen/SystemZ/cmpxchg-05.ll b/llvm/test/CodeGen/SystemZ/cmpxchg-05.ll
new file mode 100644
index 0000000..68261ef
--- /dev/null
+++ b/llvm/test/CodeGen/SystemZ/cmpxchg-05.ll
@@ -0,0 +1,81 @@
+; Test proper extension of 8-bit/16-bit cmpxchg.
+;
+; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck %s
+
+; CHECK-LABEL: f1
+; CHECK: crjlh
+; CHECK-NOT: llcr
+; CHECK-NOT: cr
+; CHECK: llgcr %r2, [[RES:%r[0-9]+]]
+; CHECK-NOT: llcr
+; CHECK-NOT: cr
+define zeroext i8 @f1(i8* nocapture, i8 zeroext, i8 zeroext) {
+ %cx = cmpxchg i8* %0, i8 %1, i8 %2 seq_cst seq_cst
+ %res = extractvalue { i8, i1 } %cx, 0
+ ret i8 %res
+}
+
+; CHECK-LABEL: f2
+; CHECK: crjlh
+; CHECK-NOT: llhr
+; CHECK-NOT: cr
+; CHECK: llghr %r2, [[RES:%r[0-9]+]]
+; CHECK-NOT: llhr
+; CHECK-NOT: cr
+define zeroext i16 @f2(i16* nocapture, i16 zeroext, i16 zeroext) {
+ %cx = cmpxchg i16* %0, i16 %1, i16 %2 seq_cst seq_cst
+ %res = extractvalue { i16, i1 } %cx, 0
+ ret i16 %res
+}
+
+; CHECK-LABEL: f3
+; CHECK: crjlh
+; CHECK-NOT: llcr
+; CHECK-NOT: cr
+; CHECK: lgbr %r2, [[RES:%r[0-9]+]]
+; CHECK-NOT: llcr
+; CHECK-NOT: cr
+define signext i8 @f3(i8* nocapture, i8 signext, i8 signext) {
+ %cx = cmpxchg i8* %0, i8 %1, i8 %2 seq_cst seq_cst
+ %res = extractvalue { i8, i1 } %cx, 0
+ ret i8 %res
+}
+
+; CHECK-LABEL: f4
+; CHECK: crjlh
+; CHECK-NOT: llhr
+; CHECK-NOT: cr
+; CHECK: lghr %r2, [[RES:%r[0-9]+]]
+; CHECK-NOT: llhr
+; CHECK-NOT: cr
+define signext i16 @f4(i16* nocapture, i16 signext, i16 signext) {
+ %cx = cmpxchg i16* %0, i16 %1, i16 %2 seq_cst seq_cst
+ %res = extractvalue { i16, i1 } %cx, 0
+ ret i16 %res
+}
+
+; Now use the comparison result.
+; CHECK-LABEL: f5
+; CHECK: llcr [[REG:%r[0-9]+]], [[RES:%r[0-9]+]]
+; CHECK: cr [[REG]], %r3
+define zeroext i8 @f5(i8* nocapture, i8 zeroext, i8 zeroext) {
+ %cx = cmpxchg i8* %0, i8 %1, i8 %2 seq_cst seq_cst
+ %res = extractvalue { i8, i1 } %cx, 1
+ %xres = sext i1 %res to i8
+ ret i8 %xres
+}
+
+; Now use the comparison result and zero-extended old value.
+; CHECK-LABEL: f6
+; CHECK: llcr [[REG:%r[0-9]+]], [[RES:%r[0-9]+]]
+; CHECK: st [[REG]], 0(%r5)
+; CHECK: cr [[REG]], %r3
+define zeroext i8 @f6(i8* nocapture, i8 zeroext, i8 zeroext, i32*) {
+ %cx = cmpxchg i8* %0, i8 %1, i8 %2 seq_cst seq_cst
+ %old = extractvalue { i8, i1 } %cx, 0
+ %xold = zext i8 %old to i32
+ store i32 %xold, i32* %3
+ %res = extractvalue { i8, i1 } %cx, 1
+ %xres = sext i1 %res to i8
+ ret i8 %xres
+}