Lower memory barriers to sync instructions.

llvm-svn: 135537
This commit is contained in:
Akira Hatanaka 2011-07-19 23:30:50 +00:00
parent e012858bd6
commit a4c09bce9b
4 changed files with 47 additions and 2 deletions

View File

@ -61,6 +61,7 @@ const char *MipsTargetLowering::getTargetNodeName(unsigned Opcode) const {
case MipsISD::ExtractElementF64: return "MipsISD::ExtractElementF64";
case MipsISD::WrapperPIC: return "MipsISD::WrapperPIC";
case MipsISD::DynAlloc: return "MipsISD::DynAlloc";
case MipsISD::Sync: return "MipsISD::Sync";
default: return NULL;
}
}
@ -159,7 +160,7 @@ MipsTargetLowering(MipsTargetMachine &TM)
// Use the default for now
setOperationAction(ISD::STACKSAVE, MVT::Other, Expand);
setOperationAction(ISD::STACKRESTORE, MVT::Other, Expand);
setOperationAction(ISD::MEMBARRIER, MVT::Other, Expand);
setOperationAction(ISD::MEMBARRIER, MVT::Other, Custom);
if (Subtarget->isSingleFloat())
setOperationAction(ISD::SELECT_CC, MVT::f64, Expand);
@ -527,6 +528,7 @@ LowerOperation(SDValue Op, SelectionDAG &DAG) const
case ISD::VASTART: return LowerVASTART(Op, DAG);
case ISD::FCOPYSIGN: return LowerFCOPYSIGN(Op, DAG);
case ISD::FRAMEADDR: return LowerFRAMEADDR(Op, DAG);
case ISD::MEMBARRIER: return LowerMEMBARRIER(Op, DAG);
}
return SDValue();
}
@ -1525,6 +1527,15 @@ LowerFRAMEADDR(SDValue Op, SelectionDAG &DAG) const {
return FrameAddr;
}
// TODO: set SType according to the desired memory barrier behavior.
SDValue MipsTargetLowering::LowerMEMBARRIER(SDValue Op,
SelectionDAG& DAG) const {
unsigned SType = 0;
DebugLoc dl = Op.getDebugLoc();
return DAG.getNode(MipsISD::Sync, dl, MVT::Other, Op.getOperand(0),
DAG.getConstant(SType, MVT::i32));
}
//===----------------------------------------------------------------------===//
// Calling Convention Implementation
//===----------------------------------------------------------------------===//

View File

@ -81,7 +81,9 @@ namespace llvm {
WrapperPIC,
DynAlloc
DynAlloc,
Sync
};
}
@ -128,6 +130,7 @@ namespace llvm {
SDValue LowerVASTART(SDValue Op, SelectionDAG &DAG) const;
SDValue LowerFCOPYSIGN(SDValue Op, SelectionDAG &DAG) const;
SDValue LowerFRAMEADDR(SDValue Op, SelectionDAG &DAG) const;
SDValue LowerMEMBARRIER(SDValue Op, SelectionDAG& DAG) const;
virtual SDValue
LowerFormalArguments(SDValue Chain,

View File

@ -41,6 +41,7 @@ def SDT_MipsThreadPointer : SDTypeProfile<1, 0, [SDTCisPtrTy<0>]>;
def SDT_MipsDynAlloc : SDTypeProfile<1, 1, [SDTCisVT<0, i32>,
SDTCisVT<1, iPTR>]>;
def SDT_Sync : SDTypeProfile<0, 1, [SDTCisVT<0, i32>]>;
// Call
def MipsJmpLink : SDNode<"MipsISD::JmpLink",SDT_MipsJmpLink,
@ -106,6 +107,8 @@ def MipsWrapperPIC : SDNode<"MipsISD::WrapperPIC", SDTIntUnaryOp>;
def MipsDynAlloc : SDNode<"MipsISD::DynAlloc", SDT_MipsDynAlloc,
[SDNPHasChain, SDNPInGlue]>;
def MipsSync : SDNode<"MipsISD::Sync", SDT_Sync, [SDNPHasChain]>;
//===----------------------------------------------------------------------===//
// Mips Instruction Predicate Definitions.
//===----------------------------------------------------------------------===//
@ -589,6 +592,15 @@ def SB : StoreM<0x28, "sb", truncstorei8>;
def SH : StoreM<0x29, "sh", truncstorei16>;
def SW : StoreM<0x2b, "sw", store>;
let hasSideEffects = 1 in
def SYNC : MipsInst<(outs), (ins i32imm:$stype), "sync $stype",
[(MipsSync imm:$stype)], NoItinerary>
{
let opcode = 0;
let Inst{25-11} = 0;
let Inst{5-0} = 15;
}
/// Load-linked, Store-conditional
let mayLoad = 1, hasDelaySlot = 1 in
def LL : FI<0x30, (outs CPURegs:$dst), (ins mem:$addr),

View File

@ -12,6 +12,7 @@ declare i8 @llvm.atomic.load.nand.i8.p0i8(i8* nocapture, i8) nounwind
declare i8 @llvm.atomic.swap.i8.p0i8(i8* nocapture, i8) nounwind
declare i8 @llvm.atomic.cmp.swap.i8.p0i8(i8* nocapture, i8, i8) nounwind
declare void @llvm.memory.barrier(i1, i1, i1, i1, i1) nounwind
@x = common global i32 0, align 4
@ -239,3 +240,21 @@ entry:
; CHECK: sll $[[R17:[0-9]+]], $[[R16]], 24
; CHECK: sra $2, $[[R17]], 24
}
@countsint = common global i32 0, align 4
define i32 @CheckSync(i32 %v) nounwind noinline {
entry:
tail call void @llvm.memory.barrier(i1 true, i1 true, i1 true, i1 true, i1 true)
%0 = tail call i32 @llvm.atomic.load.add.i32.p0i32(i32* @countsint, i32 %v)
tail call void @llvm.memory.barrier(i1 true, i1 true, i1 true, i1 true, i1 true)
ret i32 %0
; CHECK: CheckSync:
; CHECK: sync 0
; CHECK: ll
; CHECK: sc
; CHECK: beq
; CHECK: sync 0
}