// TRAP - Trapping instruction
TRAP,
+ // PREFETCH - This corresponds to a prefetch intrinsic. It takes chains are
+ // their first operand. The other operands are the address to prefetch,
+ // read / write specifier, and locality specifier.
+ PREFETCH,
+
// OUTCHAIN = MEMBARRIER(INCHAIN, load-load, load-store, store-load,
// store-store, device)
// This corresponds to the memory.barrier intrinsic.
}
break;
+ case ISD::PREFETCH:
+ assert(Node->getNumOperands() == 4 && "Invalid Prefetch node!");
+ switch (TLI.getOperationAction(ISD::PREFETCH, MVT::Other)) {
+ default: assert(0 && "This action is not supported yet!");
+ case TargetLowering::Legal:
+ Tmp1 = LegalizeOp(Node->getOperand(0)); // Legalize the chain.
+ Tmp2 = LegalizeOp(Node->getOperand(1)); // Legalize the address.
+ Tmp3 = LegalizeOp(Node->getOperand(2)); // Legalize the rw specifier.
+ Tmp4 = LegalizeOp(Node->getOperand(3)); // Legalize locality specifier.
+ Result = DAG.UpdateNodeOperands(Result, Tmp1, Tmp2, Tmp3, Tmp4);
+ break;
+ case TargetLowering::Expand:
+ // It's a noop.
+ Result = LegalizeOp(Node->getOperand(0));
+ break;
+ }
+ break;
+
case ISD::MEMBARRIER: {
assert(Node->getNumOperands() == 6 && "Invalid MemBarrier node!");
switch (TLI.getOperationAction(ISD::MEMBARRIER, MVT::Other)) {
return "<<Unknown Target Node>>";
}
+ case ISD::PREFETCH: return "Prefetch";
case ISD::MEMBARRIER: return "MemBarrier";
case ISD::ATOMIC_LCS: return "AtomicLCS";
case ISD::ATOMIC_LAS: return "AtomicLAS";
DAG.setRoot(DAG.getNode(ISD::STACKRESTORE, MVT::Other, getRoot(), Tmp));
return 0;
}
- case Intrinsic::prefetch:
- // FIXME: Currently discarding prefetches.
- return 0;
-
case Intrinsic::var_annotation:
// Discard annotate attributes
return 0;
DAG.setRoot(DAG.getNode(ISD::TRAP, MVT::Other, getRoot()));
return 0;
}
+ case Intrinsic::prefetch: {
+ SDOperand Ops[4];
+ Ops[0] = getRoot();
+ Ops[1] = getValue(I.getOperand(1));
+ Ops[2] = getValue(I.getOperand(2));
+ Ops[3] = getValue(I.getOperand(3));
+ DAG.setRoot(DAG.getNode(ISD::PREFETCH, MVT::Other, &Ops[0], 4));
+ return 0;
+ }
+
case Intrinsic::memory_barrier: {
SDOperand Ops[6];
Ops[0] = getRoot();
setOperationAction(ISD::STACKRESTORE, MVT::Other, Expand);
setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32 , Expand);
setOperationAction(ISD::MEMBARRIER , MVT::Other, Expand);
+ setOperationAction(ISD::PREFETCH , MVT::Other, Expand);
if (!Subtarget->hasV6Ops()) {
setOperationAction(ISD::SIGN_EXTEND_INREG, MVT::i16, Expand);
setOperationAction(ISD::STACKSAVE, MVT::Other, Expand);
setOperationAction(ISD::STACKRESTORE, MVT::Other, Expand);
setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i64, Expand);
+ setOperationAction(ISD::PREFETCH, MVT::Other, Expand);
// We want to legalize GlobalAddress and ConstantPool and
// ExternalSymbols nodes into the appropriate instructions to
setOperationAction(ISD::STACKRESTORE , MVT::Other, Expand);
setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32 , Expand);
setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i64 , Expand);
+ setOperationAction(ISD::PREFETCH , MVT::Other, Expand);
// Cell SPU has instructions for converting between i64 and fp.
setOperationAction(ISD::FP_TO_SINT, MVT::i64, Custom);
setOperationAction(ISD::STACKSAVE, MVT::Other, Expand);
setOperationAction(ISD::STACKRESTORE, MVT::Other, Expand);
setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i64, Expand);
+ setOperationAction(ISD::PREFETCH , MVT::Other, Expand);
// Thread Local Storage
setOperationAction(ISD::GlobalTLSAddress, MVT::i64, Custom);
setOperationAction(ISD::MEMSET, MVT::Other, Expand);
setOperationAction(ISD::MEMCPY, MVT::Other, Expand);
setOperationAction(ISD::MEMBARRIER, MVT::Other, Expand);
+ setOperationAction(ISD::PREFETCH, MVT::Other, Expand);
setOperationAction(ISD::CTPOP, MVT::i32, Expand);
setOperationAction(ISD::CTTZ , MVT::i32, Expand);
setOperationAction(ISD::MEMSET, MVT::Other, Expand);
setOperationAction(ISD::MEMCPY, MVT::Other, Expand);
setOperationAction(ISD::MEMBARRIER, MVT::Other, Expand);
+ setOperationAction(ISD::PREFETCH, MVT::Other, Expand);
// PowerPC has no SREM/UREM instructions
setOperationAction(ISD::SREM, MVT::i32, Expand);
setOperationAction(ISD::MEMSET, MVT::Other, Expand);
setOperationAction(ISD::MEMCPY, MVT::Other, Expand);
setOperationAction(ISD::MEMBARRIER, MVT::Other, Expand);
+ setOperationAction(ISD::PREFETCH, MVT::Other, Expand);
setOperationAction(ISD::FSIN , MVT::f64, Expand);
setOperationAction(ISD::FCOS , MVT::f64, Expand);
SDTCisEltOfVec<2, 1>, SDTCisSameAs<0, 1>, SDTCisPtrTy<3>
]>;
-def STDMemBarrier : SDTypeProfile<0, 5, [
+def STDPrefetch : SDTypeProfile<0, 3, [ // prefetch
+ SDTCisPtrTy<0>, SDTCisSameAs<1, 2>, SDTCisInt<1>
+]>;
+
+def STDMemBarrier : SDTypeProfile<0, 5, [ // memory barier
SDTCisSameAs<0,1>, SDTCisSameAs<0,2>, SDTCisSameAs<0,3>, SDTCisSameAs<0,4>,
SDTCisInt<0>
]>;
def ret : SDNode<"ISD::RET" , SDTNone, [SDNPHasChain]>;
def trap : SDNode<"ISD::TRAP" , SDTNone,
[SDNPHasChain, SDNPSideEffect]>;
-def membarrier : SDNode<"ISD::MEMBARRIER" , STDMemBarrier,
+
+def prefetch : SDNode<"ISD::PREFETCH" , STDPrefetch,
+ [SDNPHasChain, SDNPMayLoad, SDNPMayStore]>;
+
+def membarrier : SDNode<"ISD::MEMBARRIER" , STDMemBarrier,
[SDNPHasChain, SDNPSideEffect]>;
+
// Do not use atomic_* directly, use atomic_*_size (see below)
-def atomic_lcs : SDNode<"ISD::ATOMIC_LCS", STDAtomic3,
- [SDNPHasChain, SDNPMayStore, SDNPMayLoad]>;
-def atomic_las : SDNode<"ISD::ATOMIC_LAS", STDAtomic2,
- [SDNPHasChain, SDNPMayStore, SDNPMayLoad]>;
+def atomic_lcs : SDNode<"ISD::ATOMIC_LCS" , STDAtomic3,
+ [SDNPHasChain, SDNPMayStore, SDNPMayLoad]>;
+def atomic_las : SDNode<"ISD::ATOMIC_LAS" , STDAtomic2,
+ [SDNPHasChain, SDNPMayStore, SDNPMayLoad]>;
def atomic_swap : SDNode<"ISD::ATOMIC_SWAP", STDAtomic2,
- [SDNPHasChain, SDNPMayStore, SDNPMayLoad]>;
+ [SDNPHasChain, SDNPMayStore, SDNPMayLoad]>;
// Do not use ld, st directly. Use load, extload, sextload, zextload, store,
// and truncst (see below).
setOperationAction(ISD::MEMSET , MVT::Other, Custom);
setOperationAction(ISD::MEMCPY , MVT::Other, Custom);
+ if (!Subtarget->hasSSE1())
+ setOperationAction(ISD::PREFETCH , MVT::Other, Expand);
+
if (!Subtarget->hasSSE2())
setOperationAction(ISD::MEMBARRIER , MVT::Other, Expand);
"movmskpd\t{$src, $dst|$dst, $src}",
[(set GR32:$dst, (int_x86_sse2_movmsk_pd VR128:$src))]>;
-// Prefetching loads.
-// TODO: no intrinsics for these?
-def PREFETCHT0 : PSI<0x18, MRM1m, (outs), (ins i8mem:$src), "prefetcht0\t$src", []>;
-def PREFETCHT1 : PSI<0x18, MRM2m, (outs), (ins i8mem:$src), "prefetcht1\t$src", []>;
-def PREFETCHT2 : PSI<0x18, MRM3m, (outs), (ins i8mem:$src), "prefetcht2\t$src", []>;
-def PREFETCHNTA : PSI<0x18, MRM0m, (outs), (ins i8mem:$src), "prefetchnta\t$src", []>;
+// Prefetch intrinsic.
+def PREFETCHT0 : PSI<0x18, MRM1m, (outs), (ins i8mem:$src),
+ "prefetcht0\t$src", [(prefetch addr:$src, imm, (i32 3))]>;
+def PREFETCHT1 : PSI<0x18, MRM2m, (outs), (ins i8mem:$src),
+ "prefetcht1\t$src", [(prefetch addr:$src, imm, (i32 2))]>;
+def PREFETCHT2 : PSI<0x18, MRM3m, (outs), (ins i8mem:$src),
+ "prefetcht2\t$src", [(prefetch addr:$src, imm, (i32 1))]>;
+def PREFETCHNTA : PSI<0x18, MRM0m, (outs), (ins i8mem:$src),
+ "prefetchnta\t$src", [(prefetch addr:$src, imm, (i32 0))]>;
// Non-temporal stores
def MOVNTPSmr : PSI<0x2B, MRMDestMem, (outs), (ins i128mem:$dst, VR128:$src),
--- /dev/null
+; RUN: llvm-as < %s | llc -march=x86 -mattr=+sse1 | grep prefetchnta
+; RUN: llvm-as < %s | llc -march=x86 -mattr=+sse1 | grep prefetcht0
+; RUN: llvm-as < %s | llc -march=x86 -mattr=+sse1 | grep prefetcht1
+; RUN: llvm-as < %s | llc -march=x86 -mattr=+sse1 | grep prefetcht2
+
+define void @t(i8* %ptr) nounwind {
+entry:
+ tail call void @llvm.prefetch( i8* %ptr, i32 0, i32 1 )
+ tail call void @llvm.prefetch( i8* %ptr, i32 0, i32 2 )
+ tail call void @llvm.prefetch( i8* %ptr, i32 0, i32 3 )
+ tail call void @llvm.prefetch( i8* %ptr, i32 0, i32 0 )
+ ret void
+}
+
+declare void @llvm.prefetch(i8*, i32, i32) nounwind