Make use of isAtLeastRelease/Acquire in the ARM/AArch64 backends

[morisset]

Summary

Make use of isAtLeastRelease/Acquire in the ARM/AArch64 backends
These helper functions are introduced in D4844.
Depends D4844

[jfb]

Looks good to me beyond the comment on asserts. I'll let others OK the change since I haven't touched that code base.

lib/Target/AArch64/AArch64InstrAtomics.td

32

You probably want to keep this assert. Although it is weird that subsequent code doesn't have similar asserts (I'd expect consistency in all-asserts or none).

[morisset]

lib/Target/AArch64/AArch64InstrAtomics.td

32

I removed this assert because of this lack of consistency, but I can add it back (and to the other functions maybe) if you want.

[reames]

LGTM.

[jfb]

lib/Target/AArch64/AArch64InstrAtomics.td

32

Removing it seems fine to me since the other locations don't have similar asserts.

[morisset]

lib/Target/AArch64/AArch64InstrAtomics.td

32

Then can you accept the revision please? Thanks.

[jfb]

lib/Target/AArch64/AArch64InstrAtomics.td

32

Done.

[morisset]

Closed by commit rL215902 (authored by @morisset).

lib/Target/AArch64/AArch64ISelLowering.cpp

 bool AArch64TargetLowering::hasLoadLinkedStoreConditional() const {
   return true;
 }
 
 Value *AArch64TargetLowering::emitLoadLinked(IRBuilder<> &Builder, Value *Addr,
                                              AtomicOrdering Ord) const {
   Module *M = Builder.GetInsertBlock()->getParent()->getParent();
   Type *ValTy = cast<PointerType>(Addr->getType())->getElementType();
-  bool IsAcquire =
-      Ord == Acquire || Ord == AcquireRelease || Ord == SequentiallyConsistent;
+  bool IsAcquire = isAtLeastAcquire(Ord);
 
   // Since i128 isn't legal and intrinsics don't get type-lowered, the ldrexd
   // intrinsic must return {i64, i64} and we have to recombine them into a
   // single i128 here.
   if (ValTy->getPrimitiveSizeInBits() == 128) {
     Intrinsic::ID Int =
         IsAcquire ? Intrinsic::aarch64_ldaxp : Intrinsic::aarch64_ldxp;
     Function *Ldxr = llvm::Intrinsic::getDeclaration(M, Int);
return Builder.CreateTruncOrBitCast(
       Builder.CreateCall(Ldxr, Addr),
       cast<PointerType>(Addr->getType())->getElementType());
 }
 
 Value *AArch64TargetLowering::emitStoreConditional(IRBuilder<> &Builder,
                                                    Value *Val, Value *Addr,
                                                    AtomicOrdering Ord) const {
   Module *M = Builder.GetInsertBlock()->getParent()->getParent();
-  bool IsRelease =
-      Ord == Release || Ord == AcquireRelease || Ord == SequentiallyConsistent;
+  bool IsRelease = isAtLeastRelease(Ord);
 
   // Since the intrinsics must have legal type, the i128 intrinsics take two
   // parameters: "i64, i64". We must marshal Val into the appropriate form
   // before the call.
   if (Val->getType()->getPrimitiveSizeInBits() == 128) {
     Intrinsic::ID Int =
         IsRelease ? Intrinsic::aarch64_stlxp : Intrinsic::aarch64_stxp;
     Function *Stxr = Intrinsic::getDeclaration(M, Int);

lib/Target/AArch64/AArch64InstrAtomics.td

 // When they're actually atomic, only one addressing mode (GPR64sp) is
 // supported, but when they're relaxed and anything can be used, all the
 // standard modes would be valid and may give efficiency gains.
 
 // A atomic load operation that actually needs acquire semantics.
 class acquiring_load<PatFrag base>
   : PatFrag<(ops node:$ptr), (base node:$ptr), [{
   AtomicOrdering Ordering = cast<AtomicSDNode>(N)->getOrdering();
-  assert(Ordering != AcquireRelease && "unexpected load ordering");
-  return Ordering == Acquire || Ordering == SequentiallyConsistent;
+  return isAtLeastAcquire(Ordering);
 }]>;
 
 // An atomic load operation that does not need either acquire or release
 // semantics.
 class relaxed_load<PatFrag base>
   : PatFrag<(ops node:$ptr), (base node:$ptr), [{
   AtomicOrdering Ordering = cast<AtomicSDNode>(N)->getOrdering();
-  return Ordering == Monotonic || Ordering == Unordered;
+  return !isAtLeastAcquire(Ordering);
 }]>;
 
 // 8-bit loads
 def : Pat<(acquiring_load<atomic_load_8>  GPR64sp:$ptr), (LDARB GPR64sp:$ptr)>;
 def : Pat<(relaxed_load<atomic_load_8> (ro_Windexed8 GPR64sp:$Rn, GPR32:$Rm,
                                                      ro_Wextend8:$offset)),
           (LDRBBroW GPR64sp:$Rn, GPR32:$Rm, ro_Wextend8:$offset)>;
 def : Pat<(relaxed_load<atomic_load_8> (ro_Xindexed8 GPR64sp:$Rn, GPR64:$Rm,

 // supported, but when they're relaxed and anything can be used, all the
 // standard modes would be valid and may give efficiency gains.
 
 // A store operation that actually needs release semantics.
 class releasing_store<PatFrag base>
   : PatFrag<(ops node:$ptr, node:$val), (base node:$ptr, node:$val), [{
   AtomicOrdering Ordering = cast<AtomicSDNode>(N)->getOrdering();
   assert(Ordering != AcquireRelease && "unexpected store ordering");
-  return Ordering == Release || Ordering == SequentiallyConsistent;
+  return isAtLeastRelease(Ordering);
 }]>;
 
 // An atomic store operation that doesn't actually need to be atomic on AArch64.
 class relaxed_store<PatFrag base>
   : PatFrag<(ops node:$ptr, node:$val), (base node:$ptr, node:$val), [{
   AtomicOrdering Ordering = cast<AtomicSDNode>(N)->getOrdering();
-  return Ordering == Monotonic || Ordering == Unordered;
+  return !isAtLeastRelease(Ordering);
 }]>;
 
 // 8-bit stores
 def : Pat<(releasing_store<atomic_store_8> GPR64sp:$ptr, GPR32:$val),
           (STLRB GPR32:$val, GPR64sp:$ptr)>;
 def : Pat<(relaxed_store<atomic_store_8>
                (ro_Windexed8 GPR64sp:$Rn, GPR32:$Rm, ro_Wextend8:$extend),
                GPR32:$val),

lib/Target/ARM/ARMISelLowering.cpp

 bool ARMTargetLowering::useLoadStackGuardNode() const {
   return Subtarget->getTargetTriple().getObjectFormat() == Triple::MachO;
 }
 
 Value *ARMTargetLowering::emitLoadLinked(IRBuilder<> &Builder, Value *Addr,
                                          AtomicOrdering Ord) const {
   Module *M = Builder.GetInsertBlock()->getParent()->getParent();
   Type *ValTy = cast<PointerType>(Addr->getType())->getElementType();
-  bool IsAcquire =
-      Ord == Acquire || Ord == AcquireRelease || Ord == SequentiallyConsistent;
+  bool IsAcquire = isAtLeastAcquire(Ord);
 
   // Since i64 isn't legal and intrinsics don't get type-lowered, the ldrexd
   // intrinsic must return {i32, i32} and we have to recombine them into a
   // single i64 here.
   if (ValTy->getPrimitiveSizeInBits() == 64) {
     Intrinsic::ID Int =
         IsAcquire ? Intrinsic::arm_ldaexd : Intrinsic::arm_ldrexd;
     Function *Ldrex = llvm::Intrinsic::getDeclaration(M, Int);
return Builder.CreateTruncOrBitCast(
       Builder.CreateCall(Ldrex, Addr),
       cast<PointerType>(Addr->getType())->getElementType());
 }
 
 Value *ARMTargetLowering::emitStoreConditional(IRBuilder<> &Builder, Value *Val,
                                                Value *Addr,
                                                AtomicOrdering Ord) const {
   Module *M = Builder.GetInsertBlock()->getParent()->getParent();
-  bool IsRelease =
-      Ord == Release || Ord == AcquireRelease || Ord == SequentiallyConsistent;
+  bool IsRelease = isAtLeastRelease(Ord);
 
   // Since the intrinsics must have legal type, the i64 intrinsics take two
   // parameters: "i32, i32". We must marshal Val into the appropriate form
   // before the call.
   if (Val->getType()->getPrimitiveSizeInBits() == 64) {
     Intrinsic::ID Int =
         IsRelease ? Intrinsic::arm_stlexd : Intrinsic::arm_strexd;
     Function *Strex = Intrinsic::getDeclaration(M, Int);

lib/Target/ARM/ARMInstrInfo.td

 def : ARMPat<(stlex_1 (and GPR:$Rt, 0xff), addr_offset_none:$addr),
              (STLEXB GPR:$Rt, addr_offset_none:$addr)>;
 def : ARMPat<(stlex_2 (and GPR:$Rt, 0xffff), addr_offset_none:$addr),
              (STLEXH GPR:$Rt, addr_offset_none:$addr)>;
 
 class acquiring_load<PatFrag base>
   : PatFrag<(ops node:$ptr), (base node:$ptr), [{
   AtomicOrdering Ordering = cast<AtomicSDNode>(N)->getOrdering();
-  return Ordering == Acquire || Ordering == SequentiallyConsistent;
+  return isAtLeastAcquire(Ordering);
 }]>;
 
 def atomic_load_acquire_8  : acquiring_load<atomic_load_8>;
 def atomic_load_acquire_16 : acquiring_load<atomic_load_16>;
 def atomic_load_acquire_32 : acquiring_load<atomic_load_32>;
 
 class releasing_store<PatFrag base>
   : PatFrag<(ops node:$ptr, node:$val), (base node:$ptr, node:$val), [{
   AtomicOrdering Ordering = cast<AtomicSDNode>(N)->getOrdering();
-  return Ordering == Release || Ordering == SequentiallyConsistent;
+  return isAtLeastRelease(Ordering);
 }]>;
 
 def atomic_store_release_8  : releasing_store<atomic_store_8>;
 def atomic_store_release_16 : releasing_store<atomic_store_16>;
 def atomic_store_release_32 : releasing_store<atomic_store_32>;
 
 let AddedComplexity = 8 in {
   def : ARMPat<(atomic_load_acquire_8 addr_offset_none:$addr),  (LDAB addr_offset_none:$addr)>;