ARM: implement correct atomic operations on v7M

ARM v7M has ldrex/strex but not ldrexd/strexd. This means 32-bit
operations should work as normal, but 64-bit ones are almost certainly
doomed.

Patch by Phoebe Buckheister.

llvm-svn: 211042

2 files changed, 73 insertions, 8 deletions

diff --git a/llvm/lib/Target/ARM/ARMISelLowering.cpp b/llvm/lib/Target/ARM/ARMISelLowering.cpp
index 3350601aadc..f9fc641b12d 100644
--- a/llvm/lib/Target/ARM/ARMISelLowering.cpp
+++ b/llvm/lib/Target/ARM/ARMISelLowering.cpp
@@ -10744,14 +10744,20 @@ bool ARMTargetLowering::shouldConvertConstantLoadToIntImm(const APInt &Imm,
 bool ARMTargetLowering::shouldExpandAtomicInIR(Instruction *Inst) const {
   // Loads and stores less than 64-bits are already atomic; ones above that
   // are doomed anyway, so defer to the default libcall and blame the OS when
-  // things go wrong:
-  if (StoreInst *SI = dyn_cast<StoreInst>(Inst))
-    return SI->getValueOperand()->getType()->getPrimitiveSizeInBits() == 64;
-  else if (LoadInst *LI = dyn_cast<LoadInst>(Inst))
-    return LI->getType()->getPrimitiveSizeInBits() == 64;
-
-  // For the real atomic operations, we have ldrex/strex up to 64 bits.
-  return Inst->getType()->getPrimitiveSizeInBits() <= 64;
+  // things go wrong. Cortex M doesn't have ldrexd/strexd though, so don't emit
+  // anything for those.
+  bool IsMClass = Subtarget->isMClass();
+  if (StoreInst *SI = dyn_cast<StoreInst>(Inst)) {
+    unsigned Size = SI->getValueOperand()->getType()->getPrimitiveSizeInBits();
+    return Size == 64 && !IsMClass;
+  } else if (LoadInst *LI = dyn_cast<LoadInst>(Inst)) {
+    return LI->getType()->getPrimitiveSizeInBits() == 64 && !IsMClass;
+  }
+
+  // For the real atomic operations, we have ldrex/strex up to 32 bits,
+  // and up to 64 bits on the non-M profiles
+  unsigned AtomicLimit = IsMClass ? 32 : 64;
+  return Inst->getType()->getPrimitiveSizeInBits() <= AtomicLimit;
 }
 
 Value *ARMTargetLowering::emitLoadLinked(IRBuilder<> &Builder, Value *Addr,
diff --git a/llvm/test/CodeGen/ARM/ldstrex-m.ll b/llvm/test/CodeGen/ARM/ldstrex-m.ll
new file mode 100644
index 00000000000..b50699f4cde
--- /dev/null
+++ b/llvm/test/CodeGen/ARM/ldstrex-m.ll
@@ -0,0 +1,59 @@
+; RUN: llc < %s -mtriple=thumbv7m-none-eabi -mcpu=cortex-m4 | FileCheck %s
+
+; CHECK-LABEL: f0:
+; CHECK-NOT: ldrexd
+define i64 @f0(i64* %p) nounwind readonly {
+entry:
+  %0 = load atomic i64* %p seq_cst, align 8
+  ret i64 %0
+}
+
+; CHECK-LABEL: f1:
+; CHECK-NOT: strexd
+define void @f1(i64* %p) nounwind readonly {
+entry:
+  store atomic i64 0, i64* %p seq_cst, align 8
+  ret void
+}
+
+; CHECK-LABEL: f2:
+; CHECK-NOT: ldrexd
+; CHECK-NOT: strexd
+define i64 @f2(i64* %p) nounwind readonly {
+entry:
+  %0 = atomicrmw add i64* %p, i64 1 seq_cst
+  ret i64 %0
+}
+
+; CHECK-LABEL: f3:
+; CHECK: ldr
+define i32 @f3(i32* %p) nounwind readonly {
+entry:
+  %0 = load atomic i32* %p seq_cst, align 4
+  ret i32 %0
+}
+
+; CHECK-LABEL: f4:
+; CHECK: ldrb
+define i8 @f4(i8* %p) nounwind readonly {
+entry:
+  %0 = load atomic i8* %p seq_cst, align 4
+  ret i8 %0
+}
+
+; CHECK-LABEL: f5:
+; CHECK: str
+define void @f5(i32* %p) nounwind readonly {
+entry:
+  store atomic i32 0, i32* %p seq_cst, align 4
+  ret void
+}
+
+; CHECK-LABEL: f6:
+; CHECK: ldrex
+; CHECK: strex
+define i32 @f6(i32* %p) nounwind readonly {
+entry:
+  %0 = atomicrmw add i32* %p, i32 1 seq_cst
+  ret i32 %0
+}