_Atomic of empty struct shouldn't assert

[jfb]

Summary

An _Atomic of an empty struct is pretty silly. In general we just widen empty structs to hold a byte's worth of storage, and we represent size and alignment as 0 internally and let LLVM figure out what to do. For _Atomic it's a bit different: the memory model mandates concrete effects occur when atomic operations occur, so in most cases actual instructions need to get emitted. It's really not worth trying to optimize empty struct atomics by figuring out e.g. that a fence would do, even though sane compilers should do optimize atomics. Further, wg21.link/p0528 will fix C++20 atomics with padding bits so that cmpxchg on them works, which means that we'll likely need to do the zero-init song and dance for empty atomic structs anyways (and I think we shouldn't special-case this behavior to C++20 because prior standards are just broken).

This patch therefore makes a minor change to r176658 "Promote atomic type sizes up to a power of two": if the width of the atomic's value type is 0, just use 1 byte for width and alignment.

This fixes an assertion:

(NumBits >= MIN_INT_BITS && "bitwidth too small"), function get, file ../lib/IR/Type.cpp, line 241.

It seems like this has run into other assertions before (namely the unreachable Kind check in ImpCastExprToType), but I haven't reproduced that issue with tip-of-tree.

rdar://problem/39678063

[rjmccall]

lib/AST/ASTContext.cpp

1965

Alignment, unlike size, is definitely never 0. I think you should leave the original alignment in place; it's a weird case, but we honor over-aligned empty types in a bunch of other places.

[jfb]

• Assert on zero alignment, instead of making it always byte-aligned.

[jfb]

lib/AST/ASTContext.cpp

1965

Yeah that makes total sense. I turned it to an assert.

[rjmccall]

LGTM.

lib/AST/ASTContext.cpp

return getTypeInfo(
                   cast<AttributedType>(T)->getEquivalentType().getTypePtr());
 
   case Type::Atomic: {
     // Start with the base type information.
     TypeInfo Info = getTypeInfo(cast<AtomicType>(T)->getValueType());
     Width = Info.Width;
     Align = Info.Align;
 
+    if (!Width) {
+      // An otherwise zero-sized type should still generate an
+      // atomic operation.
+      Width = Target->getCharWidth();
+      assert(Align);
+    } else if (Width <= Target->getMaxAtomicPromoteWidth()) {
       // If the size of the type doesn't exceed the platform's max
       // atomic promotion width, make the size and alignment more
       // favorable to atomic operations:
-    if (Width != 0 && Width <= Target->getMaxAtomicPromoteWidth()) {
+
       // Round the size up to a power of 2.
       if (!llvm::isPowerOf2_64(Width))
         Width = llvm::NextPowerOf2(Width);
 
       // Set the alignment equal to the size.
       Align = static_cast<unsigned>(Width);
     }
   }

test/CodeGen/c11atomics-ios.c

_Bool test_promoted_cmpxchg(_Atomic(PS) *addr, PS *desired, PS *new) {
   // CHECK:   [[RES_BOOL8:%.*]] = zext i1 [[RES_BOOL]] to i8
   // CHECK:   store i8 [[RES_BOOL8]], i8* [[RES_ADDR]], align 1
   // CHECK:   [[RES_BOOL8:%.*]] = load i8, i8* [[RES_ADDR]], align 1
   // CHECK:   [[RETVAL:%.*]] = trunc i8 [[RES_BOOL8]] to i1
   // CHECK:   ret i1 [[RETVAL]]
 
   return __c11_atomic_compare_exchange_strong(addr, desired, *new, 5, 5);
 }
+
+struct Empty {};
+
+struct Empty testEmptyStructLoad(_Atomic(struct Empty)* empty) {
+  // CHECK-LABEL: @testEmptyStructLoad(
+  // CHECK-NOT: @__atomic_load
+  // CHECK: load atomic i8, i8* %{{.*}} seq_cst, align 1
+  return *empty;
+}
+
+void testEmptyStructStore(_Atomic(struct Empty)* empty, struct Empty value) {
+  // CHECK-LABEL: @testEmptyStructStore(
+  // CHECK-NOT: @__atomic_store
+  // CHECK: store atomic i8 %{{.*}}, i8* %{{.*}} seq_cst, align 1
+  *empty = value;
+}

test/CodeGen/c11atomics.c

_Bool test_promoted_cmpxchg(_Atomic(PS) *addr, PS *desired, PS *new) {
   // CHECK:   [[ATOMIC_NEW64:%.*]] = bitcast { %struct.PS, [2 x i8] }* [[ATOMIC_NEW]] to i64*
   // CHECK:   [[ADDR8:%.*]] = bitcast i64* [[ADDR64]] to i8*
   // CHECK:   [[ATOMIC_DESIRED8:%.*]] = bitcast i64* [[ATOMIC_DESIRED64]] to i8*
   // CHECK:   [[NEW64:%.*]] = load i64, i64* [[ATOMIC_NEW64]], align 2
   // CHECK:   [[RES:%.*]] = call arm_aapcscc zeroext i1 @__atomic_compare_exchange_8(i8* [[ADDR8]], i8* [[ATOMIC_DESIRED8]], i64 [[NEW64]], i32 5, i32 5)
   // CHECK:   ret i1 [[RES]]
   return __c11_atomic_compare_exchange_strong(addr, desired, *new, 5, 5);
 }
+
+struct Empty {};
+
+struct Empty test_empty_struct_load(_Atomic(struct Empty)* empty) {
+  // CHECK-LABEL: @test_empty_struct_load(
+  // CHECK: call arm_aapcscc zeroext i8 @__atomic_load_1(i8* %{{.*}}, i32 5)
+  return __c11_atomic_load(empty, 5);
+}
+
+void test_empty_struct_store(_Atomic(struct Empty)* empty, struct Empty value) {
+  // CHECK-LABEL: @test_empty_struct_store(
+  // CHECK: call arm_aapcscc void @__atomic_store_1(i8* %{{.*}}, i8 zeroext %{{.*}}, i32 5)
+  __c11_atomic_store(empty, value, 5);
+}