make SequentialThreadId not a template

Summary:
`SequentialThreadId` is only ever instantiated with `std::atomic` so just make it no longer a template.

This permits moving the implementation into the source file.

Reviewed By: Mizuchi

Differential Revision: D27669406

fbshipit-source-id: ec2cddbff433df31434548b61111a34ca2da5cef

folly/concurrency/CacheLocality.cpp

@@ -320,7 +320,11 @@ Getcpu::Func Getcpu::resolveVdsoFunc() {
 
 #ifdef FOLLY_CL_USE_FOLLY_TLS
 /////////////// SequentialThreadId
-template struct SequentialThreadId<std::atomic>;
+unsigned SequentialThreadId::get() {
+  static std::atomic<unsigned> global{0};
+  static FOLLY_TLS unsigned local{0};
+  return FOLLY_LIKELY(local) ? local : (local = ++global);
+}
 #endif
 
 namespace detail {

folly/concurrency/CacheLocality.h

@@ -148,32 +148,9 @@ struct Getcpu {
 };
 
 #ifdef FOLLY_CL_USE_FOLLY_TLS
-template <template <typename> class Atom>
 struct SequentialThreadId {
-  /// Returns the thread id assigned to the current thread
-  static unsigned get() {
-    auto rv = currentId;
-    if (UNLIKELY(rv == 0)) {
-      rv = currentId = ++prevId;
-    }
-    return rv;
-  }
-
- private:
-  static Atom<unsigned> prevId;
-
-  static FOLLY_TLS unsigned currentId;
+  static unsigned get();
 };
-
-template <template <typename> class Atom>
-Atom<unsigned> SequentialThreadId<Atom>::prevId(0);
-
-template <template <typename> class Atom>
-FOLLY_TLS unsigned SequentialThreadId<Atom>::currentId(0);
-
-// Suppress this instantiation in other translation units. It is
-// instantiated in CacheLocality.cpp
-extern template struct SequentialThreadId<std::atomic>;
 #endif
 
 struct HashingThreadId {
@@ -201,7 +178,7 @@ struct FallbackGetcpu {
 };
 
 #ifdef FOLLY_CL_USE_FOLLY_TLS
-typedef FallbackGetcpu<SequentialThreadId<std::atomic>> FallbackGetcpuType;
+typedef FallbackGetcpu<SequentialThreadId> FallbackGetcpuType;
 #else
 typedef FallbackGetcpu<HashingThreadId> FallbackGetcpuType;
 #endif

folly/concurrency/test/CacheLocalityBenchmark.cpp

@@ -47,7 +47,7 @@ using namespace folly;
 DECLARE_SPREADER_TAG(
     ThreadLocalTag,
     CacheLocality::system<>(),
-    folly::FallbackGetcpu<SequentialThreadId<std::atomic>>::getcpu)
+    folly::FallbackGetcpu<SequentialThreadId>::getcpu)
 DECLARE_SPREADER_TAG(
     PthreadSelfTag,
     CacheLocality::system<>(),

folly/concurrency/test/CacheLocalityTest.cpp

@@ -1044,13 +1044,12 @@ TEST(Getcpu, VdsoGetcpu) {
 #ifdef FOLLY_CL_USE_FOLLY_TLS
 TEST(ThreadId, SimpleTls) {
   unsigned cpu = 0;
-  auto rv = folly::FallbackGetcpu<SequentialThreadId<std::atomic>>::getcpu(
-      &cpu, nullptr, nullptr);
+  auto rv =
+      folly::FallbackGetcpu<SequentialThreadId>::getcpu(&cpu, nullptr, nullptr);
   EXPECT_EQ(rv, 0);
   EXPECT_TRUE(cpu > 0);
   unsigned again;
-  folly::FallbackGetcpu<SequentialThreadId<std::atomic>>::getcpu(
-      &again, nullptr, nullptr);
+  folly::FallbackGetcpu<SequentialThreadId>::getcpu(&again, nullptr, nullptr);
   EXPECT_EQ(cpu, again);
 }
 #endif