base: Add a perftest for thread_local based TLS.

thread_local is already used in Chrome, though not widely as it is still banned
by the style-guide. Add a benchmark to estimate its performance.

Detailed results below, tldr:
- On Linux: 2-3x faster than the current TLS implementation for reading, ~10x
  faster for writing
- On Android: ~4x faster for reading

On Linux (Intel Xeon "Haswell"):
[ RUN      ] ThreadLocalStoragePerfTest.ThreadLocalStorage
*RESULT TLS read throughput: ThreadLocalStorage= 154230.54381689752 operations/ms
*RESULT TLS write throughput: ThreadLocalStorage= 168171.8043152885 operations/ms
*RESULT TLS read-write throughput: ThreadLocalStorage= 82276.76257394624 operations/ms
*RESULT TLS read throughput: ThreadLocalStorage 4 threads= 155118.12245024586 operations/ms
*RESULT TLS write throughput: ThreadLocalStorage 4 threads= 171458.9441558219 operations/ms
*RESULT TLS read-write throughput: ThreadLocalStorage 4 threads= 82700.67318347971 operations/ms

[ RUN      ] ThreadLocalStoragePerfTest.PlatformTls
*RESULT TLS read throughput: PlatformTls= 318390.2190524707 operations/ms
*RESULT TLS write throughput: PlatformTls= 232018.56148491878 operations/ms
*RESULT TLS read-write throughput: PlatformTls= 144791.13878230652 operations/ms
*RESULT TLS read throughput: PlatformTls 4 threads= 255597.5871587772 operations/ms
*RESULT TLS write throughput: PlatformTls 4 threads= 195446.10573634322 operations/ms
*RESULT TLS read-write throughput: PlatformTls 4 threads= 132872.70794578793 operations/ms

[ RUN      ] ThreadLocalStoragePerfTest.Cpp11Tls
*RESULT TLS read throughput: C++ thread_local TLS= 384763.37052712584 operations/ms
*RESULT TLS write throughput: C++ thread_local TLS= 2500625.1562890722 operations/ms
*RESULT TLS read-write throughput: C++ thread_local TLS= 2430724.3558580456 operations/ms
*RESULT TLS read throughput: C++ thread_local TLS 4 threads= 342794.4604415193 operations/ms
*RESULT TLS write throughput: C++ thread_local TLS 4 threads= 1192037.1915603767 operations/ms
*RESULT TLS read-write throughput: C++ thread_local TLS 4 threads= 1774937.8771742987 operations/ms

On Android (Pixel 1, Snapdragon 821):
[ RUN      ] ThreadLocalStoragePerfTest.ThreadLocalStorage
*RESULT TLS read throughput: ThreadLocalStorage= 48410.911819524124 operations/ms
*RESULT TLS write throughput: ThreadLocalStorage= 29321.041952546824 operations/ms
*RESULT TLS read-write throughput: ThreadLocalStorage= 18727.64387512407 operations/ms
*RESULT TLS read throughput: ThreadLocalStorage 4 threads= 34612.97501981593 operations/ms
*RESULT TLS write throughput: ThreadLocalStorage 4 threads= 22686.79741824245 operations/ms
*RESULT TLS read-write throughput: ThreadLocalStorage 4 threads= 14610.463137070981 operations/ms

[ RUN      ] ThreadLocalStoragePerfTest.PlatformTls
*RESULT TLS read throughput: PlatformTls= 40691.75991861648 operations/ms
*RESULT TLS write throughput: PlatformTls= 43693.49884430696 operations/ms
*RESULT TLS read-write throughput: PlatformTls= 23500.547562758213 operations/ms
*RESULT TLS read throughput: PlatformTls 4 threads= 31334.701615303868 operations/ms
*RESULT TLS write throughput: PlatformTls 4 threads= 33404.26171570969 operations/ms
*RESULT TLS read-write throughput: PlatformTls 4 threads= 19587.027120197752 operations/ms

[ RUN      ] ThreadLocalStoragePerfTest.Cpp11Tls
*RESULT TLS read throughput: C++ thread_local TLS= 205532.9469313931 operations/ms
*RESULT TLS write throughput: C++ thread_local TLS= 19075.81491881333 operations/ms
*RESULT TLS read-write throughput: C++ thread_local TLS= 15905.078491562355 operations/ms
*RESULT TLS read throughput: C++ thread_local TLS 4 threads= 81040.56080068075 operations/ms
*RESULT TLS write throughput: C++ thread_local TLS 4 threads= 11412.84127248615 operations/ms
*RESULT TLS read-write throughput: C++ thread_local TLS 4 threads= 9210.180934004447 operations/ms

Bug: 998048
Change-Id: I018c3a7989f3ae48406ba86c282f5c4e4b205bea
Reviewed-on: https://chromium-review.googlesource.com/c/chromium/src/+/1873751Reviewed-by: François Doray <fdoray@chromium.org>
Commit-Queue: Benoit L <lizeb@chromium.org>
Cr-Commit-Position: refs/heads/master@{#708539}

base/threading/thread_local_storage_perftest.cc

@@ -62,7 +62,6 @@ class TLSThread : public SimpleThread {
 };
 
 class ThreadLocalStoragePerfTest : public testing::Test {
- public:
  protected:
   ThreadLocalStoragePerfTest() = default;
   ~ThreadLocalStoragePerfTest() override = default;
@@ -124,11 +123,13 @@ class ThreadLocalStoragePerfTest : public testing::Test {
     for (auto& thread : threads)
       thread->Join();
 
-    perf_test::PrintResult(
-        measurment, "", trace,
-        num_operation /
-            static_cast<double>(operation_duration.InMilliseconds()),
-        "operations/ms", true);
+    perf_test::PrintResult(measurment, "", trace,
+                           num_operation / operation_duration.InMillisecondsF(),
+                           "operations/ms", true);
+    size_t nanos_per_operation =
+        operation_duration.InNanoseconds() / num_operation;
+    perf_test::PrintResult(measurment, "", trace, nanos_per_operation,
+                           "ns/operation", true);
   }
 
  private:
@@ -196,5 +197,17 @@ TEST_F(ThreadLocalStoragePerfTest, PlatformTls) {
 
 #endif
 
+TEST_F(ThreadLocalStoragePerfTest, Cpp11Tls) {
+  thread_local intptr_t thread_local_variable;
+
+  auto read = [&]() { return thread_local_variable; };
+  auto write = [&](intptr_t value) {
+    reinterpret_cast<volatile intptr_t*>(&thread_local_variable)[0] = value;
+  };
+
+  Benchmark("C++ thread_local TLS", read, write, 10000000, 1);
+  Benchmark("C++ thread_local TLS 4 threads", read, write, 10000000, 4);
+}
+
 }  // namespace internal
-}  // namespace base
+}  // namespace base
\ No newline at end of file