[compiler-rt] c4cb9b6 - tsan: add more micro benchmarks

[Dmitry Vyukov via llvm-commits]

Author: Dmitry Vyukov
Date: 2021-07-28T20:11:59+02:00
New Revision: c4cb9b64dd350f0d675e12d38ea54f4e0c3ceb37

URL: https://github.com/llvm/llvm-project/commit/c4cb9b64dd350f0d675e12d38ea54f4e0c3ceb37
DIFF: https://github.com/llvm/llvm-project/commit/c4cb9b64dd350f0d675e12d38ea54f4e0c3ceb37.diff

LOG: tsan: add more micro benchmarks

1. Add a set of micro benchmarks for memory accesses,
   mem* functions and unaligned accesses.
2. Add support for multiple benchmarks in a single binary
   (or it would require 12 new benchmark binaries).
3. Remove the "clock growth" machinery,
   it affects the current tsan runtime by increasing size of
   all vector clocks, but this won't be relevant for the new
   tsan runtime.

Reviewed By: vitalybuka

Differential Revision: https://reviews.llvm.org/D106961

Added: 
    compiler-rt/test/tsan/bench_memory_access.cpp

Modified: 
    compiler-rt/test/tsan/bench.h

Removed: 
    


################################################################################
diff  --git a/compiler-rt/test/tsan/bench.h b/compiler-rt/test/tsan/bench.h
index 5ae0dd813db9..58a0a78d4598 100644
--- a/compiler-rt/test/tsan/bench.h
+++ b/compiler-rt/test/tsan/bench.h
@@ -1,18 +1,12 @@
-#include <pthread.h>
-#include <stdlib.h>
-#include <stddef.h>
-#include <unistd.h>
-#include <stdio.h>
+#include "test.h"
 #include <time.h>
 
 int bench_nthread;
 int bench_niter;
-int grow_clock_var;
-pthread_barrier_t glow_clock_barrier;
+int bench_mode;
 
 void bench();  // defined by user
 void start_thread_group(int nth, void(*f)(int tid));
-void grow_clock_worker(int tid);
 
 int main(int argc, char **argv) {
   bench_nthread = 2;
@@ -21,15 +15,8 @@ int main(int argc, char **argv) {
   bench_niter = 100;
   if (argc > 2)
     bench_niter = atoi(argv[2]);
-
-  // Grow thread's clock.
-  int clock_size = 10;
-  if (argc > 1)
-    clock_size = 1000;
-  pthread_barrier_init(&glow_clock_barrier, 0, clock_size);
-  start_thread_group(clock_size, grow_clock_worker);
-  pthread_barrier_destroy(&glow_clock_barrier);
-  __atomic_load_n(&grow_clock_var, __ATOMIC_ACQUIRE);
+  if (argc > 3)
+    bench_mode = atoi(argv[3]);
 
   timespec tp0;
   clock_gettime(CLOCK_MONOTONIC, &tp0);
@@ -50,10 +37,3 @@ void start_thread_group(int nth, void(*f)(int tid)) {
   for (int i = 0; i < nth; i++)
     pthread_join(th[i], 0);
 }
-
-void grow_clock_worker(int tid) {
-  int res = pthread_barrier_wait(&glow_clock_barrier);
-  if (res == PTHREAD_BARRIER_SERIAL_THREAD)
-    __atomic_store_n(&grow_clock_var, 0, __ATOMIC_RELEASE);
-}
-

diff  --git a/compiler-rt/test/tsan/bench_memory_access.cpp b/compiler-rt/test/tsan/bench_memory_access.cpp
new file mode 100644
index 000000000000..78bb053f130e
--- /dev/null
+++ b/compiler-rt/test/tsan/bench_memory_access.cpp
@@ -0,0 +1,89 @@
+// RUN: %clangxx_tsan %s -o %t
+// RUN: %run %t 2>&1 | FileCheck %s
+
+// bench.h needs pthread barriers which are not available on OS X
+// UNSUPPORTED: darwin
+
+#include "bench.h"
+#include <memory.h>
+
+void thread(int tid) {
+  volatile long x = 0;
+  switch (bench_mode) {
+  case 0:
+    for (int i = 0; i < bench_niter; i++)
+      *(volatile char *)&x = 1;
+    break;
+  case 1:
+    for (int i = 0; i < bench_niter; i++)
+      *(volatile short *)&x = 1;
+    break;
+  case 2:
+    for (int i = 0; i < bench_niter; i++)
+      *(volatile int *)&x = 1;
+    break;
+  case 3:
+    for (int i = 0; i < bench_niter; i++)
+      *(volatile long *)&x = 1;
+    break;
+  case 4:
+    for (int i = 0; i < bench_niter; i++)
+      *(volatile char *)&x;
+    break;
+  case 5:
+    for (int i = 0; i < bench_niter; i++)
+      *(volatile short *)&x;
+    break;
+  case 6:
+    for (int i = 0; i < bench_niter; i++)
+      *(volatile int *)&x;
+    break;
+  case 7:
+    for (int i = 0; i < bench_niter; i++)
+      *(volatile long *)&x;
+  case 8:
+    for (int i = 0; i < bench_niter / 10; i++) {
+      ((volatile long *)&x)[0];
+      ((volatile int *)&x)[0];
+      ((volatile short *)&x)[2];
+      ((volatile char *)&x)[6];
+      ((volatile char *)&x)[7];
+      ((volatile long *)&x)[0] = 1;
+      ((volatile int *)&x)[0] = 1;
+      ((volatile short *)&x)[2] = 1;
+      ((volatile char *)&x)[6] = 1;
+      ((volatile char *)&x)[7] = 1;
+    }
+    break;
+  case 9: {
+    volatile long size = sizeof(x);
+    for (int i = 0; i < bench_niter; i++)
+      memset((void *)&x, i, size);
+    break;
+  }
+  case 10: {
+    volatile long data[2] = {};
+    volatile long size = sizeof(data) - 2;
+    for (int i = 0; i < bench_niter; i++)
+      memset(((char *)data) + 1, i, size);
+    break;
+  }
+  case 11: {
+    volatile long data[2] = {};
+    for (int i = 0; i < bench_niter / 8 / 3; i++) {
+      for (int off = 0; off < 8; off++) {
+        __sanitizer_unaligned_store16(((char *)data) + off, i);
+        __sanitizer_unaligned_store32(((char *)data) + off, i);
+        __sanitizer_unaligned_store64(((char *)data) + off, i);
+      }
+    }
+    break;
+  }
+  }
+}
+
+void bench() {
+  start_thread_group(bench_nthread, thread);
+}
+
+// CHECK: DONE