tsan: improve lots_of_threads test

[dvyukov]

Summary

The current 10 threads is not particularly "lots" and not stressful.
Create 10x300 threads and ensure they all are running at the same time.

Depends on D106953.

[melver]

compiler-rt/test/tsan/lots_of_threads.c

19–26

constexpr int kIters = 10

?

21

This works because inner i is shadowing outer i, but it'd still be nicer if it was using i and j and not two i.

[dvyukov]

addressed comments

compiler-rt/test/tsan/lots_of_threads.c

 // RUN: %clang_tsan -O1 %s -o %t && %run %t 2>&1 | FileCheck %s
 #include "test.h"
 
 void *thr(void *arg) {
   // Create a sync object on stack, so there is something to free on thread end.
   volatile int x;
   __atomic_fetch_add(&x, 1, __ATOMIC_SEQ_CST);
   barrier_wait(&barrier);
   return 0;
 }
 
 int main() {
-  const int kThreads = 10;
+  const int kThreads = 300;
+  const int kIters = 10;
   barrier_init(&barrier, kThreads + 1);
-  pthread_t t[kThreads];
   pthread_attr_t attr;
   pthread_attr_init(&attr);
   pthread_attr_setstacksize(&attr, 16 << 20);
-  pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED);
-  for (int i = 0; i < kThreads; i++)
-    pthread_create(&t[i], &attr, thr, 0);
-  pthread_attr_destroy(&attr);
+  for (int iter = 0; iter < kIters; iter++) {
+    pthread_t threads[kThreads];
+    for (int t = 0; t < kThreads; t++)
+      pthread_create(&threads[t], &attr, thr, 0);
     barrier_wait(&barrier);
-  sleep(1);
+    for (int t = 0; t < kThreads; t++)
+      pthread_join(threads[t], 0);
+  }
+  pthread_attr_destroy(&attr);
   fprintf(stderr, "DONE\n");
   return 0;
 }
 
 // CHECK: DONE