[PATCH v5] benchtest: malloc tcache hotpath benchtest.

[Wilco Dijkstra]

Hi Cupertino,

We're getting there! Just one thing to ensure the benchmark loop runs
properly across all targets and compilers. It uses a division, however if
we change the 2nd loop to use 4 rather than 5 items, and force it to be
always inlined, it is optimized and then it all works perfectly.

Other than that, it looks good.

Cheers,
Wilco


+static size_t
+malloc_benchmark_loop (void **elems, size_t nr_items)

Should be __always_inline so the division can be optimized.

+{
+  size_t iters = nr_items;
+
+  while (!timeout)
+    {
+      elems[iters % nr_items] = TEST_FUNC (alloc_size);
+      iters++;
+      free (elems[iters % nr_items]);
+    }
+
+  return iters - nr_items;
+}
+
+static void
+do_benchmark (struct bench_result *res)
+{
+  timing_t start, stop;
+  void *elems[10];
+  memset (elems, 0, sizeof(void *) * 10);

I think sizeof (elems) is simpler and safer.

+
+  /* Ramp up cpu before measuring.  */
+  alarm (BENCHMARK_DURATION);
+  bench_start ();
+  TIMING_NOW (start);
+  res[0].iters = malloc_benchmark_loop (elems, 1);
+  TIMING_NOW (stop);
+  TIMING_DIFF (res[0].elapsed, start, stop);
+
+  timeout = false;
+  alarm (BENCHMARK_DURATION);
+  bench_start ();
+  TIMING_NOW (start);
+  res[1].iters = malloc_benchmark_loop (elems, 5);

This should be 4 to avoid an expensive division.