about summary refs log tree commit diff
path: root/benchtests/bench-skeleton.c
diff options
context:
space:
mode:
authorSiddhesh Poyarekar <siddhesh@redhat.com>2013-04-30 14:10:20 +0530
committerSiddhesh Poyarekar <siddhesh@redhat.com>2013-04-30 14:10:20 +0530
commitd569c6eeb48219993063f956e516704281602f7d (patch)
treeee4b52c88456865a025c5943ff6f0bb5b741ff0a /benchtests/bench-skeleton.c
parenta6a242fe7cc0d7fcac1c9741d8be2ca8c2a5c744 (diff)
downloadglibc-d569c6eeb48219993063f956e516704281602f7d.tar.gz
glibc-d569c6eeb48219993063f956e516704281602f7d.tar.xz
glibc-d569c6eeb48219993063f956e516704281602f7d.zip
Maintain runtime of each benchmark at ~10 seconds
The idea to run benchmarks for a constant number of iterations is
problematic.  While the benchmarks may run for 10 seconds on x86_64,
they could run for about 30 seconds on powerpc and worse, over 3
minutes on arm.  Besides that, adding a new benchmark is cumbersome
since one needs to find out the number of iterations needed for a
sufficient runtime.

A better idea would be to run each benchmark for a specific amount of
time.  This patch does just that.  The run time defaults to 10 seconds
and it is configurable at command line:

  make BENCH_DURATION=5 bench
Diffstat (limited to 'benchtests/bench-skeleton.c')
-rw-r--r--benchtests/bench-skeleton.c38
1 files changed, 30 insertions, 8 deletions
diff --git a/benchtests/bench-skeleton.c b/benchtests/bench-skeleton.c
index 13f986d817..bbd151b0ec 100644
--- a/benchtests/bench-skeleton.c
+++ b/benchtests/bench-skeleton.c
@@ -22,13 +22,18 @@
 #include <time.h>
 #include <inttypes.h>
 
+#define TIMESPEC_AFTER(a, b) \
+  (((a).tv_sec == (b).tv_sec) ?						      \
+     ((a).tv_nsec > (b).tv_nsec) :					      \
+        ((a).tv_sec > (b).tv_sec))
 int
 main (int argc, char **argv)
 {
-  unsigned long i, j, k;
+  unsigned long i, k;
   uint64_t total = 0, max = 0, min = 0x7fffffffffffffff;
-  struct timespec start, end;
+  struct timespec start, end, runtime;
 
+  memset (&runtime, 0, sizeof (runtime));
   memset (&start, 0, sizeof (start));
   memset (&end, 0, sizeof (end));
 
@@ -39,11 +44,15 @@ main (int argc, char **argv)
      Measurements close to the minimum clock resolution won't make much sense,
      but it's better than having nothing at all.  */
   unsigned long iters = 1000 * start.tv_nsec;
-  unsigned long total_iters = ITER / iters;
 
-  for (i = 0; i < NUM_SAMPLES; i++)
+  /* Run for approxmately DURATION seconds.  */
+  clock_gettime (CLOCK_MONOTONIC_RAW, &runtime);
+  runtime.tv_sec += DURATION;
+
+  double d_total_i = 0;
+  while (1)
     {
-      for (j = 0; j < total_iters; j ++)
+      for (i = 0; i < NUM_SAMPLES; i++)
 	{
 	  clock_gettime (CLOCK_PROCESS_CPUTIME_ID, &start);
 	  for (k = 0; k < iters; k++)
@@ -61,12 +70,25 @@ main (int argc, char **argv)
 	    min = cur;
 
 	  total += cur;
+
+	  d_total_i += iters;
 	}
+
+      struct timespec curtime;
+
+      memset (&curtime, 0, sizeof (curtime));
+      clock_gettime (CLOCK_MONOTONIC_RAW, &curtime);
+      if (TIMESPEC_AFTER (curtime, runtime))
+	goto done;
     }
 
-  double d_total_s = total * 1e-9;
-  double d_iters = iters;
-  double d_total_i = (double)ITER * NUM_SAMPLES;
+  double d_total_s;
+  double d_iters;
+
+ done:
+  d_total_s = total * 1e-9;
+  d_iters = iters;
+
   printf (FUNCNAME ": ITERS:%g: TOTAL:%gs, MAX:%gns, MIN:%gns, %g iter/s\n",
 	  d_total_i, d_total_s, max / d_iters, min / d_iters,
 	  d_total_i / d_total_s);