Merge lp:~erwan-t/sysbench/sysbench into lp:sysbench

[Erwan Velu]

This patch insure that no code that could consume CPU is between the LOG_EVENT_START and the working loop.

In the current code, a random calculation was done after the LOG_EVENT_START was setup. If for any reason this random thing is taking some time, then the complete result will be inaccurate.

[Alexey Kopytov (akopytov)]

Thank you.

review: Approve

=== modified file 'sysbench/tests/memory/sb_memory.c'
--- sysbench/tests/memory/sb_memory.c	2011-09-26 13:54:56 +0000
+++ sysbench/tests/memory/sb_memory.c	2014-09-15 11:18:34 +0000
@@ -261,11 +261,10 @@
     buf = buffers[thread_id];
   end = (int *)((char *)buf + memory_block_size);
 
-  LOG_EVENT_START(msg, thread_id);
-
   if (memory_access_rnd)
   {
     rand = sb_rnd();
+    LOG_EVENT_START(msg, thread_id);
     switch (mem_req->type) {
       case SB_MEM_OP_WRITE:
         for (i = 0; i < memory_block_size; i++)
@@ -291,6 +290,7 @@
   }
   else
   {
+    LOG_EVENT_START(msg, thread_id);
     switch (mem_req->type) {
       case SB_MEM_OP_NONE:
         for (; buf < end; buf++)