Integrate performance testing into main tree for release

This is a combination of a bunch of small fixes to get the performance
monitoring infrastructure ready for release. I've squashed the changes because
they shouldn't pollute the commit log.

Update README.performance-monitoring to use a markdown TOC

Change README.performance-monitoring to README.performance-monitoring.md so it's recognized as markdown.

change .travis.yml to use performance flags

get performance compiling with C++

es

make sure bool plays nicely with both c++ and regular c

get rid of a straggler from the rebase

add logging.h and performance.h to Makefile.am

add first draft of performance option and ensure that performance profiling does not break a regular make

check enabled and disabled with travis.yml

making both causes travis.yml to time out

fix OSX structure compile silliness and add a test case

test performance infrastructure

really automake, really?

ignore performance test case

Author: nevans

.travis.yml

@@ -1,2 +1,2 @@
 language: c
-script: ./autogen.sh && ./configure && make && make check
+script: ./autogen.sh && ./configure --enable-performance-monitoring && make && make check

README.performance-monitoring renamed to README.performance-monitoring.md

@@ -1,51 +1,31 @@
-<div id="table-of-contents">
-<h2>Table of Contents</h2>
-<div id="text-table-of-contents">
-<ul>
-<li><a href="#sec-1">1. Introduction</a></li>
-<li><a href="#sec-2">2. Quick note on memory management</a></li>
-<li><a href="#sec-3">3. Tutorial</a>
-<ul>
-<li><a href="#sec-3-1">3.1. Quickstart</a></li>
-<li><a href="#sec-3-2">3.2. Setup Functions</a>
-<ul>
-<li><a href="#sec-3-2-1">3.2.1. Internal measurements</a></li>
-<li><a href="#sec-3-2-2">3.2.2. State group management</a></li>
-</ul>
-</li>
-<li><a href="#sec-3-3">3.3. Runtime Functions</a>
-<ul>
-<li><a href="#sec-3-3-1">3.3.1. <code>void qtperf_enter_state(qtperfdata_t* data, qtperfid_t state_id)</code></a></li>
-<li><a href="#sec-3-3-2">3.3.2. <code>const char* qtperf_state_name(qtstategroup_t* group, qtperfid_t state_id)</code></a></li>
-<li><a href="#sec-3-3-3">3.3.3. <code>void qtperf_start()</code></a></li>
-<li><a href="#sec-3-3-4">3.3.4. <code>void qtperf_stop()</code></a></li>
-</ul>
-</li>
-<li><a href="#sec-3-4">3.4. Reporting and data access functions</a>
-<ul>
-<li><a href="#sec-3-4-1">3.4.1. <code>void qtperf_print_results()</code></a></li>
-<li><a href="#sec-3-4-2">3.4.2. <code>void qtperf_print_delimited(qtstategroup_t* group, const char* delim, bool print_headers, const char* prefix)</code></a></li>
-<li><a href="#sec-3-4-3">3.4.3. <code>void qtperf_print_perfdata(qtperfdata_t* perfdata, bool show_states_with_zero_time)</code></a></li>
-<li><a href="#sec-3-4-4">3.4.4. <code>void qtperf_print_group(qtstategroup_t* group)</code></a></li>
-<li><a href="#sec-3-4-5">3.4.5. <code>void qtperf_print_perfdata(qtperfdata_t* data, bool show_states_with_zero_time)</code></a></li>
-<li><a href="#sec-3-4-6">3.4.6. <code>qtperfcounter_t qtperf_total_group_time(qtstategroup_t* group)</code></a></li>
-<li><a href="#sec-3-4-7">3.4.7. <code>qtperfcounter_t qtperf_total_time(qtperfdata_t* data)</code></a></li>
-</ul>
-</li>
-<li><a href="#sec-3-5">3.5. Iterators</a>
-<ul>
-<li><a href="#sec-3-5-1">3.5.1. <code>void qtperf_iter_begin(qtperf_iterator** iter)</code></a></li>
-<li><a href="#sec-3-5-2">3.5.2. <code>qtperfdata_t* qtperf_iter_next(qtperf_iterator_t** iter)</code></a></li>
-<li><a href="#sec-3-5-3">3.5.3. <code>qtperfdata_t* qtperf_iter_deref(qtperf_iterator_t * iter)</code></a></li>
-<li><a href="#sec-3-5-4">3.5.4. <code>qtperf_iterator_t* qtperf_iter_end()</code></a></li>
-</ul>
-</li>
-<li><a href="#sec-3-6">3.6. Teardown</a></li>
-</ul>
-</li>
-</ul>
-</div>
-</div>
+# Table of contents
+1. [Introduction](#introduction)
+2. [Quick note on memory management](#quick-note-on-memory-management)
+3. [Tutorial](#tutorial)
+    1. [Quickstart](#quickstart)
+    2. [Setup Functions](#setup-functions)
+        1. [Internal measurements](#internal-measurements)
+        2. [State group management](#state-group-management)
+    3. [Runtime Functions](#runtime-functions)
+        1. `void qtperf_enter_state(qtperfdata_t* data, qtperfid_t state_id)`
+        2. `const char* qtperf_state_name(qtstategroup_t* group, qtperfid_t state_id)`
+        3. `void qtperf_start()`
+        4. `void qtperf_stop()`
+    4. [Reporting and data access functions](#report-and-data-access-functions)
+        1. `void qtperf_print_results()`
+        2. `void qtperf_print_delimited(qtstategroup_t* group, const char* delim, bool print_headers, const char* prefix)`
+        3. `void qtperf_print_perfdata(qtperfdata_t* perfdata, bool show_states_with_zero_time)`
+        4. `void qtperf_print_group(qtstategroup_t* group)`
+        5. `void qtperf_print_perfdata(qtperfdata_t* data, bool show_states_with_zero_time)`
+        6. `void qtperf_print_group(qtstategroup_t* group)`
+        7. `qtperfcounter_t qtperf_total_group_time(qtstategroup_t* group)`
+        8. `qtperfcounter_t qtperf_total_time(qtperfdata_t* data)`
+    5. [Iterators](#iterators)
+        1. `void qtperf_iter_begin(qtperf_iterator** iter)`
+        2. `qtperfdata_t* qtperf_iter_next(qtperf_iterator_t** iter)`
+        3. `qtperfdata_t* qtperf_iter_deref(qtperf_iterator_t * iter)`
+        4. `qtperf_iterator_t* qtperf_iter_end()`
+    6. [Teardown](#teardown)
 
 # Introduction
 
@@ -101,7 +81,7 @@ they are to be tracked.
 
 Here is a quick example of a tiny program that makes use of the
 basic internal logging features of qtperf:
-
+```
     #include<qthread/qthread.h>
     #include<qthread/performance.h>
     #include<qthread/logging.h>
@@ -149,9 +129,9 @@ basic internal logging features of qtperf:
     
       return 0;
     }
-
+```
 Here is a program to demonstrate how to set up and use a custom state group:
-
+```
     #include<qthread/qthread.h>
     #include<qthread/performance.h>
     #include<qthread/logging.h>
@@ -292,7 +272,7 @@ Here is a program to demonstrate how to set up and use a custom state group:
     
       return 0;
     }
-
+```
 API
 
 ## Setup Functions

configure.ac

@@ -155,6 +155,11 @@ AC_ARG_ENABLE([internal-spinlock],
               [AS_HELP_STRING([--disable-internal-spinlock],
                               [avoid using the internal spinlock])])
 
+AC_ARG_ENABLE([performance-monitoring],
+              [AS_HELP_STRING([--enable-performance-monitoring],
+                              [enable performance monitoring infrastructure])])
+
+
 AC_ARG_ENABLE([debug],
               [AS_HELP_STRING([--enable-debug=[[areas]]],
                               [prints out debugging information as programs
@@ -742,6 +747,12 @@ AC_CACHE_SAVE
 # Figure out if we need makecontext
 QTHREAD_PICK_CONTEXT_TYPE(qthread_makecontext_type)
 
+AS_IF([test "x$enable_performance_monitoring" = "xyes"],
+[AC_DEFINE([QTHREAD_PERFORMANCE], [1], [Defined if performance monitoring support desired]),
+[disable_lazy_threadids=yes]
+[enable_picky=yes]
+])
+
 AS_IF([test "x$with_gcd" = "xshift"],
       [AC_DEFINE([QTHREAD_SHIFT_GCD],[1],[use a shift-based gcd algorithm])])
 
@@ -1060,6 +1071,8 @@ AS_IF([test "x$enable_multinode" = "xyes"],
        LIBS="$LIBS $portals4_LIBS $portals4_runtime_LIBS"
       ])
 
+
+
 AS_IF([test "x$enable_lf_febs" == "xyes"],
       [AC_DEFINE([LOCK_FREE_FEBS], [1], [Define to use a lock-free hash table for FEB metadata.])],
       [enable_lf_febs=no])
@@ -1093,6 +1106,7 @@ AM_CONDITIONAL([COMPILE_EUREKAS], [test "x$enable_eurekas" = "xyes"])
 AM_CONDITIONAL([HAVE_GUARD_PAGES], [test "x$enable_guard_pages" = "xyes"])
 AM_CONDITIONAL([HAVE_PROG_TIMELIMIT], [test "x$timelimit_path" != "x"])
 AM_CONDITIONAL([COMPILE_MULTINODE], [test "$enable_multinode" = "yes"])
+AM_CONDITIONAL([QTHREAD_PERFORMANCE], [test "$enable_performance_monitoring" = "yes"])
 AM_CONDITIONAL([WANT_SINGLE_WORKER_SCHEDULER], [test "x$with_scheduler" = "xnemesis" -o "x$with_scheduler" = "xlifo" -o "x$with_scheduler" = "xmutexfifo" -o "x$with_scheduler" = "xmtsfifo" -o "x$with_scheduler" = "xmdlifo"])
 AM_CONDITIONAL([COMPILE_OMP_BENCHMARKS], [test "x$have_openmp" = "xyes"])
 AM_CONDITIONAL([COMPILE_TBB_BENCHMARKS], [test "x$have_tbb" = "xyes"])

include/qt_shepherd_innards.h

@@ -2,7 +2,7 @@
 #define QT_SHEPHERD_INNARDS_H
 
 #ifdef QTHREAD_PERFORMANCE
-#include<qthread/performance.h>
+#include <qthread/performance.h>
 #endif
 
 /* System Pre-requisites */
@@ -48,7 +48,7 @@ struct qthread_worker_s {
     qthread_worker_id_t       worker_id;
     qthread_worker_id_t       packed_worker_id;
 #ifdef QTHREAD_PERFORMANCE
-    qtperfdata_t*             performance_data;
+    struct qtperfdata_s*             performance_data;
 #endif
     Q_ALIGNED(8) uint_fast8_t QTHREAD_CASLOCK(active);
 };

include/qthread/Makefile.am

@@ -30,6 +30,8 @@ pkginclude_HEADERS = \
 	wavefront.h \
 	loop_templates.hpp \
 	loop_iter.hpp \
+	performance.h \
+	logging.h \
 	loop.hpp
 
 # These headers are generated by ./configure

include/qthread/performance.h

@@ -36,6 +36,10 @@
  * \version 0.1
  */
 
+#ifdef __cplusplus 
+extern "C" {
+#endif
+
 #ifndef QT_PERFORMANCE_H
 #define QT_PERFORMANCE_H
 #include<stddef.h>
@@ -45,7 +49,9 @@
 #  define PERFDBG 0
 #endif
 
+#ifndef __cplusplus
 typedef unsigned char bool;
+#endif
 typedef size_t qtperfid_t;
 typedef unsigned long long qtperfcounter_t;
 typedef unsigned long qttimestamp_t;
@@ -680,4 +686,9 @@ bool qtperf_check_invariants(void);
 #endif // ifdef QTHREAD_PERFORMANCE
 
 
+#endif
+
+    /* Declarations of this file */
+#ifdef __cplusplus
+}
 #endif

src/performance.c

@@ -159,9 +159,6 @@ qtperfdata_t* qtperf_create_perfdata(qtstategroup_t* state_group) {
 }
 
 void qtperf_free_perfdata_internals(qtperfdata_t* perfdata) {
-  qt_free(perfdata->perf_counters);
-  perfdata->perf_counters = NULL;
-  perfdata->state_group = NULL;
   // don't spin lock here, it will deadlock from qtperf_free_perf_list
   if(perfdata->piggybacks != NULL) {
     size_t index=0;