| 1 |
README for gtk+/perf |
| 2 |
-------------------- |
| 3 |
|
| 4 |
This is a framework for testing performance in GTK+. For GTK+, being |
| 5 |
performant does not only mean "paint widgets fast". It also means |
| 6 |
things like the time needed to set up widgets, to map and draw a |
| 7 |
window for the first time, and emitting/propagating signals. |
| 8 |
|
| 9 |
The following is accurate as of 2006/Jun/14. |
| 10 |
|
| 11 |
|
| 12 |
Background |
| 13 |
---------- |
| 14 |
|
| 15 |
A widget's lifetime looks more or less like this: |
| 16 |
|
| 17 |
1. Instantiation |
| 18 |
2. Size request |
| 19 |
3. Size allocate |
| 20 |
5. Realize |
| 21 |
4. Map |
| 22 |
5. Expose |
| 23 |
6. Destroy |
| 24 |
|
| 25 |
Some of these stages are particularly interesting: |
| 26 |
|
| 27 |
- Instantiation means creating the widget. This may be as simple as a |
| 28 |
few malloc()s and setting some fields. It could also be a |
| 29 |
complicated operation if the widget needs to contact an external |
| 30 |
server to create itself, or if it needs to read data files. |
| 31 |
|
| 32 |
- Size requisition is when GTK+ asks the widget, "how big do you want |
| 33 |
to be on the screen"? This can be an expensive operation. The |
| 34 |
widget has to measure its text, measure its icons (and thus load its |
| 35 |
icons), and generally run through its internal layout code. |
| 36 |
|
| 37 |
- Realization is when the widget creates its GDK resources, like its |
| 38 |
GdkWindow and graphics contexts it may need. This could be |
| 39 |
expensive if the widget needs to load data files for cursors or |
| 40 |
backgrounds. |
| 41 |
|
| 42 |
- Expose is when the widget gets repainted. This will happen many |
| 43 |
times throughout the lifetime of the widget: every time you drag a |
| 44 |
window on top of it, every time its data changes and it needs to |
| 45 |
redraw, every time it gets resized. |
| 46 |
|
| 47 |
GtkWidgetProfiler is a mechanism to let you get individual timings for |
| 48 |
each of the stages in the lifetime of a widget. It also lets you run |
| 49 |
some stages many times in a sequence, so that you can run a real |
| 50 |
profiler and get an adequate number of samples. For example, |
| 51 |
GtkWidgetProfiler lets you say "repaint this widget 1000 times". |
| 52 |
|
| 53 |
Why is this not as simple as doing |
| 54 |
|
| 55 |
start_timer (); |
| 56 |
for (i = 0; i < 1000; i++) { |
| 57 |
gtk_widget_queue_draw (widget); |
| 58 |
while (gtk_events_pending ()) |
| 59 |
gtk_main_iteration (); |
| 60 |
} |
| 61 |
stop_timer (); |
| 62 |
|
| 63 |
Huh? |
| 64 |
|
| 65 |
Because X is an asynchronous window system. So, when you send the |
| 66 |
"paint" commands, your program will regain control but it will take |
| 67 |
some time for the X server to actually process those commands. |
| 68 |
GtkWidgetProfiler has special code to wait for the X server and give |
| 69 |
you accurate timings. |
| 70 |
|
| 71 |
|
| 72 |
Using the framework |
| 73 |
------------------- |
| 74 |
|
| 75 |
Right now the framework is very simple; it just has utility functions |
| 76 |
to time widget creation, mapping, exposure, and destruction. To run |
| 77 |
such a test, you use the GtkWidgetProfiler object in |
| 78 |
gtkwidgetprofiler.h. |
| 79 |
|
| 80 |
The gtk_widget_profiler_profile_boot() function will emit the |
| 81 |
"create-widget" signal so that you can create your widget for |
| 82 |
testing. It will then take timings for the widget, and emit the |
| 83 |
"report" signal as appropriate. |
| 84 |
|
| 85 |
The "create-widget" signal: |
| 86 |
|
| 87 |
The handler has this form: |
| 88 |
|
| 89 |
GtkWidget *create_widget_callback (GtkWidgetProfiler *profiler, |
| 90 |
gpointer user_data); |
| 91 |
|
| 92 |
You need to create a widget in your handler, and return it. Do not |
| 93 |
show the widget; the profiler will do that by itself at the right |
| 94 |
time, and will actually complain if you show the widget. |
| 95 |
|
| 96 |
|
| 97 |
The "report" signal: |
| 98 |
|
| 99 |
This function will get called when the profiler wants to report that |
| 100 |
it finished timing an important stage in the lifecycle of your |
| 101 |
widget. The handler has this form: |
| 102 |
|
| 103 |
void report_callback (GtkWidgetProfiler *profiler, |
| 104 |
GtkWidgetProfilerReport report, |
| 105 |
GtkWidget *widget, |
| 106 |
gdouble elapsed, |
| 107 |
gpointer user_data); |
| 108 |
|
| 109 |
The "report" argument tells you what happened to your widget: |
| 110 |
|
| 111 |
GTK_WIDGET_PROFILER_REPORT_CREATE. A timer gets started right |
| 112 |
before the profiler emits the "create-widget" signal,, and it gets |
| 113 |
stopped when your callback returns with the new widget. This |
| 114 |
measures the time it takes to set up your widget, but not show it. |
| 115 |
|
| 116 |
GTK_WIDGET_PROFILER_REPORT_MAP. A timer gets started right before |
| 117 |
the profiler calls gtk_widget_show_all() on your widget, and it |
| 118 |
gets stopped when the the widget has been mapped. |
| 119 |
|
| 120 |
GTK_WIDGET_PROFILER_REPORT_EXPOSE. A timer gets started right before |
| 121 |
the profiler starts waiting for GTK+ and the X server to finish |
| 122 |
painting your widget, and it gets stopped when the widget is fully |
| 123 |
painted to the screen. |
| 124 |
|
| 125 |
GTK_WIDGET_PROFILER_REPORT_DESTROY. A timer gets started right |
| 126 |
before the profiler calls gtk_widget_destroy() on your widget, and |
| 127 |
it gets stopped when gtk_widget_destroy() returns. |
| 128 |
|
| 129 |
As a very basic example of using GtkWidgetProfiler is this: |
| 130 |
|
| 131 |
---------------------------------------------------------------------- |
| 132 |
#include <stdio.h> |
| 133 |
#include <gtk/gtk.h> |
| 134 |
#include "gtkwidgetprofiler.h" |
| 135 |
|
| 136 |
static GtkWidget * |
| 137 |
create_widget_cb (GtkWidgetProfiler *profiler, gpointer data) |
| 138 |
{ |
| 139 |
GtkWidget *window; |
| 140 |
|
| 141 |
window = gtk_window_new (GTK_WINDOW_TOPLEVEL); |
| 142 |
/* ... fill the window with widgets, and don't show them ... */ |
| 143 |
|
| 144 |
return window; |
| 145 |
} |
| 146 |
|
| 147 |
static void |
| 148 |
report_cb (GtkWidgetProfiler *profiler, GtkWidgetProfilerReport report, GtkWidget *widget, gdouble elapsed, gpointer data) |
| 149 |
{ |
| 150 |
const char *type; |
| 151 |
|
| 152 |
switch (report) { |
| 153 |
case GTK_WIDGET_PROFILER_REPORT_CREATE: |
| 154 |
type = "widget creation"; |
| 155 |
break; |
| 156 |
|
| 157 |
case GTK_WIDGET_PROFILER_REPORT_MAP: |
| 158 |
type = "widget map"; |
| 159 |
break; |
| 160 |
|
| 161 |
case GTK_WIDGET_PROFILER_REPORT_EXPOSE: |
| 162 |
type = "widget expose"; |
| 163 |
break; |
| 164 |
|
| 165 |
case GTK_WIDGET_PROFILER_REPORT_DESTROY: |
| 166 |
type = "widget destruction"; |
| 167 |
break; |
| 168 |
|
| 169 |
default: |
| 170 |
g_assert_not_reached (); |
| 171 |
type = NULL; |
| 172 |
} |
| 173 |
|
| 174 |
fprintf (stderr, "%s: %g sec\n", type, elapsed); |
| 175 |
|
| 176 |
if (report == GTK_WIDGET_PROFILER_REPORT_DESTROY) |
| 177 |
fputs ("\n", stderr); |
| 178 |
} |
| 179 |
|
| 180 |
int |
| 181 |
main (int argc, char **argv) |
| 182 |
{ |
| 183 |
GtkWidgetProfiler *profiler; |
| 184 |
|
| 185 |
gtk_init (&argc, &argv); |
| 186 |
|
| 187 |
profiler = gtk_widget_profiler_new (); |
| 188 |
g_signal_connect (profiler, "create-widget", |
| 189 |
G_CALLBACK (create_widget_cb), NULL); |
| 190 |
g_signal_connect (profiler, "report", |
| 191 |
G_CALLBACK (report_cb), NULL); |
| 192 |
|
| 193 |
gtk_widget_profiler_set_num_iterations (profiler, 100); |
| 194 |
gtk_widget_profiler_profile_boot (profiler); |
| 195 |
|
| 196 |
gtk_widget_profiler_profile_expose (profiler); |
| 197 |
|
| 198 |
return 0; |
| 199 |
} |
| 200 |
|
| 201 |
---------------------------------------------------------------------- |
| 202 |
|
| 203 |
|
| 204 |
Getting meaningful results |
| 205 |
-------------------------- |
| 206 |
|
| 207 |
Getting times for widget creation/mapping/exposing/destruction is |
| 208 |
interesting, but how do you actually find the places that need |
| 209 |
optimizing? |
| 210 |
|
| 211 |
Why, you run the tests under a profiler, of course. |
| 212 |
|
| 213 |
FIXME: document how to do this. |
| 214 |
|
| 215 |
|
| 216 |
Feedback |
| 217 |
-------- |
| 218 |
|
| 219 |
Please mail your feedback to Federico Mena-Quintero <[email protected]>. |
| 220 |
This performance framework is a work in progress. |