4 **apitrace** consists of a set of tools to:
6 * trace OpenGL, OpenGL ES, Direct3D, and DirectDraw APIs calls to a file;
8 * replay OpenGL and OpenGL ES calls from a file;
10 * inspect OpenGL state at any call while retracing;
12 * visualize and edit trace files.
14 See the [apitrace homepage](http://apitrace.github.io/) for more details.
17 Obtaining **apitrace**
18 ======================
20 To obtain apitrace either [download the latest
21 binaries](http://apitrace.github.io/#download) for your platform if
22 available, or follow the instructions in INSTALL.markdown to build it yourself.
23 On 64bits Linux and Windows platforms you'll need apitrace binaries that match
24 the architecture (32bits or 64bits) of the application being traced.
30 Run the application you want to trace as
32 apitrace trace --api API /path/to/application [args...]
34 and it will generate a trace named `application.trace` in the current
35 directory. You can specify the written trace filename by passing the
36 `--output` command line option.
38 Problems while tracing (e.g, if the application uses calls/parameters
39 unsupported by apitrace) will be reported via stderr output on Unices. On
40 Windows you'll need to run
41 [DebugView](http://technet.microsoft.com/en-us/sysinternals/bb896647) to view
44 Follow the "Tracing manually" instructions below if you cannot obtain a trace.
48 apitrace dump application.trace
50 Replay an OpenGL trace with
52 apitrace replay application.trace
54 Pass the `--sb` option to use a single buffered visual. Pass `--help` to
55 `apitrace replay` for more options.
63 qapitrace application.trace
65 You can also tell the GUI to go directly to a specific call
67 qapitrace application.trace 12345
70 Advanced command line usage
71 ===========================
77 Several tools take `CALLSET` arguments, e.g:
79 apitrace dump --calls=CALLSET foo.trace
80 apitrace dump-images --calls=CALLSET foo.trace
82 The call syntax is very flexible. Here are a few examples:
86 * `0,2,4,5` set of calls
88 * `"0 2 4 5"` set of calls (commas are optional and can be replaced with whitespace)
90 * `0-100/2` calls 1, 3, 5, ..., 99
92 * `0-1000/draw` all draw calls between 0 and 1000
94 * `0-1000/fbo` all fbo changes between calls 0 and 1000
96 * `frame` all calls at end of frames
98 * `@foo.txt` read call numbers from `foo.txt`, using the same syntax as above
107 On 64 bits systems, you'll need to determine ether the application is 64 bits
108 or 32 bits. This can be done by doing
110 file /path/to/application
112 But beware of wrapper shell scripts -- what matters is the architecture of the
115 Run the GLX application you want to trace as
117 LD_PRELOAD=/path/to/apitrace/wrappers/glxtrace.so /path/to/application
119 and it will generate a trace named `application.trace` in the current
120 directory. You can specify the written trace filename by setting the
121 `TRACE_FILE` environment variable before running.
123 For EGL applications you will need to use `egltrace.so` instead of
126 The `LD_PRELOAD` mechanism should work with the majority applications. There
127 are some applications (e.g., Unigine Heaven, Android GPU emulator, etc.), that
128 have global function pointers with the same name as GL entrypoints, living in a
129 shared object that wasn't linked with `-Bsymbolic` flag, so relocations to
130 those globals function pointers get overwritten with the address to our wrapper
131 library, and the application will segfault when trying to write to them. For
132 these applications it is possible to trace by using `glxtrace.so` as an
133 ordinary `libGL.so` and injecting it via `LD_LIBRARY_PATH`:
135 ln -s glxtrace.so wrappers/libGL.so
136 ln -s glxtrace.so wrappers/libGL.so.1
137 ln -s glxtrace.so wrappers/libGL.so.1.2
138 export LD_LIBRARY_PATH=/path/to/apitrace/wrappers:$LD_LIBRARY_PATH
139 export TRACE_LIBGL=/path/to/real/libGL.so.1
142 If you are an application developer, you can avoid this either by linking with
143 `-Bsymbolic` flag, or by using some unique prefix for your function pointers.
145 See the `ld.so` man page for more information about `LD_PRELOAD` and
146 `LD_LIBRARY_PATH` environment flags.
148 To trace the application inside gdb, invoke gdb as:
150 gdb --ex 'set exec-wrapper env LD_PRELOAD=/path/to/glxtrace.so' --args /path/to/application
154 To trace standalone native OpenGL ES applications, use
155 `LD_PRELOAD=/path/to/egltrace.so /path/to/application` like described in the
156 previous section. To trace Java applications, refer to Dalvik.markdown.
160 Run the application you want to trace as
162 DYLD_LIBRARY_PATH=/path/to/apitrace/wrappers /path/to/application
164 Note that although Mac OS X has an `LD_PRELOAD` equivalent,
165 `DYLD_INSERT_LIBRARIES`, it is mostly useless because it only works with
166 `DYLD_FORCE_FLAT_NAMESPACE=1` which breaks most applications. See the `dyld` man
167 page for more details about these environment flags.
171 When tracing third-party applications, you can identify the target
172 application's main executable, either by:
174 * right clicking on the application's icon in the _Start Menu_, choose
175 _Properties_, and see the _Target_ field;
177 * or by starting the application, run Windows Task Manager (taskmgr.exe), right
178 click on the application name in the _Applications_ tab, choose _Go To Process_,
179 note the highlighted _Image Name_, and search it on `C:\Program Files` or
180 `C:\Program Files (x86)`.
182 On 64 bits Windows, you'll need to determine ether the application is a 64 bits
183 or 32 bits. 32 bits applications will have a `*32` suffix in the _Image Name_
184 column of the _Processes_ tab of _Windows Task Manager_ window.
186 Copy the appropriate `opengl32.dll`, `d3d8.dll`, or `d3d9.dll` from the
187 wrappers directory to the directory with the application you want to trace.
188 Then run the application as usual.
190 You can specify the written trace filename by setting the `TRACE_FILE`
191 environment variable before running.
193 For D3D10 and higher you really must use `apitrace trace -a DXGI ...`. This is
194 because D3D10-11 API span many DLLs which depend on each other, and once a DLL
195 with a given name is loaded Windows will reuse it for LoadLibrary calls of the
196 same name, causing internal calls to be traced erroneously. `apitrace trace`
197 solves this issue by injecting a DLL `dxgitrace.dll` and patching all modules
198 to hook only the APIs of interest.
201 Emitting annotations to the trace
202 ---------------------------------
204 From OpenGL applications you can embed annotations in the trace file through the
205 [`GL_GREMEDY_string_marker`](http://www.opengl.org/registry/specs/GREMEDY/string_marker.txt)
207 [`GL_GREMEDY_frame_terminator`](http://www.opengl.org/registry/specs/GREMEDY/frame_terminator.txt)
210 **apitrace** will advertise and intercept these GL extensions independently of
211 the GL implementation. So all you have to do is to use these extensions when
214 For example, if you use [GLEW](http://glew.sourceforge.net/) to dynamically
215 detect and use GL extensions, you could easily accomplish this by doing:
219 if (GLEW_GREMEDY_string_marker) {
220 glStringMarkerGREMEDY(0, __FUNCTION__ ": enter");
225 if (GLEW_GREMEDY_string_marker) {
226 glStringMarkerGREMEDY(0, __FUNCTION__ ": leave");
231 This has the added advantage of working equally well with gDEBugger.
234 From OpenGL ES applications you can embed annotations in the trace file through the
235 [`GL_EXT_debug_marker`](http://www.khronos.org/registry/gles/extensions/EXT/EXT_debug_marker.txt)
239 For Direct3D applications you can follow the standard procedure for
240 [adding user defined events to Visual Studio Graphics Debugger / PIX](http://msdn.microsoft.com/en-us/library/vstudio/hh873200.aspx):
242 - `D3DPERF_BeginEvent`, `D3DPERF_EndEvent`, and `D3DPERF_SetMarker` for D3D9 applications.
244 - `ID3DUserDefinedAnnotation::BeginEvent`,
245 `ID3DUserDefinedAnnotation::EndEvent`, and
246 `ID3DUserDefinedAnnotation::SetMarker` for D3D11.1 applications.
249 Dump GL state at a particular call
250 ----------------------------------
252 You can get a dump of the bound GL state at call 12345 by doing:
254 apitrace replay -D 12345 application.trace > 12345.json
256 This is precisely the mechanism the GUI obtains its own state.
258 You can compare two state dumps by doing:
260 apitrace diff-state 12345.json 67890.json
263 Comparing two traces side by side
264 ---------------------------------
266 apitrace diff trace1.trace trace2.trace
268 This works only on Unices, and it will truncate the traces due to performance
272 Recording a video with FFmpeg/Libav
273 -----------------------------------
275 You can make a video of the output with FFmpeg by doing
277 apitrace dump-images -o - application.trace \
278 | ffmpeg -r 30 -f image2pipe -vcodec ppm -i pipe: -vcodec mpeg4 -y output.mp4
280 or Libav (which replaces FFmpeg on recent Debian/Ubuntu distros) doing
282 apitrace dump-images -o - application.trace \
283 | avconv -r 30 -f image2pipe -vcodec ppm -i - -vcodec mpeg4 -y output.mp4
285 Recording a video with gstreamer
286 --------------------------------------
288 You can make a video of the output with gstreamer by doing
290 glretrace --snapshot-format=RGB -s - smokinguns.trace | gst-launch-0.10 fdsrc blocksize=409600 ! queue \
291 ! videoparse format=rgb width=1920 height=1080 ! queue ! ffmpegcolorspace ! queue \
292 ! vaapiupload direct-rendering=0 ! queue ! vaapiencodeh264 ! filesink location=xxx.264
297 You can truncate a trace by doing:
299 apitrace trim --exact --calls 0-12345 -o trimed.trace application.trace
301 If you need precise control over which calls to trim you can specify the
302 individual call numbers a plaintext file, as described in the 'Call sets'
305 There is also experimental support for automatically trimming the calls
306 necessary for a given frame or call:
308 apitrace trim --auto --calls=12345 -o trimed.trace application.trace
309 apitrace trim --auto --frames=12345 -o trimed.trace application.trace
315 You can perform gpu and cpu profiling with the command line options:
317 * `--pgpu` record gpu times for frames and draw calls.
319 * `--pcpu` record cpu times for frames and draw calls.
321 * `--ppd` record pixels drawn for each draw call.
323 The results from this can then be read by hand or analysed with a script.
325 `scripts/profileshader.py` will read the profile results and format them into a
326 table which displays profiling results per shader.
328 For example, to record all profiling data and utilise the per shader script:
330 apitrace replay --pgpu --pcpu --ppd foo.trace | ./scripts/profileshader.py
333 Advanced usage for OpenGL implementors
334 ======================================
336 There are several advanced usage examples meant for OpenGL implementors.
342 These are the steps to create a regression test-suite around **apitrace**:
346 * obtain reference snapshots, by doing on a reference system:
348 mkdir /path/to/reference/snapshots/
349 apitrace dump-images -o /path/to/reference/snapshots/ application.trace
351 * prune the snapshots which are not interesting
353 * to do a regression test, use `apitrace diff-images`:
355 apitrace dump-images -o /path/to/test/snapshots/ application.trace
356 apitrace diff-images --output summary.html /path/to/reference/snapshots/ /path/to/test/snapshots/
359 Automated git-bisection
360 -----------------------
362 With tracecheck.py it is possible to automate git bisect and pinpoint the
363 commit responsible for a regression.
365 Below is an example of using tracecheck.py to bisect a regression in the
366 Mesa-based Intel 965 driver. But the procedure could be applied to any GL
367 driver hosted on a git repository.
369 First, create a build script, named build-script.sh, containing:
373 export PATH=/usr/lib/ccache:$PATH
376 ./autogen.sh --disable-egl --disable-gallium --disable-glut --disable-glu --disable-glw --with-dri-drivers=i965
380 It is important that builds are both robust, and efficient. Due to broken
381 dependency discovery in Mesa's makefile system, it was necessary invoke `make
382 clean` in every iteration step. `ccache` should be installed to avoid
383 recompiling unchanged source files.
388 export LIBGL_DEBUG=verbose
389 export LD_LIBRARY_PATH=$PWD/lib
390 export LIBGL_DRIVERS_DIR=$PWD/lib
392 6491e9593d5cbc5644eb02593a2f562447efdcbb 71acbb54f49089b03d3498b6f88c1681d3f649ac \
393 -- src/mesa/drivers/dri/intel src/mesa/drivers/dri/i965/
394 git bisect run /path/to/tracecheck.py \
395 --precision-threshold 8.0 \
396 --build /path/to/build-script.sh \
397 --gl-renderer '.*Mesa.*Intel.*' \
398 --retrace=/path/to/glretrace \
399 -c /path/to/reference/snapshots/ \
400 topogun-1.06-orc-84k.trace
402 The trace-check.py script will skip automatically when there are build
405 The `--gl-renderer` option will also cause a commit to be skipped if the
406 `GL_RENDERER` is unexpected (e.g., when a software renderer or another GL
407 driver is unintentionally loaded due to missing symbol in the DRI driver, or
408 another runtime fault).
411 Side by side retracing
412 ----------------------
414 In order to determine which draw call a regression first manifests one could
415 generate snapshots for every draw call, using the `-S` option. That is, however,
416 very inefficient for big traces with many draw calls.
418 A faster approach is to run both the bad and a good GL driver side-by-side.
419 The latter can be either a previously known good build of the GL driver, or a
420 reference software renderer.
422 This can be achieved with retracediff.py script, which invokes glretrace with
423 different environments, allowing to choose the desired GL driver by
424 manipulating variables such as `LD_LIBRARY_PATH`, `LIBGL_DRIVERS_DIR`, or
427 For example, on Linux:
429 ./scripts/retracediff.py \
430 --ref-env LD_LIBRARY_PATH=/path/to/reference/GL/implementation \
431 --retrace /path/to/glretrace \
432 --diff-prefix=/path/to/output/diffs \
437 python scripts\retracediff.py --retrace \path\to\glretrace.exe --ref-env TRACE_LIBGL=\path\to\reference\opengl32.dll application.trace
443 qapitrace has rudimentary support for replaying traces on a remote
444 target device. This can be useful, for example, when developing for an
445 embedded system. The primary GUI will run on the local host, while any
446 replays will be performed on the target device.
448 In order to target a remote device, use the command-line:
450 qapitrace --remote-target <HOST> <trace-file>
452 In order for this to work, the following must be available in the
453 system configuration:
455 1. It must be possible for the current user to initiate an ssh session
456 that has access to the target's window system. The command to be
457 exectuted by qapitrace will be:
461 For example, if the target device is using the X window system, one
462 can test whether an ssh session has access to the target X server
467 If this command fails with something like "cannot open display"
468 then the user will have to configure the target to set the DISPLAY
469 environment variable, (for example, setting DISPLAY=:0 in the
470 .bashrc file on the target or similar).
472 Also, note that if the ssh session requires a custom username, then
473 this must be configured on the host side so that ssh can be
474 initiated without a username.
476 For example, if you normally connect with `ssh user@192.168.0.2`
477 you could configure ~/.ssh/config on the host with a block such as:
483 And after this you should be able to connect with `ssh target` so
484 that you can also use `qapitrace --remote-target target`.
486 2. The target host must have a functional glretrace binary available
488 3. The target host must have access to <trace-file> at the same path
489 in the filesystem as the <trace-file> path on the host system being
490 passed to the qapitrace command line.
493 [![githalytics.com alpha](https://cruel-carlota.pagodabox.com/c1062ad633aa7a458e9d7520021307e4 "githalytics.com")](http://githalytics.com/apitrace/apitrace)