| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
|
|
Smaller footprint, and simplifies refactoring.
|
|
- allow the user to choose and pull a file from device and use open it
in ApiTrace GUI.
- allow the user to choose a local (EGL) file and push it to the device
for retrace.
- allow the user link a local file with a file on the device.
- retrace on android.
|
|
Retracer now knows what thumbnails are needed. This information needs
to be communicated to the glretrace command.
Construct a string that contains the call numbers for all missing thumbnails.
The constructed string is then passed to glretrace with the "-S" argument.
TODO: we want to be able to detect contiguous sequences of calls and emit
them as a range. However, I wasn't able to get this working correctly and
will defer this to a future patch.
|
|
The previous implementation assumed that all frame thumbnails would be
provided to the trace by the retracer. Now, not only are just a subset of
frame thumbnails being provided, but call thumbnails may also be present,
depending on where the user was looking.
Therefore, rather than process a list of thumbnails, one for each frame,
this patch modifies the code to process a hash table of thumbnails which
is addressed by the call index.
In addition, the code now needs to extract the call index that was encoded
by glretrace as an image comment. Therefore, the PNM parser will now
attempt to interpret the comment in the PNM header as a number and return
it in addition to the other header data.
In order for Qt to manage the modified message data, main.cpp now has to
define a different data packet as well (ImageHash). I tried to not define
that new type but the Qt preprocessor didn't like the two type parameters
I was using.
Finally, ApiTrace::bindThumbnails is made more generic to deal with the
sparse data in the hash table as well as potentially assign call
thumbnails as well as frame thumbnails.
v2: Attach inner thumbnails to the right call. Calls may be filtered,
so it's not possible to find the right call merely by indexing. -- Jose
|
|
Now that the trace knows what thumbnails are needed, that information
needs to be communicated to the retracer to actually capture them.
The trace is modified to iterate through the missing thumbnails. For
each of these thumbnail indices, the trace calls a callback to notify
it of the thumbnail.
I don't really like my solution to this but I couldn't figure out a
better one. What I wanted to do was have the trace hand out an iterator
along the lines of what Java or C# do, but I couldn't figure out how
to create one. If the reviewers of this code have a suggestion, I'm all
ears.
The main window provides the callback and acts as an intermediate to the
retracer. The callback receives the missing index and passes it to the
retracer, which then adds the index to the collection of missing thumbnail
indices it is gathering.
|
|
This was preventing using qapitrace to debug issues using Mesa. The
application being debugged used GL 3.3, but since a core profile wasn't
requested, the max Mesa would give was 3.0.
|
|
Some traces fail to replay (hang) if glretrace does not have
--singelthread on the command line, this change makes a UI option
available.
|
|
This is supported with a new command-line argument:
qapitrace --remote-target <HOST> <trace-file>
See README.markdown for documentation on usage and current limitations
of this feature.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
This patch sets the stage for displaying frame thumbnails in
qapitrace by first capturing a snaphot stream.
We first prepare qapitrace to capture a snapshot stream. We enhance
replayTrace() to accept a second parameter for specifying thumbnail
capture. This then enables us to notify the retracer so set up command
line parameters to glretrace, which runs out of process. We specify that
glretrace must generate snapshots and then emit them to stdout.
After glretrace returns, we extract PNM images from the output stream.
We parse PNM the ASCII header to extract number channels, the width and
the height of the following binary image data. We then extract the
binary data of the image on a per-scanline copy. We create a QImage of
the correct size and insert the image data into the QImage. Finally, we
scale the size of the image down to the size of a thumbnail.
Lastly, we create a QList collection to hold the sequence of thumbnails.
This list is return to the main window object of qapitrace and will
later be used to display these thumbnails.
|
|
|
|
On glretrace crash, both error and finished QProcess events are emitted, by
this order. Members were reset on error, causing null pointer dereference
on replayFinished.
Fix this by handling abnormal termination on replayFinished too.
replayError should probably removed/merged into replayFinished.
|
|
|
|
This gets down the memory usage down quite a bit.
|
|
|
|
|
|
preparing for adding options to benchmark and use double buffering
from the gui.
|
