| 4adc1515f867b26c19c2f7498e9de93a230a234d |
24-Oct-2014 |
Pyry Haulos <phaulos@google.com> |
Fix JNI usage issues in Android TestLogParser
* Android TestLogParser used CallObjectMethod() on Java methods returning
void, which causes CheckJNI to throw an error. Changed to use correct
CallVoidMethod().
* TestLogListener wrapper was not freeing local jstring references,
which given large parse buffer size could easily lead to ART local
reference table overflow. Fixed by freeing all local references as soon
as the objects are passed to the Java / VM side.
Change-Id: I19811be6fdc0f3d77bc7a0d99d243529031d3b08
cuTestLogParserJNI.cpp
|
| fd1df7eaea78d2c9c6bdb3da51d5f1255c479694 |
23-Oct-2014 |
Pyry Haulos <phaulos@google.com> |
Remove invalid assert from RenderActivity::setThread()
It is valid to call setThread() with null thread as part of activity
teardown, so the assert was invalid.
Change-Id: I0b3a89db1f5aff04c97def32b1881e0d66e5d250
cuAndroidRenderActivity.cpp
|
| 217426c2d19d5247e094e363615d7a0b84daf3a7 |
02-Oct-2014 |
Pyry Haulos <phaulos@google.com> |
Make Android integration more robust for API >= 11
Recent targetSdkLevel bump exposed robustness issues in the Android
port which were hidden by pre-honeycomb workarounds in
android.app.ActivityThread. For honeycomb and up when launching activity
with lockscreen on or screen off the system for some reason first starts
the activity once, puts it back to stopped state and then resumes again.
Original Android port assumed that stop event signaled that the test
execution should always stop and process should kill itself. This might
have been necessary workaround on some older Android versions where
onDestroy was never called even if app called finish().
This commit improves Android integration in multiple ways:
* Android port can continue test execution even across stop-start
boundary in the Activity life-cycle.
* eglu::GLRenderContext can recover from window loss by recreating
the window if NativeWindow::postIterate() throws WindowDestroyedError.
* eglu::GLRenderContext refreshes window surface size in postIterate()
call. Note that most test cases might not like if surface is resized
during execution though.
* Temporary window loss (recreation) no longer leads to test run abort,
but rather Android NativeWindow correctly signals that event allowing
eglu::GLRenderContext to recover.
Change-Id: Iaa240f16a2ae6cea7fb894c5772d1e7959a1cfac
cuAndroidPlatform.cpp
cuAndroidPlatform.hpp
cuAndroidRenderActivity.cpp
cuAndroidRenderActivity.hpp
cuAndroidTestActivity.cpp
cuAndroidTestActivity.hpp
cuAndroidWindow.cpp
cuAndroidWindow.hpp
|
| 67e56c9a76571b87c58a174bdad44fa1425d40fe |
30-Sep-2014 |
Pyry Haulos <phaulos@google.com> |
Clean up msg queue implementation in RenderActivity
tcu::Android::Message type (which RenderActivity uses to communicate
with the render thread) didn't validate payload type against the message
type. This change fixes that.
Change-Id: I2f120783e4f1bb5b7d0122bb8849c429577c267c
cuAndroidRenderActivity.cpp
cuAndroidRenderActivity.hpp
|
| 0d71d061a51cfdcb78b44e10a5c9fac87eb16cf7 |
30-Sep-2014 |
Brian Carlstrom <bdc@google.com> |
Add missing jclass argument to static native methods
Bug: 17691683
Change-Id: Ibfa039ea23ec25ab4974f2c1675d427638c66367
cuTestLogParserJNI.cpp
|
| 3c827367444ee418f129b2c238299f49d3264554 |
02-Sep-2014 |
Jarkko Poyry <jpoyry@google.com> |
Import dEQP.
Import drawElements Quality Program from an internal repository.
Bug: 17388917
Change-Id: Ic109fe4a57e31b2a816113d90fbdf51a43e7abeb
cuAndroidAssets.cpp
cuAndroidAssets.hpp
cuAndroidExecService.cpp
cuAndroidExecService.hpp
cuAndroidJNI.cpp
cuAndroidMain.cpp
cuAndroidNativeActivity.cpp
cuAndroidNativeActivity.hpp
cuAndroidPlatform.cpp
cuAndroidPlatform.hpp
cuAndroidRenderActivity.cpp
cuAndroidRenderActivity.hpp
cuAndroidTestActivity.cpp
cuAndroidTestActivity.hpp
cuAndroidUtil.cpp
cuAndroidUtil.hpp
cuAndroidWindow.cpp
cuAndroidWindow.hpp
cuTestLogParserJNI.cpp
|