2a25d97e40dbd5dd9195195a3a269095a4932097 |
|
05-Oct-2010 |
Glenn Kasten <gkasten@google.com> |
Object an interface initialization/destroy cleanup Prior to this, object and interface fields were being initialized and deinitialized in random places. This resulted in some real and potential leaks, especially in the case of partial object construction. There were even some places where fields were not being initialized at all. Details: Move destructor for AndroidEffectCapabilities from CEngine::Destroy to IAndroidEffectCapabilities deinit hook. Move destructor for equalizer preset names from CEngine::Destroy to IEngine deinit hook. In Object::Destroy, call deinitializer for each initialized interface, regardless whether it was ever exposed. Remove double initialization of mStepSize in IPlaybackRate constructor. Remove obsolete logs. Add expose hooks that are called when interface is requested in Engine::Create... or by DynamicInterface::AddInterface. Always initialize all interfaces, even if not exposed Fix typos in same-line comments on INTERFACE_ constants. Add new interface state INITIALIZED: uninitialized -> initialized -> exposed. DynamicInterface::Remove sets interface state to initialized (but not exposed), instead of uninitialized. Add explicit constructor calls to initialize fields in C structs (e.g. smart pointers), using placement new. See also build warning which was turned off with -Wno-invalid-offsetof. Add comments about per-instance fields not associated with an interface. Add default initializations for all per-instance fields that are not associated with an interface, in case the object is only partially constructed and then needs to be destroyed due to an error. Add placement new comments. Update comments on private per-instance fields not associated with an interface. Move effect clears from the containing object destructor (CAudioPlayer_Destroy and COutputMix_Destroy) to the interface destructor. In this process this caught a bug that environmental reverb and preset reverb were not being destroyed properly for an audio player. Add comments on interface hooks. Add entries for interface destructors. Add placement new (explicit constructor) and explicit destructor calls for C++ fields in C struct. Fix template instantiation error. Move explicit calls to IBufferQueue_Destroy (now renamed to IBufferQueue_deinit) from CAudioPlayer_Destroy and CAudioRecorder_Destroy to be implicit via the BufferQueue interface deinit hook. ThreadSync interface destructor now logs a warning if ThreadSync::EnterCriticalSection was active at Engine::Destroy Add some comments. Clean up gMonitorFp close code. Fix typo in log message. Fix DynamicInterface to handle distinction between UNINITIALIZED and INITIALIZED states for an interface, and to call the optional expose hook. Do not use memset per interface, because the size of the last interface is unknown. Set the v-table pointer while lock is held. Add curlies to if statements. Log errors if interfaces active during Object::Destroy, instead of asserting. Log a warning if interface requests an interface but it is not available, and is optional. Add log with the class name and address of each new object. Mark only interfaces with init hooks as INITIALIZED; others are UNINITIALIZED. Change-Id: Ibb9c1b5f21e6195af6b20b2f7a5c7157905b735b
/system/media/opensles/libopensles/CAudioPlayer.c
|
182f822e7110bac4947ba8bded1c90ce8050c0fd |
|
30-Sep-2010 |
Glenn Kasten <gkasten@google.com> |
Add new test, fix bugs found by test & code review Bugs: - EnvironmentalReverb::GetEnvironmentalReverbProperties left mutex unlocked at end. - buffer queue configuration test revealed not checking enough combinations - would have failed conformance test in a couple places - replace peek locks by shared locks for SMP - undocumented symbols were being exported by library - fix handling of EffectSend for MIDI player New test: - Add reverb test Miscellaneous: - Move some FIXME to to-do list where they are easier to organize - Fix typos in name of profile constant - EffectSend::IsEnabled sets enabled = false for unsupported aux effect. - Add curlies to if statements. - Add more error logging. - Fix indentation in switch statement. - Add comments to makefile. Details: - Reduce visibility of libOpenSLESUT symbols. - Use LOCAL_STATIC_LIBRARIES := libOpenSLESUT where needed in applications. - Move FIXME for OEM configuration, interface, SDL, Sndfile, buffer queue, destroy to to-do list - For SDL, disallow buffer queue with source not 44.1 kHz because it is hard-wired to 44.1 kHz and there is no sample rate converter. - return successful result for enabled effect send on unsupported MIDI player for future conformance testing Change-Id: I39cb543169996f78c16e3ac363ea6157a9242743
/system/media/opensles/libopensles/CAudioPlayer.c
|
ec3a5a5555af3ed612b70dc54ef96998a9256b9a |
|
28-Sep-2010 |
Glenn Kasten <gkasten@google.com> |
Add MuteSolo checks for audio player and fix bugs The specification says "This [SLMuteSoloItf] interface cannot be exposed on a player whose audio format is mono." This has a lot of implications ... Deny an explicit interface request for SLMuteSoloItf at audio player creation time if the player is known pre-realize to have mono channel count (e.g. PCM buffer queue). Deny a GetInterface on MuteSolo if channel count is known to be mono, but allow it if the channel count is still unknown. When DynamicSource (or its replacement) is implemented, it will be possible for the application to have done a GetInterface successfully, and then to change the source such that the MuteSolo interface is no longer valid. In that case, return SL_RESULT_FEATURE_UNSUPPORTED for most MuteSolo operations, if called when channel count is unknown or is mono. We _do_ allow MuteSolo::GetChannelCount to succeed if channel count is known to be mono. Enhance the URI-based mute solo test to check some of the above, and the interactive buffer queue test to get and test the channel count. Add curly braces to if statements. Other bug fixes: - Do not rely on peek locks; they don't work on SMP. - Channel count and sample rate were not being set correctly for buffer queue data source audio players. - CreateAudioPlayer and AudioRecorder were both referring to application pointers more than once, with the possibility of reading inconsistent data. Change-Id: Ie0109cbddc0aff8a56c0b53d989fb2be823e627b
/system/media/opensles/libopensles/CAudioPlayer.c
|
3ac54e4d6e699467b608a4f0eda6023d50d3cd61 |
|
23-Sep-2010 |
Glenn Kasten <gkasten@google.com> |
Fix bug found by BufferQueue automated test Fix bug for destroying partially constructed audio player. Remove obsolete logging. Change-Id: Ib186153bf50ed1311c021d53711ec90d000bdee2
/system/media/opensles/libopensles/CAudioPlayer.c
|
51cb31b09f6af53402b3fbe7e9de29badc1155a2 |
|
21-Sep-2010 |
Glenn Kasten <gkasten@google.com> |
As part of API review, disable incomplete profiles Don't compile dead code relating to unsupported features, but keep profiles for CT. Move whole file conditionals from the source code to Android.mk, by just not compiling the module for unsupported features. There must an interface initialization hook present to expose an interface implicitly or explicitly at object creation time, or to dynamically add an interface after the object is created. As a safety check, we also double-check for the initialization hook when the application does a GetInterface. Miscellaneous bug fixes: - Fix warning "variable may be used uninitialized in this function". - Fix build errors on SDL. - Was using the wrong conditional for SDL. - Fix typo mInterfaceContinued -> mInterfaceStates2 in AudioRecorder. - Fix typo in IEngineCapabilities_init. Other cleanup: - Move ugly SDL-specific stuff out of the portable code. - Combine duplicated code in CAudioPlayerDestroy and CAudioRecorder_Destroy for freeing buffer queue memory. - Add curlies to if and do statements. - Line length 100. - Decrease engine voices from 32 to 30 (based on max of 32 objects, 1 engine, 1 output mix). - Disable IRecord_init log. - Add symbolic link for Android configuration header file. - Enable trace on Android. - sl_int64_t is now unconditionally defined rather than only if ANDROID defined, because it is used by OpenSLES_Android.h which might be included by SDL also. - Removed symlink for Android headers Change-Id: I1a6389257bca09e3819923129a1193473e16669b
/system/media/opensles/libopensles/CAudioPlayer.c
|
31df22b193ea7c7c331d26a27fa6756a89c8ec3c |
|
17-Sep-2010 |
Glenn Kasten <gkasten@google.com> |
Improve reliability of Object::Destroy / Create Separated object construction from publishing, for better safety with respect to the sync and debugger thread. Unpublished objects exist and have an ID slot reserved, but the ID is not assigned. Destroy of an unpublished object does the right thing. Add pre-destroy checks: - COutputMix all platforms: add check for players attached to output mix. Object:: Destroy on an output mix that has at least one attached player is a no-op, but will log an error. - COutputMix for SDL: now asks the mixer callback to unlink output mix from engine, and then blocks until SDL mixer callback acknowledges the request; this avoids a race condition where the SDL thread calls the mixer during destroy. - C3DGroup: add check for players in group. Object::Destroy on a 3D group that contains at least one player is a no-op, but will log an error. Object::Destroy on an audio player that is part of a 3D group now removes the player from the 3D group during destroy. Remove redundant field: - CAudioPlayer.mOutputMix duplicated a field in DataSink.DataLocator_OutputMix; removed mOutputMix and replaced by function CAudioPlayer_GetOutputMix which gets the output mix from the data sink. Remove dead code: - IOutputMixExt_Destroy was obsolete. Miscellaneous changes not related to destroy: - Fixed an ancient bug in the class table for Listener object - was using wrong offset for interfaces. - Add curly braces around if, else, for bodies - Add parentheses around complex conditionals - Add some logging - Fix indentation. - Line length 100 Change-Id: I0035fe356e2a28c1932b5d265f881377aa77dc58
/system/media/opensles/libopensles/CAudioPlayer.c
|
4f924ff768d761f53db6fa2dbfb794ba7a65e776 |
|
10-Sep-2010 |
Glenn Kasten <gkasten@google.com> |
Fix some race conditions and deadlocks. Use strong reference counts to fix race conditions in these object dependencies: - audio player on 3D Grouping - MIDI player (or more specifically, the data sink locator) depends on LED or Vibra device - audio player (or more specifically, the data sink locator) depends on output mix Fix build errors on non-Android platforms. Fix incorrect result code for EnvironmentalReverb::SetDensity. Continue work on making Destroy more reliable: - Destroy calls a pre-destroy hook, e.g. for Android audio players. - Add new internal object state DESTROYING which is set when the pre-destroy hook returns. - Abort (called by Destroy) now blocks until all async operations complete or are aborted. Add curly braces around simple if and for statements. Add extra parentheses around complex if expression in if statements. Add more error logging in checkDataLocator. Add Doxygen comments. Remove spurious log entry when SL_DATALOCATOR_ANDROIDFD used. Disable the unused Resume hooks, which are currently unused. Remove some dead code in audio player create for Android. Line length 100. Prepare for fixing race condition in OutputMixExt by moving mActiveMask and mTracks from IOutputMix to IOutputMixExt. Update comments. Continued work on Destroy reliability: Do not poke around in mixer's data structures directly during Object::Destroy for an audio player; instead ask the mixer nicely and wait for it to do the job. Move master copy of computed gains from Track to AudioPlayer, and then let the mixer copy the gains locally when it needs them. Add resume and predestroy hooks everywhere. Predestroy hook returns a boolean.
/system/media/opensles/libopensles/CAudioPlayer.c
|
66f75c45c9aea410b1f913d76995661e72571b67 |
|
31-Aug-2010 |
Glenn Kasten <gkasten@google.com> |
Fix problems found by the mute solo test. Bug fixes for USE_OUTPUTMIXEXT (desktop): - SL_PLAYEVENT_HEADATNEWPOS was broken - GetPosition is now accurate, with or without seeking - Fix race condition between AudioPlayer and Track - Duration is updated if initial estimate is too low (e.g. if unknown is reported as zero) - GetPosition while pending Seek now returns that value so the Seek appears synchronous - Remove useless code to retry Enqueues; now drop decoded data if Enqueue fails For all source files touched, add curlies to the "if" statements. Change-Id: I1f70d8e683bc92831848af20857347d45c6407a9
/system/media/opensles/libopensles/CAudioPlayer.c
|
9c03f04a9c6cc2a821182c8be8f2efe964a27efe |
|
13-Aug-2010 |
Glenn Kasten <gkasten@google.com> |
Make internal implicit interfaces safer. Add IsInterfaceInitialized. Don't directly manipulate a buffer queue. Call the interface via it's named entry points, not the v-table as it may not be initialized/available. GetInterface enables the v-table not init. Update Doxygen comments. Line length 100. Allow bool, true, and false in C code. Add a verbose option to seek torture test. Change-Id: I37c42834422759b881d2ffe4b171d9aa284fde98
/system/media/opensles/libopensles/CAudioPlayer.c
|
6aa121077e8981b7d9f46471bdaa1a11211057e8 |
|
27-Jul-2010 |
Glenn Kasten <gkasten@google.com> |
Fix leak of track slot when audio player destroyed Move the link from audio player to associated output mix / track from the effect send interface up to the audio player class. Add some more comments. Line length 100. Change-Id: I4e2de5e673709e686f066d0bd852bac17996810f
/system/media/opensles/libopensles/CAudioPlayer.c
|
10ee2bc6119445f4339ecef998c40c9cc95057cd |
|
24-Jul-2010 |
Glenn Kasten <gkasten@google.com> |
Add and update comments and use typedefs more. Add Doxygen-style comments and convert some existing comments to Doxygen style. Use typedef instead of struct as type name, e.g. rename struct Track to Track. Change-Id: I251476c68304c9734210fcc5fe01d89544345af2
/system/media/opensles/libopensles/CAudioPlayer.c
|
23c38816f7c210afae5072fd44658c98fec7e119 |
|
17-Jul-2010 |
Glenn Kasten <gkasten@google.com> |
Buffer queue and audio player maintenance. Known problems on Android only (SDL is ok): High latency on first enqueue, or first after buffer underrun Play after stop does not resume the current buffer. Improvements: Implement BufferQueue::Clear for both Android and SDL. Play::SetPlayState to Stopped now works for SDL (not yet for Android), and sets play cursor to start of current buffer. Play::Pause and Play::Stop while playing now block until done for SDL. SDL now calls buffer queue callback after each buffer completion, instead of when mixer needs a buffer. Move SF_INFO sfinfo to CAudioPlayer so we always have good stuff like the sample rate etc. BufferQueue::Enqueue on a successful enqueue to a formerly empty queue now kicks off an attribute change followed by a synchronous call to a hook when first buffer enqueued. Currently unused but could be a way to reduce latency. Minor comment changes. Add interface_cond_broadcast. More logging of errors. Include source file and line number in logs. sles_to_android.cpp: needs code for #if 0 sections. Disable trace logging of BufferQueue::Enqueue and GetState. Fix printf warnings. Fix typo in number of interfaces requested at CreateAudioPlayer. Remove spurious printf and while loop. Line length 100. Add intermediate play states which need to be re-mapped. Fix printf format warnings. Re-organize the buffer queue initialization. Implement true seeking based on position for libsndfile. SndFile is working again, but just barely. Make it build on non-Android platforms. BufferQueue::RegisterCallback now checks for STOPPED play state. Include function name in trace logs. Enhance test to check buffer queue callbacks, get buffer queue state, get play state. Fix bug with Stop while Paused not resetting buffer pointer. Remove obsolete comments and #if 0. Set enqueue attribute is state is PLAYING and the first buffer is enqueued. Added slutPrintIIDs. Update comments. Fix build warnings. Fix compile error introduced by the merge. Fix runtime problems introduced during the merge. Address code review comments: - BufferQueue changes apply to AudioRecord also. - Rename to conform to new naming conventions. - Add comments. Fix build warning. Add interactive buffer queue test application. Change-Id: I5b00545a599fd9dba96efb834fca2369d149eaaa
/system/media/opensles/libopensles/CAudioPlayer.c
|
f66a508b11e327670a6bf5a8cf9c106f215e3b39 |
|
21-Jul-2010 |
Jean-Michel Trivi <jmtrivi@google.com> |
Harmonize naming scheme for Android AudioPlayer related functions with what is used for AudioRecorder. Change-Id: I96ed88ff17a5fff12b2e17dd6cb84630cba8fad8
/system/media/opensles/libopensles/CAudioPlayer.c
|
158dbd1a71a890d29490198c8e5019081ac94ac3 |
|
14-Jul-2010 |
Glenn Kasten <gkasten@google.com> |
Logging and comments to make testing easier. Add utility function to print an IID. checkDataFormat now prints a detailed log on invalid format. Use SL_LOGE and SL_LOGV for logging instead of fprintf, we can easily redirect to standard Android logging eventually. Add Doxygen-style comments. Log more specific error message on invalid data format. Use the full copyright header in one file. Change-Id: I2342e29f57e97c3b5d94cb219d00e767a75e93fa
/system/media/opensles/libopensles/CAudioPlayer.c
|
f51f5c6ec99ebc8f2c833a68f232bc1c874a7f2f |
|
09-Jul-2010 |
Glenn Kasten <gkasten@google.com> |
Bug fix and general code cleanup. Trace debug now uses Android logging by LOGV/LOGE. Added SL Utility Toolkit for OpenSL ES, similar to OpenAL UT. Add a -DUSE_TRACE instead of -DNDEBUG and add it commented out to Android.mk. LOGE for error returns, LOGV for all entry and exit. Rename debug.c to trace.c. slutPrintIID prints the symbolic name if known. Fix bug in slQueryNumSupportedEngineInterfaces and slQuerySupportedEngineInterfaces which were not skipping over unavailable interfaces. Use USE_CONFORMANCE to return SL_RESULT_FEATURE_UNSUPPORTED when not supported and make some interfaces unavailable depending on USE_CONFORMANCE. Move most FIXME to separate bug/to-do list. Fix possible intermediate overflow in duration computation for sndfile. Line length 100. Address code review comments, disable LOG.
/system/media/opensles/libopensles/CAudioPlayer.c
|
9c2a7b1e9f8203bf9de26efca0f1f805038b1f2b |
|
29-Jun-2010 |
Glenn Kasten <gkasten@google.com> |
Restore previous version of CAudioPlayer fields. Addressed code review comments: Move channel mute, channel solo, and channel count from IMuteSolo to CAudioPlayer. Deferred support for IMuteSolo to CMIDIPlayer deferred until we hear back from working group. Move comment marker to correct location. We discussed the code review comments and agreed to restore the previous handling of channel mute, channel solo, and channel count. We will probably have to review this again later, depending on what the working group says about MIDI Player. To make it easier to examine the history later, I'm submitting the previous change as is, followed immediately by this change which addresses the code review comments. Change-Id: I4d5594d70c1c2d5fb90980a8a0414177ff29d4a7
/system/media/opensles/libopensles/CAudioPlayer.c
|
15f9f5e609265dff9d6036af38bea084c42a702a |
|
24-Jun-2010 |
Glenn Kasten <gkasten@google.com> |
Gain and transport attributes at mutex unlock. Update both interface and player copies of mute/solo. Move synchronous gain and transport updates to mutex unlock exclusive. Update sync thread logic. Keep track of player position using tracks in PCM frame units. Implement solo. Various volume-related fixes. Mute solo fields are only in IMuteSolo, but IMuteSolo is immediately after IVolume to make it easier to find. Move Android volume fields into audio player. Fix printf missing newline. Only one copy of mute and solo. Clean up comments. Line length 100. SDL gain and transport updates are synchronous. Fix unused variable warnings. Use symbol for number of stereo channels. Rearrange fields to save memory. Add comments. Fix problems introduced during the last merge. Change-Id: I121fcbe4189b06850983941f540942a745fc51fb
/system/media/opensles/libopensles/CAudioPlayer.c
|
da18bcf20d4cceb715904717a5a8b1585f935080 |
|
24-Jun-2010 |
Jean-Michel Trivi <jmtrivi@google.com> |
Implement the SLMuteSoloItf interface. Change-Id: I5ed3112046c4317fe752bc2545b3e168f7e8a18a
/system/media/opensles/libopensles/CAudioPlayer.c
|
9db8335361c67c4ed877f68218835df96ce43a4b |
|
21-Jun-2010 |
Jean-Michel Trivi <jmtrivi@google.com> |
For Android code, use ANDROID flag rather then USE_ANDROID. Change-Id: I4b5877d8b6aee37bb472c817a2a863104271a03b
/system/media/opensles/libopensles/CAudioPlayer.c
|
3cdc48ac145e12a40b0ff0f90e8c01bec269bc38 |
|
16-Jun-2010 |
Glenn Kasten <gkasten@google.com> |
Add limited volume support for USE_SNDFILE platform. Fix build warnings. Remove obsolete FIXME.
/system/media/opensles/libopensles/CAudioPlayer.c
|
95364617b16ca2e322a4e0faeda93a2750d29fd6 |
|
11-Jun-2010 |
Glenn Kasten <gkasten@google.com> |
Make test apps work on host platform. Change-Id: Ia5ed73b0ff77eb16959b3a539ac33ea9e89a9f4c
/system/media/opensles/libopensles/CAudioPlayer.c
|
09f49a84b259d8b8c49a4d71b42ea801b2a9e8f2 |
|
02-Jun-2010 |
Glenn Kasten <gkasten@google.com> |
Improve resource recovery. Better cleanup in AudioPlayer.Destroy. Avoid mallocs in LEDArray and Equalizer. Line length 100. Change-Id: I2d22880b27ccf7e9038b398cae9287781fd82253 Continued work on threading. Lay the groundwork for true asynchronous realize. Call all callbacks with mutex unlocked. C volatile is meaningless. Start adding support for suspend/resume on interfaces. Line length 100. Change-Id: I631cb4f123143e4ef79c6c491d12b1e559857cab
/system/media/opensles/libopensles/CAudioPlayer.c
|
f6f90fc5202248e2ae0abde6e655d7186cebc6af |
|
29-May-2010 |
Glenn Kasten <gkasten@google.com> |
Move platform-specific code to separate files. Use sles_allinclusive.h to really mean all inclusive. MuteSolo check channel count. Change-Id: I96caab5be55a1f04342eb67b511309babd5a6916 Make a copy of data source and sink parameter to avoid dangling inputs. Pre-process URI scheme and MIME type using a perfect hash. Line length 100. Change-Id: Icb1e1a7a9c6370eda1e15e61f0dff35c1d59d7fa Fix merge problem on Android. Change-Id: I70b724273147f3981550272990729a4a409e1f72 MIME type needs to be completely initialized. Change-Id: I3766acfb170abc976506cf305fabd5548107f592 Remove MIME lookup and URI scheme lookup. Change-Id: I7058dfef01c9db509ad5915d9f8099a720e6736b
/system/media/opensles/libopensles/CAudioPlayer.c
|
0263d22efc4ad9420b367ec61b5a1d1ec72cebb9 |
|
27-May-2010 |
Jean-Michel Trivi <jmtrivi@google.com> |
First pass at implementing support of SL_DATALOCATOR_URI in an SL ES audio player data source. Change-Id: Ie8dc3b88c667035cbb20d2180eed6f699b6e7c2b
/system/media/opensles/libopensles/CAudioPlayer.c
|
188a1af5a1d2e2c339cded3dc5bbd99c2fd89fef |
|
29-May-2010 |
Glenn Kasten <gkasten@google.com> |
Continue splitting up source code. Change-Id: Iad6caa5adf111f50ad1a3aa14fc31efc7d5b285e
/system/media/opensles/libopensles/CAudioPlayer.c
|