2135 Commits

This Branch

This Branch

All Branches

Author	SHA1	Message	Date
Cameron Gutman	b603bf4887	Add a checkbox to favor image quality over performance	2013-11-26 20:03:55 -05:00
Cameron Gutman	8e22a125e2	Remove AvcDecoder.SLICE_THREADING flag which was making performance worse. Scale the thread count higher on low performing devices. Use a constant 30 fps target since rendering tends to have very little effect on decoding speed. Only wait when the wait time will exceed 8 ms in order to reduce the likelyhood that the wait causes us to render late.	2013-11-26 14:36:06 -05:00
Cameron Gutman	55d534c8cf	Fix off-by-one error in prefix matching	2013-11-25 21:48:35 -05:00
Cameron Gutman	734d80b281	Blacklist the Rockchip H264 decoders. Blacklist another software decoder that doesn't support H264 L5.0 high profile	2013-11-24 19:55:12 -05:00
Cameron Gutman	5360d6c9b2	Hint the GC to run at the completion of each frame. This forces concurrent garbage collection which means that the likelyhood of a long GC pause while other threads are running is lower.	2013-11-24 19:32:50 -05:00
Cameron Gutman	dd82573d08	Switch from ArrayBlockingQueues to LinkedBlockingQueues for increased throughput. Set thread priorities for the decoder and ping threads.	2013-11-24 19:31:50 -05:00
Cameron Gutman	d9527a3810	Optimize performance flags.	2013-11-24 19:29:26 -05:00
Cameron Gutman	9be14c3d92	Add thread names for easier debugging	2013-11-24 17:57:00 -05:00
Cameron Gutman	732ea8a3ac	Optimize the reinitialization of the buffer descriptor in the receive threads	2013-11-24 17:29:12 -05:00
Cameron Gutman	9a3c5dac0a	Suppress connection failure warnings when closing gracefully	2013-11-24 21:52:33 -05:00
Cameron Gutman	7d39ac2a45	Reuse a single AvByteBufferDescriptor when depacketizing H264 instead of allocating many of them per packet	2013-11-24 21:51:09 -05:00
Cameron Gutman	fb8b6fd7f5	Add configurable ffmpeg performance options	2013-11-24 21:46:47 -05:00
Cameron Gutman	c8b94217cf	Set the datagram destination manually for ping packets rather than connecting the datagram socket explicitly. This prevents the ICMP port unreachable message from being propagated to the receive thread which causes an IOException.	2013-11-23 17:07:33 -05:00
Cameron Gutman	0fe7549942	Loosen the connection timeouts. Remove the HTTP request timeout because it depends on how long it takes for the requested app to start.	2013-11-23 16:48:46 -05:00
Cameron Gutman	022866288d	Add more logging for exceptions.	2013-11-23 14:06:51 -05:00
Cameron Gutman	125bcb370f	Limit the size of queues. Switch to a blacklist instead of a whitelist for hardware decoding. Currently, Nvidia, TI, and Google H264 codecs are on the blacklist. Reenable large heap since we're allocating more objects now.	2013-11-23 02:44:10 -05:00
Cameron Gutman	4d5849f448	Add UI indications of the connection status and failure	2013-11-23 01:03:37 -05:00
Cameron Gutman	c88d23001e	Commit forgotten ConnectionStatusListener interface. Thanks to danielb7390 for pointing this out.	2013-11-22 22:13:39 -05:00
Cameron Gutman	2946966456	Whitelist the Exynos hardware H264 decoder	2013-11-22 02:13:02 -05:00
Cameron Gutman	d00474c1cd	Fix warning.	2013-11-22 00:31:39 -05:00
Cameron Gutman	49b2332f14	Make the IP textview single line with no suggestions. Set focus to the start streaming button by default.	2013-11-22 00:31:25 -05:00
Cameron Gutman	db78af916d	Handle decode units larger than the size of the decode unit buffer	2013-11-22 00:29:48 -05:00
Cameron Gutman	e1a8f86d4f	Make the virtual and hardware back buttons on the devices themselves work as the in-game back button. This is required for Shield's back button to work.	2013-11-21 19:56:44 -05:00
Cameron Gutman	80fdae3673	Request an IDR frame if a previous frame was lost or	2013-11-21 19:55:40 -05:00
Cameron Gutman	e52b85a883	Restore the performance level fed into the AvcDecoder	2013-11-21 16:59:04 -05:00
Cameron Gutman	6bb215eddf	Remove pooling code due to higher concurrency latency that resulted in a net loss of performance.	2013-11-21 16:40:27 -05:00
Cameron Gutman	1d7460e8b3	Revert "WIP optimization of object allocation and pools. There's currently a double free on the AvByteBufferDescriptor that I'm trying to track down." ... This reverts commit 6600329580aace81321220d2e4000099ad27f4f7.	2013-11-21 16:25:10 -05:00
Cameron Gutman	3efa356bb8	Revert "Add double-free detection code to the pools" ... This reverts commit 9cdcfd2522febcc2eb7f65d926a37cfa05f21323.	2013-11-21 16:25:01 -05:00
Cameron Gutman	45120e79e4	Revert "Fix the double-free bug and disable double free debugging." ... This reverts commit 9bb395501172f754baacc1efe6f09af55627fa15.	2013-11-21 16:24:52 -05:00
Cameron Gutman	ff50b5fdab	Revert "Fix memory leaks of special NAL sequences" ... This reverts commit d9987105d0d9c27e920b4113875e82b315d050a4.	2013-11-21 16:24:47 -05:00
Cameron Gutman	e6959e66e4	Fix memory leaks of special NAL sequences	2013-11-21 14:36:17 -05:00
Cameron Gutman	d5f59bb972	Remove large heap flag.	2013-11-21 13:08:07 -05:00
Cameron Gutman	9e4364f672	Fix the double-free bug and disable double free debugging.	2013-11-21 13:05:01 -05:00
Cameron Gutman	acb3dc8881	Add double-free detection code to the pools	2013-11-21 12:55:05 -05:00
Cameron Gutman	7951c0fa13	WIP optimization of object allocation and pools. There's currently a double free on the AvByteBufferDescriptor that I'm trying to track down.	2013-11-21 12:24:45 -05:00
Cameron Gutman	c42d40b8f6	Remove an extra frame copy that wasn't needed. Only update the stored frame if the previous frame had been displayed. Don't render the same frame again if there's no new frame to update.	2013-11-21 09:55:57 -05:00
Cameron Gutman	ab9b7b5164	Discard the alpha channel of the surface since it's not used. Always use MED_PERF for the AvcDecoder configuration because it now works well enough on Tegra 3 that the image quality improvements are worth it.	2013-11-21 08:45:25 -05:00
Cameron Gutman	45664dac2a	Draw directly to the surface buffer. Improve amount of decoding and rendering that can be done in parallel. Add performance levels and choose them by cpuinfo. Improves Tegra 3 performance significantly.	2013-11-21 08:38:49 -05:00
Cameron Gutman	54839e672d	Fix multithreaded decoding to significantly improve Tegra 3 decoding performance. It's not as smooth as Tegra 4 CPU decoding or Snapdragon hardware decoding, but it's much better than it was before.	2013-11-20 03:07:52 -05:00
Cameron Gutman	9b46155af2	Log a message based on what type of decoding is being used	2013-11-20 01:35:39 -05:00
Cameron Gutman	5a4a0d55e3	Enable both landscape orientations	2013-11-20 01:33:58 -05:00
Cameron Gutman	bd6338730d	Don't decode video on the emulator for performance reasons.	2013-11-20 01:33:38 -05:00
Cameron Gutman	f672befaa7	Revert an earlier commit that introduced a bug that caused initialization to crash	2013-11-20 00:59:12 -05:00
Cameron Gutman	34905dd01e	Fix crashes caused by rendering to a surface on teardown	2013-11-20 00:32:37 -05:00
Cameron Gutman	16d629f5cf	Remove jlibrtp (finally). CPU decoding support (works great on Tegra 4, not so well on Tegra 3)	2013-11-20 00:00:00 -05:00
Cameron Gutman	0504bed5e9	Fix ffmpeg library loading. Create a DecoderRenderer interface and use it to move the MediaCodec code into for Qualcomm devices.	2013-11-19 02:49:33 -05:00
Cameron Gutman	2c2e713166	Rework our JNI interface and add ffmpeg bindings for AVC decoding (WIP)	2013-11-18 18:58:49 -05:00
Cameron Gutman	cca7df6c25	Reassemble AVC NALs when padding is reached, rather than waiting through several packets until the next AVC frame start NAL. Improves video latency and vulnerability to packet loss.	2013-11-17 16:50:45 -05:00
Cameron Gutman	10165f3166	Fix frame pacing issues on Snapdragon. Improve latency on Tegra (not perfect still).	2013-11-15 23:42:00 -05:00
Cameron Gutman	79db2ed584	Fix a bug with MAC address selection that caused it to always select the first adapter	2013-11-14 17:30:40 -05:00