moonlight-stream/moonlight-android
1
0
Fork 0
mirror of https://github.com/moonlight-stream/moonlight-android.git synced 2025-07-19 11:03:01 +00:00
Code Issues Packages Projects Releases Wiki Activity

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.

Browse Source
This commit is contained in:
Cameron Gutman 2013-11-26 14:36:06 -05:00
parent 55d534c8cf
commit 8e22a125e2
1 changed files with 32 additions and 30 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

62
src/com/limelight/nvstream/av/video/CpuDecoderRenderer.java
View File

@ -15,12 +15,17 @@ public class CpuDecoderRenderer implements DecoderRenderer {
private Surface renderTarget;
private ByteBuffer decoderBuffer;
private Thread rendererThread;
private int perfLevel;
private int targetFps;
// Only sleep if the difference is above this value
private static final int WAIT_CEILING_MS = 8;
private static final int LOW_PERF = 1;
private static final int MED_PERF = 2;
private static final int HIGH_PERF = 3;
private int cpuCount = Runtime.getRuntime().availableProcessors();
private int findOptimalPerformanceLevel() {
StringBuilder cpuInfo = new StringBuilder();
BufferedReader br = null;
@ -71,27 +76,40 @@ public class CpuDecoderRenderer implements DecoderRenderer {
@Override
public void setup(int width, int height, Surface renderTarget) {
this.renderTarget = renderTarget;
this.perfLevel = findOptimalPerformanceLevel();
this.targetFps = 30;
int perfLevel = findOptimalPerformanceLevel();
int threadCount;
int flags = 0;
switch (perfLevel) {
case HIGH_PERF:
// Single threaded low latency decode is ideal but hard to acheive
flags = AvcDecoder.LOW_LATENCY_DECODE;
threadCount = 1;
break;
case LOW_PERF:
// Disable the loop filter for performance reasons
flags = AvcDecoder.DISABLE_LOOP_FILTER |
AvcDecoder.FAST_BILINEAR_FILTERING |
AvcDecoder.FAST_DECODE |
AvcDecoder.SLICE_THREADING;
AvcDecoder.FAST_DECODE;
// Use plenty of threads to try to utilize the CPU as best we can
threadCount = cpuCount - 1;
break;
default:
case MED_PERF:
flags = AvcDecoder.BILINEAR_FILTERING |
AvcDecoder.FAST_DECODE |
AvcDecoder.SLICE_THREADING;
break;
case HIGH_PERF:
flags = AvcDecoder.LOW_LATENCY_DECODE;
AvcDecoder.FAST_DECODE;
// Only use 2 threads to minimize frame processing latency
threadCount = 2;
break;
}
int err = AvcDecoder.init(width, height, flags, 2);
int err = AvcDecoder.init(width, height, flags, threadCount);
if (err != 0) {
throw new IllegalStateException("AVC decoder initialization failure: "+err);
}
@ -106,38 +124,22 @@ public class CpuDecoderRenderer implements DecoderRenderer {
rendererThread = new Thread() {
@Override
public void run() {
int frameRateTarget;
long nextFrameTime = System.currentTimeMillis();
switch (perfLevel) {
case HIGH_PERF:
frameRateTarget = 45;
break;
case MED_PERF:
frameRateTarget = 30;
break;
case LOW_PERF:
default:
frameRateTarget = 15;
break;
}
while (!isInterrupted())
{
long diff = nextFrameTime - System.currentTimeMillis();
if (diff > 0) {
// Sleep until the frame should be rendered
if (diff > WAIT_CEILING_MS) {
try {
Thread.sleep(diff);
} catch (InterruptedException e) {
return;
}
}
nextFrameTime = computePresentationTimeMs(frameRateTarget);
nextFrameTime = computePresentationTimeMs(targetFps);
AvcDecoder.redraw(renderTarget);
System.gc();
}
}
};