Add the ability to start a voice at a particular time

4 files changed, 49 insertions, 9 deletions

diff --git a/al/source.cpp b/al/source.cpp
index 6daa3f4e..93f3285b 100644
--- a/al/source.cpp
+++ b/al/source.cpp
@@ -644,6 +644,7 @@ bool SetVoiceOffset(Voice *oldvoice, const VoicePos &vpos, ALsource *source, ALC
     newvoice->mPosition.store(vpos.pos, std::memory_order_relaxed);
     newvoice->mPositionFrac.store(vpos.frac, std::memory_order_relaxed);
     newvoice->mCurrentBuffer.store(vpos.bufferitem, std::memory_order_relaxed);
+    newvoice->mStartTime = oldvoice->mStartTime;
     newvoice->mFlags.reset();
     if(vpos.pos > 0 || vpos.frac > 0 || vpos.bufferitem != &source->mQueue.front())
         newvoice->mFlags.set(VoiceIsFading);
@@ -3198,6 +3199,7 @@ START_API_FUNC
         voice->mPosition.store(0, std::memory_order_relaxed);
         voice->mPositionFrac.store(0, std::memory_order_relaxed);
         voice->mCurrentBuffer.store(&source->mQueue.front(), std::memory_order_relaxed);
+        voice->mStartTime = std::chrono::nanoseconds::min();
         voice->mFlags.reset();
         /* A source that's not playing or paused has any offset applied when it
          * starts playing.
diff --git a/alc/alu.cpp b/alc/alu.cpp
index bc39b108..647b163b 100644
--- a/alc/alu.cpp
+++ b/alc/alu.cpp
@@ -105,6 +105,7 @@ static_assert(!(MaxResamplerPadding&1), "MaxResamplerPadding is not a multiple o
 namespace {
 
 using uint = unsigned int;
+using namespace std::chrono;
 
 constexpr uint MaxPitch{10};
 
@@ -1841,6 +1842,9 @@ void ProcessContexts(DeviceBase *device, const uint SamplesToDo)
 {
     ASSUME(SamplesToDo > 0);
 
+    const nanoseconds curtime{device->ClockBase +
+        nanoseconds{seconds{device->SamplesDone}}/device->Frequency};
+
     for(ContextBase *ctx : *device->mContexts.load(std::memory_order_acquire))
     {
         const EffectSlotArray &auxslots = *ctx->mActiveAuxSlots.load(std::memory_order_acquire);
@@ -1861,7 +1865,7 @@ void ProcessContexts(DeviceBase *device, const uint SamplesToDo)
         {
             const Voice::State vstate{voice->mPlayState.load(std::memory_order_acquire)};
             if(vstate != Voice::Stopped && vstate != Voice::Pending)
-                voice->mix(vstate, ctx, SamplesToDo);
+                voice->mix(vstate, ctx, curtime, SamplesToDo);
         }
 
         /* Process effects. */
diff --git a/core/voice.cpp b/core/voice.cpp
index 349c3430..bf13d563 100644
--- a/core/voice.cpp
+++ b/core/voice.cpp
@@ -60,6 +60,7 @@ Resampler ResamplerDefault{Resampler::Linear};
 namespace {
 
 using uint = unsigned int;
+using namespace std::chrono;
 
 using HrtfMixerFunc = void(*)(const float *InSamples, float2 *AccumSamples, const uint IrSize,
     const MixHrtfFilter *hrtfparams, const size_t BufferSize);
@@ -456,12 +457,16 @@ void DoNfcMix(const al::span<const float> samples, FloatBufferLine *OutBuffer, D
 
 } // namespace
 
-void Voice::mix(const State vstate, ContextBase *Context, const uint SamplesToDo)
+void Voice::mix(const State vstate, ContextBase *Context, const nanoseconds deviceTime,
+    const uint SamplesToDo)
 {
     static constexpr std::array<float,MAX_OUTPUT_CHANNELS> SilentTarget{};
 
     ASSUME(SamplesToDo > 0);
 
+    DeviceBase *Device{Context->mDevice};
+    const uint NumSends{Device->NumAuxSends};
+
     /* Get voice info */
     int DataPosInt{mPosition.load(std::memory_order_relaxed)};
     uint DataPosFrac{mPositionFrac.load(std::memory_order_relaxed)};
@@ -478,13 +483,37 @@ void Voice::mix(const State vstate, ContextBase *Context, const uint SamplesToDo
         return;
     }
 
-    DeviceBase *Device{Context->mDevice};
-    const uint NumSends{Device->NumAuxSends};
+    uint Counter{mFlags.test(VoiceIsFading) ? SamplesToDo : 0};
+    uint OutPos{0u};
 
-    ResamplerFunc Resample{(increment == MixerFracOne && DataPosFrac == 0) ?
-                           Resample_<CopyTag,CTag> : mResampler};
+    /* Check if we're doing a delayed start, and we start in this update. */
+    if(mStartTime > deviceTime)
+    {
+        /* If the start time is too far ahead, don't bother. */
+        auto diff = mStartTime - deviceTime;
+        if(diff >= seconds{1})
+            return;
+
+        /* Get the number of samples ahead of the current time that output
+         * should start at. Skip this update if it's beyond the output sample
+         * count.
+         */
+        seconds::rep sampleOffset{duration_cast<seconds>(diff * Device->Frequency).count()};
+        if(sampleOffset >= SamplesToDo)
+            return;
+
+        /* Round the start position down to a multiple of 4, which some mixers
+         * want. This makes the start time accurate to 4 samples. This could be
+         * made sample-accurate by forcing non-SIMD functions on the first run.
+         *
+         * Also ensure any fading completes within this update (though don't go
+         * less than 64 samples, or the fade could be too fast).
+         */
+        OutPos = static_cast<uint>(sampleOffset) & ~3u;
+        if(Counter > 0)
+            Counter = maxu(Counter-OutPos, 64);
+    }
 
-    uint Counter{mFlags.test(VoiceIsFading) ? SamplesToDo : 0};
     if(!Counter)
     {
         /* No fading, just overwrite the old/current params. */
@@ -517,9 +546,10 @@ void Voice::mix(const State vstate, ContextBase *Context, const uint SamplesToDo
     std::transform(Device->mSampleData.end() - mChans.size(), Device->mSampleData.end(),
         MixingSamples.begin(), offset_bufferline);
 
+    const ResamplerFunc Resample{(increment == MixerFracOne && DataPosFrac == 0) ?
+        Resample_<CopyTag,CTag> : mResampler};
     const uint PostPadding{MaxResamplerEdge + mDecoderPadding};
     uint buffers_done{0u};
-    uint OutPos{0u};
     do {
         /* Figure out how many buffer samples will be needed */
         uint DstBufferSize{SamplesToDo - OutPos};
diff --git a/core/voice.h b/core/voice.h
index cf558341..df0c8c9e 100644
--- a/core/voice.h
+++ b/core/voice.h
@@ -4,6 +4,7 @@
 #include <array>
 #include <atomic>
 #include <bitset>
+#include <chrono>
 #include <memory>
 #include <stddef.h>
 #include <string>
@@ -209,6 +210,8 @@ struct Voice {
      */
     std::atomic<VoiceBufferItem*> mLoopBuffer;
 
+    std::chrono::nanoseconds mStartTime{};
+
     /* Properties for the attached buffer(s). */
     FmtChannels mFmtChannels;
     FmtType mFmtType;
@@ -262,7 +265,8 @@ struct Voice {
     Voice(const Voice&) = delete;
     Voice& operator=(const Voice&) = delete;
 
-    void mix(const State vstate, ContextBase *Context, const uint SamplesToDo);
+    void mix(const State vstate, ContextBase *Context, const std::chrono::nanoseconds deviceTime,
+        const uint SamplesToDo);
 
     void prepare(DeviceBase *device);