Pass a span of pointers to the UHJ/SuperStereo decoder

4 files changed, 29 insertions, 23 deletions

diff --git a/core/device.h b/core/device.h
index 111e26ce..e9596bb4 100644
--- a/core/device.h
+++ b/core/device.h
@@ -183,8 +183,9 @@ struct DeviceBase {
     /* Temp storage used for mixer processing. */
     static constexpr size_t MixerLineSize{BufferLineSize + MaxResamplerPadding +
         UhjDecoder::sFilterDelay};
+    static constexpr size_t MixerChannelsMax{16};
     using MixerBufferLine = std::array<float,MixerLineSize>;
-    alignas(16) std::array<MixerBufferLine,16> mSampleData;
+    alignas(16) std::array<MixerBufferLine,MixerChannelsMax> mSampleData;
 
     alignas(16) float ResampledData[BufferLineSize];
     alignas(16) float FilteredData[BufferLineSize];
diff --git a/core/uhjfilter.cpp b/core/uhjfilter.cpp
index 584718e3..a56e25bf 100644
--- a/core/uhjfilter.cpp
+++ b/core/uhjfilter.cpp
@@ -101,15 +101,15 @@ void UhjEncoder::encode(float *LeftOut, float *RightOut, const FloatBufferLine *
  * where j is a +90 degree phase shift. 3-channel UHJ excludes Q, while 2-
  * channel excludes Q and T.
  */
-void UhjDecoder::decode(const al::span<BufferLine> samples, const size_t offset,
-    const size_t samplesToDo, const size_t forwardSamples)
+void UhjDecoder::decode(const al::span<float*> samples, const size_t samplesToDo,
+    const size_t forwardSamples)
 {
     ASSUME(samplesToDo > 0);
 
     {
-        const float *RESTRICT left{samples[0].data() + offset};
-        const float *RESTRICT right{samples[1].data() + offset};
-        const float *RESTRICT t{samples[2].data() + offset};
+        const float *RESTRICT left{al::assume_aligned<16>(samples[0])};
+        const float *RESTRICT right{al::assume_aligned<16>(samples[1])};
+        const float *RESTRICT t{al::assume_aligned<16>(samples[2])};
 
         /* S = Left + Right */
         for(size_t i{0};i < samplesToDo+sFilterDelay;++i)
@@ -124,9 +124,9 @@ void UhjDecoder::decode(const al::span<BufferLine> samples, const size_t offset,
             mT[i] = t[i];
     }
 
-    float *RESTRICT woutput{samples[0].data() + offset};
-    float *RESTRICT xoutput{samples[1].data() + offset};
-    float *RESTRICT youtput{samples[2].data() + offset};
+    float *RESTRICT woutput{al::assume_aligned<16>(samples[0])};
+    float *RESTRICT xoutput{al::assume_aligned<16>(samples[1])};
+    float *RESTRICT youtput{al::assume_aligned<16>(samples[2])};
 
     /* Precompute j(0.828331*D + 0.767820*T) and store in xoutput. */
     auto tmpiter = std::copy(mDTHistory.cbegin(), mDTHistory.cend(), mTemp.begin());
@@ -154,7 +154,7 @@ void UhjDecoder::decode(const al::span<BufferLine> samples, const size_t offset,
 
     if(samples.size() > 3)
     {
-        float *RESTRICT zoutput{samples[3].data() + offset};
+        float *RESTRICT zoutput{al::assume_aligned<16>(samples[3])};
         /* Z = 1.023332*Q */
         for(size_t i{0};i < samplesToDo;++i)
             zoutput[i] = 1.023332f*zoutput[i];
@@ -174,14 +174,14 @@ void UhjDecoder::decode(const al::span<BufferLine> samples, const size_t offset,
  * where j is a +90 degree phase shift. w is a variable control for the
  * resulting stereo width, with the range 0 <= w <= 0.7.
  */
-void UhjDecoder::decodeStereo(const al::span<BufferLine> samples, const size_t offset,
-    const size_t samplesToDo, const size_t forwardSamples)
+void UhjDecoder::decodeStereo(const al::span<float*> samples, const size_t samplesToDo,
+    const size_t forwardSamples)
 {
     ASSUME(samplesToDo > 0);
 
     {
-        const float *RESTRICT left{samples[0].data() + offset};
-        const float *RESTRICT right{samples[1].data() + offset};
+        const float *RESTRICT left{al::assume_aligned<16>(samples[0])};
+        const float *RESTRICT right{al::assume_aligned<16>(samples[1])};
 
         for(size_t i{0};i < samplesToDo+sFilterDelay;++i)
             mS[i] = left[i] + right[i];
@@ -212,9 +212,9 @@ void UhjDecoder::decodeStereo(const al::span<BufferLine> samples, const size_t o
         mCurrentWidth = wtarget;
     }
 
-    float *RESTRICT woutput{samples[0].data() + offset};
-    float *RESTRICT xoutput{samples[1].data() + offset};
-    float *RESTRICT youtput{samples[2].data() + offset};
+    float *RESTRICT woutput{al::assume_aligned<16>(samples[0])};
+    float *RESTRICT xoutput{al::assume_aligned<16>(samples[1])};
+    float *RESTRICT youtput{al::assume_aligned<16>(samples[2])};
 
     /* Precompute j*D and store in xoutput. */
     auto tmpiter = std::copy(mDTHistory.cbegin(), mDTHistory.cend(), mTemp.begin());
diff --git a/core/uhjfilter.h b/core/uhjfilter.h
index 9e692599..11058700 100644
--- a/core/uhjfilter.h
+++ b/core/uhjfilter.h
@@ -66,7 +66,7 @@ struct UhjDecoder : public UhjFilterBase {
      * reconstructed from 2-channel UHJ should not be run through a normal
      * B-Format decoder, as it needs different shelf filters.
      */
-    void decode(const al::span<BufferLine> samples, const size_t offset, const size_t samplesToDo,
+    void decode(const al::span<float*> samples, const size_t samplesToDo,
         const size_t forwardSamples);
 
     /**
@@ -75,11 +75,11 @@ struct UhjDecoder : public UhjFilterBase {
      * should contain 3 channels, the first two being the left and right stereo
      * channels, and the third left empty.
      */
-    void decodeStereo(const al::span<BufferLine> samples, const size_t offset,
-        const size_t samplesToDo, const size_t forwardSamples);
+    void decodeStereo(const al::span<float*> samples, const size_t samplesToDo,
+        const size_t forwardSamples);
 
-    using DecoderFunc = void (UhjDecoder::*)(const al::span<BufferLine> samples,
-        const size_t offset, const size_t samplesToDo, const size_t forwardSamples);
+    using DecoderFunc = void (UhjDecoder::*)(const al::span<float*> samples,
+        const size_t samplesToDo, const size_t forwardSamples);
 
     DEF_NEWDEL(UhjDecoder)
 };
diff --git a/core/voice.cpp b/core/voice.cpp
index 924a8446..424184af 100644
--- a/core/voice.cpp
+++ b/core/voice.cpp
@@ -53,6 +53,7 @@ struct CopyTag;
 
 static_assert(!(sizeof(DeviceBase::MixerBufferLine)&15),
     "DeviceBase::MixerBufferLine must be a multiple of 16 bytes");
+static_assert(!(MaxResamplerEdge&3), "MaxResamplerEdge is not a multiple of 4");
 
 Resampler ResamplerDefault{Resampler::Linear};
 
@@ -627,9 +628,13 @@ void Voice::mix(const State vstate, ContextBase *Context, const uint SamplesToDo
 
             if(mDecoder)
             {
+                std::array<float*,DeviceBase::MixerChannelsMax> chanptrs;
+                std::transform(MixingSamples.begin(), MixingSamples.end(), chanptrs.begin(),
+                    [](DeviceBase::MixerBufferLine &bufline) noexcept -> float*
+                    { return bufline.data() + MaxResamplerEdge; });
                 const size_t srcOffset{(increment*DstBufferSize + DataPosFrac)>>MixerFracBits};
                 SrcBufferSize = SrcBufferSize - PostPadding + MaxResamplerEdge;
-                ((*mDecoder).*mDecoderFunc)(MixingSamples, MaxResamplerEdge, SrcBufferSize,
+                ((*mDecoder).*mDecoderFunc)({chanptrs.data(), MixingSamples.size()}, SrcBufferSize,
                     srcOffset * likely(vstate == Playing));
             }
         }