Improve a couple algorithms

3 files changed, 39 insertions, 29 deletions

diff --git a/Alc/mastering.cpp b/Alc/mastering.cpp
index 2cf6acf5..00b79e67 100644
--- a/Alc/mastering.cpp
+++ b/Alc/mastering.cpp
@@ -295,28 +295,30 @@ void GainCompressor(Compressor *Comp, const ALsizei SamplesToDo)
  */
 void SignalDelay(Compressor *Comp, const ALsizei SamplesToDo, FloatBufferLine *OutBuffer)
 {
-    static constexpr ALsizei mask{BUFFERSIZE - 1};
     const ALsizei numChans{Comp->mNumChans};
-    const ALsizei indexIn{Comp->mDelayIndex};
-    const ALsizei indexOut{Comp->mDelayIndex - Comp->mLookAhead};
+    const ALsizei lookAhead{Comp->mLookAhead};
 
     ASSUME(SamplesToDo > 0);
     ASSUME(numChans > 0);
+    ASSUME(lookAhead > 0);
 
     for(ALsizei c{0};c < numChans;c++)
     {
-        ALfloat *RESTRICT inout{al::assume_aligned<16>(OutBuffer[c].data())};
-        ALfloat *RESTRICT delay{al::assume_aligned<16>(Comp->mDelay[c])};
-        for(ALsizei i{0};i < SamplesToDo;i++)
-        {
-            const ALfloat sig{inout[i]};
+        ALfloat *inout{al::assume_aligned<16>(OutBuffer[c].data())};
+        ALfloat *delaybuf{al::assume_aligned<16>(Comp->mDelay[c].data())};
 
-            inout[i] = delay[(indexOut + i) & mask];
-            delay[(indexIn + i) & mask] = sig;
+        auto inout_end = inout + SamplesToDo;
+        if(LIKELY(SamplesToDo >= lookAhead))
+        {
+            auto delay_end = std::rotate(inout, inout_end - lookAhead, inout_end);
+            std::swap_ranges(inout, delay_end, delaybuf);
+        }
+        else
+        {
+            auto delay_start = std::swap_ranges(inout, inout_end, delaybuf);
+            std::rotate(delaybuf, delay_start, delaybuf + lookAhead);
         }
     }
-
-    Comp->mDelayIndex = (indexIn + SamplesToDo) & mask;
 }
 
 } // namespace
@@ -354,9 +356,10 @@ std::unique_ptr<Compressor> CompressorInit(const ALsizei NumChans, const ALuint
                            const ALfloat Ratio, const ALfloat KneeDb,
                            const ALfloat AttackTime, const ALfloat ReleaseTime)
 {
-    auto lookAhead = static_cast<ALsizei>(
+    const auto lookAhead = static_cast<ALsizei>(
         clampf(std::round(LookAheadTime*SampleRate), 0.0f, BUFFERSIZE-1));
-    auto hold = static_cast<ALsizei>(clampf(std::round(HoldTime*SampleRate), 0.0f, BUFFERSIZE-1));
+    const auto hold = static_cast<ALsizei>(
+        clampf(std::round(HoldTime*SampleRate), 0.0f, BUFFERSIZE-1));
 
     size_t size{sizeof(Compressor)};
     if(lookAhead > 0)
@@ -398,15 +401,15 @@ std::unique_ptr<Compressor> CompressorInit(const ALsizei NumChans, const ALuint
     {
         if(hold > 1)
         {
-            Comp->mHold = new (reinterpret_cast<void*>(Comp.get() + 1)) SlidingHold{};
+            Comp->mHold = new (static_cast<void*>(Comp.get() + 1)) SlidingHold{};
             Comp->mHold->mValues[0] = -std::numeric_limits<float>::infinity();
             Comp->mHold->mExpiries[0] = hold;
             Comp->mHold->mLength = hold;
-            Comp->mDelay = reinterpret_cast<ALfloat(*)[BUFFERSIZE]>(Comp->mHold + 1);
+            Comp->mDelay = new (static_cast<void*>(Comp->mHold + 1)) FloatBufferLine{};
         }
         else
         {
-            Comp->mDelay = reinterpret_cast<ALfloat(*)[BUFFERSIZE]>(Comp.get() + 1);
+            Comp->mDelay = new (static_cast<void*>(Comp.get() + 1)) FloatBufferLine{};
         }
     }
 
@@ -422,6 +425,9 @@ Compressor::~Compressor()
     if(mHold)
         mHold->~SlidingHold();
     mHold = nullptr;
+    if(mDelay)
+        mDelay->~FloatBufferLine();
+    mDelay = nullptr;
 }
 
 
diff --git a/Alc/mastering.h b/Alc/mastering.h
index 6cc40ba3..31d0ef97 100644
--- a/Alc/mastering.h
+++ b/Alc/mastering.h
@@ -50,8 +50,7 @@ struct Compressor {
     alignas(16) ALfloat mCrestFactor[BUFFERSIZE]{};
 
     SlidingHold *mHold{nullptr};
-    ALfloat (*mDelay)[BUFFERSIZE]{nullptr};
-    ALsizei mDelayIndex{0};
+    FloatBufferLine *mDelay{nullptr};
 
     ALfloat mCrestCoeff{0.0f};
     ALfloat mGainEstimate{0.0f};
diff --git a/Alc/uhjfilter.cpp b/Alc/uhjfilter.cpp
index acdf6f0b..c2657f9f 100644
--- a/Alc/uhjfilter.cpp
+++ b/Alc/uhjfilter.cpp
@@ -66,15 +66,17 @@ void Uhj2Encoder::encode(FloatBufferLine &LeftOut, FloatBufferLine &RightOut, Fl
 
     ASSUME(SamplesToDo > 0);
 
+    auto winput = InSamples[0].cbegin();
+    auto xinput = InSamples[1].cbegin();
+    auto yinput = InSamples[2].cbegin();
     for(ALsizei base{0};base < SamplesToDo;)
     {
-        ALsizei todo = mini(SamplesToDo - base, MAX_UPDATE_SAMPLES);
+        const ALsizei todo{mini(SamplesToDo - base, MAX_UPDATE_SAMPLES)};
         ASSUME(todo > 0);
 
         /* D = 0.6554516*Y */
-        const ALfloat *RESTRICT input{al::assume_aligned<16>(InSamples[2].data()+base)};
-        for(ALsizei i{0};i < todo;i++)
-            temp[i] = 0.6554516f*input[i];
+        std::transform(yinput+base, yinput+todo, std::begin(temp),
+            [](const float y) noexcept -> float { return 0.6554516f*y; });
         allpass_process(&mFilter1_Y[0], temp, temp, Filter1CoeffSqr[0], todo);
         allpass_process(&mFilter1_Y[1], temp, temp, Filter1CoeffSqr[1], todo);
         allpass_process(&mFilter1_Y[2], temp, temp, Filter1CoeffSqr[2], todo);
@@ -89,10 +91,9 @@ void Uhj2Encoder::encode(FloatBufferLine &LeftOut, FloatBufferLine &RightOut, Fl
         mLastY = temp[todo-1];
 
         /* D += j(-0.3420201*W + 0.5098604*X) */
-        const ALfloat *RESTRICT input0{al::assume_aligned<16>(InSamples[0].data()+base)};
-        const ALfloat *RESTRICT input1{al::assume_aligned<16>(InSamples[1].data()+base)};
-        for(ALsizei i{0};i < todo;i++)
-            temp[i] = -0.3420201f*input0[i] + 0.5098604f*input1[i];
+        std::transform(winput, winput+todo, xinput, std::begin(temp),
+            [](const float w, const float x) noexcept -> float
+            { return -0.3420201f*w + 0.5098604f*x; });
         allpass_process(&mFilter2_WX[0], temp, temp, Filter2CoeffSqr[0], todo);
         allpass_process(&mFilter2_WX[1], temp, temp, Filter2CoeffSqr[1], todo);
         allpass_process(&mFilter2_WX[2], temp, temp, Filter2CoeffSqr[2], todo);
@@ -101,8 +102,9 @@ void Uhj2Encoder::encode(FloatBufferLine &LeftOut, FloatBufferLine &RightOut, Fl
             D[i] += temp[i];
 
         /* S = 0.9396926*W + 0.1855740*X */
-        for(ALsizei i{0};i < todo;i++)
-            temp[i] = 0.9396926f*input0[i] + 0.1855740f*input1[i];
+        std::transform(winput, winput+todo, xinput, std::begin(temp),
+            [](const float w, const float x) noexcept -> float
+            { return 0.9396926f*w + 0.1855740f*x; });
         allpass_process(&mFilter1_WX[0], temp, temp, Filter1CoeffSqr[0], todo);
         allpass_process(&mFilter1_WX[1], temp, temp, Filter1CoeffSqr[1], todo);
         allpass_process(&mFilter1_WX[2], temp, temp, Filter1CoeffSqr[2], todo);
@@ -121,6 +123,9 @@ void Uhj2Encoder::encode(FloatBufferLine &LeftOut, FloatBufferLine &RightOut, Fl
         for(ALsizei i{0};i < todo;i++)
             right[i] += (S[i] - D[i]) * 0.5f;
 
+        winput += todo;
+        xinput += todo;
+        yinput += todo;
         base += todo;
     }
 }