Implement sfizz oversampler using new code

src/sfizz/Oversampler.cpp

@@ -5,6 +5,7 @@
 // If not, contact the sfizz maintainers at https://github.com/sfztools/sfizz
 
 #include "Oversampler.h"
+#include "OversamplerHelpers.h"
 #include "Buffer.h"
 #include "AudioSpan.h"
 #include "AudioReader.h"
@@ -14,52 +15,6 @@
 template <class T, std::size_t A = sfz::config::defaultAlignment>
 using aligned_vector = std::vector<T, jsl::aligned_allocator<T, A>>;
 
-constexpr std::array<double, 12> coeffsStage2x {
-    0.036681502163648017,
-    0.13654762463195771,
-    0.27463175937945411,
-    0.42313861743656667,
-    0.56109869787919475,
-    0.67754004997416162,
-    0.76974183386322659,
-    0.83988962484963803,
-    0.89226081800387891,
-    0.9315419599631839,
-    0.96209454837808395,
-    0.98781637073289708
-};
-
-constexpr std::array<double, 4> coeffsStage4x {
-    0.042448989488488006,
-    0.17072114107630679,
-    0.39329183835224008,
-    0.74569514831986694
-};
-
-constexpr std::array<double, 3> coeffsStage8x {
-    0.055748680811302048,
-    0.24305119574153092,
-    0.6466991311926823
-};
-
-
-#if SFIZZ_HAVE_SSE
-#include "hiir/Upsampler2xSse.h"
-using Upsampler2x = hiir::Upsampler2xSse<coeffsStage2x.size()>;
-using Upsampler4x = hiir::Upsampler2xSse<coeffsStage4x.size()>;
-using Upsampler8x = hiir::Upsampler2xSse<coeffsStage8x.size()>;
-#elif SFIZZ_HAVE_NEON
-#include "hiir/Upsampler2xNeon.h"
-using Upsampler2x = hiir::Upsampler2xNeon<coeffsStage2x.size()>;
-using Upsampler4x = hiir::Upsampler2xNeon<coeffsStage4x.size()>;
-using Upsampler8x = hiir::Upsampler2xNeon<coeffsStage8x.size()>;
-#else
-#include "hiir/Upsampler2xFpu.h"
-using Upsampler2x = hiir::Upsampler2xFpu<coeffsStage2x.size()>;
-using Upsampler4x = hiir::Upsampler2xFpu<coeffsStage4x.size()>;
-using Upsampler8x = hiir::Upsampler2xFpu<coeffsStage8x.size()>;
-#endif
-
 sfz::Oversampler::Oversampler(sfz::Oversampling factor, size_t chunkSize)
 : factor(factor), chunkSize(chunkSize)
 {
@@ -74,36 +29,10 @@ void sfz::Oversampler::stream(AudioSpan<float> input, AudioSpan<float> output, s
     const auto numFrames = input.getNumFrames();
     const auto numChannels = input.getNumChannels();
 
-    aligned_vector<Upsampler2x> upsampler2x;
+    aligned_vector<Upsampler> upsampler(numChannels);
-    aligned_vector<Upsampler4x> upsampler4x;
-    aligned_vector<Upsampler8x> upsampler8x;
 
-    switch(factor)
+    // Intermediate buffer
-    {
+    sfz::Buffer<float> temp { std::max<size_t>(128, Upsampler::recommendedBuffer(16, chunkSize)) };
-    case Oversampling::x8:
-        upsampler8x.resize(numChannels);
-        for (auto& upsampler: upsampler8x)
-            upsampler.set_coefs(coeffsStage8x.data());
-        // fallthrough
-    case Oversampling::x4:
-        upsampler4x.resize(numChannels);
-        for (auto& upsampler: upsampler4x)
-            upsampler.set_coefs(coeffsStage4x.data());
-        // fallthrough
-    case Oversampling::x2:
-        upsampler2x.resize(numChannels);
-        for (auto& upsampler: upsampler2x)
-            upsampler.set_coefs(coeffsStage2x.data());
-        break;
-    case Oversampling::x1:
-        break;
-    }
-
-    // Intermediate buffers
-    sfz::Buffer<float> buffer1 { chunkSize * 2 };
-    sfz::Buffer<float> buffer2 { chunkSize * 4 };
-    auto span1 = absl::MakeSpan(buffer1);
-    auto span2 = absl::MakeSpan(buffer2);
 
     size_t inputFrameCounter { 0 };
     size_t outputFrameCounter { 0 };
@@ -115,23 +44,10 @@ void sfz::Oversampler::stream(AudioSpan<float> input, AudioSpan<float> output, s
         for (size_t chanIdx = 0; chanIdx < numChannels; chanIdx++) {
             const auto inputChunk = input.getSpan(chanIdx).subspan(inputFrameCounter, thisChunkSize);
             const auto outputChunk = output.getSpan(chanIdx).subspan(outputFrameCounter, outputChunkSize);
-            switch (factor) {
+            upsampler[chanIdx].process(
-            case Oversampling::x1:
+                static_cast<int>(factor),
-                copy<float>(inputChunk, outputChunk);
+                inputChunk.data(), outputChunk.data(), static_cast<int>(inputChunk.size()),
-                break;
+                temp.data(), static_cast<int>(temp.size()));
-            case Oversampling::x2:
-                upsampler2x[chanIdx].process_block(outputChunk.data(), inputChunk.data(), static_cast<long>(thisChunkSize));
-                break;
-            case Oversampling::x4:
-                upsampler2x[chanIdx].process_block(span1.data(), inputChunk.data(), static_cast<long>(thisChunkSize));
-                upsampler4x[chanIdx].process_block(outputChunk.data(), span1.data(), static_cast<long>(thisChunkSize * 2));
-                break;
-            case Oversampling::x8:
-                upsampler2x[chanIdx].process_block(span1.data(), inputChunk.data(), static_cast<long>(thisChunkSize));
-                upsampler4x[chanIdx].process_block(span2.data(), span1.data(), static_cast<long>(thisChunkSize * 2));
-                upsampler8x[chanIdx].process_block(outputChunk.data(), span2.data(), static_cast<long>(thisChunkSize * 4));
-                break;
-            }
         }
         inputFrameCounter += thisChunkSize;
         outputFrameCounter += outputChunkSize;
@@ -149,47 +65,18 @@ void sfz::Oversampler::stream(AudioReader& input, AudioSpan<float> output, std::
     const auto numFrames = static_cast<size_t>(input.frames());
     const auto numChannels = input.channels();
 
-    aligned_vector<Upsampler2x> upsampler2x;
+    aligned_vector<Upsampler> upsampler(numChannels);
-    aligned_vector<Upsampler4x> upsampler4x;
-    aligned_vector<Upsampler8x> upsampler8x;
-
-    switch(factor)
-    {
-    case Oversampling::x8:
-        upsampler8x.resize(numChannels);
-        for (auto& upsampler: upsampler8x)
-            upsampler.set_coefs(coeffsStage8x.data());
-        // fallthrough
-    case Oversampling::x4:
-        upsampler4x.resize(numChannels);
-        for (auto& upsampler: upsampler4x)
-            upsampler.set_coefs(coeffsStage4x.data());
-        // fallthrough
-    case Oversampling::x2:
-        upsampler2x.resize(numChannels);
-        for (auto& upsampler: upsampler2x)
-            upsampler.set_coefs(coeffsStage2x.data());
-        break;
-    case Oversampling::x1:
-        break;
-    }
 
     // Intermediate buffers
     sfz::Buffer<float> fileBlock { chunkSize * numChannels };
-    sfz::Buffer<float> buffer1 { chunkSize * 2 };
+    sfz::Buffer<float> channelBlock { chunkSize };
-    sfz::Buffer<float> buffer2 { chunkSize * 4 };
+    sfz::Buffer<float> temp { std::max<size_t>(128, Upsampler::recommendedBuffer(16, chunkSize)) };
-    auto span1 = absl::MakeSpan(buffer1);
-    auto span2 = absl::MakeSpan(buffer2);
 
-    auto upsample2xFromInterleaved = [numChannels](
+    auto deinterleave = [numChannels](
-        Upsampler2x& upsampler, float* output, const float* input,
+        float* output, const float* input, size_t numFrames, unsigned chanIdx)
-        size_t numInputFrames, unsigned chanIdx)
     {
-        for (size_t i = 0; i < numInputFrames; ++i) {
+        for (size_t i = 0; i < numFrames; ++i)
-            float* outp = &output[2 * i];
+            output[i] = input[i * numChannels + chanIdx];
-            const float* inp = &input[i * numChannels + chanIdx];
-            upsampler.process_sample(outp[0], outp[1], inp[0]);
-        }
     };
 
     size_t inputFrameCounter { 0 };
@@ -211,23 +98,15 @@ void sfz::Oversampler::stream(AudioReader& input, AudioSpan<float> output, std::
 
         for (size_t chanIdx = 0; chanIdx < numChannels; chanIdx++) {
             const auto outputChunk = output.getSpan(chanIdx).subspan(outputFrameCounter, outputChunkSize);
-            switch (factor) {
+
-            case Oversampling::x1:
+            if (factor == Oversampling::x1)
-                for (size_t i = 0; i < thisChunkSize; ++i)
+                deinterleave(outputChunk.data(), fileBlock.data(), thisChunkSize, chanIdx);
-                    outputChunk[i] = fileBlock[i * numChannels + chanIdx];
+            else {
-                break;
+                deinterleave(channelBlock.data(), fileBlock.data(), thisChunkSize, chanIdx);
-            case Oversampling::x2:
+                upsampler[chanIdx].process(
-                upsample2xFromInterleaved(upsampler2x[chanIdx], outputChunk.data(), fileBlock.data(), thisChunkSize, chanIdx);
+                    static_cast<int>(factor),
-                break;
+                    channelBlock.data(), outputChunk.data(), static_cast<int>(thisChunkSize),
-            case Oversampling::x4:
+                    temp.data(), static_cast<int>(temp.size()));
-                upsample2xFromInterleaved(upsampler2x[chanIdx], span1.data(), fileBlock.data(), thisChunkSize, chanIdx);
-                upsampler4x[chanIdx].process_block(outputChunk.data(), span1.data(), static_cast<long>(thisChunkSize * 2));
-                break;
-            case Oversampling::x8:
-                upsample2xFromInterleaved(upsampler2x[chanIdx], span1.data(), fileBlock.data(), thisChunkSize, chanIdx);
-                upsampler4x[chanIdx].process_block(span2.data(), span1.data(), static_cast<long>(thisChunkSize * 2));
-                upsampler8x[chanIdx].process_block(outputChunk.data(), span2.data(), static_cast<long>(thisChunkSize * 4));
-                break;
             }
         }
         inputFrameCounter += thisChunkSize;