Added a panning SIMD helper and a benchmark against block operations and a scalar computation

benchmarks/BM_pan.cpp

@@ -0,0 +1,96 @@
+// Copyright (c) 2019, Paul Ferrand
+// All rights reserved.
+
+// Redistribution and use in source and binary forms, with or without
+// modification, are permitted provided that the following conditions are met:
+
+// 1. Redistributions of source code must retain the above copyright notice, this
+//    list of conditions and the following disclaimer.
+// 2. Redistributions in binary form must reproduce the above copyright notice,
+//    this list of conditions and the following disclaimer in the documentation
+//    and/or other materials provided with the distribution.
+
+// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+// ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+// WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+// DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR
+// ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+// (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+// LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+// ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+// SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+
+#include <benchmark/benchmark.h>
+#include <random>
+#include <numeric>
+#include <vector>
+#include <cmath>
+#include <iostream>
+#include "../sfizz/SIMDHelpers.h"
+#include "../sfizz/Config.h"
+#include "absl/types/span.h"
+
+class PanArray : public benchmark::Fixture {
+public:
+  void SetUp(const ::benchmark::State& state) {
+    std::random_device rd { };
+    std::mt19937 gen { rd() };
+    std::uniform_real_distribution<float> dist { 0.001, 1 };
+    pan = std::vector<float>(state.range(0));
+    left = std::vector<float>(state.range(0));
+    right = std::vector<float>(state.range(0));
+    std::generate(pan.begin(), pan.end(), [&]() { return dist(gen); });
+    std::generate(right.begin(), right.end(), [&]() { return dist(gen); });
+    std::generate(left.begin(), left.end(), [&]() { return dist(gen); });
+    temp1 = std::vector<float>(state.range(0));
+    temp2 = std::vector<float>(state.range(0));
+    span1 = absl::MakeSpan(temp1);
+    span2 = absl::MakeSpan(temp2);
+  }
+
+  void TearDown(const ::benchmark::State& state [[maybe_unused]]) {
+
+  }
+
+  std::vector<float> pan;
+  std::vector<float> left;
+  std::vector<float> right;
+  std::vector<float> temp1;
+  std::vector<float> temp2;
+  absl::Span<float> span1;
+  absl::Span<float> span2;
+};
+
+
+BENCHMARK_DEFINE_F(PanArray, Scalar)(benchmark::State& state) {
+    for (auto _ : state)
+    {
+        ::pan<float, false>(pan, absl::MakeSpan(left), absl::MakeSpan(right));
+    }
+}
+
+BENCHMARK_DEFINE_F(PanArray, SIMD)(benchmark::State& state) {
+    for (auto _ : state)
+    {
+        ::pan<float, true>(pan, absl::MakeSpan(left), absl::MakeSpan(right));
+    }
+}
+
+BENCHMARK_DEFINE_F(PanArray, BlockOps)(benchmark::State& state) {
+    for (auto _ : state)
+    {
+        ::fill<float>(span2, 1.0f);
+        ::add<float>(span1, span2);
+        ::applyGain<float>(piFour<float>, span2);
+        ::cos<float>(span2, span1);
+        ::sin<float>(span2, span2);
+        ::applyGain<float>(span1, absl::MakeSpan(left));
+        ::applyGain<float>(span2, absl::MakeSpan(right));
+    }
+}
+
+BENCHMARK_REGISTER_F(PanArray, Scalar)->RangeMultiplier(4)->Range(1 << 2, 1 << 12);
+BENCHMARK_REGISTER_F(PanArray, SIMD)->RangeMultiplier(4)->Range(1 << 2, 1 << 12);
+BENCHMARK_REGISTER_F(PanArray, BlockOps)->RangeMultiplier(4)->Range(1 << 2, 1 << 12);
+BENCHMARK_MAIN();

benchmarks/CMakeLists.txt

@@ -66,6 +66,9 @@ target_link_libraries(bm_subtract benchmark absl::span absl::algorithm)
 add_executable(bm_copy BM_copy.cpp ${SFIZZ_SIMD_SOURCES})
 target_link_libraries(bm_copy benchmark absl::span absl::algorithm)
 
+add_executable(bm_pan BM_pan.cpp ${SFIZZ_SIMD_SOURCES})
+target_link_libraries(bm_pan benchmark absl::span absl::algorithm)
+
 add_custom_target(sfizz_benchmarks)
 add_dependencies(sfizz_benchmarks 
 	bm_opf_high_vs_low 
@@ -79,6 +82,7 @@ add_dependencies(sfizz_benchmarks
 	bm_ramp
 	bm_ADSR
 	bm_add
+	bm_pan
 	bm_subtract
 	bm_multiplyAdd
 )

sfizz/Config.h

@@ -57,4 +57,5 @@ namespace SIMDConfig {
     constexpr bool subtract { false };
     constexpr bool multiplyAdd { false };
     constexpr bool copy { false };
+    constexpr bool pan { true };
 }

sfizz/SIMDDummy.cpp

@@ -124,4 +124,10 @@ template <>
 void copy<float, true>(absl::Span<const float> input, absl::Span<float> output) noexcept
 {
     copy<float, false>(input, output);
+}
+
+template <>
+void pan<float, true>(absl::Span<const float> panEnvelope, absl::Span<float> leftBuffer, absl::Span<float> rightBuffer) noexcept
+{
+    pan<float, false>(panEnvelope, leftBuffer, rightBuffer);
 }

sfizz/SIMDHelpers.h

@@ -388,4 +388,28 @@ void copy(absl::Span<const T> input, absl::Span<T> output) noexcept
 }
 
 template <>
-void copy<float, true>(absl::Span<const float> input, absl::Span<float> output) noexcept;
+void copy<float, true>(absl::Span<const float> input, absl::Span<float> output) noexcept;
+
+template <class T>
+inline void snippetPan(const T*& pan, T*& left, T*& right)
+{
+    const auto circlePan = piFour<float> * (static_cast<T>(1.0) + *pan++);
+    *left++ *= std::cos(circlePan);
+    *right++ *= std::sin(circlePan);
+}
+
+template <class T, bool SIMD = SIMDConfig::pan>
+void pan(absl::Span<const T> panEnvelope, absl::Span<T> leftBuffer, absl::Span<T> rightBuffer) noexcept
+{
+    ASSERT(leftBuffer.size() >= panEnvelope.size());
+    ASSERT(rightBuffer.size() >= panEnvelope.size());
+    auto* pan = panEnvelope.begin();
+    auto* left = leftBuffer.begin();
+    auto* right = rightBuffer.begin();
+    auto* sentinel = pan + min(panEnvelope.size(), leftBuffer.size(), rightBuffer.size());
+    while (pan < sentinel)
+        snippetPan(pan, left, right);
+}
+
+template <>
+void pan<float, true>(absl::Span<const float> panEnvelope, absl::Span<float> leftBuffer, absl::Span<float> rightBuffer) noexcept;

sfizz/SIMDSSE.cpp

@@ -564,4 +564,40 @@ void copy<float, true>(absl::Span<const float> input, absl::Span<float> output)
 
     while (out < sentinel)
         snippetCopy<float>(in, out);
+}
+
+template <>
+void pan<float, true>(absl::Span<const float> panEnvelope, absl::Span<float> leftBuffer, absl::Span<float> rightBuffer) noexcept
+{
+    ASSERT(leftBuffer.size() >= panEnvelope.size());
+    ASSERT(rightBuffer.size() >= panEnvelope.size());
+    auto* pan = panEnvelope.begin();
+    auto* left = leftBuffer.begin();
+    auto* right = rightBuffer.begin();
+    auto* sentinel = pan + min(panEnvelope.size(), leftBuffer.size(), rightBuffer.size());
+    const auto* lastAligned = prevAligned(sentinel);
+
+    while (unaligned(pan, left, right) && pan < lastAligned)
+        snippetPan(pan, left, right);
+
+    const auto mmOne = _mm_set_ps1(1.0f);
+    const auto mmPiFour = _mm_set_ps1(piFour<float>);
+    __m128 mmCos; 
+    __m128 mmSin; 
+    while (pan < lastAligned) {
+        auto mmPan = _mm_load_ps(pan);
+        mmPan = _mm_add_ps(mmOne, mmPan);
+        mmPan = _mm_mul_ps(mmOne, mmPiFour);
+        sincos_ps(mmPan, &mmSin, &mmCos);
+        auto mmLeft = _mm_mul_ps(mmCos, _mm_load_ps(left));
+        auto mmRight = _mm_mul_ps(mmCos, _mm_load_ps(left));
+        _mm_store_ps(left, mmLeft);
+        _mm_store_ps(right, mmRight);
+        left += TypeAlignment;
+        right += TypeAlignment;
+        pan += TypeAlignment;
+    }
+
+    while (pan < sentinel)
+        snippetPan(pan, left, right);
 }

sfizz/Voice.cpp

@@ -234,6 +234,7 @@ void sfz::Voice::processMono(AudioSpan<float> buffer) noexcept
 
     panEnvelope.getBlock(span1);
     // We assume that the pan envelope is already normalized between -1 and 1
+    // Check bm_pan for your architecture to check if it's interesting to use the pan helper instead
     ::fill<float>(span2, 1.0f);
     ::add<float>(span1, span2);
     ::applyGain<float>(piFour<float>, span2);