Merge branch 'develop' into simpler-file-loading

.travis.yml

@@ -5,6 +5,9 @@ dist: bionic
 cache:
   directories:
   - vst/download/
+  # macOS Homebrew
+  - $HOME/Library/Caches/Homebrew
+  - /usr/local/Homebrew
 
 jobs:
   include:
@@ -48,6 +51,11 @@ jobs:
     install: .travis/install_osx.sh
     script: .travis/script_osx.sh
     after_success: .travis/prepare_osx.sh
+    before_cache:
+      - brew cleanup
+      # Credit https://discourse.brew.sh/t/best-practice-for-homebrew-on-travis-brew-update-is-5min-to-build-time/5215/9
+      # Cache only .git files under "/usr/local/Homebrew" so "brew update" does not take 5min every build
+      - find /usr/local/Homebrew \! -regex ".+\.git.+" -delete
 
   - name: "MOD devices arm"
     env:

.travis/install_osx.sh

@@ -3,6 +3,7 @@
 set -ex
 
 sudo ln -s /usr/local /opt/local
+brew cask uninstall --force java
 brew update
 brew upgrade cmake
 brew install python ||  brew link --overwrite python

benchmarks/BM_pan_arm.cpp

@@ -0,0 +1,134 @@
+// SPDX-License-Identifier: BSD-2-Clause
+
+// This code is part of the sfizz library and is licensed under a BSD 2-clause
+// license. You should have receive a LICENSE.md file along with the code.
+// If not, contact the sfizz maintainers at https://github.com/sfztools/sfizz
+
+#include "Panning.h"
+#include "simd/Common.h"
+#include <benchmark/benchmark.h>
+#include <random>
+#include <absl/algorithm/container.h>
+#include "absl/types/span.h"
+#include <arm_neon.h>
+
+#include <jsl/allocator>
+template <class T, std::size_t A = 16>
+using aligned_vector = std::vector<T, jsl::aligned_allocator<T, A>>;
+
+// Number of elements in the table, odd for equal volume at center
+constexpr int panSize = 4095;
+
+// Table of pan values for the left channel, extra element for safety
+static const auto panData = []()
+{
+    std::array<float, panSize + 1> pan;
+    int i = 0;
+
+    for (; i < panSize; ++i)
+        pan[i] = std::cos(i * (piTwo<double>() / (panSize - 1)));
+
+    for (; i < static_cast<int>(pan.size()); ++i)
+        pan[i] = pan[panSize - 1];
+
+    return pan;
+}();
+
+float _panLookup(float pan)
+{
+    // reduce range, round to nearest
+    int index = lroundPositive(pan * (panSize - 1));
+    return panData[index];
+}
+
+void panScalar(const float* panEnvelope, float* leftBuffer, float* rightBuffer, unsigned size) noexcept
+{
+    const auto sentinel = panEnvelope + size;
+    while (panEnvelope < sentinel) {
+        auto p =(*panEnvelope + 1.0f) * 0.5f;
+        p = clamp(p, 0.0f, 1.0f);
+        *leftBuffer *= _panLookup(p);
+        *rightBuffer *= _panLookup(1 - p);
+        incrementAll(panEnvelope, leftBuffer, rightBuffer);
+    }
+}
+
+void panSIMD(const float* panEnvelope, float* leftBuffer, float* rightBuffer, unsigned size) noexcept
+{
+    const auto sentinel = panEnvelope + size;
+    int32_t indices[4];
+    while (panEnvelope < sentinel) {
+        float32x4_t mmPan = vld1q_f32(panEnvelope);
+        mmPan = vaddq_f32(mmPan, vdupq_n_f32(1.0f));
+        mmPan = vmulq_n_f32(mmPan, 0.5f * panSize);
+        mmPan = vaddq_f32(mmPan, vdupq_n_f32(0.5f));
+        mmPan = vminq_f32(mmPan, vdupq_n_f32(panSize));
+        mmPan = vmaxq_f32(mmPan, vdupq_n_f32(0.0f));
+        int32x4_t mmIdx = vcvtq_s32_f32(mmPan);
+        vst1q_s32(indices, mmIdx);
+
+        leftBuffer[0] *= panData[indices[0]];
+        rightBuffer[0] *= panData[panSize - indices[0] - 1];
+        leftBuffer[1] *= panData[indices[1]];
+        rightBuffer[1] *= panData[panSize - indices[1]- 1];
+        leftBuffer[2] *= panData[indices[2]];
+        rightBuffer[2] *= panData[panSize - indices[2]- 1];
+        leftBuffer[3] *= panData[indices[3]];
+        rightBuffer[3] *= panData[panSize - indices[3]- 1];
+
+        incrementAll<4>(panEnvelope, leftBuffer, rightBuffer);
+    }
+}
+
+class PanFixture : public benchmark::Fixture {
+public:
+  void SetUp(const ::benchmark::State& state) {
+    std::random_device rd { };
+    std::mt19937 gen { rd() };
+    std::uniform_real_distribution<float> dist { -1.0f, 1.0f };
+    pan.resize(state.range(0));
+    right.resize(state.range(0));
+    left.resize(state.range(0));
+
+    if (!willAlign<16>(pan.data(), left.data(), right.data()))
+        std::cout << "Will not align!" << '\n';
+    absl::c_generate(pan, [&]() { return dist(gen); });
+    absl::c_generate(left, [&]() { return dist(gen); });
+    absl::c_generate(right, [&]() { return dist(gen); });
+  }
+
+  void TearDown(const ::benchmark::State& /* state */) {
+
+  }
+
+    aligned_vector<float> pan;
+    aligned_vector<float> right;
+    aligned_vector<float> left;
+};
+
+BENCHMARK_DEFINE_F(PanFixture, PanScalar)(benchmark::State& state) {
+    for (auto _ : state)
+    {
+        panScalar(pan.data(), left.data(), right.data(), state.range(0));
+    }
+}
+
+BENCHMARK_DEFINE_F(PanFixture, PanSIMD)(benchmark::State& state) {
+    for (auto _ : state)
+    {
+        panSIMD(pan.data(), left.data(), right.data(), state.range(0));
+    }
+}
+
+BENCHMARK_DEFINE_F(PanFixture, PanSfizz)(benchmark::State& state) {
+    for (auto _ : state)
+    {
+        sfz::pan(pan.data(), left.data(), right.data(), state.range(0));
+    }
+}
+
+// Register the function as a benchmark
+BENCHMARK_REGISTER_F(PanFixture, PanScalar)->RangeMultiplier(4)->Range((1 << 4), (1 << 12));
+BENCHMARK_REGISTER_F(PanFixture, PanSIMD)->RangeMultiplier(4)->Range((1 << 4), (1 << 12));
+BENCHMARK_REGISTER_F(PanFixture, PanSfizz)->RangeMultiplier(4)->Range((1 << 4), (1 << 12));
+BENCHMARK_MAIN();

benchmarks/CMakeLists.txt

@@ -147,6 +147,12 @@ if (TARGET bm_resample)
 	add_dependencies(sfizz_benchmarks bm_resample)
 endif()
 
+if (SFIZZ_SYSTEM_PROCESSOR MATCHES "armv7l")
+    sfizz_add_benchmark(bm_pan_arm BM_pan_arm.cpp ../src/sfizz/Panning.cpp)
+    target_link_libraries(bm_pan_arm PRIVATE sfizz-jsl)
+    add_dependencies(sfizz_benchmarks bm_pan_arm)
+endif()
+
 configure_file("sample.wav" "${CMAKE_BINARY_DIR}/benchmarks/sample1.wav" COPYONLY)
 configure_file("sample.wav" "${CMAKE_BINARY_DIR}/benchmarks/sample2.wav" COPYONLY)
 configure_file("sample.wav" "${CMAKE_BINARY_DIR}/benchmarks/sample3.wav" COPYONLY)

cmake/SfizzConfig.cmake

@@ -60,6 +60,9 @@ if (CMAKE_CXX_COMPILER_ID MATCHES "GNU|Clang")
     add_compile_options(-Werror=return-type)
     if (SFIZZ_SYSTEM_PROCESSOR MATCHES "^(i.86|x86_64)$")
         add_compile_options(-msse2)
+    elseif(SFIZZ_SYSTEM_PROCESSOR MATCHES "^(arm.*)$")
+        add_compile_options(-mfpu=neon)
+        add_compile_options(-mfloat-abi=hard)
     endif()
 elseif (CMAKE_CXX_COMPILER_ID MATCHES "MSVC")
     set(CMAKE_CXX_STANDARD 17)
@@ -76,6 +79,10 @@ endfunction()
 # The sndfile library
 add_library(sfizz-sndfile INTERFACE)
 
+# The jsl utility library for C++
+add_library(sfizz-jsl INTERFACE)
+target_include_directories(sfizz-jsl INTERFACE "external/jsl/include")
+
 if (SFIZZ_USE_VCPKG OR CMAKE_CXX_COMPILER_ID MATCHES "MSVC")
     find_package(SndFile CONFIG REQUIRED)
     find_path(SNDFILE_INCLUDE_DIR sndfile.hh)

cmake/SfizzSIMDSourceFiles.cmake

@@ -3,6 +3,7 @@ macro(sfizz_add_simd_sources SOURCES_VAR PREFIX)
 
     list (APPEND ${SOURCES_VAR}
         ${PREFIX}/sfizz/SIMDHelpers.cpp
+        ${PREFIX}/sfizz/simd/HelpersNEON.cpp
         ${PREFIX}/sfizz/simd/HelpersSSE.cpp
         ${PREFIX}/sfizz/simd/HelpersAVX.cpp)
 

cmake/VSTConfig.cmake

@@ -30,10 +30,10 @@ if(NOT VST3_PACKAGE_ARCHITECTURE)
             else()
                 set(VST3_PACKAGE_ARCHITECTURE "i386")
             endif()
-        elseif(VST3_SYSTEM_PROCESSOR MATCHES "^(armv7l)$")
-                set(VST3_PACKAGE_ARCHITECTURE "armv7l")
+        elseif(VST3_SYSTEM_PROCESSOR MATCHES "^(armv[0-9]+)")
+            string(REGEX REPLACE "^(armv[0-9]+).*$" "\\1hl" VST3_PACKAGE_ARCHITECTURE "${VST3_SYSTEM_PROCESSOR}")
         elseif(VST3_SYSTEM_PROCESSOR MATCHES "^(aarch64)$")
-                set(VST3_PACKAGE_ARCHITECTURE "aarch64")
+            set(VST3_PACKAGE_ARCHITECTURE "aarch64")
         else()
             message(FATAL_ERROR "We don't know this architecture for VST3: ${VST3_SYSTEM_PROCESSOR}.")
         endif()