there are issues with SIMD, lets fix them!

.github/workflows/ci.yml

@@ -31,6 +31,26 @@ jobs:
       - name: Run tox
         run: tox
 
+  test-arm:
+    runs-on: ubuntu-22.04-arm
+    steps:
+      - uses: actions/checkout@v3
+        with:
+          lfs: true
+          submodules: true
+      - name: Initialize lfs
+        run: git lfs pull
+      - name: Set up Python 3.11
+        uses: actions/setup-python@v5
+        with:
+          python-version: '3.11'
+      - name: Install dependencies
+        run: |
+          python -m pip install --upgrade pip setuptools wheel
+          pip install -U tox tox-gh-actions poetry
+      - name: Run tox
+        run: tox
+
   build:
     runs-on: ubuntu-20.04
     steps:

hvcc/generators/ir2c/static/HvSignalPhasor.c

@@ -60,8 +60,8 @@ static void sPhasor_k_updatePhase(SignalPhasor *o, hv_uint32_t p) {
 static void sPhasor_k_updateFrequency(SignalPhasor *o, float f, double r) {
 #if HV_SIMD_AVX
   o->step.f2sc = (float) (f/r);
-  o->inc = _mm256_set1_ps((float) (8.0f*f/r));
-  sPhasor_k_updatePhase(o, o->phase[0]);
+  o->inc = _mm256_set1_epi32((int)(f/r * 8388608.0f));  // 2^23 scaling
+  sPhasor_k_updatePhase(o, _mm256_castsi256_ps(o->phase));  // Adjust if needed
 #elif HV_SIMD_SSE
   o->step.s = (hv_int32_t) (f*(HV_PHASOR_2_32/r));
   o->inc = _mm_set1_epi32(4*o->step.s);

hvcc/generators/ir2c/static/HvSignalPhasor.h

@@ -25,8 +25,8 @@ extern "C" {
 
 typedef struct SignalPhasor {
 #if HV_SIMD_AVX
-  __m256 phase; // current phase
-  __m256 inc;   // phase increment
+  __m256i phase; // current phase
+  __m256i inc;   // phase increment
 #elif HV_SIMD_SSE
   __m128i phase;
   __m128i inc;
@@ -108,11 +108,14 @@ static inline void __hv_phasor_f(SignalPhasor *o, hv_bInf_t bIn, hv_bOutf_t bOut
 
 static inline void __hv_phasor_k_f(SignalPhasor *o, hv_bOutf_t bOut) {
 #if HV_SIMD_AVX
-  *bOut = _mm256_sub_ps(o->phase, _mm256_set1_ps(1.0f));
-  o->phase = _mm256_or_ps(_mm256_andnot_ps(
-      _mm256_set1_ps(-INFINITY),
-      _mm256_add_ps(o->phase, o->inc)),
-      _mm256_set1_ps(1.0f));
+  // Output current phase as float in [0,1)
+  *bOut = _mm256_sub_ps(_mm256_castsi256_ps(
+    _mm256_or_si256(_mm256_srli_epi32(o->phase, 9),
+    _mm256_set1_epi32(0x3F800000))),
+    _mm256_set1_ps(1.0f));
+
+  // Update phase as integer
+  o->phase = _mm256_add_epi32(o->phase, o->inc);
 #elif HV_SIMD_SSE
   *bOut = _mm_sub_ps(_mm_castsi128_ps(
       _mm_or_si128(_mm_srli_epi32(o->phase, 9),

hvcc/generators/ir2c/static/HvUtils.h

@@ -55,7 +55,7 @@
 #if !(HV_SIMD_NONE || HV_SIMD_NEON || HV_SIMD_SSE || HV_SIMD_AVX)
   #define HV_SIMD_NEON __ARM_NEON__
   #define HV_SIMD_SSE (__SSE__ && __SSE2__ && __SSE3__ && __SSSE3__ && __SSE4_1__)
-  #define HV_SIMD_AVX (__AVX__ && HV_SIMD_SSE)
+  #define HV_SIMD_AVX (__AVX2__ && HV_SIMD_SSE)
 #endif
 #ifndef HV_SIMD_FMA
   #define HV_SIMD_FMA __FMA__

tests/framework/base_signal.py

@@ -48,7 +48,7 @@ def compile_and_run(
             str(block_size or 480),
             str(num_iterations or 100)])
 
-        return wav_path
+        return exe_path, wav_path
 
     def _compare_wave_output(self, out_dir, c_sources, golden_path, flag=None):
         # http://stackoverflow.com/questions/10580676/comparing-two-numpy-arrays-for-equality-element-wise

tests/framework/base_test.py

@@ -112,6 +112,9 @@ def _compile_and_run(
                 source_files=source_files,
                 out_path=exe_path))
 
+        # run the clean command
+        subprocess.check_output(["make", "-C", os.path.dirname(makefile_path), "clean"])
+
         # run the compile command
         subprocess.check_output(["make", "-C", os.path.dirname(makefile_path), "-j"])
 

tests/framework/template/Makefile

@@ -1,6 +1,6 @@
 CC=clang
 CXX=clang++
-COMMONFLAGS=-Werror -Wno-unused-function -Wno-\#warnings {{" ".join(simd_flags)}}
+COMMONFLAGS=-Werror -Wno-unused-function -gdwarf-4 -Wno-\#warnings {{" ".join(simd_flags)}}
 CFLAGS=-std=c11 $(COMMONFLAGS)
 CXXFLAGS=-std=c++11 -fno-exceptions -fno-rtti $(COMMONFLAGS)
 

tests/src/test_control.c

@@ -33,7 +33,7 @@ int main(int argc, const char *argv[]) {
   HeavyContextInterface *context = hv_heavy_new(48000.0);
   hv_setPrintHook(context, &printHook);
 
-  float *outBuffers = (float *) malloc(numOutputChannels * BLOCK_SIZE * sizeof(float));
+  float *outBuffers = (float *) hv_malloc(numOutputChannels * BLOCK_SIZE * sizeof(float));
 
   for (int i = 0; i < numIterations; ++i) {
     hv_processInline(context, NULL, outBuffers, BLOCK_SIZE);

tests/src/test_midi.cpp

@@ -79,7 +79,7 @@ int main(int argc, const char *argv[]) {
 
   MidiEvent* mev;
 
-  float *outBuffers = (float *) malloc(numOutputChannels * BLOCK_SIZE * sizeof(float));
+  float *outBuffers = (float *) hv_malloc(numOutputChannels * BLOCK_SIZE * sizeof(float));
 
   for (int i = 0; i < numIterations; ++i) {
     for (int event=0; event < midifile[0].size(); event++) {

tests/src/test_signal.c

@@ -20,6 +20,15 @@
 #include "tinywav.h"
 
 int main(int argc, const char *argv[]) {
+  #if HV_SIMD_AVX
+    printf("AVX!\n");
+  #elif HV_SIMD_SSE
+    printf("SSE!\n");
+  #elif HV_SIMD_NEON
+    printf("NEON!\n");
+  #else // HV_SIMD_NONE
+    printf("NONE!\n");
+  #endif
   if (argc < 5) return -1;
   const char *outputPath = argv[1];
   const double sampleRate = atof(argv[2]);
@@ -40,7 +49,7 @@ int main(int argc, const char *argv[]) {
       (int32_t) hv_getSampleRate(context),
       TW_FLOAT32, TW_INLINE, outputPath);
 
-  float *outBuffers = (float *) malloc(
+  float *outBuffers = (float *) hv_malloc(
       hv_getNumOutputChannels(context) * blockSize * sizeof(float));
 
   for (int i = 0; i < numIterations; ++i) {

tests/test_signal.py

@@ -30,6 +30,9 @@ def test_line(self):
     def test_phasor_control(self):
         self._test_signal_patch("test-phasor-control.pd")
 
+    def test_phasor_signal(self):
+        self._test_signal_patch("test-phasor-signal.pd")
+
 
 def main():
     parser = argparse.ArgumentParser(