there are issues with SIMD, lets fix them!

hvcc/generators/ir2c/static/HvSignalPhasor.h

@@ -108,11 +108,30 @@ 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));
+  // *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));
+
+  __m128 phase_lo = _mm256_extractf128_ps(o->phase, 0);
+  __m128 phase_hi = _mm256_extractf128_ps(o->phase, 1);
+  __m128 inc_lo = _mm256_extractf128_ps(o->inc, 0);
+  __m128 inc_hi = _mm256_extractf128_ps(o->inc, 1);
+
+  for (int i = 0; i < 4; i++) {
+    phase_lo = _mm_add_ps(phase_lo, inc_lo);
+    phase_lo = _mm_sub_ps(phase_lo, _mm_and_ps(_mm_cmpge_ps(phase_lo, _mm_set1_ps(2.0f)), _mm_set1_ps(2.0f)));
+    bOut[i] = _mm256_castps128_ps256(phase_lo);
+
+    phase_hi = _mm_add_ps(phase_hi, inc_hi);
+    phase_hi = _mm_sub_ps(phase_hi, _mm_and_ps(_mm_cmpge_ps(phase_hi, _mm_set1_ps(2.0f)), _mm_set1_ps(2.0f)));
+    bOut[i+4] = _mm256_insertf128_ps(_mm256_castps128_ps256(phase_hi), phase_hi, 1);
+  }
+
+  o->phase = _mm256_set_m128(phase_hi, phase_lo);
+
+
 #elif HV_SIMD_SSE
   *bOut = _mm_sub_ps(_mm_castsi128_ps(
       _mm_or_si128(_mm_srli_epi32(o->phase, 9),

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

@@ -108,6 +108,9 @@ 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(