Use _mm_cvtepi32_ps in place of rint on SSE2

crsib · Oct 2, 2023 · 60c3f81 · 60c3f81
1 parent cdb21e3
commit 60c3f81
Show file tree

Hide file tree

Showing 2 changed files with 8 additions and 1 deletion.
diff --git a/libraries/lib-time-and-pitch/StaffPad/SimdTypes_sse2.h b/libraries/lib-time-and-pitch/StaffPad/SimdTypes_sse2.h
@@ -133,6 +133,12 @@ __finl float_x4 __vecc sqrt(const float_x4 &a)
   return _mm_sqrt_ps(a.s);
 }
 
+__finl float __vecc rint(float x)
+{
+  __m128i A = _mm_cvtps_epi32(_mm_set_ss(x));
+  return _mm_cvtss_f32(_mm_cvtepi32_ps(A));
+}
+
 __finl float_x4 __vecc rint(const float_x4 &a)
 {
   __m128i A = _mm_cvtps_epi32(a.s);

diff --git a/libraries/lib-time-and-pitch/StaffPad/TimeAndPitch.cpp b/libraries/lib-time-and-pitch/StaffPad/TimeAndPitch.cpp
@@ -187,7 +187,8 @@ namespace {
 // wrap a phase value into -PI..PI
 inline float _unwrapPhase(float arg)
 {
-  return arg - int(arg * 0.15915494309f + 0.5f) * 6.283185307f;
+  using namespace audio::simd;
+  return arg - rint(arg * 0.15915494309f) * 6.283185307f;
 }
 
 void _unwrapPhaseVec(float* v, int n)