staffpad::audio::simd Namespace Reference

Classes
struct	float_x4

Functions
void *	aligned_malloc (size_t required_bytes, size_t alignment)
	reserve aligned memory. Needs to be freed with aligned_free() More...
void	aligned_free (void *p)
	free memory allocated with aligned_malloc More...
template<typename cls >
cls *	aligned_new (int alignment)
	create a c++ class at an memory-aligned spot that needs to be deleted using aligned_delete More...
template<typename cls >
void	aligned_delete (cls *obj)
template<typename T >
bool	is_aligned (T *obj, int alignment)
template<typename fnc >
__finl void	perform_parallel_simd_aligned (float *a, float *b, int n, const fnc &f)
template<typename fnc >
__finl void	perform_parallel_simd_aligned (float *a, int n, const fnc &f)
	template for applying math to one data buffer More...
__finl float_x4 __vecc	float_x4_from_float (float x)
__finl float_x4 __vecc	float_x4_load_aligned (const float *x)
__finl void __vecc	store_aligned (const float_x4 &a, float *x)
__finl float_x4 __vecc	unzip1 (const float_x4 &a, const float_x4 &b)
__finl float_x4 __vecc	unzip2 (const float_x4 &a, const float_x4 &b)
__finl float_x4 __vecc	operator+ (float_x4 a, float_x4 b)
__finl float_x4 __vecc	operator- (float_x4 a, float_x4 b)
__finl float_x4 __vecc	operator* (float_x4 a, float_x4 b)
__finl float_x4 __vecc	sqrt (const float_x4 &a)
__finl float __vecc	rint (float a)
__finl float_x4 __vecc	rint (const float_x4 &a)

Function Documentation

aligned_delete()

template<typename cls >

void staffpad::audio::simd::aligned_delete ( cls *  obj )

inline

delete objects created using aligned_new

Definition at line 59 of file SimdTypes.h.

{
  if (obj != nullptr)
  {
    obj->~cls();
    aligned_free((void *)obj);
  }
}

References aligned_free().

Here is the call graph for this function:

aligned_free()

void staffpad::audio::simd::aligned_free ( void *  p )

inline

free memory allocated with aligned_malloc

Definition at line 43 of file SimdTypes.h.

{
  if (p)
    free(((void **)p)[-1]);
}

References staffpad::vo::free().

Referenced by aligned_delete(), and staffpad::SamplesFloat< T >::dealloc().

Here is the call graph for this function:

Here is the caller graph for this function:

aligned_malloc()

void * staffpad::audio::simd::aligned_malloc ( size_t  required_bytes, size_t  alignment  )

inline

reserve aligned memory. Needs to be freed with aligned_free()

Definition at line 29 of file SimdTypes.h.

{
  auto offset = alignment - 1 + sizeof(void *);
  auto p1 = std::malloc(required_bytes + offset);
  if (p1 == nullptr)
    return nullptr;
  // figure out aligned position
  void *p2 = (void *)(((size_t)(p1) + offset) & ~(alignment - 1));
  // write malloced pointer in front of aligned data
  ((void **)p2)[-1] = p1;
  return p2;
}

Referenced by aligned_new(), and staffpad::SamplesFloat< T >::alloc().

Here is the caller graph for this function:

aligned_new()

template<typename cls >

cls * staffpad::audio::simd::aligned_new ( int  alignment )

inline

create a c++ class at an memory-aligned spot that needs to be deleted using aligned_delete

Definition at line 51 of file SimdTypes.h.

{
  void *mem = aligned_malloc(sizeof(cls), alignment);
  return new (mem) cls();
}

References aligned_malloc().

Here is the call graph for this function:

float_x4_from_float()

__finl float_x4 __vecc staffpad::audio::simd::float_x4_from_float

(

float

x

)

Definition at line 104 of file SimdTypes_neon.h.

{
  return vdupq_n_f32(x);
}

float_x4_load_aligned()

__finl float_x4 __vecc staffpad::audio::simd::float_x4_load_aligned

(

const float *

x

)

Definition at line 109 of file SimdTypes_neon.h.

{
  return vld1q_f32(x);
}

Referenced by perform_parallel_simd_aligned().

Here is the caller graph for this function:

is_aligned()

template<typename T >

bool staffpad::audio::simd::is_aligned ( T *  obj, int  alignment  )

inline

Definition at line 69 of file SimdTypes.h.

{
    return (((size_t)obj) & (alignment - 1)) == 0;
}

Referenced by perform_parallel_simd_aligned().

Here is the caller graph for this function:

operator*()

__finl float_x4 __vecc staffpad::audio::simd::operator*	(	float_x4	a,
		float_x4	b
	)

Definition at line 139 of file SimdTypes_neon.h.

{
  return vmulq_f32(a.s, b.s);
}

References staffpad::audio::simd::float_x4::s.

operator+()

__finl float_x4 __vecc staffpad::audio::simd::operator+	(	float_x4	a,
		float_x4	b
	)

Definition at line 129 of file SimdTypes_neon.h.

{
  return vaddq_f32(a.s, b.s);
}

References staffpad::audio::simd::float_x4::s.

operator-()

__finl float_x4 __vecc staffpad::audio::simd::operator-	(	float_x4	a,
		float_x4	b
	)

Definition at line 134 of file SimdTypes_neon.h.

{
  return vsubq_f32(a.s, b.s);
}

References staffpad::audio::simd::float_x4::s.

perform_parallel_simd_aligned() [1/2]

template<typename fnc >

__finl void staffpad::audio::simd::perform_parallel_simd_aligned	(	float *	a,
		float *	b,
		int	n,
		const fnc &	f
	)

this template allows to write float SIMD code with the supported operators the following way:

float *a_vec; const float *b_vec; perform_parallel_simd_aligned(a_vec, b_vec, 512, [](auto &a, auto &b) { auto t = a; a = 3.f * a + 12.f * b; b = 0.25f * a + 3.f * b; });

Definition at line 87 of file SimdTypes.h.

{
  // fnc& f needs to be a lambda of type [](auto &a, auto &b){}.
  // the autos will be float_x4/float
  constexpr int N = 4;
  constexpr int byte_size = sizeof(float);
 
  assert(is_aligned(a, N * byte_size) && is_aligned(b, N * byte_size));
 
  for (int i = 0; i <= n - N; i += N)
  {
    auto x = float_x4_load_aligned(a + i);
    auto y = float_x4_load_aligned(b + i);
    f(x, y);
    store_aligned(x, a + i);
    store_aligned(y, b + i);
  }
  // deal with last partial packet
  for (int i = n & (~(N - 1)); i < n; ++i)
    f(a[i], b[i]);
}

References float_x4_load_aligned(), is_aligned(), and store_aligned().

Referenced by staffpad::anonymous_namespace{TimeAndPitch.cpp}::_fft_shift(), staffpad::anonymous_namespace{TimeAndPitch.cpp}::_lr_to_ms(), staffpad::anonymous_namespace{TimeAndPitch.cpp}::_ms_to_lr(), and staffpad::anonymous_namespace{TimeAndPitch.cpp}::_unwrapPhaseVec().

Here is the call graph for this function:

Here is the caller graph for this function:

perform_parallel_simd_aligned() [2/2]

template<typename fnc >

__finl void staffpad::audio::simd::perform_parallel_simd_aligned	(	float *	a,
		int	n,
		const fnc &	f
	)

template for applying math to one data buffer

Definition at line 111 of file SimdTypes.h.

{
  // fnc& f needs to be a lambda of type [](auto &a){}.
  constexpr int N = 4;
  constexpr int byte_size = sizeof(float);
  assert(is_aligned(a, N * byte_size));
 
  for (int i = 0; i <= n - N; i += N)
  {
    auto x = float_x4_load_aligned(a + i);
    f(x);
    store_aligned(x, a + i);
  }
  // deal with last partial packet
  for (int i = n & (~(N - 1)); i < n; ++i)
    f(a[i]);
}

References float_x4_load_aligned(), is_aligned(), and store_aligned().

Here is the call graph for this function:

rint() [1/2]

__finl float_x4 __vecc staffpad::audio::simd::rint

(

const float_x4 &

a

)

Definition at line 154 of file SimdTypes_neon.h.

{
  return vrndnq_f32(a.s);
}

References staffpad::audio::simd::float_x4::s.

Referenced by rint().

Here is the caller graph for this function:

rint() [2/2]

__finl float __vecc staffpad::audio::simd::rint

(

float

a

)

Definition at line 149 of file SimdTypes_neon.h.

{
  return std::rint(a);
}

References rint().

Referenced by staffpad::anonymous_namespace{TimeAndPitch.cpp}::_unwrapPhase(), staffpad::anonymous_namespace{TimeAndPitch.cpp}::_unwrapPhaseVec(), WaveTrackShifter::HintOffsetLarger(), WaveTrackShifter::QuantizeOffset(), and LWSlider::ValueToPosition().

Here is the call graph for this function:

Here is the caller graph for this function:

sqrt()

__finl float_x4 __vecc staffpad::audio::simd::sqrt

(

const float_x4 &

a

)

Definition at line 144 of file SimdTypes_neon.h.

{
  return vsqrtq_f32(a.s);
}

References staffpad::audio::simd::float_x4::s.

Referenced by AmpStat(), LegacyCompressorBase::AvgCircle(), Biquad::CalcChebyshevType1Filter(), Biquad::CalcChebyshevType2Filter(), SqliteSampleBlock::CalcSummary(), SpectrumAnalyst::Calculate(), VoiceKey::CalibrateNoise(), BassTrebleBase::Instance::Coefficients(), SpectrumAnalyst::CubicMaximize(), DistortionBase::Instance::CubicTable(), graphics::Distance(), ExportMixerPanel::Distance(), SqliteSampleBlock::DoGetMinMaxRMS(), anonymous_namespace{SpectrumView.cpp}::DrawClipSpectrum(), SelectedRegion::fc(), anonymous_namespace{WaveDataCache.cpp}::AppendBufferHelper::FillBlocksFromAppendBuffer(), ScienFilterBase::FilterMagnAtFreq(), ContrastBase::GetDB(), WaveChannelUtilities::GetRMS(), Sequence::GetRMS(), WaveDataCache::InitializeElement(), SpectralSelectionBar::ModifySpectralSelection(), SelectHandle::MoveSnappingFreqSelection(), graphics::Norm(), LoudnessBase::Process(), NyquistBase::Process(), PaulStretch::process(), NoiseBase::ProcessBlock(), SpectralSelectionBar::SetFrequencies(), SelectHandle::SnapCenterOnce(), MeterPanel::UpdateDisplay(), audacity::cloud::audiocom::sync::anonymous_namespace{WavPackCompressor.cpp}::UpdateRMS(), and LinearOutputRateTimeWarper::Warp().

Here is the caller graph for this function:

store_aligned()

__finl void __vecc staffpad::audio::simd::store_aligned	(	const float_x4 &	a,
		float *	x
	)

Definition at line 114 of file SimdTypes_neon.h.

{
  vst1q_f32(x, a.s);
}

References staffpad::audio::simd::float_x4::s.

Referenced by perform_parallel_simd_aligned().

Here is the caller graph for this function:

unzip1()

__finl float_x4 __vecc staffpad::audio::simd::unzip1	(	const float_x4 &	a,
		const float_x4 &	b
	)

Definition at line 119 of file SimdTypes_neon.h.

{
  return vuzp1q_f32(a.s, b.s);
}

References staffpad::audio::simd::float_x4::s.

unzip2()

__finl float_x4 __vecc staffpad::audio::simd::unzip2	(	const float_x4 &	a,
		const float_x4 &	b
	)

Definition at line 124 of file SimdTypes_neon.h.

{
  return vuzp2q_f32(a.s, b.s);
}

References staffpad::audio::simd::float_x4::s.