olb::cpu::simd Namespace Reference

Implementations of vector CPU specifics. More...

Namespaces
namespace	concepts

Classes
class	Cell
	Implementation of the Cell concept for vectorized collision operators. More...
class	Column
	Plain column for SIMD CPU targets. More...
class	ConcreteDynamics
	Implementation of cpu::Dynamics for concrete DYNAMICS on SIMD blocks. More...
class	CyclicColumn
	Virtual memory based cyclic column for usage in ColumnVector. More...
class	FieldPtr
	SIMD-specific pointer to a pack of rows of a D-dimensional field. More...
class	Mask
class	Mask< double >
class	Mask< float >
class	Pack
class	Pack< double >
class	Pack< float >

Functions
template<typename T >
Pack< T >	pow (Pack< T > base, Pack< T > exp)
template<typename T >
Pack< T >	min (Pack< T > rhs, Pack< T > lhs)
template<typename T >
Pack< T >	max (Pack< T > rhs, Pack< T > lhs)
template<typename T >
Pack< T >	fabs (Pack< T > x)
template<typename T >
void	maskstore (T *target, Mask< T > mask, Pack< T > value)
template<>
void	maskstore< double > (double *target, Mask< double > mask, Pack< double > value)
template<>
void	maskstore< float > (float *target, Mask< float > mask, Pack< float > value)
template<typename T >
void	store (T *target, Pack< T > value)
template<>
void	store< double > (double *target, Pack< double > value)
template<>
void	store< float > (float *target, Pack< float > value)
template<typename T >
void	store (T *target, Pack< T > value, const typename Pack< T >::index_t *indices)
template<>
void	store< double > (double *target, Pack< double > value, const Pack< double >::index_t *indices)
template<>
void	store< float > (float *target, Pack< float > value, const Pack< float >::index_t *indices)
template<>
void	maskstore< double > (double *target, Mask< double > mask, Pack< double > value)
template<>
void	maskstore< float > (float *target, Mask< float > mask, Pack< float > value)
template<>
void	store< double > (double *target, Pack< double > value)
template<>
void	store< float > (float *target, Pack< float > value)
template<>
void	store< double > (double *target, Pack< double > value, const Pack< double >::index_t *indices)
template<>
void	store< float > (float *target, Pack< float > value, const Pack< float >::index_t *indices)
template<typename T >
std::size_t	getPageAlignedCount (std::size_t count)
template<std::floating_point T, concepts::convertible_to_pack S>
Pack< T >	operator+ (S lhs, Pack< T > rhs)
template<std::floating_point T, concepts::convertible_to_pack S>
Pack< T >	operator+ (Pack< T > lhs, S rhs)
template<std::floating_point T, concepts::convertible_to_pack S>
Pack< T >	operator- (S lhs, Pack< T > rhs)
template<std::floating_point T, concepts::convertible_to_pack S>
Pack< T >	operator- (Pack< T > lhs, S rhs)
template<std::floating_point T, concepts::convertible_to_pack S>
Pack< T >	operator* (Pack< T > lhs, S rhs)
template<std::floating_point T, concepts::convertible_to_pack S>
Pack< T >	operator* (S lhs, Pack< T > rhs)
template<std::floating_point T, concepts::convertible_to_pack S>
Pack< T >	operator/ (Pack< T > lhs, S rhs)
template<std::floating_point T, concepts::convertible_to_pack S>
Pack< T >	operator/ (S lhs, Pack< T > rhs)
template<std::floating_point T>
Pack< T >	sqrt (Pack< T > x)

Variables
const int	PROT_RW = PROT_READ | PROT_WRITE

Detailed Description

Implementations of vector CPU specifics.

Function Documentation

fabs()

template<typename T >

Pack< T > olb::cpu::simd::fabs

(

Pack< T >

x

)

Definition at line 426 of file 256.h.

{
  return max(x, -x);
}

References max().

Here is the call graph for this function:

Here is the caller graph for this function:

getPageAlignedCount()

template<typename T >

std::size_t olb::cpu::simd::getPageAlignedCount

(

std::size_t

count

)

Definition at line 134 of file column.h.

{
  const std::size_t page_size = sysconf(_SC_PAGESIZE);
  const std::size_t size = ((count * sizeof(T) - 1) / page_size + 1) * page_size;
  const std::size_t volume = size / sizeof(T);
 
  if (size % page_size != 0) {
    throw std::invalid_argument("Buffer size must be multiple of PAGE_SIZE");
  }
 
  return volume;
};

maskstore()

template<typename T >

void olb::cpu::simd::maskstore	(	T *	target,
		Mask< T >	mask,
		Pack< T >	value )

Here is the caller graph for this function:

maskstore< double >() [1/2]

template<>

void olb::cpu::simd::maskstore< double >	(	double *	target,
		Mask< double >	mask,
		Pack< double >	value )

Definition at line 435 of file 256.h.

{
  _mm256_maskstore_pd(target, mask, value);
}

maskstore< double >() [2/2]

template<>

void olb::cpu::simd::maskstore< double >	(	double *	target,
		Mask< double >	mask,
		Pack< double >	value )

Definition at line 407 of file 512.h.

{
  _mm512_mask_storeu_pd(target, mask, value);
}

maskstore< float >() [1/2]

template<>

void olb::cpu::simd::maskstore< float >	(	float *	target,
		Mask< float >	mask,
		Pack< float >	value )

Definition at line 441 of file 256.h.

{
  _mm256_maskstore_ps(target, mask, value);
}

maskstore< float >() [2/2]

template<>

void olb::cpu::simd::maskstore< float >	(	float *	target,
		Mask< float >	mask,
		Pack< float >	value )

Definition at line 413 of file 512.h.

{
  _mm512_mask_storeu_ps(target, mask, value);
}

max()

template<typename T >

Pack< T > olb::cpu::simd::max	(	Pack< T >	rhs,
		Pack< T >	lhs )

Definition at line 416 of file 256.h.

{
  if constexpr (std::is_same_v<T,double>) {
    return _mm256_max_pd(rhs, lhs);
  } else {
    return _mm256_max_ps(rhs, lhs);
  }
}

Here is the caller graph for this function:

min()

template<typename T >

Pack< T > olb::cpu::simd::min	(	Pack< T >	rhs,
		Pack< T >	lhs )

Definition at line 406 of file 256.h.

{
  if constexpr (std::is_same_v<T,double>) {
    return _mm256_min_pd(rhs, lhs);
  } else {
    return _mm256_min_ps(rhs, lhs);
  }
}

Here is the caller graph for this function:

operator*() [1/2]

template<std::floating_point T, concepts::convertible_to_pack S>

Pack< T > olb::cpu::simd::operator*	(	Pack< T >	lhs,
		S	rhs )

Definition at line 74 of file pack.h.

{
  return lhs * Pack<T>(rhs);
}

operator*() [2/2]

template<std::floating_point T, concepts::convertible_to_pack S>

Pack< T > olb::cpu::simd::operator*	(	S	lhs,
		Pack< T >	rhs )

Definition at line 80 of file pack.h.

{
  return Pack<T>(lhs) * rhs;
}

operator+() [1/2]

template<std::floating_point T, concepts::convertible_to_pack S>

Pack< T > olb::cpu::simd::operator+	(	Pack< T >	lhs,
		S	rhs )

Definition at line 56 of file pack.h.

{
  return lhs + Pack<T>(rhs);
}

operator+() [2/2]

template<std::floating_point T, concepts::convertible_to_pack S>

Pack< T > olb::cpu::simd::operator+	(	S	lhs,
		Pack< T >	rhs )

Definition at line 50 of file pack.h.

{
  return Pack<T>(lhs) + rhs;
}

operator-() [1/2]

template<std::floating_point T, concepts::convertible_to_pack S>

Pack< T > olb::cpu::simd::operator-	(	Pack< T >	lhs,
		S	rhs )

Definition at line 68 of file pack.h.

{
  return lhs - Pack<T>(rhs);
}

operator-() [2/2]

template<std::floating_point T, concepts::convertible_to_pack S>

Pack< T > olb::cpu::simd::operator-	(	S	lhs,
		Pack< T >	rhs )

Definition at line 62 of file pack.h.

{
  return Pack<T>(lhs) - rhs;
}

operator/() [1/2]

template<std::floating_point T, concepts::convertible_to_pack S>

Pack< T > olb::cpu::simd::operator/	(	Pack< T >	lhs,
		S	rhs )

Definition at line 86 of file pack.h.

{
  return lhs / Pack<T>(rhs);
}

operator/() [2/2]

template<std::floating_point T, concepts::convertible_to_pack S>

Pack< T > olb::cpu::simd::operator/	(	S	lhs,
		Pack< T >	rhs )

Definition at line 92 of file pack.h.

{
  return Pack<T>(lhs) / rhs;
}

pow()

template<typename T >

Pack< T > olb::cpu::simd::pow	(	Pack< T >	base,
		Pack< T >	exp )

Definition at line 395 of file 256.h.

{
  // TODO: Replace by more efficient implementation
  Pack<T> result;
  for (unsigned i=0; i < Pack<T>::size; ++i) {
    result[i] = util::pow(base[i], exp[i]);
  }
  return result;
}

References olb::util::pow().

Here is the call graph for this function:

Here is the caller graph for this function:

sqrt()

template<std::floating_point T>

Pack< T > olb::cpu::simd::sqrt

(

Pack< T >

x

)

Definition at line 98 of file pack.h.

{
  return x.sqrt();
}

store() [1/2]

template<typename T >

void olb::cpu::simd::store	(	T *	target,
		Pack< T >	value )

Here is the caller graph for this function:

store() [2/2]

template<typename T >

void olb::cpu::simd::store	(	T *	target,
		Pack< T >	value,
		const typename Pack< T >::index_t *	indices )

store< double >() [1/4]

template<>

void olb::cpu::simd::store< double >	(	double *	target,
		Pack< double >	value )

Definition at line 451 of file 256.h.

{
  _mm256_storeu_pd(target, value);
}

store< double >() [2/4]

template<>

void olb::cpu::simd::store< double >	(	double *	target,
		Pack< double >	value )

Definition at line 423 of file 512.h.

{
  _mm512_storeu_pd(target, value);
}

store< double >() [3/4]

template<>

void olb::cpu::simd::store< double >	(	double *	target,
		Pack< double >	value,
		const Pack< double >::index_t *	indices )

Definition at line 467 of file 256.h.

{
#ifdef __AVX512F__
  _mm256_i32scatter_pd(target, _mm_loadu_si128(reinterpret_cast<const __m128i*>(indices)), value, sizeof(double));
#else
  __m256d reg = value;
  for (unsigned i=0; i < simd::Pack<double>::size; ++i) {
    target[indices[i]] = reg[i];
  }
#endif
}

store< double >() [4/4]

template<>

void olb::cpu::simd::store< double >	(	double *	target,
		Pack< double >	value,
		const Pack< double >::index_t *	indices )

Definition at line 439 of file 512.h.

{
  _mm512_i32scatter_pd(target, _mm256_loadu_si256(reinterpret_cast<const __m256i*>(indices)), value, sizeof(double));
}

store< float >() [1/4]

template<>

void olb::cpu::simd::store< float >	(	float *	target,
		Pack< float >	value )

Definition at line 457 of file 256.h.

{
  _mm256_storeu_ps(target, value);
}

store< float >() [2/4]

template<>

void olb::cpu::simd::store< float >	(	float *	target,
		Pack< float >	value )

Definition at line 429 of file 512.h.

{
  _mm512_storeu_ps(target, value);
}

store< float >() [3/4]

template<>

void olb::cpu::simd::store< float >	(	float *	target,
		Pack< float >	value,
		const Pack< float >::index_t *	indices )

Definition at line 480 of file 256.h.

{
#ifdef __AVX512F__
  _mm256_i32scatter_ps(target, _mm256_loadu_si256(reinterpret_cast<const __m256i*>(indices)), value, sizeof(float));
#else
  __m256 reg = value;
  for (unsigned i=0; i < simd::Pack<float>::size; ++i) {
    target[indices[i]] = reg[i];
  }
#endif
}

store< float >() [4/4]

template<>

void olb::cpu::simd::store< float >	(	float *	target,
		Pack< float >	value,
		const Pack< float >::index_t *	indices )

Definition at line 446 of file 512.h.

{
  _mm512_i32scatter_ps(target, _mm512_loadu_si512(reinterpret_cast<const __m512i*>(indices)), value, sizeof(float));
}

Variable Documentation

PROT_RW

const int olb::cpu::simd::PROT_RW = PROT_READ | PROT_WRITE

Definition at line 131 of file column.h.