CBlockView< ScalarType > Struct Template Reference

View of one matrix block, const-correct via the ScalarType template parameter. CBlockView<const ScalarType> is read-only; CBlockView<ScalarType> is mutable and exposes apply<Overwrite>(f) for writing with on-the-fly quantized encoding. Evaluates to false if the block is absent from the sparsity pattern. More...

#include <CSysMatrix.hpp>

Public Types
using	QuantType = std::conditional_t< std::is_const_v< ScalarType >, const int8_t, int8_t >
using	QuantScaleType = std::conditional_t< std::is_const_v< ScalarType >, const uint8_t, uint8_t >

Public Member Functions
	operator bool () const
	False when the block is not present in the sparsity pattern.
std::remove_const_t< ScalarType >	operator() (unsigned long i, unsigned long j) const
	Return entry (row i, col j), decoding quantization if necessary.
template<bool Overwrite, class F , class S = ScalarType, su2enable_if<!std::is_const_v< S > > = 0>
void	apply (const F &f) const
	Write the block from callable f(i,j). Overwrite=true overwrites (or quantizes for Q_LU_SGS off-diagonal blocks); Overwrite=false accumulates into non-quantized storage only — accumulating into quantized storage would require decode-accumulate-encode and is a silent no-op. Only enabled for mutable (non-const ScalarType) views.

Public Attributes
ScalarType *	ptr = nullptr
	Full-precision block; non-null iff not quantized.
QuantScaleType *	qs = nullptr
	Per-row biased binary exponent; non-null iff quantized.
QuantType *	qv = nullptr
	Quantized values (row-major); non-null iff quantized.
unsigned long	nVar = 0

Detailed Description

template<class ScalarType>
struct CBlockView< ScalarType >

View of one matrix block, const-correct via the ScalarType template parameter. CBlockView<const ScalarType> is read-only; CBlockView<ScalarType> is mutable and exposes apply<Overwrite>(f) for writing with on-the-fly quantized encoding. Evaluates to false if the block is absent from the sparsity pattern.

Member Typedef Documentation

QuantScaleType

template<class ScalarType >

using CBlockView< ScalarType >::QuantScaleType = std::conditional_t<std::is_const_v<ScalarType>, const uint8_t, uint8_t>

QuantType

template<class ScalarType >

using CBlockView< ScalarType >::QuantType = std::conditional_t<std::is_const_v<ScalarType>, const int8_t, int8_t>

Member Function Documentation

apply()

template<class ScalarType >

template<bool Overwrite, class F , class S = ScalarType, su2enable_if<!std::is_const_v< S > > = 0>

void CBlockView< ScalarType >::apply ( const F &  f ) const

inline

Write the block from callable f(i,j). Overwrite=true overwrites (or quantizes for Q_LU_SGS off-diagonal blocks); Overwrite=false accumulates into non-quantized storage only — accumulating into quantized storage would require decode-accumulate-encode and is a silent no-op. Only enabled for mutable (non-const ScalarType) views.

operator bool()

template<class ScalarType >

CBlockView< ScalarType >::operator bool ( ) const

inlineexplicit

False when the block is not present in the sparsity pattern.

operator()()

template<class ScalarType >

std::remove_const_t< ScalarType > CBlockView< ScalarType >::operator() ( unsigned long  i, unsigned long  j  ) const

inline

Return entry (row i, col j), decoding quantization if necessary.

Member Data Documentation

nVar

template<class ScalarType >

unsigned long CBlockView< ScalarType >::nVar = 0

ptr

template<class ScalarType >

ScalarType* CBlockView< ScalarType >::ptr = nullptr

Full-precision block; non-null iff not quantized.

qs

template<class ScalarType >

QuantScaleType* CBlockView< ScalarType >::qs = nullptr

Per-row biased binary exponent; non-null iff quantized.

qv

template<class ScalarType >

QuantType* CBlockView< ScalarType >::qv = nullptr

Quantized values (row-major); non-null iff quantized.

---

The documentation for this struct was generated from the following file:

• Common/include/linear_algebra/CSysMatrix.hpp