fml  0.1-0
Fused Matrix Library
fml::gpumat< REAL > Class Template Reference

Matrix class for data held on a single GPU. More...

#include <gpumat.hh>

Inheritance diagram for fml::gpumat< REAL >:
fml::unimat< REAL >

Public Member Functions

 gpumat (std::shared_ptr< card > gpu)
 
 gpumat (std::shared_ptr< card > gpu, len_t nrows, len_t ncols)
 
 gpumat (std::shared_ptr< card > gpu, REAL *data, len_t nrows, len_t ncols, bool free_on_destruct=false)
 
 gpumat (const gpumat &x)
 
void resize (len_t nrows, len_t ncols)
 Resize the internal object storage. More...
 
void resize (std::shared_ptr< card > gpu, len_t nrows, len_t ncols)
 
void inherit (std::shared_ptr< card > gpu, REAL *data, len_t nrows, len_t ncols, bool free_on_destruct=false)
 Set the internal object storage to the specified array. More...
 
gpumat< REAL > dupe () const
 Duplicate the object in a deep copy.
 
void print (uint8_t ndigits=4, bool add_final_blank=true) const
 Print all values in the object. More...
 
void info () const
 Print some brief information about the object.
 
void fill_zero ()
 Set all values to zero.
 
void fill_val (const REAL v)
 Set all values to input value. More...
 
void fill_linspace ()
 Set values to linearly spaced numbers. More...
 
void fill_linspace (const REAL start, const REAL stop)
 
void fill_eye ()
 Set diagonal entries to 1 and non-diagonal entries to 0.
 
void fill_diag (const gpuvec< REAL > &v)
 Set diagonal entries of the matrix to those in the vector. More...
 
void fill_runif (const uint32_t seed, const REAL min=0, const REAL max=1)
 Set diagonal entries to 1 and non-diagonal entries to 0. More...
 
void fill_runif (const REAL min=0, const REAL max=1)
 This is an overloaded member function, provided for convenience. It differs from the above function only in what argument(s) it accepts.
 
void fill_rnorm (const uint32_t seed, const REAL mean=0, const REAL sd=1)
 Set diagonal entries to 1 and non-diagonal entries to 0. More...
 
void fill_rnorm (const REAL mean=0, const REAL sd=1)
 This is an overloaded member function, provided for convenience. It differs from the above function only in what argument(s) it accepts.
 
void diag (gpuvec< REAL > &v)
 Get the diagonal entries. More...
 
void antidiag (gpuvec< REAL > &v)
 Get the anti-diagonal entries, i.e. those on the bottom-left to top-right. More...
 
void scale (const REAL s)
 Multiply all values by the input value. More...
 
void rev_rows ()
 Reverse the rows of the matrix.
 
void rev_cols ()
 Reverse the columns of the matrix.
 
bool any_inf () const
 Are any values infinite?
 
bool any_nan () const
 
REAL get (const len_t i) const
 Get the specified value. More...
 
REAL get (const len_t i, const len_t j) const
 Get the specified value. More...
 
void set (const len_t i, const REAL v)
 Set the storage at the specified index with the provided value. More...
 
void set (const len_t i, const len_t j, const REAL v)
 Set the storage at the specified index with the provided value. More...
 
void get_row (const len_t i, gpuvec< REAL > &v) const
 Get the specified row. More...
 
void set_row (const len_t i, const gpuvec< REAL > &v)
 Set the specified row. More...
 
void get_col (const len_t j, gpuvec< REAL > &v) const
 Get the specified column. More...
 
void set_col (const len_t j, const gpuvec< REAL > &v)
 Set the specified column. More...
 
bool operator== (const gpumat< REAL > &x) const
 See if the two objects are the same. More...
 
bool operator!= (const gpumat< REAL > &x) const
 See if the two objects are not the same. Uses same internal logic as the == method. More...
 
gpumat< REAL > & operator= (const gpumat< REAL > &x)
 Operator that sets the LHS to a shallow copy of the input. Desctruction of the LHS object will not result in the internal array storage being freed. More...
 
std::shared_ptr< cardget_card () const
 
dim3 get_blockdim () const
 
dim3 get_griddim () const
 
- Public Member Functions inherited from fml::unimat< REAL >
bool is_square () const
 Is the matrix square?
 
len_t nrows () const
 Number of rows.
 
len_t ncols () const
 Number of columns.
 
REAL * data_ptr ()
 Pointer to the internal array.
 
REAL * data_ptr () const
 

Protected Attributes

std::shared_ptr< cardc
 
- Protected Attributes inherited from fml::unimat< REAL >
len_t m
 
len_t n
 
REAL * data
 
bool free_data
 

Additional Inherited Members

- Protected Member Functions inherited from fml::unimat< REAL >
bool should_free () const
 
void check_index (const len_t i) const
 
void check_index (const len_t i, const len_t j) const
 
void printval (const REAL val, uint8_t ndigits) const
 
void printval (const int val, uint8_t ndigits) const
 

Detailed Description

template<typename REAL>
class fml::gpumat< REAL >

Matrix class for data held on a single GPU.

Template Parameters
REALshould be '__half', 'float', or 'double'.

Member Function Documentation

◆ antidiag()

template<typename REAL >
void fml::gpumat< REAL >::antidiag ( gpuvec< REAL > &  v)

Get the anti-diagonal entries, i.e. those on the bottom-left to top-right.

Parameters
[out]vThe anti-diagonal. Length should match the length of the diagonal of the input (minimum of the matrix dimensions). If not, the vector will automatically be resized.

Memory Allocations
If the output dimension is inappropriately sized, it will automatically be re-allocated.

Exceptions
If a reallocation is triggered and fails, a bad_alloc exception will be thrown.

◆ diag()

template<typename REAL >
void fml::gpumat< REAL >::diag ( gpuvec< REAL > &  v)

Get the diagonal entries.

Parameters
[out]vThe diagonal. Length should match the length of the diagonal of the input (minimum of the matrix dimensions). If not, the vector will automatically be resized.

Memory Allocations
If the output dimension is inappropriately sized, it will automatically be re-allocated.

Exceptions
If a reallocation is triggered and fails, a bad_alloc exception will be thrown.

◆ fill_diag()

template<typename REAL >
void fml::gpumat< REAL >::fill_diag ( const gpuvec< REAL > &  v)

Set diagonal entries of the matrix to those in the vector.

If the vector is smaller than the matrix diagonal, the vector will recycle until the matrix diagonal is filled. If the vector is longer, then not all of it will be used.

Parameters
[in]vVector of values to set the matrix diagonal to.

◆ fill_linspace()

template<typename T >
void fml::gpumat< T >::fill_linspace

Set values to linearly spaced numbers.

Parameters
[in]start,stopBeginning/ending numbers. If not supplied, the matrix will be filled with whole numbers from 1 to the total number of elements.

◆ fill_rnorm()

template<typename REAL >
void fml::gpumat< REAL >::fill_rnorm ( const uint32_t  seed,
const REAL  mean = 0,
const REAL  sd = 1 
)

Set diagonal entries to 1 and non-diagonal entries to 0.

Parameters
[in]seedSeed for the rng.
[in]mean,sdParameters for the generator.

Memory Allocations
Some internal GPU data has to be allocated.

Exceptions
If something goes wrong with the GPU generator, a runtime_error exception will be thrown.

◆ fill_runif()

template<typename REAL >
void fml::gpumat< REAL >::fill_runif ( const uint32_t  seed,
const REAL  min = 0,
const REAL  max = 1 
)

Set diagonal entries to 1 and non-diagonal entries to 0.

Parameters
[in]seedSeed for the rng.
[in]min,maxParameters for the generator.

Memory Allocations
Some internal GPU data has to be allocated.

Exceptions
If something goes wrong with the GPU generator, a runtime_error exception will be thrown.

◆ fill_val()

template<typename REAL >
void fml::gpumat< REAL >::fill_val ( const REAL  v)

Set all values to input value.

Parameters
[in]vValue to set all data values to.

◆ get() [1/2]

template<typename REAL >
REAL fml::gpumat< REAL >::get ( const len_t  i) const

Get the specified value.

Parameters
[in]iThe index of the desired value, 0-indexed. The numbering considers the internal storage as a 1-dimensional array.

Exceptions
If indices are out of bounds, the method will throw a runtime_error exception.

◆ get() [2/2]

template<typename REAL >
REAL fml::gpumat< REAL >::get ( const len_t  i,
const len_t  j 
) const

Get the specified value.

Parameters
[in]i,jThe indices of the desired value, 0-indexed.

Exceptions
If indices are out of bounds, the method will throw a runtime_error exception.

◆ get_col()

template<typename REAL >
void fml::gpumat< REAL >::get_col ( const len_t  j,
fml::gpuvec< REAL > &  v 
) const

Get the specified column.

Parameters
[in]jThe desired column, 0-indexed.
[out]vThe column values.

Memory Allocations
If the output dimension is inappropriately sized, it will automatically be re-allocated.

Exceptions
If j is an inappropriate value (i.e. does not refer to a matrix column), then the method will throw a logic_error exception. If a reallocation is triggered and fails, a bad_alloc exception will be thrown.

◆ get_row()

template<typename REAL >
void fml::gpumat< REAL >::get_row ( const len_t  i,
fml::gpuvec< REAL > &  v 
) const

Get the specified row.

Parameters
[in]iThe desired row, 0-indexed.
[out]vThe row values.

Memory Allocations
If the output dimension is inappropriately sized, it will automatically be re-allocated.

Exceptions
If i is an inappropriate value (i.e. does not refer to a matrix row), then the method will throw a logic_error exception. If a reallocation is triggered and fails, a bad_alloc exception will be thrown.

◆ inherit()

template<typename REAL >
void fml::gpumat< REAL >::inherit ( std::shared_ptr< card gpu,
REAL *  data,
len_t  nrows,
len_t  ncols,
bool  free_on_destruct = false 
)

Set the internal object storage to the specified array.

Parameters
[in]gpuShared pointer to GPU card object.
[in]dataValue storage.
[in]nrows,ncolsNumber rows/columns of the matrix. The product of these should be the length of the input data.
[in]free_on_destructShould the object destructor free the internal array data?

Exceptions
If the input values are invalid, a runtime_error exception will be thrown.

◆ operator!=()

template<typename REAL >
bool fml::gpumat< T >::operator!= ( const gpumat< REAL > &  x) const

See if the two objects are not the same. Uses same internal logic as the == method.

Parameters
[in]Comparisonobject.

◆ operator=()

template<typename REAL >
fml::gpumat< REAL > & fml::gpumat< REAL >::operator= ( const gpumat< REAL > &  x)

Operator that sets the LHS to a shallow copy of the input. Desctruction of the LHS object will not result in the internal array storage being freed.

Parameters
[in]xSetter value.

◆ operator==()

template<typename REAL >
bool fml::gpumat< T >::operator== ( const gpumat< REAL > &  x) const

See if the two objects are the same.

Parameters
[in]Comparisonobject.
Returns
If the dimensions mismatch, then false is necessarily returned. Next, if the card objects have different ordinal IDs, then false is returned. Next, if the pointer to the internal storage arrays match, then true is returned. Otherwise the objects are compared value by value.

◆ print()

template<typename REAL >
void fml::gpumat< REAL >::print ( uint8_t  ndigits = 4,
bool  add_final_blank = true 
) const

Print all values in the object.

Parameters
[in]ndigitsNumber of decimal digits to print.
[in]add_final_blankShould a final blank line be printed?

◆ resize()

template<typename REAL >
void fml::gpumat< REAL >::resize ( len_t  nrows,
len_t  ncols 
)

Resize the internal object storage.

Parameters
[in]nrows,ncolsNumber rows/columns needed.

Memory Allocations
Resizing triggers a re-allocation.

Exceptions
If the reallocation fails, a bad_alloc exception will be thrown. If the input values are invalid, a runtime_error exception will be thrown.

◆ scale()

template<typename REAL >
void fml::gpumat< REAL >::scale ( const REAL  s)

Multiply all values by the input value.

Parameters
[in]sScaling value.

◆ set() [1/2]

template<typename REAL >
void fml::gpumat< REAL >::set ( const len_t  i,
const len_t  j,
const REAL  v 
)

Set the storage at the specified index with the provided value.

Parameters
[in]i,jThe indices of the desired value, 0-indexed.
[in]vSetter value.

Exceptions
If indices are out of bounds, the method will throw a runtime_error exception.

◆ set() [2/2]

template<typename REAL >
void fml::gpumat< REAL >::set ( const len_t  i,
const REAL  v 
)

Set the storage at the specified index with the provided value.

Parameters
[in]iThe index of the desired value, 0-indexed. The numbering considers the internal storage as a 1-dimensional array.
[in]vSetter value.

Exceptions
If indices are out of bounds, the method will throw a runtime_error exception.

◆ set_col()

template<typename REAL >
void fml::gpumat< REAL >::set_col ( const len_t  i,
const gpuvec< REAL > &  v 
)

Set the specified column.

Parameters
[in]iThe desired row, 0-indexed.
[in]vThe column values.

Exceptions
If i is an inappropriate value (i.e. does not refer to a matrix row), then the method will throw a logic_error exception. If the vector is inappropriately sized, a runtime_error exception will be thrown.

◆ set_row()

template<typename REAL >
void fml::gpumat< REAL >::set_row ( const len_t  i,
const gpuvec< REAL > &  v 
)

Set the specified row.

Parameters
[in]iThe desired row, 0-indexed.
[in]vThe row values.

Exceptions
If i is an inappropriate value (i.e. does not refer to a matrix row), then the method will throw a logic_error exception. If the vector is inappropriately sized, a runtime_error exception will be thrown.


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