steve
/
factor
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Vsub word for slicing BLAS vectors. Documentation improvements

db4
Joe Groff 2008-07-05 18:28:53 -07:00
parent 954737dfeb
commit 761fbfeb8f
4 changed files with 95 additions and 19 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

65
extra/math/blas/matrices/matrices-docs.factor
View File

@ -1,8 +1,17 @@
USING: alien byte-arrays help.markup help.syntax math.blas.vectors sequences ;
IN: math.blas.matrices
ARTICLE: "math.blas" "Basic Linear Algebra Subroutines (BLAS) interface"
ARTICLE: "math.blas-summary" "Basic Linear Algebra Subroutines (BLAS) interface"
"Factor provides an interface to high-performance vector and matrix math routines available in the system's BLAS library. A set of specialized types are provided for handling packed, unboxed vector data:"
{ $subsection "math.blas-types" }
"Scalar-vector and vector-vector operations are available in the " { $vocab-link "math.blas.vectors" } " vocabulary:"
{ $subsection "math.blas.vectors" }
"Vector-matrix and matrix-matrix operations are available in the " { $vocab-link "math.blas.matrices" } " vocabulary:"
{ $subsection "math.blas.matrices" }
"The low-level BLAS C interface can be accessed directly through the " { $vocab-link "math.blas.cblas" } " vocabulary." ;
ARTICLE: "math.blas-types" "BLAS interface types"
"BLAS vectors come in single- and double-precision, real and complex flavors:"
{ $subsection float-blas-vector }
{ $subsection double-blas-vector }
{ $subsection float-complex-blas-vector }
@ -11,30 +20,54 @@ ARTICLE: "math.blas" "Basic Linear Algebra Subroutines (BLAS) interface"
{ $subsection float-blas-matrix }
{ $subsection double-blas-matrix }
{ $subsection float-complex-blas-matrix }
{ $subsection double-complex-blas-matrix }
"Matrices can be transposed, broken down into sequences of row or column vectors, or sliced into rectangular submatrices:"
{ $subsection double-complex-blas-matrix }
"Syntax words are provided for constructing literal vectors and matrices in the " { $vocab-link "math.blas.syntax" } " vocabulary:"
{ $subsection "math.blas.syntax" }
"There are BOA constructors for all vector and matrix types, which provide the most flexibility in specifying memory layout:"
{ $subsection <float-blas-vector> }
{ $subsection <double-blas-vector> }
{ $subsection <float-complex-blas-vector> }
{ $subsection <double-complex-blas-vector> }
{ $subsection <float-blas-matrix> }
{ $subsection <double-blas-matrix> }
{ $subsection <float-complex-blas-matrix> }
{ $subsection <double-complex-blas-matrix> }
"For the simple case of creating a dense, zero-filled vector or matrix, simple empty object constructors are provided:"
{ $subsection <empty-vector> }
{ $subsection <empty-matrix> } ;
ARTICLE: "math.blas.matrices" "BLAS interface matrix operations"
"Transposing and slicing matrices:"
{ $subsection Mtranspose }
{ $subsection Mrows }
{ $subsection Mcols }
{ $subsection Msub }
"Scalar-vector and vector-vector operations are available in the " { $vocab-link "math.blas.vectors" } " vocabulary:"
{ $subsection Vnorm }
{ $subsection Vasum }
{ $subsection Viamax }
{ $subsection n*V }
{ $subsection V+ }
{ $subsection V- }
{ $subsection V. }
{ $subsection V.conj }
"Vector-matrix and matrix-matrix operations are available in the " { $vocab-link "math.blas.matrices" } " vocabulary:"
"Matrix-vector products:"
{ $subsection n*M.V+n*V-in-place }
{ $subsection n*M.V+n*V }
{ $subsection n*M.V }
{ $subsection M.V }
"Vector outer products:"
{ $subsection n*V(*)V+M-in-place }
{ $subsection n*V(*)Vconj+M-in-place }
{ $subsection n*V(*)V+M }
{ $subsection n*V(*)Vconj+M }
{ $subsection n*V(*)V }
{ $subsection n*V(*)Vconj }
{ $subsection V(*) }
{ $subsection V(*)conj }
"Matrix products:"
{ $subsection n*M.M+n*M-in-place }
{ $subsection n*M.M+n*M }
{ $subsection n*M.M }
{ $subsection M. }
"Scalar-matrix products:"
{ $subsection n*M-in-place }
{ $subsection n*M }
"The above operations only operate on the BLAS vector and matrix types. You cannot mix element types in a BLAS operation; for example, you can't use " { $link V. } " to take the dot product of a " { $link float-blas-vector } " and " { $link double-blas-vector } "."
$nl
"The low-level BLAS C interface can be accessed directly through the " { $vocab-link "math.blas.cblas" } " vocabulary." ;
{ $subsection M*n }
{ $subsection M/n } ;
ABOUT: "math.blas.matrices"
HELP: blas-matrix-base
{ $class-description "The base class for all BLAS matrix types. Objects of this type should not be created directly; instead, instantiate one of the typed subclasses:"

30
extra/math/blas/vectors/vectors-docs.factor
View File

@ -1,6 +1,32 @@
USING: alien byte-arrays help.markup help.syntax sequences ;
IN: math.blas.vectors
ARTICLE: "math.blas.vectors" "BLAS interface vector operations"
"Slicing vectors:"
{ $subsection Vsub }
"Taking the norm (magnitude) of a vector:"
{ $subsection Vnorm }
"Summing and taking the maximum of elements:"
{ $subsection Vasum }
{ $subsection Viamax }
{ $subsection Vamax }
"Scalar-vector products:"
{ $subsection n*V-in-place }
{ $subsection n*V }
{ $subsection V*n }
{ $subsection V/n }
{ $subsection Vneg }
"Vector addition:"
{ $subsection n*V+V-in-place }
{ $subsection n*V+V }
{ $subsection V+ }
{ $subsection V- }
"Vector inner products:"
{ $subsection V. }
{ $subsection V.conj } ;
ABOUT: "math.blas.vectors"
HELP: blas-vector-base
{ $class-description "The base class for all BLAS vector types. Objects of this type should not be created directly; instead, instantiate one of the typed subclasses:"
{ $list
@ -100,4 +126,6 @@ HELP: V/n
{ n*V+V-in-place n*V-in-place n*V+V n*V V+ V- Vneg V*n V/n } related-words
HELP: Vsub
{ $values { "v" "a BLAS vector inheriting from " { $link blas-vector-base } } { "start" "The index of the first element of the slice" } { "length" "The length of the slice" } }
{ $description "Slice a subvector out of " { $snippet "v" } " with the given length. The subvector will share storage with the parent vector." } ;

7
extra/math/blas/vectors/vectors-tests.factor
View File

@ -171,3 +171,10 @@ unit-test
[ -6.0 ] [ dvector{ 1.0 -5.0 4.0 -6.0 -1.0 } Vamax ] unit-test
[ C{ 2.0 -5.0 } ] [ cvector{ C{ 2.0 -5.0 } 4.0 -6.0 -1.0 } Vamax ] unit-test
[ C{ 2.0 -5.0 } ] [ zvector{ C{ 2.0 -5.0 } 4.0 -6.0 -1.0 } Vamax ] unit-test
! Vsub
[ svector{ -5.0 4.0 -6.0 } ] [ svector{ 1.0 -5.0 4.0 -6.0 -1.0 } 1 3 Vsub ] unit-test
[ dvector{ -5.0 4.0 -6.0 } ] [ dvector{ 1.0 -5.0 4.0 -6.0 -1.0 } 1 3 Vsub ] unit-test
[ cvector{ -5.0 C{ 4.0 3.0 } -6.0 } ] [ cvector{ 1.0 -5.0 C{ 4.0 3.0 } -6.0 -1.0 } 1 3 Vsub ] unit-test
[ zvector{ -5.0 C{ 4.0 3.0 } -6.0 } ] [ zvector{ 1.0 -5.0 C{ 4.0 3.0 } -6.0 -1.0 } 1 3 Vsub ] unit-test

12
extra/math/blas/vectors/vectors.factor
View File

@ -135,7 +135,7 @@ PRIVATE>
[ length>> 0 ]
[ (blas-vector-like) ] tri ;
: <empty-vector> ( length exemplar -- <empty-vector> )
: <empty-vector> ( length exemplar -- vector )
[ element-type <c-array> ]
[ 1 swap ] 2bi
(blas-vector-like) ;
@ -285,5 +285,13 @@ METHOD: Viamax { float-complex-blas-vector }
METHOD: Viamax { double-complex-blas-vector }
(prepare-nrm2) cblas_izamax ;
: Vamax ( v -- max )
: Vamax ( x -- max )
[ Viamax ] keep nth ; inline
: Vsub ( v start length -- vsub )
rot [
[
nip [ inc>> ] [ element-type heap-size ] [ data>> ] tri
[ * * ] dip <displaced-alien>
] [ swap 2nip ] [ 2nip inc>> ] 3tri
] keep (blas-vector-like) ;