Skip to content
Projects
Groups
Snippets
Help
Loading...
Help
Support
Submit feedback
Contribute to GitLab
Sign in
Toggle navigation
tnl-dev
Project
Project
Details
Activity
Releases
Cycle Analytics
Repository
Repository
Files
Commits
Branches
Tags
Contributors
Graph
Compare
Charts
Issues
22
Issues
22
List
Boards
Labels
Milestones
Merge Requests
1
Merge Requests
1
Wiki
Wiki
Snippets
Snippets
Members
Members
Collapse sidebar
Close sidebar
Activity
Graph
Charts
Create a new issue
Commits
Issue Boards
Open sidebar
TNL
tnl-dev
Commits
496d4e77
Commit
496d4e77
authored
Aug 29, 2019
by
Tomáš Oberhuber
Browse files
Options
Browse Files
Download
Email Patches
Plain Diff
Writting documentation on reduction.
parent
48b7a7e0
Changes
1
Hide whitespace changes
Inline
Side-by-side
Showing
1 changed file
with
162 additions
and
0 deletions
+162
-0
Reduction.h
src/TNL/Containers/Algorithms/Reduction.h
+162
-0
No files found.
src/TNL/Containers/Algorithms/Reduction.h
View file @
496d4e77
...
...
@@ -22,12 +22,59 @@ namespace TNL {
namespace
Containers
{
namespace
Algorithms
{
/**
* \brief Reduction implements [(parallel) reduction](https://en.wikipedia.org/wiki/Reduce_(parallel_pattern)) for vectors and arrays.
*
* Reduction can be used for operations having one or more vectors (or arrays) elements is input and returning
* one number (or element) as output. Some examples of such operations can be vectors/arrays comparison,
* vector norm, scalar product of two vectors or computing minimum or maximum. If one needs to know even
* position of the smallest or the largest element, reduction with argument can be used.
*
* \tparam Device this parameter says on what device the reduction is gonna be performed.
*
* See \ref Reduction< Devices::Host > and \ref Reduction< Devices::Cuda >.
*/
template
<
typename
Device
>
struct
Reduction
;
template
<>
struct
Reduction
<
Devices
::
Host
>
{
/**
* \brief Computes reduction on CPU.
*
* \tparam Index is a type for indexing.
* \tparam Result is a type of the reduction result.
* \tparam ReductionOperation is a lambda function performing the reduction.
* \tparam DataFetcher is a lambda function for fetching the input data.
*
* \param size is number of elements to be reduced.
* \param reduction is a lambda function defining the reduction operation.
* \param dataFetcher is a lambda function fetching the input data.
* \param zero is the idempotent element for the reduction operation, i.e. element which
* does not change the result of the reduction.
* \return result of the reduction
*
* The dataFetcher lambda function takes one argument which is index of the element to be fetched:
*
* ```
* auto dataFetcher1 = [=] __cuda_callable__ ( Index i ) { return ... };
* ```
*
* The reduction lambda function takes two variables which are supposed to be reduced:
*
* ```
* auto reduction = [] __cuda_callable__ ( const Result& a, const Result& b ) { return ... };
* ```
*
* \par Example
*
* \include ReductionAndScan/SumExample.cpp
*
* \par Output
*
* \include SumExample.out
*/
template
<
typename
Index
,
typename
Result
,
typename
ReductionOperation
,
...
...
@@ -38,6 +85,46 @@ struct Reduction< Devices::Host >
DataFetcher
&
dataFetcher
,
const
Result
&
zero
);
/**
* \brief Computes reduction on CPU and returns position of an element of interest.
*
* For example in case of computing minimal or maximal element in array/vector,
* the position of the element having given value can be obtained. The use of this method
* is, however, more flexible.
*
* \tparam Index is a type for indexing.
* \tparam Result is a type of the reduction result.
* \tparam ReductionOperation is a lambda function performing the reduction.
* \tparam DataFetcher is a lambda function for fetching the input data.
*
* \param size is number of elements to be reduced.
* \param reduction is a lambda function defining the reduction operation and managing the elements positions.
* \param dataFetcher is a lambda function fetching the input data.
* \param zero is the idempotent element for the reduction operation, i.e. element which
* does not change the result of the reduction.
* \return result of the reduction in a form of std::pair< Index, Result> structure. `pair.first'
* is the element position and `pair.second` is the reduction result.
*
* The dataFetcher lambda function takes one argument which is index of the element to be fetched:
*
* ```
* auto dataFetcher1 = [=] __cuda_callable__ ( Index i ) { return ... };
* ```
*
* The reduction lambda function takes two variables which are supposed to be reduced:
*
* ```
* auto reduction = [] __cuda_callable__ ( Index& aIdx, const Index& bIdx, const Result& a, const Result& b ) { return ... };
* ```
*
* \par Example
*
* \include ReductionAndScan/ReductionWithArgument.cpp
*
* \par Output
*
* \include ReductionWithArgument.out
*/
template
<
typename
Index
,
typename
Result
,
typename
ReductionOperation
,
...
...
@@ -52,6 +139,41 @@ struct Reduction< Devices::Host >
template
<>
struct
Reduction
<
Devices
::
Cuda
>
{
/**
* \brief Computes reduction on GPU.
*
* \tparam Index is a type for indexing.
* \tparam Result is a type of the reduction result.
* \tparam ReductionOperation is a lambda function performing the reduction.
* \tparam DataFetcher is a lambda function for fetching the input data.
*
* \param size is number of elements to be reduced.
* \param reduction is a lambda function defining the reduction operation.
* \param dataFetcher is a lambda function fetching the input data.
* \param zero is the idempotent element for the reduction operation, i.e. element which
* does not change the result of the reduction.
* \return result of the reduction
*
* The dataFetcher lambda function takes one argument which is index of the element to be fetched:
*
* ```
* auto dataFetcher1 = [=] __cuda_callable__ ( Index i ) { return ... };
* ```
*
* The reduction lambda function takes two variables which are supposed to be reduced:
*
* ```
* auto reduction = [] __cuda_callable__ ( const Result& a, const Result& b ) { return ... };
* ```
*
* \par Example
*
* \include ReductionAndScan/SumExample.cpp
*
* \par Output
*
* \include SumExample.out
*/
template
<
typename
Index
,
typename
Result
,
typename
ReductionOperation
,
...
...
@@ -62,6 +184,46 @@ struct Reduction< Devices::Cuda >
DataFetcher
&
dataFetcher
,
const
Result
&
zero
);
/**
* \brief Computes reduction on GPU and returns position of an element of interest.
*
* For example in case of computing minimal or maximal element in array/vector,
* the position of the element having given value can be obtained. The use of this method
* is, however, more flexible.
*
* \tparam Index is a type for indexing.
* \tparam Result is a type of the reduction result.
* \tparam ReductionOperation is a lambda function performing the reduction.
* \tparam DataFetcher is a lambda function for fetching the input data.
*
* \param size is number of elements to be reduced.
* \param reduction is a lambda function defining the reduction operation and managing the elements positions.
* \param dataFetcher is a lambda function fetching the input data.
* \param zero is the idempotent element for the reduction operation, i.e. element which
* does not change the result of the reduction.
* \return result of the reduction in a form of std::pair< Index, Result> structure. `pair.first'
* is the element position and `pair.second` is the reduction result.
*
* The dataFetcher lambda function takes one argument which is index of the element to be fetched:
*
* ```
* auto dataFetcher1 = [=] __cuda_callable__ ( Index i ) { return ... };
* ```
*
* The reduction lambda function takes two variables which are supposed to be reduced:
*
* ```
* auto reduction = [] __cuda_callable__ ( Index& aIdx, const Index& bIdx, const Result& a, const Result& b ) { return ... };
* ```
*
* \par Example
*
* \include ReductionAndScan/ReductionWithArgument.cpp
*
* \par Output
*
* \include ReductionWithArgument.out
*/
template
<
typename
Index
,
typename
Result
,
typename
ReductionOperation
,
...
...
Write
Preview
Markdown
is supported
0%
Try again
or
attach a new file
Attach a file
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment