CODEWITHSUNDEEP
c++visual-c++visual-studio-2012gpgpuc++-amp
Matthew Crews
Matthew Crews

Reputation: 4315

C++ AMP iterating over array_views with different dimensions

I am using C++ AMP with Visual Studio 2012 on Windows 8. I have a case where I have 2D array_view, experimentData and a 1D array_view, experimentFactors. I want to iterate over the first dimension of the 2D array_view and the 1 dimension of the 1D array_view. This is what I have but I constantly get errors saying there aren't overloads for this. I am wanting to iterate over the numberOfTests which is the first dimension. I am wanting the line:

auto test = experimentData[idx];

to return the entire row of data. I think the section method on the array_view is the way to do this but I can't figure out how.

array_view<int_2, 2> experimentData(numberOfTests, numberOfSolutions, initialConditionSet);
array_view<float_2, 1> experimentFactors(numberOfTests, factorData);
extent<1> e_size(numberOfTests);

parallel_for_each(e_size, [=] (index<1> idx) restrict(amp) {
    auto test = experimentData.section(idx);
    auto factors = experimentFactors[idx];

    analysisAlgorithm(test, factors);
});

The test object should be a 1xN section of the experimentData array_view. The factors object should be a single item from the experimentFactors array_view.

Clarification

The experimentData array_view has M rows and N columns

The experimentFactors array_view has M rows

Upvotes: 4

Views: 844

Answers (1)

Matthew Crews
Matthew Crews

Reputation: 4315

If you are someone working on C++ and you have questions like this one or with dealing projecting data to different dimensions stop and read this article:

http://blogs.msdn.com/b/nativeconcurrency/archive/2012/02/17/projections-in-c-amp.aspx

It deals with this exact problem in a beautiful way. Here is what my parallel_for loop looks like now:

parallel_for_each(e_size, [=] (index<1> idx) restrict(amp) {
    array_view<int_2, 1> test = experimentData[idx[0]];
    auto factors = experimentFactors[idx];

    analysisAlgorithm(test, factors);
});

Notice how the creation of the test array has changed. I state that it is going to be a 1D array of type int_2. Then I use the projection technique discussed in the article to select the entire row of data with = experimentData[idx[0]];. When you only give one index to a multidimensional array it selects all of the data held within that dimension. The article makes this clearer.

Upvotes: 4

Related Questions