Draft: Matrix Transpose Lesson 11 for Kernel #40
Conversation
| RAJA::View<double, RAJA::Layout<DIM>> A_t(a_t, M, N); | ||
|
|
||
| RAJA::TypedRangeSegment<int> row_range(0, N); | ||
| RAJA::TypedRangeSegment<int> col_range(0, M); |
There was a problem hiding this comment.
Where is the TODO part of the exercise?
| RAJA::KernelPolicy< | ||
| RAJA::statement::CudaKernel< | ||
| RAJA::statement::For<1, RAJA::cuda_thread_y_loop, | ||
| RAJA::statement::For<0, RAJA::cuda_thread_x_loop, |
There was a problem hiding this comment.
we should coordinate with @johnbowen42 to make sure we use the same policies
There was a problem hiding this comment.
I also think that it would be useful to explain the basic differences between 'loop' and 'direct' policies. We've had issues in the past when users apply loop policies following our basic examples and see performance that is less than expected. I would recommend to not use loop policies for an introduction.
There was a problem hiding this comment.
I think we should use the global policies here, and part of slides could be an overview of different options. I was planning include that in my slide deck.
|
We need a companion readme in this PR |
| RAJA::KernelPolicy< | ||
| RAJA::statement::CudaKernel< | ||
| RAJA::statement::For<1, RAJA::cuda_global_size_y_loop<8>, | ||
| RAJA::statement::For<0, RAJA::cuda_global_size_x_direct<32>, |
There was a problem hiding this comment.
Do you think we will have to explain the block sizes at the tutorial?
|
I think we can close this PR, we have decided to leave this lesson out of the intro tutorial. |
|
No longer needed |
No description provided.