CuArray view has different variable type outside x inside the cuda kernel

[leoherz]

Hey guys,

Not sure if this is bug or not, but just in case, i am posting it here so we can discuss.

When I define a variable that is a view of CuArray outside the kernel, I get the expected result (considering the column major order storage).

However, inside the cuda kernel, i apply the exactly same command and get different results.

Here is a sample code to reproduce the behavior.

Can you please enlight me if this is expected or if it is indeed a problem ?

Thanks in advance

#-------------------------------------------

using CUDA

function my_kernel(matrix)

i = threadIdx().x + (blockIdx().x - 1) * blockDim().x

if i == 1

row_view = @view matrix[1, :]
col_view = @view matrix[:, 1]

@cuprintln("Type of row_view inside the kernel: ", typeof(row_view))
@cuprintln("Type of col_view inside the kernel: ", typeof(col_view))

end

matrix[i, i] += 1

return
end

matrix = CUDA.zeros( Float32 , 1024 , 1024 ) 

row_view = @view matrix[1, :]
col_view = @view matrix[:, 1]

println("Type of row_view outside the kernel: ", typeof(row_view))
println("Type of col_view outside the kernel: ", typeof(col_view))
println("")

@cuda threads=256 blocks=4 my_kernel(matrix)

result = Array(matrix)

#-------------------------------------------

[maleadt]

This is expected.

Not sure if this is bug or not, but just in case, i am posting it here so we can discuss.

There's "Discussions" for that :-)