On Wednesday, 27 May 2020 at 16:53:37 UTC, jmh530 wrote:
On Wednesday, 27 May 2020 at 16:07:58 UTC, welkam wrote:
On Wednesday, 27 May 2020 at 01:31:23 UTC, data pulverizer
wrote:
column major
Cute puppies die when people access their arrays in column
major.
Not always true...many languages support column-major order
(Fortran, most obviously). The Eigen C++ library allows the
user to specify row major or column major. I had brought this
up with Ilya early on in mir and he thought it would increase
complexity to allow both and could also require more memory. So
mir is row major.
Actually it is a question of notation. For example, mir-lapack
uses ndslice as column-major Fortran arrays. This may cause some
headaches because the data needs to be transposed in mind. We can
think about ndslice as about column-major nd-arrays with the
reversed order of indexing.
The current template looks like
Slice(Iterator, size_t N = 1, SliceKind kind = 1)
If we add a special column-major notation, then it will look like
Slice(Iterator, size_t N = 1, SliceKind kind = Contiguous,
PayloadOrder = RowMajor)
A PR that adds this feature will be accepted.
