how to convert a slice to a standard D array (T[]) ? about mir-algorithm HOT 14 CLOSED

this snippet (copy sl to the buffer) seems to indicate no such utility to flatten elements into an array?

You can use flattened to use flat representation (it is not an array).
http://docs.algorithm.dlang.io/latest/mir_ndslice_topology.html#.flattened

not sure why median above doesn't use ndarray

median uses preallocated buffer.

from mir-algorithm.

BTW, for Slice!(Contiguous, packs, T*) you can always do sl.iterator[0 .. sl.elementsCount].

from mir-algorithm.

... this will return a common D array without allocation

from mir-algorithm.

How about adding such common questions into the main README or the wiki?
I hope that @timotheecour won't be the last one to try to transition ;-)

from mir-algorithm.

Looks good

from mir-algorithm.

@9il

• arrayview

BTW, for Slice!(Contiguous, packs, T*) you can always do sl.iterator[0 .. sl.elementsCount].

IMO this should be exposed as a method, it's a common case to interoperate between mir and non-mir code.

How about:

auto arrayview(S)(S slice) if (isContiguous!S){
  return sl.iterator[0 .. sl.elementsCount];
}

• ndarray/flattened with existing buffer

not sure why median above doesn't use ndarray
median uses preallocated buffer.

Likewise, ndarray/flattened using a preallocated buffer (or using Appender!T or an arbitrary allocator) is a common case, and should be exposed in the library. What do you suggest?
How about overloads:

auto darray=myslice.ndarray(Mallocator.instance);
auto darray=myslice.flattened(Mallocator.instance);

(or maybe using ndarrayAlloc flattenedAlloc if name clash is an issue)

• is there a plan to upgrade the reference example in https://dlang.org/phobos/std_experimental_ndslice.html with the new mir? (hopefully with a better buf[n++] = e; replaced as discussed above)

from mir-algorithm.

is there a plan to upgrade the reference example in https://dlang.org/phobos/std_experimental_ndslice.html with the new mir? (hopefully with a better buf[n++] = e; replaced as discussed above)

ndslice will be removed with the next release

from mir-algorithm.

auto arrayview(S)(S slice) if (isContiguous!S){
return sl.iterator[0 .. sl.elementsCount];
}

auto asArray(size_t[] packs, T)(Slice!(Contiguous, packs, T*) sl)
{
    return sl.unpack.iterator[0..sl.elementsCount];
}

auto darray=myslice.ndarray(Mallocator.instance);
auto darray=myslice.flattened(Mallocator.instance);

makeSlice already allows to create new Contiguous slice with a data from another slice.
Then asArray can be used to get array representation.

hopefully with a better buf[n++] = e; replaced as discussed above

We need other method, something like copy to an array.

from mir-algorithm.

Is there a plan to upgrade the reference example in https://dlang.org/phobos/std_experimental_ndslice.html with the new mir?
ndslice will be removed with the next release

Just sent out the request for removal:
dlang/phobos#5187

from mir-algorithm.

is there a plan to upgrade the reference example in https://dlang.org/phobos/std_experimental_ndslice.html with the new mir? (hopefully with a better buf[n++] = e; replaced as discussed above)

ndslice will be removed with the next release

ya but adding this example (or whatever other similar good example snippet) and upgrading it in the new mir repo

from mir-algorithm.

@9il
asArray sounds good assuming all the asX do not allocate as well.

To be clear, that doesn't exist yet right? (couldn't find it)

makeSlice already allows to create new Contiguous slice with a data from another slice.
Then asArray can be used to get array representation.
We need other method, something like copy to an array.

is that the simplest?

import std.algorithm:copy;
myslice.makeSlice.asArray.copy(buffer);

// or something like: myslice.flattened(Allocater from buffer); ?

from mir-algorithm.

@9il
asArray sounds good assuming all the asX do not allocate as well.

To be clear, that doesn't exist yet right? (couldn't find it)

makeSlice already allows to create new Contiguous slice with a data from another slice.
Then asArray can be used to get array representation.
We need other method, something like copy to an array.

is that the simplest?

import std.algorithm:copy;
myslice.makeSlice.asArray.copy(buffer);

makeSlice allocates data. So you do not need to use it here.

I do not understand the problem with buffer filling and current API. The code in the example will remain the same because the buffer size is very small (like 4x4 pixels), so manual coping will be faster in this case.

Current API allows

buf.ptr.sliced(sl.shape)[] = sl;

This looks pretty clear and simple, IMO

from mir-algorithm.

BTW, for Slice!(Contiguous, packs, T*) you can always do sl.iterator[0 .. sl.elementsCount]

should that be simply sl.iterator.array ? (ie a.toArray can be done via a.iterator.array)

from mir-algorithm.

BTW, for Slice!(Contiguous, packs, T*) you can always do sl.iterator[0 .. sl.elementsCount]
should that be simply sl.iterator.array ? (ie a.toArray can be done via a.iterator.array)

For Slice!(Contiguous, packs, T*) slices it is just sl.field (without data reallocation).

For all slices sl.field.array allocates new array and copies slice data to it, array is from std.array.

from mir-algorithm.