DIP80: phobos additions

[Ilya Yaroshenko via Digitalmars-d]

On Tuesday, 9 June 2015 at 08:50:16 UTC, John Colvin wrote:
> On Tuesday, 9 June 2015 at 06:59:07 UTC, Andrei Alexandrescu 
> wrote:
>> On 6/8/15 8:26 PM, Ilya Yaroshenko wrote:
>>> On Sunday, 7 June 2015 at 18:27:16 UTC, Robert burner Schadek 
>>> wrote:
>>>> Phobos is awesome, the libs of go, python and rust only have 
>>>> better
>>>> marketing.
>>>> As discussed on dconf, phobos needs to become big and blow 
>>>> the rest
>>>> out of the sky.
>>>>
>>>> http://wiki.dlang.org/DIP80
>>>>
>>>> lets get OT, please discuss
>>>
>>> There are
>>> https://github.com/9il/simple_matrix and
>>> https://github.com/9il/cblas .
>>> I will try to rework them for Phobos.
>>>
>>> Any ideas and suggestions?
>>>
>>> Some notes about portability:
>>>   1. OS X has Accelerated framework builtin.
>>>   2. Linux has blast by default or it can be easily 
>>> installed. However
>>> default blast is very slow. The openBLAS is preferred.
>>>   3. Looks like there is no simple way to have BLAS support 
>>> on Windows.
>>>
>>> Should we provide BLAS library with DMD for Windows and maybe 
>>> Linux?
>>
>> I think licensing matters would make this difficult. What I do 
>> think we can do is:
>>
>> (a) Provide standard data layouts in std.array for the typical 
>> shapes supported by linear algebra libs: row major, column 
>> major, alongside with striding primitives.
>
> I don't think this is quite the right approach. 
> Multidimensional arrays and matrices are about accessing and 
> iteration over data, not data structures themselves. The 
> standard layouts are common special cases.

Probably we need both approaches:

[1]. Multidimensional random access slices (ranges, not only 
arrays)

We can do it easily:

size_t anyNumber;
auto ar = new int[3 * 8 * 9 + anyNumber];
auto slice = Slice[0..3, 4..8, 1..9];
assert(ar.canBeSlicedWith(slice)); //checks that ar.length <= 3 * 
8 * 9

auto tensor = ar.sliced(slice);
tensor[0, 1, 2] = 4;

auto matrix = tensor[0..$, 1, 0..$];
assert(matrix[0, 2] == 4);


[2]. BLAS Transposed.no/yes and Major.row/column (naming can be 
changed) flags for plain 2D matrixes based on
   2.1 D arrays (both GC and manual memory management)
   2.2 std.container.array (RefCounted)
RowMajor and RowMinor are not needed if Transposed is already 
defined. But this stuff helps engineers implement software in 
terms of corresponding mathematical documentation.

I hope to create @nogc versions for 2.1 and 2.2 (because GC is 
not needed for slices).
Furthermore [2] can be based on [1].


>> (b) Provide signatures for C and Fortran libraries so people 
>> who have them can use them easily with D.
>>
>> (c) Provide high-level wrappers on top of those functions.
>>
>>
>> Andrei
>
> That is how e.g. numpy works and it's OK, but D can do better.
>
> Ilya, I'm very interested in discussing this further with you. 
> I have a reasonable idea and implementation of how I would want 
> the generic n-dimensional types in D to work, but you seem to 
> have more experience with BLAS and LAPACK than me* and of 
> course interfacing with them is critical.
>
> *I rarely interact with them directly.

John, please describe your ideas and use cases. I think github 
issues is more convenient place. You have opened 
https://github.com/kyllingstad/scid/issues/24 . So I think we can 
past our code examples at SciD issue.