On Saturday, 24 September 2016 at 18:15:30 UTC, ZombineDev wrote:
> On Saturday, 24 September 2016 at 17:46:07 UTC, Ilya Yaroshenko wrote:
>> On Saturday, 24 September 2016 at 13:49:35 UTC, Andrei Alexandrescu wrote:
>>> [...]
>>
>> Eigen was added (but only data, still need to write text). Relative charts was added. You was added "Acknowledgements" section --Ilya
>
> It would also be interesting to compare the results to Blaze [1]. According to https://www.youtube.com/watch?v=hfn0BVOegac it is faster than Eigen and on some instances faster than even Intel MKL.
>
> [1]: https://bitbucket.org/blaze-lib/blaze

It has not CBLAS interface like Eigen, so additional efforts are required.

On Saturday, 24 September 2016 at 19:01:47 UTC, Andrei Alexandrescu wrote:
> On 09/24/2016 01:46 PM, Ilya Yaroshenko wrote:
>> [...]
>
> Looks awesome. Couple more nits after one more pass:
>
> "numerical and scientific projects" -> "numeric and scientific projects"
>
> [...]

Thank for the review! I have added notes about Eigen and CBLAS interface example.

Ilya

On 09/25/2016 03:45 AM, Ilya Yaroshenko wrote:
> On Saturday, 24 September 2016 at 19:01:47 UTC, Andrei Alexandrescu wrote:
>> On 09/24/2016 01:46 PM, Ilya Yaroshenko wrote:
>>> [...]
>>
>> Looks awesome. Couple more nits after one more pass:
>>
>> "numerical and scientific projects" -> "numeric and scientific projects"
>>
>> [...]
>
> Thank for the review! I have added notes about Eigen and CBLAS interface
> example.
>
> Ilya

Some more:

In the same time, CBLAS interface is unwieldy ->
On the other hand, CBLAS interface is unwieldy
(Or something better?)

GLAS calling conversion ->
GLAS calling convention

single precisions ->
single precision
(Several occurrences)

double precisions ->
double precision
(Several occurrences)

Stay in touch with the lastest developments in scientific computing for D. ->
(I will let others recommend something better there but neither "stay in touch" nor "lastest" sounds right to my ears. :) )

Ali

On Sunday, 25 September 2016 at 23:03:27 UTC, Ali Çehreli wrote:
> On 09/25/2016 03:45 AM, Ilya Yaroshenko wrote:
>> On Saturday, 24 September 2016 at 19:01:47 UTC, Andrei Alexandrescu wrote:
>>> [...]
>>
>> Thank for the review! I have added notes about Eigen and CBLAS interface
>> example.
>>
>> Ilya
>
> Some more:
>
> In the same time, CBLAS interface is unwieldy ->
> On the other hand, CBLAS interface is unwieldy
> (Or something better?)
>
> GLAS calling conversion ->
> GLAS calling convention
>
> single precisions ->
> single precision
> (Several occurrences)
>
> double precisions ->
> double precision
> (Several occurrences)
>
> Stay in touch with the lastest developments in scientific computing for D. ->
> (I will let others recommend something better there but neither "stay in touch" nor "lastest" sounds right to my ears. :) )
>
> Ali

Thank you, fixed

On Sunday, 25 September 2016 at 23:03:27 UTC, Ali Çehreli wrote:
> Stay in touch with the lastest developments in scientific computing for D. ->
> (I will let others recommend something better there but neither "stay in touch" nor "lastest" sounds right to my ears. :) )

"lastest" -> "latest" ... ?

On Sunday, 25 September 2016 at 10:45:35 UTC, Ilya Yaroshenko wrote:
> Thank for the review! I have added notes about Eigen and CBLAS interface example.

One extra suggestion:

"Mir GLAS has native mir.ndslice interface" -> "Mir GLAS has a native mir.ndslice interface"

I would also suggest adding a small note on what `ndslice` is, e.g.

"mir.ndslice is a development version of std.experimental.ndslice, which provides an N-dimensional equivalent of D's built-in array slicing."

On Monday, 26 September 2016 at 08:57:06 UTC, Joseph Rushton Wakeling wrote:
> On Sunday, 25 September 2016 at 10:45:35 UTC, Ilya Yaroshenko wrote:
>> Thank for the review! I have added notes about Eigen and CBLAS interface example.
>
> One extra suggestion:
>
> "Mir GLAS has native mir.ndslice interface" -> "Mir GLAS has a native mir.ndslice interface"
>
> I would also suggest adding a small note on what `ndslice` is, e.g.
>
> "mir.ndslice is a development version of std.experimental.ndslice, which provides an N-dimensional equivalent of D's built-in array slicing."

Thank you, added

On Saturday, 24 September 2016 at 09:14:38 UTC, Ilya Yaroshenko wrote:
> Please use `dub build ...` and then run report at least 2 times, and choice a better one.

Is this what you mean by your description of the results as e.g. "single precision numbers x2", "double precision numbers x2", etc.?

Might be better, instead of the "x2", to offer a small one-sentence description, e.g. "Each benchmark was run twice for each matrix size, and the better of the two runs was chosen in each case."

On Monday, 26 September 2016 at 09:46:50 UTC, Joseph Rushton Wakeling wrote:
> On Saturday, 24 September 2016 at 09:14:38 UTC, Ilya Yaroshenko wrote:
>> Please use `dub build ...` and then run report at least 2 times, and choice a better one.
>
> Is this what you mean by your description of the results as e.g. "single precision numbers x2", "double precision numbers x2", etc.?
>
> Might be better, instead of the "x2", to offer a small one-sentence description, e.g. "Each benchmark was run twice for each matrix size, and the better of the two runs was chosen in each case."

I mean that for single precision numbers I have 2 charts (normal and normalized).

On Monday, 26 September 2016 at 10:01:44 UTC, Ilya Yaroshenko wrote:
> I mean that for single precision numbers I have 2 charts (normal and normalized).

Ah, OK.  Would still be nice to have a note, though, on how the numbers in the charts are generated, i.e. are they the result of a single run, best of N, average of N ... ?