www.digitalmars.com         C & C++   DMDScript  

digitalmars.D.announce - Mir Blog: Writing efficient numerical code in D

reply Relja Ljubobratovic <ljubobratovic.relja gmail.com> writes:
Hey guys,

We have just published another post on "Writing efficient 
numerical code in D", to Mir's Blog[1]. In this post we are 
focusing on mir.ndslice.algorithm usage in DCV[2], computer 
vision library that has recently joined libmir organization. 
We've had great success in optimizing DCV's algorithms mostly by 
replacing loop-style code with mir.ndslice.algorithm 
multidimensional iteration utilities[3], and we thought it was 
worth blogging about.

Any grammar (or any other type of) fix report is greatly 
appreciated!

Also, we'd like to ask Mike Parker to forward this to reddit if 
it's not too much trouble? :) Maybe only wait few days so we 
could weed out some mistakes from the text.

Cheers,
Relja

[1] 
http://blog.mir.dlang.io/ndslice/algorithm/optimization/2016/12/12/writing-efficient-numerical-code.html
[2] https://github.com/libmir/dcv
[3] https://github.com/libmir/dcv/pull/58
Dec 12 2016
next sibling parent reply Johan Engelen <j j.nl> writes:
On Monday, 12 December 2016 at 21:58:23 UTC, Relja Ljubobratovic 
wrote:

 Hey guys,

 We have just published another post on "Writing efficient 
 numerical code in D", to Mir's Blog[1].


Nice :)

Some comments:
- You can make the numbers in the table more readable by adding 
commas?
For example, 1624687 --> 1,624,687. Probably even better is to 
only show the most significant digits. 1624687 us --> 1.625 s for 
example.
- Add a sentence about the benchmarking set up. " The numbers 
shown are the best times of 10 trials, with LDC commandline 
options `ldc2 -O3 -release`" Something like that. Or add it to 
the configuration table.
- For the "convolution" example, could you elaborate on why the 
Mir-based implementation is so much faster than the C-like 
implementation? From the big table, I see that the speedup for 
DCV's convolution algorithms is not as large as 10x.
- The "threshold" algorithm has a 19x speed up?! (I would remove 
"threshold" from the average improvement calculation. The numbers 
are already awesome, no need to oversell it!)
- Overall, does this mean that DCV became more than twice as 
fast? How does it stack up to other vision libs after this 
massive speed boost?

cheers,
   Johan
Dec 12 2016
next sibling parent Relja Ljubobratovic <ljubobratovic.relja gmail.com> writes:
On Monday, 12 December 2016 at 22:34:32 UTC, Johan Engelen wrote:

 On Monday, 12 December 2016 at 21:58:23 UTC, Relja 
 Ljubobratovic wrote:

 Hey guys,

 We have just published another post on "Writing efficient 
 numerical code in D", to Mir's Blog[1].


 Nice :)


Thanks! :)


 Some comments:
 - You can make the numbers in the table more readable by adding 
 commas?
 For example, 1624687 --> 1,624,687. Probably even better is to 
 only show the most significant digits. 1624687 us --> 1.625 s 
 for example.


Great suggestion - I was focusing on the speedup, so I didn't pay 
attention to run time, but obviously other people will.


 - Add a sentence about the benchmarking set up. " The numbers 
 shown are the best times of 10 trials, with LDC commandline 
 options `ldc2 -O3 -release`" Something like that. Or add it to 
 the configuration table.


Thanks, will do. I agree there's an explanation missing on how 
benchmark is configured.


 - For the "convolution" example, could you elaborate on why the 
 Mir-based implementation is so much faster than the C-like 
 implementation? From the big table, I see that the speedup for 
 DCV's convolution algorithms is not as large as 10x.


DCV's convolution is parallelized, which I believe, could be the 
reason. Ilya is probably the best man to elaborate the speedup vs 
C implementation.


 - The "threshold" algorithm has a 19x speed up?! (I would 
 remove "threshold" from the average improvement calculation. 
 The numbers are already awesome, no need to oversell it!)


I agree. I was thrilled when I saw the comparison at first, but 
to be honest, I'm not sure what's the main reason. Here's the 
commit diff if anyone is interested to explain:
https://github.com/libmir/dcv/pull/58/commits/0c744a7bb7308f07bd45d041c6e88e906c905fcf


 - Overall, does this mean that DCV became more than twice as 
 fast? How does it stack up to other vision libs after this 
 massive speed boost?


I'm not sure. But I'm afraid we're not even close to opencv, 
since its base image processing (and other) algorithms have been 
intensively (manually) optimized with SIMD, but also have almost 
100% coverage with OpenCL (and CUDA from 3.x). Anyhow, I'd like 
not to bother with comparisons like this until we are satisfied 
with DCV - there's much work yet to be done, IMHO.


 cheers,
   Johan


Thanks for the feedback Johan!
Dec 12 2016
prev sibling parent rikki cattermole <rikki cattermole.co.nz> writes:
On 13/12/2016 11:34 AM, Johan Engelen wrote:

 On Monday, 12 December 2016 at 21:58:23 UTC, Relja Ljubobratovic wrote:

 Hey guys,

 We have just published another post on "Writing efficient numerical
 code in D", to Mir's Blog[1].


 Nice :)

 Some comments:
 - You can make the numbers in the table more readable by adding commas?
 For example, 1624687 --> 1,624,687. Probably even better is to only show
 the most significant digits. 1624687 us --> 1.625 s for example.


Oh underscores, then it is valid D!
Dec 12 2016
prev sibling next sibling parent reply Andrei Alexandrescu <SeeWebsiteForEmail erdani.org> writes:
On 12/12/16 4:58 PM, Relja Ljubobratovic wrote:

 Hey guys,

 We have just published another post on "Writing efficient numerical code
 in D", to Mir's Blog[1]. In this post we are focusing on
 mir.ndslice.algorithm usage in DCV[2], computer vision library that has
 recently joined libmir organization. We've had great success in
 optimizing DCV's algorithms mostly by replacing loop-style code with
 mir.ndslice.algorithm multidimensional iteration utilities[3], and we
 thought it was worth blogging about.

 Any grammar (or any other type of) fix report is greatly appreciated!

 Also, we'd like to ask Mike Parker to forward this to reddit if it's not
 too much trouble? :) Maybe only wait few days so we could weed out some
 mistakes from the text.

 Cheers,
 Relja

 [1]
 http://blog.mir.dlang.io/ndslice/algorithm/optimization/2016/12/12/writing-efficient-numerical-code.html

 [2] https://github.com/libmir/dcv
 [3] https://github.com/libmir/dcv/pull/58


This is awesome! Mike, could you please post to reddit about 11 hours 
from now if possible.

I'm about to fall asleep, but I made a quick pass through the piece. 
Here are a few nits:

* "In this post I’d like give" -> "This post gives"

* "module equipped with" -> "module that provides"

* "it’s usage" -> "its usage"

* "Note: It is assumed the reader is already somewhat familiar with 
ndslice package [1]." -> "This article assumes the reader has a cursory 
understanding of the ndslice package. If not, please consult this 
post/doc/webpage/etc before returning here."

* "What does it offer?" -> "What does mir.ndslice.algorithm offer?"

* "its integrated seamlessly" -> "it is integrated seamlessly"

* "by that allowing more natural flow of processing pipeline" -> "which 
makes for elegantly flowing processing pipelines"

* "One of key components to make code based" -> "One key component that 
makes code based"

* "LLVM based D compiler" -> "the LLVM based D compiler"

* "kernels you write" -> "computation kernels written by the end user"

* "else turned on" -> "turned on"

* ", by Johan Engelen" -> remove the comma

* "we’ve been actively refactoring" -> define the team; so far the 
article referred to the author as a single person

* "with mir.ndslice.algorithm equivalent" -> "with mir.ndslice.algorithm 
equivalents"

* align numbers in html tables to the right

* "As you can see, speedups" -> "Speedups"

* "as I’ll show you in this post" -> "as shown below"

* "take a look at PR" -> "take a look at the pull request"

* "Discalmer" -> "Disclaimer"

* "mind DCV project" -> "mind the DCV project"

* "showing power" -> "showing the power"

* "dive into implementation" -> "dive into the implementation"

* "without extensive optimization techniques applied" -> "without 
extensive optimizations"

* "In future we’ll focus" -> "A future post will focus"

* "first we’ll take a look" -> "first let's take a look"

* "basic principle how it can efficiently replace loop-based code" -> 
"basic principle of replacing loop-based code with pipelines efficiently"

* "… This code" -> "This code"

* "apply given kernel" -> "apply the given kernel"

* "To bring this example more down to earth" -> "To make the example 
more concrete"

* "As you can see, we are replacing" -> "The pipeline version replaces"

* "with few magic calls" -> "with a few magic calls"

* "And as you may have noticed, at" -> "At"

* "I get ~10x speedup" -> there's been a bunch of "we" and now getting 
back to "I" is surprising. You may want to make a pass making sure the 
use of person is consistent.

* "that variant written" -> "that the variant written"

* "syntax, and are" -> "syntax and are"

* "this is most basic binarization" -> "this is a most basic binarization"

* "Special part to care about, related to lockstep, is that byElement 
call on each input Slice" -> "The call to byElement inside lockstep is 
worth special attention"

* "We can notice right away there’s a change" -> "First off, there's a 
change"

* "requires this so" -> "assumes that is the case"

* "We can also notice byElement" -> "Notably, byElement"

* "examine performance differences" -> "compare the impact of these 
changes on performance"

* "On my machine" -> "On a Core i7 laptop with xyz GB or RAM and xyz 
other relevant parameters..."

* "result is following" -> "the results are as follows"

* "with very little effort using" -> "with very little effort by using"

* "and it’s submodules" -> "and its submodules"

* "coming for some numerical computing" -> "having an interest in 
numerical computing"

* "would have to be introduced to ndslice" -> "should take a close look 
at ndslice"

I suggest making two passes through the piece: (a) one pass should 
reduce fillers that are abundant ("as you can see", "as I'll show you", 
"as said in the docs"...) A few are fine (perhaps 2-3 in the whole 
article. Then (b) one pass looking for unified use of "I" (most likely: 
the author, use sparingly), "you" (the reader, use sparingly), and "we" 
(most likely: the author and reader looking together at something).

The results are terrific, congrats Ilya and Relja!


Andrei
Dec 12 2016
next sibling parent Relja Ljubobratovic <ljubobratovic.relja gmail.com> writes:
On Tuesday, 13 December 2016 at 05:27:01 UTC, Andrei Alexandrescu 
wrote:

 I'm about to fall asleep, but I made a quick pass through the 
 piece. Here are a few nits:


Thanks for an extensive feedback, Andrei! I've followed up 99% of 
fixes given here[1], but unfortunately I'll be unavailable for 
the rest of the day, so I'm passing the torch (or the hot potato) 
to Ilya.


 * "we’ve been actively refactoring" -> define the team; so far 
 the article referred to the author as a single person


I've done the coding, but couldn't have done it without Ilya's 
consulting, so I put the 'Mir team' there. Hope thats ok.


 * "On my machine" -> "On a Core i7 laptop with xyz GB or RAM 
 and xyz other relevant parameters..."


There are specs shown at the top of the post, so I thought it 
would be best just to remove the "On my machine part". Again, I 
hope that's alright.


 I suggest making two passes through the piece: (a) one pass 
 should reduce fillers that are abundant ("as you can see", "as 
 I'll show you", "as said in the docs"...) A few are fine 
 (perhaps 2-3 in the whole article. Then (b) one pass looking 
 for unified use of "I" (most likely: the author, use 
 sparingly), "you" (the reader, use sparingly), and "we" (most 
 likely: the author and reader looking together at something).


As I've demonstrated, I'm quite unexperienced with writing. :) 
So, thanks for the feedback, I've learned a lot! If I catch some 
time during the day, I'll be sure to take another glance at it!


 The results are terrific, congrats Ilya and Relja!


 Andrei


Thanks a lot Andrei!
Relja

[1] https://github.com/libmir/blog/pull/32
Dec 12 2016
prev sibling parent reply Mike Parker <aldacron gmail.com> writes:
On Tuesday, 13 December 2016 at 05:27:01 UTC, Andrei Alexandrescu 
wrote:


 This is awesome! Mike, could you please post to reddit about 11 
 hours from now if possible.


Yes, I intend to post it to both reddit and FB, but it will have 
to be a couple of hours earlier than that (Korea is UTC+9).
Dec 13 2016
parent reply Mike Parker <aldacron gmail.com> writes:
On Tuesday, 13 December 2016 at 09:58:09 UTC, Mike Parker wrote:

 On Tuesday, 13 December 2016 at 05:27:01 UTC, Andrei 
 Alexandrescu wrote:


 This is awesome! Mike, could you please post to reddit about 
 11 hours from now if possible.


 Yes, I intend to post it to both reddit and FB, but it will 
 have to be a couple of hours earlier than that (Korea is UTC+9).


https://www.reddit.com/r/programming/comments/5i42vc/writing_efficient_numerical_code_in_d/
Dec 13 2016
parent Relja Ljubobratovic <ljubobratovic.relja gmail.com> writes:
On Tuesday, 13 December 2016 at 15:31:00 UTC, Mike Parker wrote:
https://www.reddit.com/r/programming/comments/5i42vc/writing_efficient_numerical_code_in_d/

Thank you, Mike!
Dec 13 2016
prev sibling next sibling parent =?UTF-8?B?Tm9yZGzDtnc=?= <per.nordlow gmail.com> writes:
On Monday, 12 December 2016 at 21:58:23 UTC, Relja Ljubobratovic 
wrote:

 http://blog.mir.dlang.io/ndslice/algorithm/optimization/2016/12/12/writing-efficient-numerical-code.html


Awesome!
Dec 13 2016
prev sibling parent Mark <smarksc gmail.com> writes:
On Monday, 12 December 2016 at 21:58:23 UTC, Relja Ljubobratovic 
wrote:

 [1] 
 http://blog.mir.dlang.io/ndslice/algorithm/optimization/2016/12/12/writing-efficient-numerical-code.html
 [2] https://github.com/libmir/dcv
 [3] https://github.com/libmir/dcv/pull/58



Very impressive work.
Dec 13 2016