A look at things from a Bird's eye view

The ramblings of a realistic-optimist, recounting the events and mishaps of technology driven life. var _gaq = _gaq || []; _gaq.push(['_setAccount', 'UA-31265622-1']); _gaq.push(['_trackPageview']); (function() { var ga = document.createElement('script'); ga.type = 'text/javascript'; ga.async = true; ga.src = ('https:' == document.location.protocol ? 'https://ssl' : 'http://www') + '.google-analytics.com/ga.js'; var s = document.getElementsByTagName('script')[0]; s.parentNode.insertBefore(ga, s); })();

[rfbird-blog]

OS X Maverick

Yesterday, the world saw the release of OS X Mavericks, but what sort of drop in productivity did the world see today?

I can't help but wonder how the release of OS X Mavericks effected the global productivity as a whole. Now of course if you're a sensible corporation you won't be rolling out Mavericks today to all end users, and certainly not in office hours, but for those lucky enough to managed their own OS X system, how many of us took the time out today to install it?

With an install time of around an hour ... and add in _some _extra time for post-boot configuration (more on that some other time perhaps), and that could lead to a fair amount of down time globally. With the millions of macs out there, I don't dare estate the worst-case time "cost" of this, but I would say in the $millions....

[rfbird-blog]

Vim Find/Replace

Recently I found myself wanting to comment out all lines of code that started with a keyword. (In my case, "MPI")

Being a vim advocate, I decided I could do this automagically in one line and came up with something like this:

%s/^\(\( \)\+\)\([^\/]\+\(WORD\)\)/\1\/\/\3/gc

Of course it is easy to find all matching lines and append the comment characters, but my first version ended up commenting out some already commented out lines (not a problem exactly, but not great!). A few small changes later and you can match line start, then NOT comments, then the string.

Not bad for a moments work, and work well it did :)

NB: The first bit could of course be tidied up to use inbuilt regex white-space matching, but in our code base I knew lines would start with some multiple of 4 spaces, so for my use this was sufficient!

[rfbird-blog]

Automatically detecting website changes

Recently I wanted to automatically get a notification when a new rental property was added to Rightmove. I began by using their built in notifications which initially worked quite well, but I still found myself manually checking the site as those updates only came every 24 hours.

The solution? As with many web-based automation tasks, I turned to the much loved ifttt.com. However initially there is no way to detect changes. To do this you have to team it with a third party site; I chose to use page2rss.com.

Page2RSS does what you might expect from the name, and makes a feed out of any site. This can then be used with IFTTT's Feed action to notify you when it changes.

Sadly this still isn't quite real time, and I suspect with a different choice of third party site that could be improved upon (although how happy that would make rightmove, I don't know..). But this served my needs well, all without having to make any additional accounts

[rfbird-blog]

Latex / Paper Author List

For any one struggling to format with the author list of a paper because there are too many institutions, the following trick may help.

Whilst it's not too special, I found it hard to find anything concrete on-line when I looked for a way to make the ieeetrans template play nice. If you are able to use the conference format (in order to use \and to separate columns) the something like the following may work well:

\author{ \IEEEauthorblockN{ \makebox[.99\textwidth]{ Author~One\IEEEauthorrefmark{1}, Author~Two\IEEEauthorrefmark{2}, Author~Three\IEEEauthorrefmark{3}, } } \and \IEEEauthorblockA{ Company One\IEEEauthorrefmark{1}\\ Address Line 1\\ Address Line 2\\ } \and \IEEEauthorblockA{ Company Two\IEEEauthorrefmark{2}\\ Address Line 1\\ Address Line 2\\ } \and \IEEEauthorblockA{ Company Three\IEEEauthorrefmark{3}\\ Address Line 1\\ Address Line 2\\ } }

The idea is to make the author list wide enough to fill the page, so all other elements are new-lined.

[rfbird-blog]

Bus Checker UK

Just a quick fix for any one who uses Bus Checker UK, and like me was only receiving updates when connected to wifi (and refusing to use my data plan), here is the solution:

In settings tick: Disable internet warnings. (towards the bottom)

This presumably stops it checking to see if it thinks you're online..and just goes for it. Interestingly this fixed the problem, so even if I untick this option it is now happy to use my dataplan...

Much more useful than Please check your internet connection then try again are a screen that goes round in a loop saying try again

[rfbird-blog]

`strace`

It took me far too long as a linux user before I realised the power of strace, so I though it was worth a mention!

It's a neat little tool that lets you see the system calls made a program, particularly useful if it keeps crashing and you want to see why (or rather, what it was doing at the time). It's one of the many trace systems available in unix, but if you suspect that something is failing because of a system call (like it not being able to get at the correct file - I'm looking at you google-chrome.) then strace is the tool for the job!

[rfbird-blog]

OpenCL Spinlock

After some digging around on google, and some umming and arrring, I finally got a spinlock (semaphore) working in OpenCL.

Whilst it's not pretty, nor is it particularly portable (as the compiler may chose to break / "optimise" it at any time), here it is:

inline int getSemaphore(__global int *semaphore); inline int releaseSemaphore(__global int *semaphore); inline int getSemaphore(__global int *semaphore) { // This previous method could actually cause a deadlock on GPU like devices that move in lockstep. //int occupied = atom_xchg(semaphore, 1); //while(occupied > 0) //{ return atom_xchg(semaphore, 1); //} //return 1; } inline int releaseSemaphore(__global int *semaphore) { int prevVal = atom_xchg(semaphore, 0); return 1; }

In kernel:

int waiting = 1; while (waiting) { if (!getSemaphore(sem)) { // Do Critical work waiting = 0; }

There are a few things I didn't see when looking for this on line, the main one being that you want to set your semaphore variable as a kernel argument in order to get it in global memory.

Remember that this should generally be avoided, but if you want a quick hack like me to validate some results, this could be handy :)

EDIT: make sure sem comes in with a sensible value .. like 0!

[rfbird-blog]

Pi

I have recently been reading "Alex's Adventures in Numberland" by Alex Bellos, and whilst I've been enjoying it all, one chapter in particular really stood out out for me, and it was all about Pi!

The highlight of which, was an amusing way to remember pie to good few digits, by simply remembering the following phrase:

How I need a drink, alcoholic of course.

How does this help, simply count the number of characters in each word.

3 . 1 4 1 5 9 2 6

Throw in a decimal point and there you have it, Pi. (to an accuracy good enough to calculate the circumference of the earth to within centimetres, no less.)

This also opened my eyes to a whole new side of literature, and even philosophy. So called piems, are poems whose word length match the digits of pie, similar to that which we see above. Clever and amusing. Who doesn't like literature could often use more maths?

[rfbird-blog]

Mac Hostname

When opening a terminal under OSX, my hostname (as returned by `hostname`) was showing as my "unknown" followed by my MAC address (Hint: Media Access Control, nothing to do with it being a Mac computer!). Curious behaviour that had been made apparent to me by my beloved Zsh showing my hostname incorrectly at the prompt after switching to the nice "aussiegeek" theme. Obviously this had to be fixed! After a little look around, and poking around with some commands I eventually found what I had been looking for: sudo scutil --set HostName Familiar to any BSD user, this allows you to set the HostName for your mac, which it not the same as your computer name (as set in sharing) as I first suspected!

[rfbird-blog]

Replacing Branching

Today, whilst thinking about how to minimise branching in some code, I stumbled back across something that I hadn't thought about in far too long! A neat way to remove from branching by storing values in a hash/array and just accessing it directly based on values, no if statements needed. It had been so long in fact it had fallen from the fronts of my mine, to a place where I had to basically rediscover it when I needed it again! Firstly, we all know removing branching is important where you can do it for free! Helps with vectorisation, removes the possibility of the branch predictor making a mistake and can buy you back some cycles! So on my naive first pass of writing some code, I foolishly did something like this: if (condition) then if (otherCondition) i = someTrueVariable else if (someOtherconditon) i = someOtherTrueVariable else if (someOtherconditon) i = someFalseVariable else if (someOtherconditon) i = someOtherFalseVariable end while (someConditoon) do ... i = someNewValue end # Now what happens if I want to write the changed value of i back? Now when trying to write i back, what do you have to do? Re-do the logic? Normally you could just use i as a pointer, but on this occasion the language I was using (Ruby) didn't support that! So what is a boy to do... Of course! Why not re-write it as: array[condition][someOtherCondition] = [someTrueValue/someOtherTrueValue...] depending on the values of the conditions, that way you can just do `array[condition][someOtherCondition] ` in place of i in the loop. Not necessarily the best way to do things all the time, but it's good to keep neat tricks like this at the top of your brain. NB: Yes, I am assuming the more expensive lookup is worth it :)

[rfbird-blog]

Assembly Inspection

When writing some fairly performance critical code earlier, as happens more often than not now I've started working in HPC, I decided it was probably wise to see what assembly was actually coming out. The immediate thing to try, is to get the compiler to write out the assembly it's going to output, right? Something like adding a `-S` to the compile line used by gcc? A process I've done successfully before, but only on simple code. Well you would think that's still the thing to do for all codes, but it turns out that what it gives you for larger codes really isn't that visually pleasing at all! And needless to say, if you try to read it, it's just going to hurt your eyes! So after some searching around online, I did something that seemed odd at the time, but after the success I wouldn't be surprised if it happens in a large percent of the cases! I simply compiled the binary, and then decompiled it again! This can be done by a variety of specialist tools, and to do this I used objdump (that comes packaged with binutils). Objdump was able to give me much, much, nicer results that were pretty much readable! (after all assembly is an acquired taste..) Never again will I trust gcc to output Assembly! From this day forward (or, at least until the gcc option gets better), its disassembly for me.

[rfbird-blog]

Fortran Bounds Checking

Admittedly, this is more of a note to self than anything! But I thought I would share a few things I didn't realise about Fortran. I found these out first time I seriously tried to use and abuse a fair size Fortran code, and whilst they're simple they might be of use to some one! Some where between playing around with a Fortran Array, moving some data around, getting it ready to be passed into C (and then back again), I managed to fall out the end of the array, and get a not so helpful seg fault. Now Fortran being the dotty old man that he is, won't exactly offer up too much useful information about this without being provoked. Luckily, as with many things it's easy enough to do when you know how! It's simply a matter of changing a compile time flag to add support for bounds checking, using: -fbounds-check or -check bounds This of course will have some performance penalties, as it will have to spend a few operations each array access making sure you're still in bounds, but is perfect for situations like the one I found myself in! Some more digging around this subject unveiled another feature I miss from the sluggish-but-cuddly nature of high level languages, a decent functional equivalent to myArray.length. Enter **UBOUND**. Used to print out the upper bound of an array, rather useful and saves making those pesky extra variables to track such things. (Although one does wonder if they should almost always be tracked in constants any way?) As I said before, simple things but its good to know them!

[rfbird-blog]

More a joke than a quote but: I never wanted to believe that my Dad was stealing from his job as a road worker. But when I got home, all the signs were there.

(who doesn't love a cheeky joke that you can come out with should you be pressed for one!)

[rfbird-blog]

Ram Failure #Resolved

All be-it a little overdue, I resolved my ram failure. Restoring my PC to full stability.

The answer was to swap to DDR3, faster, cheaper and easy to get hold of larger single sticks. Swapping to 2 sticks of DDR3 seems to have done the job nicely! And more interestingly removed my issue with sometimes failing to cold boot...good new all round! Now the question is, is it worth chasing the life time warranty to on the more questionable sticks? (And what to do with the working stuff! As it doesn't work in my ECC box)

[rfbird-blog]

Ram Failure

For far too long now, my system stability has been a tad .. questionable. A fault to which I attribute the instability of my memory controller. This motherboard, being ~4 years old, has put a good fight but started to run with 4 sticks in - at least not with any degree of stability. My initial work around was to under-clock the ram to ease the load on the memory controller, but this only worked for so long. After a while, either due to ram failure or the memory controller dying further (who knows!) I could only boot with 2 sticks in. 

What ever could be the problem with that? Running with half the ram you're used to is inideal for start. Suddenly having grindingly slow performance where you never had it before. Not only do you have that but that nagging thought in your head that your memory controller may topple over at any time? Or what killed your lifetime guaranteed ram in the first place..especially when it was running under-clocked so not likely to be heat.. :(

What to do next? Who knows.

[rfbird-blog]

Rails for Zombies

So this weekend has seen me make use of the free weekend over at code school, and led to me getting my ruby on!

The rails for zombies series is great, if not for its amusing pun on teaching rails from the ground up (cheeky eh?), nor for it's semi amusing zombie puns but for the actual ruby content. Overall it has been great (so far, anyway.)! A little basic for any one who has done much ruby, and the error descriptions aren't great for those who haven't, but a nice reminder and/or chance to practise some rails stuff. It does a good job of covering most bases, and was a great lead onto the video on rails best practices.

Not only is it great that it was free for the weekend, but also they let you download the resources and for completing courses let you receive rewards such as free pod casts from other great place, or discounts on their own products .. as much as I hate to admit it, they may well end up getting some money out of me somewhere! (and I'll be taking them up on their free month over a railscasts!)

[rfbird-blog]

A round up of some vim things #2

As all vim users will know, you pick up little keystrokes that are simply amazing at the time - swearing to yourself that you will remember them, and they inevitably fade away. Here is a little round up of some of the more interesting ones that have been good enough to make a note of in the last few months! (vim ninjas should know all of these already)

Ctrl-o lets you pop out of insert mode to issue a single command, then puts you back in insert mode. Give if you just want to quickly move around the screen or do some quick tidying up!

g-j lets you move by screen line instead of by text line, great for when those long lines start getting wrapped (not so much use for the Fortran developers out there)  

* with a word selected lets you jump to the next instance of that word, great for quickly hoping around text after doing an initial search

When in insert mode Ctrl-n pops up an autocomplete box that many vim users underuse, of course there are plug-ins to extend this functionality but useful to avoid retyping those long frequently used words

~ over a character changes it's case. Surprisingly useful.

% in some code skips to the matching closing curly brace, well, works with most bracket types actually

Ctrl-x decrements a number (Ctrl-a to increments)

Thats all for now, just had to get them those few gems out there.