Rpg's Corner

Installing Ubuntu fonts in Fedora

2011-11-25T11:25:00.001+05:30

I like the new Ubuntu fonts. Here's how to get them working in Fedora.

Sage: A first experience

2011-09-12T08:41:00.003+05:30

I had heard about sage a while ago. I had figured that I'd give it a try if I needed it over Mathematica and friends. That need arose a short while ago. And I tried using sage for some hybrid symbolic/numerical computation. So, how was it? Read on.

a) The online sage notebook experience is simply awesome. A very good idea. May be, somebody should do an Android/iOS app for this. It could be quite useful.

b) The symbolic manipulation capabilities are decent. Mathematica's are demonstrably better at this time, but the difference I saw was that the expressions were simplified better in Mathematica. This is hardly a exhaustive review, but it wasn't a deal breaker in any way.

c) I tried evaluating the error function for a complex argument. Sage went poof. The expression was complicated, so I didn't wan't to get my hands dirty with it. But the milestone is set to be the next version, so it's not that bad.

Bottom line, Sage is a very good step in a very good direction. But it will be a while before it crosses the reliability threshold. My guess is 3 years with decent amount of funding.

C++11 : a great addition

2011-08-25T09:28:00.003+05:30

I have always liked a certain subset of C++. I never cared much for OOP, but I have always liked functional paradigm. The C++ standard has just undergone a major revision and it has now added support for lambda functions.

Now, C++11 is almost my favorite language. It offers precise control over memory, instructions and data structures while almost letting you use functional composition and functional abstraction and offers type inference to boot.

There a few brittle points, and the new STL still seems to lack a version/variant of reduce operation.

If only C++ had a garbage collected and interpreted subset ass well :)

Writing a ray tracer

2011-07-19T23:20:00.005+05:30

I have written earlier about the simple rasterizer I wrote some time ago. Well, I have been feeling somewhat more adventurous of late, and I decided to write a ray tracer. The good news is that it is mostly working. The bad news is that making pretty pictures is more work than fun and messing around with the internals is lots more fun, so nothing worth posting here as yet.

Things I learnt,

a) You might need to link with -lgomp with gcc, in addition to -fopenmp, if your code was initially compiled with -c option.

b) Parallelizing code can make you clear up your understanding of your own code by a lot.

c) OpenMP rocks. But GPUs rock even more.

It needs some secondary rays goodness and some more work for it's acceleration structure. I think I'll do the latter first.

Text rendering in Firefox 5

2011-07-19T23:13:00.003+05:30

This was a nice surprise. In FF 3.6, at least on linux, text rendering left a lot to be desired, especially as you zoomed in or out. With Firefox 5, rendering quality is positively cute. With awesome AA. I just keep liking FF more and more.

At this point, I only miss two things from chrome. Auto translate and H.264 video playback with HTML5.

Any suggestions?

Upgraded to Ubuntu Unity, and I like it

2011-06-26T21:27:00.008+05:30

I am running Natty with Unity. I am kinda smitten by the new UI Ubuntu has come up with. With gpu acceleration, this thing really flies. These shortcuts are really useful.

I have upgraded to Firefox 5, and frankly, I am not missing chromium, which is quite surprising, considering my experience with Firefox in the past. A few addons have helped, like omnibar, downthemall, and torbutton helped swing the decision. Firefox is sitting behind polipo, which is good, as I can filter all the nasty ad serving URLs. It helps that I need polipo for tor anyway. I highly recommend tor for anyone who cares about his privacy and free speech online.

The default nouveau driver was a bit of a pain (I have gt330m), as it kept hanging. Got rid of it and with the new nvidia driver, there is no problem at all.

There are a few downsides though. The firewall situation needs to be looked at. I had to set selinux in permissive mode, else it wouldn't let me run the nvidia driver.

Got my rasterizer working

2011-05-21T15:06:00.004+05:30

Long time, no see huh :|

Anyways, the big news is that I have crossed my rite of passage. See, most people would write a simple ray tracer when they are fooling around with graphics. But I for some reason, felt that writing a rasterizer would be so much cooler. The thing is, this technique is so looked down upon that even the word rasterize not in default Firefox dictionary. Talk about being an outcast. :P

But, I managed to write it in the end. So here's the obligatory screenshot.

It has vertex shaders, pixel shaders, bilinear filtering of rectangular textures (but no mip mapping) and perspective correct interpolation. It still has a few rough edges and sharp corners though. The rasterization process is anything but robust and for some reason, I decided to defer implementing the z buffer itself. Yeah, I am going to hell for this...

There's a very illuminating story behind the history of z buffer itself. It was discovered by Edward Catmull, who among other things discovered texturing. In a review of 10 visible surface determination algorithms (PDF) by the legendary Ivan Sutherland, it was characterized as ridiculously expensive. And yet, here it is, having virtually displaced everything else.

Bottom line, amenability to hardware implementation and Moore's law put together can change the cost of an algorithm.

Maverick's font

2010-10-12T10:00:00.001+05:30

Just upgraded to Ubuntu Maverick on my work machine. The new font is pretty cool.

Perhaps the best way to paste code in blogger

2010-10-01T22:28:00.003+05:30

Github Gists.

Here's a simple example

Syntax highlighting for a gazillion langugaes, forkable, editable, and revisions to boot. What's not to like compared to methods like these?

Honestly, I think any dev worth his salt should switch to this method or it's equivalent.

Testing out Github Pages

2010-09-05T15:08:00.002+05:30

Github's really cool. They have some awesome services on top of the already awesome git based version control. I just made a test page to checkout the possibility of hosting a blog over there.

Link.

Another compiler in haskell series

2010-07-29T21:20:00.002+05:30

Here's another compiler in Haskell series. This one's based on Crenshaw's famous series, and is quite a a bit simpler than the prior one. But it takes quite a different route as well.

Compiler construction series: all of it in Haskell

2010-07-23T00:43:00.002+05:30

Seems pretty nice to me.

Dear IITB

2010-07-17T13:07:00.000+05:30

I'll miss you.

GCC: Now you are going to unroll the loop or should I?

2010-06-15T21:50:00.003+05:30

A codelet to calculate the log (to base 2) of a number which is known to be a power of 2.

unsigned int log2_pow2(unsigned int v) // 32-bit value to find the log2 of

{

static const unsigned int b[] = {0xAAAAAAAA, 0xCCCCCCCC, 0xF0F0F0F0,

0xFF00FF00, 0xFFFF0000};

int i;

for (i = 4; i > 0; i--) // unroll for speed...

{

r |= ((v & b[i]) != 0) <<>

}

return r;

}

Let's see what do we get with GCC 4.4 at O3,

400550: 31 c0 xor %eax,%eax 400552: f7 c7 aa aa aa aa test $0xaaaaaaaa,%edi 400558: 0f 95 c0 setne %al 40055b: 31 d2 xor %edx,%edx 40055d: f7 c7 00 00 ff ff test $0xffff0000,%edi 400563: 0f 95 c2 setne %dl 400566: 31 c9 xor %ecx,%ecx 400568: c1 e2 04 shl $0x4,%edx 40056b: f7 c7 00 ff 00 ff test $0xff00ff00,%edi 400571: 0f 95 c1 setne %cl 400574: c1 e1 03 shl $0x3,%ecx 400577: 09 ca or %ecx,%edx 400579: 09 c2 or %eax,%edx 40057b: 31 c0 xor %eax,%eax 40057d: f7 c7 f0 f0 f0 f0 test $0xf0f0f0f0,%edi 400583: 0f 95 c0 setne %al 400586: c1 e0 02 shl $0x2,%eax 400589: 09 c2 or %eax,%edx 40058b: 31 c0 xor %eax,%eax 40058d: f7 c7 cc cc cc cc test $0xcccccccc,%edi 400593: 0f 95 c0 setne %al 400596: 01 c0 add %eax,%eax 400598: 09 d0 or %edx,%eax 40059a: c3 retq 40059b: 0f 1f 44 00 00 nopl 0x0(%rax,%rax,1)

So GCC unrolled the code for us. Nice. But how about this bit of assistance?

unsigned int log2_pow2_2(unsigned int v) { register unsigned int r = (v & 0xAAAAAAAA) != 0; r |= ((v & 0xFFFF0000) != 0) <<>

Shouldn't make much difference, right? Let's ask the GCC gods for their verdict.

4005a0: 89 f8 mov %edi,%eax 4005a2: 89 fa mov %edi,%edx 4005a4: 66 31 c0 xor %ax,%ax 4005a7: 83 f8 01 cmp $0x1,%eax 4005aa: 19 c0 sbb %eax,%eax 4005ac: 81 e2 00 ff 00 ff and $0xff00ff00,%edx 4005b2: f7 d0 not %eax 4005b4: 83 e0 10 and $0x10,%eax 4005b7: 83 fa 01 cmp $0x1,%edx 4005ba: 89 fa mov %edi,%edx 4005bc: 19 f6 sbb %esi,%esi 4005be: 81 e2 f0 f0 f0 f0 and $0xf0f0f0f0,%edx 4005c4: f7 d6 not %esi 4005c6: 83 e6 08 and $0x8,%esi 4005c9: 83 fa 01 cmp $0x1,%edx 4005cc: 89 fa mov %edi,%edx 4005ce: 19 c9 sbb %ecx,%ecx 4005d0: 81 e2 cc cc cc cc and $0xcccccccc,%edx 4005d6: f7 d1 not %ecx 4005d8: 83 e1 04 and $0x4,%ecx 4005db: 83 fa 01 cmp $0x1,%edx 4005de: 19 d2 sbb %edx,%edx 4005e0: f7 d2 not %edx 4005e2: 83 e2 02 and $0x2,%edx 4005e5: 81 e7 aa aa aa aa and $0xaaaaaaaa,%edi 4005eb: 40 0f 95 c7 setne %dil 4005ef: 40 0f b6 ff movzbl %dil,%edi 4005f3: 09 f8 or %edi,%eax 4005f5: 09 f0 or %esi,%eax 4005f7: 09 c8 or %ecx,%eax 4005f9: 09 d0 or %edx,%eax 4005fb: c3 retq 4005fc: 0f 1f 40 00 nopl 0x0(%rax)

Gee, that was unexpected. The code size grew. And is probably slower because of it. I do not claim to know much of non-SSEx assembly, but the manual unrolling seems to produce worse code.

Who knew? :|

Quote of the day.

2010-06-03T12:16:00.001+05:30

"Cache coherence certainly can scale -- for codes designed to scale on it. This generally means codes designed to not use coherence very much"

-Greg Pfister

link

Emacs and unicode

2010-05-26T23:28:00.002+05:30

Of late, my Emacs had been throwing up at the quote marks in the compile window. Here's how to fix it.

BTW, I didn't need the clipboard-coding-system bit. My emacs works fine without it.

Eclipse with Arduino

2010-05-07T15:02:00.001+05:30

This is a very good guide.

Indian electronic voting machines are vulnerable to fraud

2010-05-04T12:46:00.003+05:30

Full paper

Video

Broadly speaking, I see three problems.

1) Belief in the magic of security through obscurity.

2) Near-total absence of understanding of electronic technology and it's security related aspects.

3) Confusing the comfort of Powerpoint presentations for a reliable and robust defence against hardcore cracking by motivated and well-resourced individuals.

Web 2.0 Calc

2010-04-30T14:38:00.003+05:30

A really nice web app. Very useful at times when you don't have a physical calculator at hand and the problem is too small to fire up an ipython/scipy console.

Build a compiler series

2010-04-28T13:15:00.004+05:30

A great 15 part series here. Definitely worth bookmarking.

The approach taken is an interesting one. Ruthlessly eliminate features. Relentlessly punt to user. Just focus on building a working, complete - and functionally correct - baseline. By functionally correct, I mean it should correctly - and in all cases completely - raise the red flag whenever it encounters anything illegal according the syntax or semantics.

Perhaps it deserves to be given a name. How about "The Crenshaw approach"?

Apple's latest

2010-04-10T15:57:00.001+05:30

Have a look.

Great job, Apple. HTML5 can't land here fast enough. I wonder why then, they seem to be interested in HTML5 so much by their participation in WHATWG.

An eye opener...

2010-04-02T20:52:00.001+05:30

Have a look here.

"Indeed, the government predicts the jobless rate will average 9.8 percent next year and 8.4 percent in 2012 before falling to 5 percent in 2016."

~3 years to reduce unemployment by ~1%. Really eye opening.

Python reference management

2010-03-28T19:54:00.005+05:30

I ran into a crazy Python reference management behavior today. I wanted to do

x=y=[None for i in xrange(10)]

and then add elements to both x and y separately. Basically, wanted to avoid creating an empty list and then append stuff to each. No particular reason, just for the heck of it.

Turns out there is some kind of weird reference management policy in Python, which it so happens that I don't have the time to dive into right now. What this does is that it couples x to y, in such a way that changing elements of one affects the other.

So, the solution is to change the above line into

x=[None for i in xrange(10)]

y=[None for i in xrange(10)]

x86-64 unsupported.

2010-03-22T19:36:00.002+05:30

Sorry we don't support the X86_64 build environment (we never use it internally). An AMD employee, who presumably works on open64, in Dec 2009.

I am not sure whether to laugh or cry.

New/old style classes in Python

2010-03-21T21:15:00.001+05:30

I just got bitten by the new/old style classes in Python. :P

Here's a reasonably good description/discussion/summary/...

GPU acceleration in IE 9

2010-03-18T23:01:00.004+05:30

This is definitely something exciting.

It's nice to see GPU acceleration coming to consumer apps other than games. :) In the next couple of years, we can expect pervasive support for fast texturing and drawing in web browsers. Even though it leverages just the classic "drawing lines and triangles" bit of gpu's and not their emerging compute side, with imminent merger of cpu's and gpu's, I think gpu's might have found their killer app for the non-gaming folks. With on-die gpu's and faster JS, I think web apps are going to get quite cool, especially in the hands of competent web developers.

At any rate, it seems doubtful that web developers will stress even the lamest gpu's of today. They are just too ridiculously fast for these kind of workloads. And JS is going to be dog slow compared to C++ for ages to come. Unless, of course, somebody hires Mike Pall to speed up their JS engines, at which point all bets are off. :)

More generally, better web standards support from IE is great to see.

From the hardware accelerated video decoding point of view, I think Firefox going to be a loser here. H.264 seems to have won the web video codec war and so far Mozilla's policy has been to avoid it like plague.

Overall, these are exciting times for all those who browse the web, as your vehicle is about to get a serious overhaul.

I have only one question with that article. Since when has Apple solicited any kind of consumer feedback in the design phase for any of their products?

Swiftfox

2010-03-08T12:36:00.004+05:30

Swiftfox, an optimized build of Firefox for Linux is pretty good with speed. I am running the 64 bit build of 3.6 and it's cpu and mem usage seems pretty good to me. They have prebuilt binaries, but only for Ubuntu.

Check it out.

Porting numpy to Python 3 : STATUS

2010-02-28T21:28:00.001+05:30

This is good news.

Chromium begins to suck :(

2010-02-19T15:53:00.002+05:30

I had no idea it will come so close to my praise for Chromium over Firefox. But it has happened.

I use the Chromium daily PPA for Ubuntu. And for a few days now, Chromium has been showing abnormally high CPU usage. It basically locks up a core, and when you have only a dual core machine, things get bad and ugly. The only saving grace is that I can shutdown chromium rather painlessly as usually, only one process is responsible for the CPU lockdown,.

My cribs against Firefox still stand, but a quirky browser is better than a useless browser.

Back to Firefox then, :|

Emacs theme based customization

2010-02-08T23:58:00.002+05:30

Many softwares today ofter theme support to increase aesthetic attractiveness. Here's a good way to get it done in my favorite editor, emacs.

Chrome gobbling up Firefox users

2010-02-07T18:56:00.004+05:30

This is an interesting bit of news.

Overall, Firefox seems to have stalled while Chrome is growing. Amongst Ars Technica users, IE and Firefox are losing share to Safari and Chrome. A 14% share amongst a tech-savvy community within a year of it's public unveiling is good by any standards. At this rate, on this website atleast, by the end of summer, Chrome would be in the number 2 position.

While IE's decay is hardly unsurprising, the loss of mind share amongst a savvy community does not bode well for Firefox. I have been a chrome convert for a long time now. I was also unaware of the general stall in Firefox's growth

Speaking about overall internet aware community is pretty hard. After all some 20% of IE users are still using IE6. :P Why did I switch over to Chrome? Well actually I use the non trademarked, third party provided builds, so it is probably more accurate to call it Chromium.

Firefox has always been pretty good with it's rendering accuracy. FF2 and 3 were pretty good in this regard. However, where they really messed up was Javascript speed and tab isolation. All versions of FF so far have been a single process design and it used to lock up with many tabs/heavy javascript often. Many times, it would be just a single bad tab taking down the entire browser. The only way out was to kill the entire browser itself, losing your entire session.

The memory usage was another problem. I use my browsers for long periods of time often opening lots of tabs and the closing them and this usage cycle repeats often. All through this, FF's memory consumption keeps increasing due to memory fragmentation. I have seen a single Gmail tab use up more than 300 MB of memory after a long period of usage. In FF 3 and earlier, it was pretty common after just a couple of such usage cycles to have a single tab use up about 200 MB memory.

With these problems at hand, I switched over to Chromium for it's speed and scalability under heavy loads. For the most part, the rendering was fine but it did have occasional rendering SNAFU's. Within a couple of months they were sorted out too. It's not like Chromium never jams up. It does. But the failure rate is easily an order of magnitude less that FF3. FF3.5 has, for the most part, has competitive JS performance and the lock up issues occur much less frequently. I haven't looked at FF3.6 yet.

With those two bugbears off it's back, what is holding FF back for me? In no particular order, they are as follows.

First, it wastes a lot of window space, which Chromium reuses for actual web pages. The rest of browser window is unused 99% of the time, why waste screen real estate for the 1% case.

Second, it does not, yet, have flexible tab manipulation. By that, I mean easy (drag-and-drop) shuffling of tabs across browser windows. May be it is already there in FF and I don't know how to use it. For me, it is the browser's equivalent of multiple desktops in Linux world.

Third, the first option when you right click on a link is 'open in new tab' in Chromium. Juggling multiple browser windows is clumsy, which is why we have moved on to tabbed browsing today. Then what is the point of having 'open in new window' as the first option in right-click menus?

Fourth, Chromium's integrated address and search bar is enormously useful. It suits my browsing style so much that I positively detest the split bars in all other browsers today. I am not going back to any other browser if it does not have this feature.

Fifth, Chromium's another useful feature (largely overlooked by the competition) is that when you close off a tab, the rest of the tabs do not appropriate the whole horizontal window space immediately. There is a small pause, which is very useful in closing adjacent tabs, as the next tab's cross button smoothly slides into the now-closed tab's cross button's position. Try closing many tabs with your mouse in FF. You'll often have to move your pointer as you close off tabs. This is a big irritant for my daily use.

Which browser I am gonna be using in the future? Chromium, definitely. It won over FF because it was better in many (but not all) regards. Today it is better in all but one. No DownThemAll :(. I should mention that I didn't use many FF extensions. The one that I like (a lot) is DownThemAll. Sometimes, I often start up FF just to use this. But that's pretty much it.

It's the same situation with Google vs Bing. Initially people switched over to Google because it was better than everyone else in search. Now MS/Bing has been trying hard to compete with Google. And at the moment it is almost as good as Google. But it won't gain market/mind share until it becomes better than Google. And I won't go back to FF until it becomes better that Chromium.

On a side note, it appears that Ubuntu does not upgrade packages in it's update repositories. Although Ubuntu 9.10 has been keeping my FF updated with all the security and other bug fixes, it has not updated my FF 3.5 to 3.6. Apparently, it is their policy to provide feature updates in the next distribution release only. I like Fedora more on this count. :P

Semi-rant against python multiprocessing module

2010-02-07T00:37:00.005+05:30

Go here. And now look at issue 13 in that bug track. Or just have a look at this.

This is bad news. The unpicklability of lambdas has forced me to make some ugly changes to the parallel code I was using to drive my python extensions. I really-really wish it was doable.

For a while, I have been thinking that use of python's multiprocessing module to parallelize your C extensions akin to solving the problem at the wrong abstraction level. This nails it.

OTOH, opencl has a wonderful parallelization and synchronization model. In future, I am looking to write OCL code only for compute intensive tasks. Implementations from AMD and nvidia are out of beta, but they are immature atm. Intel is missing in action.

I expect it will take them about a year more to mature.

WebGL : updates

2010-01-28T23:13:00.004+05:30

The draft spec is out. And here's a great place to learn more about it. Looks like I'll be tempted into learning web programming and JS after all. :)

Apple iPad

2010-01-28T23:10:00.003+05:30

Meh....

The concept has potential. But this iteration is bloody useless for my purposes. Let's hope Apple can do much better in the future versions.

Google Book Search - a personal experience

2010-01-28T22:49:00.004+05:30

I have naturally been excited at the prospect of having all, (well, almost all) of the world's books scanned and made full text search searchable, with Google-level smarts in them.

I was kind of indifferent to the legal battle going on (go here for an update on that).

A few days ago, I had a profoundly useful experience with Google Books. I came across a solution to a problem I have been tackling (for a little hobby of mine) for a long time now. I had searched long and hard for it when I was actively working on it, but couldn't make much progress.

And then, a few days ago, when I gave it another shot, ~70% of the solution turned up in a book in Google Books. I had a look at the book title and author, and voila, a copy of it was available in my university library.

I had long thought that perhaps nobody had tried to tackle this before so a lot of heavy lifting would have to be done. But Google Books saved the day.

Which is why now I am *strongly* in favor of allowing the proposed deal to go through. I urge all the concerned decision makers to let this this wonderful thing happen. Newton stood on the shoulders of giants. Over the last ~350 years, we have accumulated the knowledge and the technology to let even folks like me stand on Mount Everest of human knowledge and make useful contributions, in whatever small way.

As for the various concerns, I have just one thing to say. Google made all the investment in technology, money, risk and effort. They have every right to milk it fwiw. However, I have just one quib. If exclusivity granted to Google were junked, it would be a Great Leap Forward for the entire humanity. In a really good way. After all, all of us stand to gain from the competition between search engines.

The way it's meant to be seen.

2010-01-26T16:47:00.003+05:30

Saw Avatar.

In IMAX 3D.

It rocks. :)

Apparently, many people agree.

No really, it changes the way movies-are-meant-to-be-seen.

How to screw the Avatar 3D experience and alienate customers

2009-12-20T18:59:00.002+05:30

How exactly might one go about doing that? After all, it is a James Cameron's MASTERPIECE, which will redefine CGI forever, a worthy sequel to Titanic.

Here's how Odeon, Connaught Place, a proud member of Anil Ambani's empire went about it.

You have a cinema screen with totally screwed up contrast ratio, color range that makes ugly cam prints look good and a sound system worse than a cheap pair of earphones.

And you charge people 200/300 bucks for it. Not to mention the 10x priced food to go with it.

Mmap-ing temporary files

2009-11-01T23:42:00.003+05:30

There is a cool way to allocate memory without actually doing malloc. It uses temporary files.

Here's how to do it,

1) Make a temporary file, and then use fileno() to get it's file descriptor.

2) Seek to the the end of the region which you want mmap-ed and the write a dummy data (just a "") to make the file of that size

3) Actually mmap this file with this size itself.

Why bother with all this, you ask. It can be useful to inter-operate between C land and numpy while sharing memory while avoiding memory leaks. This can be achieved by using numpy.memmap class and then passing in the numpy array to obtain the pointer.

This will in general need some amount of reworking in your library's memory allocation routine, but avoids the troubles associated with making sure that the C object is deallocated after the associated numpy object is freed.

I wave, you wave, we all wave

2009-10-31T13:00:00.002+05:30

I just got an invite for Google Wave, but there aren't many people around at the moment. Also I can't see any invite other folks link either. Not bad for 100th post, eh..?

Tweeting from python

2009-10-27T23:17:00.003+05:30

Stumbled across this. Really cool. Now I can send myself a private tweet to keep track of my long running jobs, or just about any damn thing. I don't have to be present at the keyboard to monitor the progress for this.

The only problem, is this. No proxy support, no.... :(

Hopefully, it'll materialize soon.

Parallel nirvana on the cheap

2009-10-24T19:49:00.003+05:30

I just had a production run for this, it took 48 hours. Now time for some parallel goodness from python side. :) So, after much experimentation and frustration, here is a great module, and here is some great advice.

Of course, you can get it in C/C++ land too, just use this.

Tegra2 netbooks

2009-10-17T20:31:00.006+05:30

It seems Tegra2 based netbooks will come out next year. If it has a dual core ARM Cortex A9, running at atleast 1.5 GHz, it'll probably be faster than my present laptop in CPU power. If it has a GPU based on the 9400M chipset, it'll be faster than the gpu in my laptop too, while supporting CUDA, OpenGL 3.2, OpenCL 1.1, WebGL etc. Not to mention that it'll have dedicated hardware for video decode acceleration as well. Tegra1 devices are supposed to be capable of playing back HD video for over 10 hours on a single battery charge. Hell, if this thing can give me 6 hours of HD video playback, I am gonna love it. On a smaller process, it should get more power efficient as well, but that is probably asking for too much. Considering the amount of integration it has, a netbook based on it will likely be cheaper as well... :) :)

If this thing comes with 2GB ram, then I am definitely gonna pick one up. With Chrome OS, it could prove mighty useful as well.

4GB ram is highly unlikely since ARM doesn't have a 64 bit CPU core out as yet, AFAIK. But Tegra3, :D

So my wishlist for Tegra2 would be,

1. Dual core ARM Cortex A9 with NEON at 2GHz (it'll be fun to write an vector backend for eigen on ARM)

2. 9400M GPU

3. hardware accelerated video decode for H.264, VC1, MPEG-4, DivX

4. 2GB RAM

5. minimum 6 hour battery life while playing 1080p HD video

6. nVidia supporting Chrome OS on it. (Let's hope Chrome OS will come with some nice hacking tools with it too, or atleast there will be some community support around building distributions that allow hacking it.)

Getting colors in git output on opensuse 11.1

2009-10-16T16:24:00.003+05:30

I just installed opensuse 11.1 on my lab machine. Git was behaving very oddly here, displaying color codes instead of color-ed output with git-diff and other commands.

A bit of googling later, I found this, exactly the same issue, but on Mac OS X. It was useful to me. Problem fixed.

Broad comparison of Larrabee and AMD and nVidia GPU's

2009-10-11T22:27:00.003+05:30

Jawed, from B3D, in an excellent post describes the broad architectural features of Larrabee and GPU's from nVidia and AMD. Worth a read for anyone who is interested in high performance hardware of tomorrow and for those who are looking to tap into the cheap teraflops of these beasts.

Looks like I did it

2009-10-07T12:34:00.004+05:30

It seems like that I have managed to graft on a few changes to my magnum opus successfully.

I am just waiting for second opinion and then, merge to master. :)

It's a bit slower than what I'd prefer, need to profile it.

Dropbox, file syncing service

2009-09-28T11:36:00.004+05:30

Somebody recently recommended to me the dropbox file sharing service. It is really cool. You just see a local folder, you make changes to it and it is auto synced across multiple machines.

They provide an opensource plugin for nautilus, and a closed source deamon to actually talk to their servers. You install the opensource plugin, and then do a dropbox start -i, and the binary deamon is downloaded and installed. After that, it could not be simpler to use.

The package (even the opensource part) will never be part of fedora proper, and here's why. But yeah, it could be a useful inclusion in rpmfusion.

Python extensions, wrapping with swig and bug hunting

2009-09-19T14:44:00.001+05:30

If you use swig to wrap your c/c++ code into a python extension, you will sometimes come across bugs when you try to import your freshly minted module.

Errors which go something like,

_mod, attribute referenced before assignment.

They can be pretty nasty to hunt, since you don't know if the bug is in your code, python interpreter (HOLY COW! that's not where I tweeted ;) ), swig (please no!!) or your own code (damn, I'll have debug it instead of passing it off on someone else..).

Chances are that the error is in your code. And if you get error like the one showed above, then python interpreter is throwing an error while importing your extension. To debug it, try the following,

1) if your module is named FOO, then do

import _FOO

as you have to prepend the underscore for the real module while swig sugarcoats it to make it look a little less ugly.

There is a very good chance that it is happening because there is a function in your code which is unknown to swig or the python interpreter is attempting to call a function that does not exist.

The path from now on is a little harder.

2) do

objdump -dS _FOO.so > some-text-file

And dive into that disassembly looking for the offending function. This isn't for the faint-hearted, but it pretty straightforward, quick and simple if you are used to it. :)

Happy bug hunting

Dual quaternions in eigen.

2009-09-18T15:09:00.003+05:30

I have been trying to add dual quaternion support (mainly for rigid transformations) to my project. That is done, but now I am trying to push it into eigen. To follow the discussion, head here and there.

Benefits of flatter/more balanced expression trees

2009-09-11T22:35:00.004+05:30

I am rewriting (partly) my magum opus. This time, I have overloaded operators of a simple class wrapping the __m128i datatype. This allows me to generate much flatter expression trees, and boy, the code generated is fantastic. I have never seen so awesome assembly code generated by gcc from my code.

Here is the assembly dump for the brave and/or the foolish :)

0000000000400ba0 <_z11addressegenpkspkis0_s0_s0_s0_ssss>:

400ba0: 66 44 0f 6e 5c 24 08 movd 0x8(%rsp),%xmm11

400ba7: 44 0f bf 54 24 20 movswl 0x20(%rsp),%r10d

400bad: 66 0f 6e 74 24 10 movd 0x10(%rsp),%xmm6

400bb3: 4c 8b 1d 06 18 20 00 mov 0x201806(%rip),%r11 # 6023c0

400bba: 66 45 0f 61 db punpcklwd %xmm11,%xmm11

400bbf: 66 0f 61 f6 punpcklwd %xmm6,%xmm6

400bc3: 66 44 0f 6e 74 24 18 movd 0x18(%rsp),%xmm14

400bca: 41 f7 da neg %r10d

400bcd: 66 45 0f ef ff pxor %xmm15,%xmm15

400bd2: 66 41 0f 70 e3 00 pshufd $0x0,%xmm11,%xmm4

400bd8: 66 0f 70 de 00 pshufd $0x0,%xmm6,%xmm3

400bdd: 66 44 0f 6e 6c 24 20 movd 0x20(%rsp),%xmm13

400be4: 66 45 0f 61 f6 punpcklwd %xmm14,%xmm14

400be9: 66 41 0f fd 63 40 paddw 0x40(%r11),%xmm4

400bef: 66 45 0f 61 ed punpcklwd %xmm13,%xmm13

400bf4: 66 44 0f 6f d4 movdqa %xmm4,%xmm10

400bf9: 66 41 0f fd 5b 50 paddw 0x50(%r11),%xmm3

400bff: 44 89 54 24 d4 mov %r10d,-0x2c(%rsp)

400c04: 66 44 0f 6f cb movdqa %xmm3,%xmm9

400c09: 66 0f 6e 44 24 d4 movd -0x2c(%rsp),%xmm0

400c0f: 66 41 0f 70 ee 00 pshufd $0x0,%xmm14,%xmm5

400c15: 66 41 0f 6f 53 70 movdqa 0x70(%r11),%xmm2

400c1b: f3 44 0f 10 f8 movss %xmm0,%xmm15

400c20: 66 45 0f 70 e5 00 pshufd $0x0,%xmm13,%xmm12

400c26: 66 44 0f db ca pand %xmm2,%xmm9

400c2b: 66 45 0f 61 ff punpcklwd %xmm15,%xmm15

400c30: 66 44 0f 6f f3 movdqa %xmm3,%xmm14

400c35: 66 44 0f db d2 pand %xmm2,%xmm10

400c3a: 66 41 0f fd ec paddw %xmm12,%xmm5

400c3f: 66 41 0f 71 f1 02 psllw $0x2,%xmm9

400c45: 66 44 0f 6f e4 movdqa %xmm4,%xmm12

400c4a: 66 41 0f 71 d6 0f psrlw $0xf,%xmm14

400c50: 66 41 0f 70 ff 00 pshufd $0x0,%xmm15,%xmm7

400c56: 66 0f 6f f3 movdqa %xmm3,%xmm6

400c5a: 66 41 0f 71 f2 04 psllw $0x4,%xmm10

400c60: 66 45 0f eb d1 por %xmm9,%xmm10

400c65: 66 41 0f 71 d4 0f psrlw $0xf,%xmm12

400c6b: 66 41 0f ef 7b 60 pxor 0x60(%r11),%xmm7

400c71: 66 0f fd ef paddw %xmm7,%xmm5

400c75: 66 44 0f 6f c5 movdqa %xmm5,%xmm8

400c7a: 66 44 0f 6f ed movdqa %xmm5,%xmm13

400c7f: 66 0f 6f fd movdqa %xmm5,%xmm7

400c83: 66 44 0f db c2 pand %xmm2,%xmm8

400c88: 66 41 0f 71 d5 0f psrlw $0xf,%xmm13

400c8e: 66 0f 6f d5 movdqa %xmm5,%xmm2

400c92: 66 45 0f eb d0 por %xmm8,%xmm10

400c97: 66 0f 71 d2 04 psrlw $0x4,%xmm2

400c9c: 66 44 0f 6f c6 movdqa %xmm6,%xmm8

400ca1: 44 0f 29 17 movaps %xmm10,(%rdi)

400ca5: 0f 29 22 movaps %xmm4,(%rdx)

400ca8: 0f 29 19 movaps %xmm3,(%rcx)

400cab: 41 0f 29 28 movaps %xmm5,(%r8)

400caf: 48 8b 05 0a 17 20 00 mov 0x20170a(%rip),%rax # 6023c0

400cb6: 66 0f 6f 80 a0 00 00 movdqa 0xa0(%rax),%xmm0

400cbd: 00

400cbe: 66 44 0f 6f b8 90 00 movdqa 0x90(%rax),%xmm15

400cc5: 00 00

400cc7: 66 0f f9 c3 psubw %xmm3,%xmm0

400ccb: 66 0f 71 d0 0f psrlw $0xf,%xmm0

400cd0: 66 44 0f db f0 pand %xmm0,%xmm14

400cd5: 66 0f 6f c2 movdqa %xmm2,%xmm0

400cd9: 66 44 0f f9 fc psubw %xmm4,%xmm15

400cde: 66 41 0f 71 d7 0f psrlw $0xf,%xmm15

400ce4: 66 45 0f db e7 pand %xmm15,%xmm12

400ce9: 66 0f 6f 88 b0 00 00 movdqa 0xb0(%rax),%xmm1

400cf0: 00

400cf1: 66 0f 71 f0 06 psllw $0x6,%xmm0

400cf6: 66 0f f9 cd psubw %xmm5,%xmm1

400cfa: 66 45 0f db e6 pand %xmm14,%xmm12

400cff: 66 0f 71 d1 0f psrlw $0xf,%xmm1

400d04: 66 44 0f db e9 pand %xmm1,%xmm13

400d09: 66 0f 6f ec movdqa %xmm4,%xmm5

400d0d: 66 0f ef c9 pxor %xmm1,%xmm1

400d11: 66 0f 71 d5 04 psrlw $0x4,%xmm5

400d16: 66 45 0f db e5 pand %xmm13,%xmm12

400d1b: 66 0f 61 e9 punpcklwd %xmm1,%xmm5

400d1f: 66 0f 61 c1 punpcklwd %xmm1,%xmm0

400d23: 66 44 0f eb a0 80 00 por 0x80(%rax),%xmm12

400d2a: 00 00

400d2c: 45 0f 29 21 movaps %xmm12,(%r9)

400d30: 4c 8b 1d 89 16 20 00 mov 0x201689(%rip),%r11 # 6023c0

400d37: 66 45 0f 6f 8b c0 00 movdqa 0xc0(%r11),%xmm9

400d3e: 00 00

400d40: 66 41 0f db f9 pand %xmm9,%xmm7

400d45: 66 45 0f db c1 pand %xmm9,%xmm8

400d4a: 66 44 0f db cc pand %xmm4,%xmm9

400d4f: 66 0f 6f e3 movdqa %xmm3,%xmm4

400d53: 66 0f 71 d7 02 psrlw $0x2,%xmm7

400d58: 66 0f 71 d4 04 psrlw $0x4,%xmm4

400d5d: 66 0f 61 e1 punpcklwd %xmm1,%xmm4

400d61: 66 41 0f 71 f1 02 psllw $0x2,%xmm9

400d67: 66 45 0f eb c1 por %xmm9,%xmm8

400d6c: 66 44 0f eb c7 por %xmm7,%xmm8

400d71: 66 44 0f 61 c1 punpcklwd %xmm1,%xmm8

400d76: 44 0f 29 06 movaps %xmm8,(%rsi)

400d7a: 4c 8b 1d 3f 16 20 00 mov 0x20163f(%rip),%r11 # 6023c0

400d81: 0f 29 6c 24 e8 movaps %xmm5,-0x18(%rsp)

400d86: 41 8b bb d0 00 00 00 mov 0xd0(%r11),%edi

400d8d: 0f af 7c 24 e8 imul -0x18(%rsp),%edi

400d92: 01 3e add %edi,(%rsi)

400d94: 48 8d 7e 04 lea 0x4(%rsi),%rdi

400d98: 41 8b 8b d4 00 00 00 mov 0xd4(%r11),%ecx

400d9f: 0f af 4c 24 ec imul -0x14(%rsp),%ecx

400da4: 01 0f add %ecx,(%rdi)

400da6: 41 8b 93 d8 00 00 00 mov 0xd8(%r11),%edx

400dad: 48 8d 4e 08 lea 0x8(%rsi),%rcx

400db1: 0f af 54 24 f0 imul -0x10(%rsp),%edx

400db6: 0f 29 44 24 d8 movaps %xmm0,-0x28(%rsp)

400dbb: 01 11 add %edx,(%rcx)

400dbd: 48 8d 56 0c lea 0xc(%rsi),%rdx

400dc1: 45 8b 93 dc 00 00 00 mov 0xdc(%r11),%r10d

400dc8: 44 0f af 54 24 f4 imul -0xc(%rsp),%r10d

400dce: 44 01 12 add %r10d,(%rdx)

400dd1: 0f 29 64 24 e8 movaps %xmm4,-0x18(%rsp)

400dd6: 45 8b 8b e0 00 00 00 mov 0xe0(%r11),%r9d

400ddd: 44 0f af 4c 24 e8 imul -0x18(%rsp),%r9d

400de3: 44 01 0e add %r9d,(%rsi)

400de6: 44 8b 4c 24 d8 mov -0x28(%rsp),%r9d

400deb: 45 8b 83 e4 00 00 00 mov 0xe4(%r11),%r8d

400df2: 44 0f af 44 24 ec imul -0x14(%rsp),%r8d

400df8: 44 01 07 add %r8d,(%rdi)

400dfb: 41 8b 83 e8 00 00 00 mov 0xe8(%r11),%eax

400e02: 0f af 44 24 f0 imul -0x10(%rsp),%eax

400e07: 01 01 add %eax,(%rcx)

400e09: 45 8b 93 ec 00 00 00 mov 0xec(%r11),%r10d

400e10: 44 0f af 54 24 f4 imul -0xc(%rsp),%r10d

400e16: 44 01 12 add %r10d,(%rdx)

400e19: 0f 29 44 24 e8 movaps %xmm0,-0x18(%rsp)

400e1e: 44 01 0e add %r9d,(%rsi)

400e21: 44 8b 44 24 ec mov -0x14(%rsp),%r8d

400e26: 44 01 07 add %r8d,(%rdi)

400e29: 8b 74 24 f0 mov -0x10(%rsp),%esi

400e2d: 01 31 add %esi,(%rcx)

400e2f: 8b 44 24 f4 mov -0xc(%rsp),%eax

400e33: 01 02 add %eax,(%rdx)

400e35: c3 retq

400e36: 66 2e 0f 1f 84 00 00 nopw %cs:0x0(%rax,%rax,1)

400e3d: 00 00 00

To be fair, I am comparing gcc 4.3 to gcc4.4.

The compiler runs out of steam at around 400d86, but still this is great. I am using eigen after that. I have no idea why the operations have not been vectorized. Need to look into that as well.

But still, +1 to this.

EDIT: The vectorization for the code after 400d86 can be fixed if you follow this and this.

AMD's profiler for linux

2009-09-03T20:24:00.000+05:30

I had no idea that there was an RPM for AMD's code analyst sitting in fedora's repositories until I came across this page.

I immediately got hold of the package and now it is sitting nicely on my laptop.

Time to grind my software against it. :)

[update] short support in eigen

2009-08-30T11:39:00.003+05:30

There was recently some good news on the vectorization for short operations front. I may be able to do it with a little less work then. If you want to follow the work that's being done on this, follow this thread. Help welcome, of course!

Now we have native 64 bit builds for chromium in fedora. Thanks to the chromium team and Tom, who has been maintaining those builds. This post was written using 64 bit chromium on fedora.

Adding short support to eigen

2009-08-24T14:39:00.004+05:30

I have started work on adding short support to eigen. I think this would help me a lot as I don't need an int in my work. Short should be just fine. You can follow the progress of my work here. There is no vectorization support yet, but it should be able to generate scalar code just fine. For more in depth discussion on this, follow this thread.

Writing unit tests is up next.

Emacs 23

2009-08-16T20:44:00.004+05:30

Is it just me, or emacs 23 (the one that ships with fedora, not ubuntu) does kinda look cool?

BTW, it broke cuda mode, dunno why? :(

Google Wave goes live on 30th sept.

2009-08-16T13:58:00.003+05:30

Sept is shaping up to be an exciting month. AMD's dx11 class gpu's launch on sept. 10, and Google Wave will go live on 30th. It seems that Wave will be by-invitation-only, just like gmail was initially. :(

But still, let's hope it will open up faster.

Installing AMD's opencl implementation for CPU

2009-08-06T17:15:00.002+05:30

I just installed AMD's implementation of OpenCL for it's CPU's on Fedora 11, x86-64. Here's how to do it.

1. Download the SDK from here. and unpack the sdk.

1A. Move the decompressed archive into the place where you want to install it. I chose /home/rpg/bin. I don't like binary crap arbitrarily polluting my system directories and settings.

2. Enter the directory of the sdk and do

$make

You could use make -j2 or even make -j4 if you have multiple cores to burn.

3. Now you need to add a few places to your paths. Add this to your .bashrc and .bash_profile files. Both of them should be in your home folder.

ATISTREAMSDKROOT=/home/rpg/bin/ati-stream-sdk-v2.0-beta2-lnx64

export ATISTREAMSDKROOT

export LD_LIBRARY_PATH=$ATISTREAMSDKROOT/lib/x86_64:$LD_LIBRARY_PATH

The installation notes provided by SDK have some typos. For a start, it is lib and not bin and you add different folders for different architectures. And you definitely don't need root access for this.

Lucky me though, I have a SSE3 capable CPU. It is ironic to note that the inventor of x86-64 is leaving behind the classic x86-64. I wonder why they felt SSE3 was mandatory.

After all this, this is what I get to see,

$ ./BlackScholes

For test only: Expires on Wed Sep 30 00:00:00 2009

deviceCallPrice

0.118873 7.04261 0 0.00272289 0 33.9944 5.90269 0.0174322 15.6142 31.8218 0.208182 31.4897 36.4303 4.70751 0 0 0.180582 0 23.1549 0 56.4034 19.635 38.0241 0.00524368 2.9981 52.4114 33.1443 50.3665 0 36.6461 0.167021 0.00245133

devicePutPrice

9.18383 0.216879 56.2858 23.5745 51.5608 3.39307e-06 0.0396819 14.4737 0.00709584 0 5.18874 0 0 3.79125 63.0257 49.3472 17.1783 51.754 0.191513 16.1561 0 0 0 29.3654 0.00833774 0 0 4.21006e-06 17.8919 6.60636e-05 4.04936 31.1893

Option samples Time taken(sec) Options / sec

4096 1.565 2617.25

The exact paths have been scrubbed from the output here though. Enjoy!

Chrome working again

2009-08-03T22:38:00.003+05:30

Chrome is working again. Inside IITB. :)

Going home

2009-07-13T19:17:00.001+05:30

I am writing this from inside the great firewall of China. :)

No chrome for me :(

2009-07-09T23:46:00.003+05:30

I knew this was coming, but I didn't it would come so fast and in this way. Latest svn build just hit a SNAFU. Ah, the perils of living on the bleeding edge. Thank God, I still have firefox. And 3.5 is pretty good with it's speed too.

Compiling open64 with gcc4.4

2009-07-06T05:34:00.003+05:30

After this, God knows what got into me and I tried to compile open64 with gcc4.4 that comes with fedora 11. There were a few bugs along the way. I included cstring, cstdlib here and there and they were fixed. The last one that I got stuck on was in c-parse.c. Now I am officially giving up on my install-as-many-compilers-as-you-can binge. LLVM should be fine, for now.

EDIT0: Ok, I broke my promise to myself. I installed it using the binary rpm. :)

installing LLVM-GCC in fedora 11, 64 bit

2009-07-06T04:02:00.006+05:30

Fedora comes with a prepackaged llvm. To install that, you just need to do yum install llvm. But to compile any thing with llvm, you need to install llvm-gcc. Here's how to do it.

1. install llvm by

$ yum install llvm

2. download llvm-gcc source code from here. I downloaded llvm-gcc4.2-2.5.source.tar.gz. Let's say you downloaded it to /path/to/llvm-gcc

3. now extract the tarball by

$ cd /path/to/llvm-gcc

$ tar -zxvf llvm-gcc-version.tar.gz

4. make temp directories for compilation

$ cd /path/to/llvm-gcc/llvm-gcc4.2-2.5.source

$ mkdir install;mkdir obj

5. now comes the fun part of customizing everything to suit your needs. I did

$ cd obj

$ ../configure --prefix=`pwd`/../install --program-prefix=llvm- --enable-llvm=/usr --enable-languages=c,c++,fortran --disable-multilib

The extras here are fortran, disabled multilib for native x86-64 builds and the prefix directory for installation.

6. Now compile and install

$ make -j2

On your machine, it might be a little different depending on how many cpus you can spare for installation.

7. Finally, install it with

$ make install

Now all your stuff will be installed int the $prefix/install/bin directory. Here, $prefix is the path you used to run ./configure, and not some arbitrary environment variable. You may place symlinks at your preferred places.

Visit to Alcatraz

2009-07-06T00:17:00.000+05:30

Getting to San Francisco: $3.70

Ferry to Alcatraz, $26.00

Having your camera battery die on you after you get there: Priceless

No chrome, :(

2009-07-04T12:08:00.004+05:30

Have a look at this.

I will have to give up chrome when I return to IITB. I am disappointed, but hopefully, this feature will get implemented soon.

More updates

2009-07-02T07:51:00.002+05:30

After falling to alomst 30, my to-watch list of movies has swelled to 42 again. :) Time to go back to India is approaching too. Mumabi would be as usual swamped with rains, no doubt. :(

Chrome on Fedora 11, 64 bit

2009-07-01T00:08:00.005+05:30

I just installed google chrome using this. Even though I am running a 64 bit version. Install using yum and it will run perfectly. I am all over this thing. It is awesome, it runs like a rocket, GAAAWWWD. Every one should try this. Thanks for making this repo available. This thing is going to kick ass when done in gentoo. !!! Thank God for the competition.

EDIT0: I wrote this using chrome. The repo maintainer seems to be updating his repo fairly quickly (for now atleast).

EDIT1: As correctly pointed out here, the rpms are not native 64 bit binaries. They are 32 bit ones, which work correctly for 32 bit ones too.

Why I love Python

2009-06-19T07:56:00.002+05:30

Yesterday, my advisor was talking to me. I was telling him about the progress we have made.

Me: This is a python script to drive every thing

Advisor: What is Python?

M: It is a scripting language.

... 5 minutes later

A: Let me look into the script.

.....He reads for a minute or two.

A: Now I see why people program in Python. I can actually understand it.

:)

Disappointed

2009-06-15T03:03:00.003+05:30

We lost. :(

But honestly though, we didn't deserve to win considering the way we played. 14 wides, top 4 playing like they are holidaying on a tropical beach, paid for by God know who.

But it keeps up the pattern we have been seeing since 1987. Did well, reached semis. 1991, a disaster. 1996, again reached semis and 1999, made a hash of it all. 2003, reached the finals and in 2007, suffered humiliation at the hands of Bangladesh. In 2007 T20, we actually won for a change, and in 2009..........

We'll lose to SA too on Tuesday, rest assured.

Knives are going to be out for Dhoni and co now, considering the Sehwag fiasco.

Reset signals

2009-06-09T10:15:00.002+05:30

I had quite a moment today. I have been bamging my head against a problem for quite some time now. Today, it got solved. I had generous help from Michal in it BTW.

The problem. The mother code works uses a active high reset and the module for my chip uses active low. Epic fail!

Software updates and linux distros

2009-06-05T07:50:00.007+05:30

Today, I just out of curiosity did a uname -a and gcc -v on th elinux server that cnnects to my hardware. And horror of horrors, what do I find? Gcc 3.4 and kernel 2.6.8. Python v2.3.

Now come on guys. Update your software atleast once a year. I know everyone does not run Gentoo. But still.

Speaking of Gentoo, I must confess that I have been drawn to it. I really need the customizability for some applications/libraries atleast. The usual complaints apply, however. I am told (by some one who is a serious linux pro) that gentoo offers too much choice, such as letting you install a bootloader yourself (aka, not by default).

I am coming around to the view of installing Sabayon and using portage to get my optimization-compilation kick. BTW, IMHO, Sabayon is the coolest looking distribution out there. I have downloaded it's iso, just in case. I'll probably use the lab machine as the guinea pig in this experiment. :)

Honestly though, I think that with a Core i7, compilation would be much less of a problem. I really like hardware multithreading. I think, all new CPU's should have it though. :) Wish AMD will go this route as well.

I have been checking out the portage tree over at gentoo-portage.com and it seems that they have done a really thourough job. I mean, apart from the llvm, vTune and sage, I could find ebuilds for every thing. Even flash, skype, imkl, acml, icc etc. The first two are a mystery, but with sage, I am not surprised. With their philosophy of making a giant static binary (the source tarball is >300M, I think), and bundling all the open source math software ever created (including their shiny python interpreter) into it, I am amazed that somebody managed to get it into Debian. I hope the fedora guys will have some luck as well. But it'll probably will always be way behind on updates.

I'll get the videos done by the weekend hopefully. Promise.

Pics

2009-06-02T10:49:00.001+05:30

Ok, the pics got done earlier than I expected. So enjoy. But the videos will probably take more time.

Pics/videos from Maker's fair 2009

2009-06-02T10:17:00.002+05:30

I'll be putting up stuff I grabbed from Maker's Fair 2009. It's gonna take a while, but I really want to annotate the stuff real well so that onlookers have some context around it. Promise.

Running skype on fedora 64 bit

2009-05-31T03:49:00.002+05:30

I just installed skype on my fedora 10 laptop (64 bit) using this and it works like a charm.

EDIT0: It is broken in fedora 11. I haven't got around to fixing it yet.

In Berkeley :)

2009-05-20T05:36:00.003+05:30

I am back at the lab (LBNL) now. I have settled down here in the last couple of days. The weather is nice here. Though I have not been doing the stuff had planned I would do yet. Really need to get started on it soon.

Fedora 11 got delayed. :(

Will update this blog soon. Stay tuned.

Updates 2

2009-04-28T22:23:00.004+05:30

This month, I have been really bad with my blogging. I did very few updates (in fact, this is the second one this month). But now I hope to turn a corner. The exams are over. The project (Sriram and mine) is done. And I am in the mood for some relaxation now. Though I still can't believe we managed to finish our project some 24 hours before the deadline. I mean, when was the last time it happened?

Never, dude.

We borrowed a lot of code from kgllib. It's a fantastically written llibrary for opengl convenience functions. Too tied to Qt for my taste, but the Qt stuff was easy to remove. Initially, we wrote a lot of nice code, but soon after degenerated into bad programming practices. Global variables all over the place, messy code in between. We avoided the deprected bits of opengl functionality wherever we could, which, IMHO, is a good thing. We had a nice time hunting for segmentation faults, and removing them. :)

Updates

2009-04-17T20:33:00.005+05:30

1) Exams are going on. But I have a bit of a break here.

2) My IIT life is coming to an end. Not immediately, but yes the end has begun.

3) Our (Sriram and mine) opengl project is coming along. Need to fix a few bugs before we can get it to work. Had a few issues with git while collaborating on it. For now, we are just screwing any optimizations. :)

4) I was recently put up as a contributor to Eigen on their webpage. Seems wonderful to see my name in print. :)

Render to texture

2009-03-16T10:27:00.004+05:30

We, (that is Sriram and myself) just got our render to texture facility working. We drew a quad and and then generated some funny patterns in the pixel shader. It was fun. My first gpu side hacking. More of it is coming along soon, I promise.

Haskell and fractals.

2009-03-12T17:12:00.004+05:30

I recently came across Fraqtive. Very nice program. And well optimized and well written too, by the looks of it. It automatically takes care of multithreading and vectorization. And it uses some very nice algorithms to to handle zoom ins etc. The big news for me came when I stumbled across this and realized that it uses template metaprogramming to generate fast C++ code. Bit more googling and what do I find, Haskell bindings for LLVM!!!

I remember that in Lisp one can get access to the code of the functions to manipulate on them. If I can get them in Haskell, it would be great. Generate LLVM code from there, and then run it under a JIT. and after that, send it off to opengl and use shaders to generate nice colors for them....

May be I am getting too ahead of myself. New language, new libraries. May be, I should start from Python first :) But there's no denying that it would make for a very good learning project in Haskell.

Vectorized integer multiplication

2009-03-11T12:23:00.003+05:30

Integer operations in vector ISA's rarely get much love (compared to floating point operations), unless it has got something to do with video decode-encode. But I needed something like that for my needs. So I wrote one, and it turns out to be faster than the one in eigen, so it was committed.

Open source developer

2009-03-08T11:35:00.002+05:30

Now I am an official opensource developer. Yay!!!!. Feels great to see some stuff from my side being accepted there. Small beginning, I admit. But hopefully, as we go along, I will contribute more and more. BTW, this community is very friendly and responsive.

I wrote this because I wanted to use quaternions and it not being vectorized is obviously a shame. I have also sent a better routine for vectorized integer multiplication. Let's hope it gets in too. The code is only somewhat faster, but hey something is better than nothing.

Good news

2009-03-03T14:40:00.003+05:30

Finally, some sanity is restored. Look here for a rant. Lets' hope nvidia drivers won't break this.

Updates

2009-02-25T21:19:00.003+05:30

Lot's of 'em here today.

I came across this site. Hack LLVM from python. Wow. I really liked it. I even went through some of the tutorials up there which tell you how to hack llvm to make your own language. The lexer and the parser I am obviously going to generate using automated tools (if I ever do it). Making AST from parsed input is obviously going to take some doing. But if you can get to LLVM IR, after that it is a joy ride.

Another nice thing about his website is that it is made using asciidoc. I immediately recognized it from it's similiarity to the git documentation. Seems way better suited to generating HTML than LATEX and you can do nice code formatting and insert math formulas.

And the first preview (to my knowledge) of RV740 just got out here. Seems really good. Good luck AMD. And please don't go bankrupt.

OpenGL Assignment

2009-02-22T20:01:00.005+05:30

It's done. I managed to do the whole thing, even as much as for 150% credit, though it is useful only if I am stuck at the borderline. The mid sem was half a washout, but I think he gives papers like this. Of course, there was no time to optimize anything.

After the grades for this come out, I'll post it here depending on how embarrassing was it. :) Though I hope using Python for it wasn't a problem.

Code-athon

2009-02-20T21:40:00.001+05:30

The exams are over. But the assignment I referred to here still needs doing. Thank God I have python. With this sword of damocles hanging on my head, all thought of optimization, and writing it as a python extension has gone miles away from my head.

Let's Rock.

Exams, again

2009-02-14T19:32:00.003+05:30

Midsems have begun, again. And immediately after them I am supposed to do an assignment for which the specs were changed midway. And now, I have to rewrite it all. I have a few ideas, and hopefully, will be able to implement them in time. Thank God for python, where would I be without you. And yes, atm, the usual aim of having it run blazing fast is out of the window.

POVray videos

2009-02-05T22:00:00.004+05:30

I just made a few videos from my POVray demo. I thought I'll put them up here. So here it is, at 800x600 resolution, encoded at 15 fps.

Wetting my feet

2009-02-05T18:41:00.002+05:30

It seems a long while ago that I wanted to learn Haskell. And I haven't made any progress yet. But it seems to be changing. I came across this. It's a very nice book at a cursory glance. And I also got this book from the library. It seems to be pretty good as well.

POVray assigment done

2009-02-04T22:09:00.009+05:30

The POVray assignment I wrote about earlier is now complete. Got some nice pics out of it. Here's one of them.

Now I am rendering them at higher resolution and will be making a movie out of them. Right now I have made a mpeg-4 one but I am looking to encode them in theora format. FFmpeg, I love you. Though this movie was made with mencoder, they both basically use the libavcodec and libavformat libraries.

It is somewhat surprising that in ray tracing too, texture mapping is used extensively.

Anyways, the pics look pretty cool to me.

And yes, I figured out how to get rid of that irritating warning. Here' how.

In the vfe/unix folder of the source tar ball for unix, there's a file called unixconsole.cpp. IN it, the following code appears,

if (user_code != current_code)

{

fprintf(stderr, "%s: this pre-release version of POV-Ray for Unix %s\n",PACKAGE,current_week <>);
exit(RETURN_ERROR);

}

Comment it out. Afterwards, it will whine about expiring in a while, but those warnings are harmless, it won't come to pass.

Happy ray tracing.

The road to GLSL: Part II

2009-01-30T14:11:00.004+05:30

Well, where's the part I dude? You may ask. Well, it was here, when I realized what I had with me for 2 years. And, in my 50th post, I am happy to tell you that we are nearly there now. Have a look.

[rpg@rpg ~]$ glxinfo|grep NVIDIA
server glx vendor string: NVIDIA Corporation
client glx vendor string: NVIDIA Corporation
OpenGL vendor string: NVIDIA Corporation
OpenGL version string: 2.1.2 NVIDIA 180.25
OpenGL shading language version string: 1.20 NVIDIA via Cg compiler
[rpg@rpg ~]$

No, my prompt isn't like that. I reinstalled Fedora 10 because I was fed up with KDE. Now I have both GNOME and KDE for future proofing. I'll just keep upgrading and use whatever seems better at the time.

The install went horribly though. Long story. Some data loss has occurred too, though can't say how much. Hopefully, it is in can-be-easily-mitigated category. My mistake. Should have acted more prudently.

Anyway, now the fun begins.

Rant against povray

2009-01-28T21:25:00.005+05:30

Recently, we were given an assignment to render some scenes using POVray and demonstrate to what we had learned about ray tracing. All right then, let's install it. OK. it may not be OSI compliant license but as the page there states, it's because of it's birth time. We are sure to have packages for it in ubuntu. It's a just a simple matter of running

$sudo apt-get install povray

I couldn't have been more wrong.

This project is quite screwed up as it turns out.

There has been no stable release for more than 4 years now. The project's mailinglists distinctly convey a sense of lack of developers. But this is the least of the troubles. Sure, I could have just stuck to the old but stable version. But it is not multithreaded.

Yup. That's right. A serial ray tracer.

If I didn't know better, I would have said that was a malapropism (GRE side effects, :-) ). People have been using SMP machines (if not multi core) for a while now and since POVray has been ported to many different machines before, it is incomprehensible that it's developers haven't threaded it yet.

The most irritating thing about it is their concept of beta timeout. Their betas usually expire after a fixed time. And on top of it, these guys can't even release betas in time. Look here.

They can't release the updated betas on time, aren't releasing new stable versions at all and everyone else is supposed to update every month or so.

I somehow manged to patch the vfe.cpp file referenced in the above link. I even managed to render a few things like this.

And, then I went to dinner. What a mistake on my part.

And now, I have to fight this.

~/povray-3.7.0.beta.29/unix@rpg-lab> ./povray
povray: this pre-release version of POV-Ray for Unix has expired

Now only the betacode hack is working, in spite of modding the sources as mentioned in the link above to remove the timeout.

God help me.

Just look at Blender. It has a huge community behind it. Scriptable in python (yay!), supports GLSL shaders, renders all right, imports/exports from/to every file format imaginable, can interface to other ray tracers as well. I think I am begining to like it. Further, it can export to anything that ffmpeg supports.

I am running it on a machine with SSE3, but gcc can hardly be expected to make use of any of it's horizontal math goodness.

GRE over, finally

2009-01-27T17:11:00.003+05:30

My GRE is over. Thank God so much. The last few days have been full of nervous tension. But it is over. And it went well.

And now, let's install Fedora 10, and chase shaders, GLSL shaders. And yeah, before that, we need to backup our data, watch some movies, etc etc.

GRE

2009-01-22T22:23:00.003+05:30

Just wanna get over with it. Feeling good about it. Let's see how it goes.

Fingers crossed.

PS: Slumdog Millionaire gets 10 Oscar nominations. AR Rehman gets 3 of them. That guy is a genius. He's gonna win atleast one of them. I am ecstatic about it.

Thread Safety

2009-01-15T19:02:00.002+05:30

OpenGL ain't thread safe. What a shame.

Turns out few libraries are thread safe. How the hell are average Joe programmers supposed to write multithreaded code when the libraries they rely upon aren't. Beats me.

Exams and system calls

2009-01-10T19:52:00.004+05:30

I have got my GRE comping up on 27th. It's really keeping me away from my crazy shaders. I wanted to get into system call level programming earlier, as I find it to be a powerful tool. I looked at the mmap system call first and found it to be very nice.

The GLSL examples I have looked at usually define a read from file function which stat's a file for it's size, allocates memory reads and passes the char * to compiler. Better still, just stat the file to get it's size, mmap those many bytes and then, just pass the mmap-ed pointer to the compiler!

PPM images lend themselves to even better use. Just open the file, read the header info and close it. Then, mmap the file at the offset such that the header is read off and use the size from prior header read, and voila, you can stream the texture to the GPU from hard disk asynchronously. Other texture loading libraries typically require you to allocate memory temporarily while they read the files. This way, we get rid of memory bugs, (which can get nasty BTW) and this is something we can do easily in the background, ideal use for multi-threading. And it's simple too, as they it requires no inter-thread I/O.

I know PPM images are too big to be practically used, but hey, OpenGL needs images in that format, and we will give it in that format. Compressed textures, we'll look into, but later.

My GPU supports SM3.0

2008-12-28T12:31:00.007+05:30

I just realized that my laptop's GPU has been a SM3.0 capable one all along. I have learnt about using programmable graphics pipeline this year itself. I am drooling at the thought of running vertex and pixel shaders (I am going to call fragment shaders as pixel shaders only) of my own. OK it is not a top of the line GPU, but hey, better than none.

When I first learnt graphics programming, it was using the lowly Turbo C++'s graphis capabilities. Then I moved onto OpenGL 1.1. I wrote a demo using display lists and stuff but then it seemed that it was more than enough for anything that I might need.

Then I discovered the joys of programmble graphics pipeline. It appeared to combine the power of Turbo C++'s low level and highly flexible power with the SGI's grpahics pipeline's inherent advantages.

Now, my Geforce 6150 (I never said it was top of the line) is ready to serve me in my pursuit of crazy shaders. Of course, I'll start small and then I plan to move the control into Python and leave the hevy lifting to C.

This also appears to be a great place to move into multithreading. Raw pthreads from C seems unwise. I would prefer to orchestrate the entire stuff from Python. Atleast the loading/storing of textures is anticipated to be simple enough. This also is a place where multithreading can help hide latencies.

Now I found the DevIL library for loading it seems very nice. Though I don't know if it is thread-safe. I hate it when I open docs for some library, only to find that it is not threadsafe. It is essential for concurrency that the libraries be threadsafe.

2008: A Review

2008-12-25T18:04:00.006+05:30

Today as I sit down to write this, one thought runs through my mind. That this is the year of change. Sure, stuff changes all the time and

Change is the only constant.

Yet I cannot shake off the feeling that this year, somehow, more things seem to have hanged than they usually do. Moreover, I see the changes in trends not just in new incidents. To summarize, (in no particular order)

1. GPGPU went mainstream. It got lots of attention from the tech press. I too read some stuff about programmable graphics pipeline. OpenCL spec was released, which I am sure is just the catalyst needed for gpu's to come in their own.

2. I wrote a program in two different languages, ie Python and C++. I suffered a lot on the way, but I had an optimized, vectorized (in parts) and cache aware BFS ready when I was done. And yes, I learnt about valgrind along the way.

3. India made history in many ways. We landed on the moon, had our best Olympics ever, had our break-out moment and won emphatically in Australia. Today, I see India, Australia and South Africa as forming a triad which will compete for the No. 1 slot into the next year.

4. Obama became the US president. It just had to make the list. And you know what it means.

5. I got an opportunity to go abroad (which is a first in my family), so a big deal for me.

6. The war on terror started going right for US and upside down for India. Yankees realized that ISI has been milking them all along and is actually busy screwing them royally in Afghanistan. Terror struck India like never before, and that's without 26/11.

7. Public coming out on streets to protest was a welcome change. However, the public anger needs to be better directed if we are to achieve our long term goals.

A quiet December

2008-12-22T14:04:00.004+05:30

It has been a quiet december this year so far. I mean nothing on the blog, and Have been too lazy so far to bother doing much else so far in the holidays. The internet connection here has been acting up lately, in part because of the cable outage. Lot's of stuff has been flowing through my mind these past few days. Will summrize it soon.

Though need to decide on the electives I will be taking next semester soon.

Going home

2008-11-26T15:59:00.002+05:30

Exams got over yesterday. Though a presentation is remaining. Leaving for home on day after tommorrow.

Super PC's now availaible off-the-shelf

2008-11-23T01:22:00.003+05:30

It seems that the trend of building supercomputers out of multiple gpu's in one pc is catching on. I really like this idea and now nvidia is offering pre-built and qualified systems on those lines. But yes, they come with Tesla series of cards instead of the cheaper geforce cards.

Come on AMD, get your OpenCL implementation working so that we can enjoy much faster super-pc's.

Good texture compression review

2008-11-22T22:24:00.002+05:30

I cam across this very nice review of texture compression. He covers the need for texture compression, and the typical techniques used for texture compression, their pros and cons. Though there are a few spelling mistakes here and there and he builds his case for texture compression around AGP, an extinct bus. All in all, a nice review to aid your understanding of texture compression.

AMD's FASTRA now possible

2008-11-18T23:39:00.004+05:30

I have written before on how better tools are needed on the AMD side to do gpgpu stuff, in some detail. I had particularly pointed out the need for their drivers to support 4 4870x2 cards in a PC as it makes for a nice 9.6T toaster.

This announcement certainly means that they are stepping up their efforts, though a true C compiler remains in the foggy future. Further, this thread says that they expect drivers to recognize them (the four cards that is) allright.

I saw a few people eager to make such a toaster. Well, with that announcement, good luck guys.

And this suggests that a C-to-IL compiler may not be so far into the future either. Well, good luck guys, give us a C compiler quickly.

India on the moon

2008-11-14T22:10:00.002+05:30

Now there's a tricolour on the moon. Great job, ISRO.