Dislaimer: I'm not good at videos :-)
Short Summary
I'm a software developer working solely on F/OSS software and Linux since 1997, a proud member of Debian community since 1999 (originally a debian-powerpc porter and now one of the armhf porters), a non-mainstream architectures advocate (personal favourites PowerPC and now ARM), a software optimization freak and an SIMD/algorithm vectorization fanatic. More details on my site http://freevec.org.
During the past years, I've had to work on porting algorithms for several SIMD engines, including PowerPC Altivec/VMX in the beginning (some Cell SPU inbetween), and lately ARM NEON and Intel SSE* (and their variations). While I had to work on one architecture only, things were easy, however afterwards I had to constantly use Google or several reference manuals to find out that bit of extra information about this particular instruction on eg. SSSE4.*. And I found myself constantly searching on Amazon and elsewhere, for a comparative book or even online reference about all/most SIMD engines. For a simple example, how to do a vector unpacking/shuffling in those 3 major SIMD engines? Is there more than one way on each platform? And what are the benefits of each method? Which is the most performant? Which utilizes the SIMD engine better and leaves no bubbles in the pipeline? Etcetera, etcetera.
So, after a lot of search and eventual frustration, I've decided to give it a shot myself. I plan to write one or two books on all 3 major SIMD engines:
The first will be a plain reference, with comparisons and showing equivalent instructions on each engine, on both assembly and C intrinsics.
The second book will be more like a cookbook, covering simple and complex algorithms -some of which can be found online already, some not- again for all 3 SIMD engines.
Whether I write one or two books will depend on the funding I receive. If it covers just the requested funding, then I'll write a reference book with some clever tricks/recipes included for several algorithms and stuff like optimizations, pipelines and comparisons between the internals of each SIMD engine. If funding surpasses my expectations, I'll definitely write a second book and give it the attention it deserves, covering more algorithms and more advanced topics (like autovectorization perhaps). But you will definitely get at least one book.
Why you should fund this?
Well, if, like me, you've had to walk around the complete maze that is SIMD engine implementations and want to port your algorithm to one or more of those architectures without spending half your day on Google and hard to reach sites, then you might want to help this effort. I know I would definitely help if something like that was available 3-4 years ago. Right now, I'm just at the point that I don't want to wait any longer and I want to "Just Do It!".
I expect this to take approximately 3-4 months to complete, perhaps less if I really push it, but I prefer to do this properly rather than hastily. I will definitely complete it regardless of the funds received as I want it for myself anyway. But a successful campaign would help me finish it sooner rather than later, as I wouldn't have to worry about real life expenses and I could devote myself to this.
What will you get back?
Besides my eternal gratitude? Depending on the amount, you'll either get your name in the credits, a free t-shirt, an e-book or also a printed book or a conference call explaining the book -any parts you would like.
