Chrome To Support Simd.js
Wednesday, 13 May 2015

With Firefox support and Microsoft working on it in Chakra, the news that Chrome is to get Simd.js support completes the set - apart from Safari that is. 



SIMD - or Single Instruction Multiple Data - is one of the simplest and easiest parallel processing mechanisms. Essentially what it comes down to is packing multiple values into a single register and then performing the operation as if it was a single value. Of course, there are some overheads - you have to pack and unpack the data - but in most cases these can be minimized. 

It isn't as good as using the GPU, but it is very much simpler. It is also something that a lot of C/C++ programs make use of to speed up things like codecs and game computations.

Given that JavaScript is the target of compilers like Emscripten, which have been used to convert games and other graphics programs to so that they run in the browser, not having SIMD support in JavaScript limits the range of programs that can be compiled without manual modification. There is an official polyfill that emulates SIMD using typed arrays, but it is slow. 

RealSIMD can produce a 3x or better speed up of programs that make heavy use of it. 

In the chart below you can see Mozilla's figures for running the polyfill v simd.js as part of Asm.js.



The process of adding SIMD instructions to JavaScript is surprisingly easy. All you need are three new vector types: float32x4, float64x2, and int32x4, and some typed array views. Once the data is loaded into the new data types, the arithmetic operators work as parallel operations. That is, if you multiply two float32x4 variables, all four floats are multiplied in one operation. 

There is a small problem that different hardware supports different types of SIMD instructions. The good news is that there is a core of operations that are supported by SSE on Intel and AMD  x86 and by NEON on ARM. Hence there are no 256- and 512-bit vector operations. What this means is that the simd.js extensions will work on x86 and ARM platforms and Google is planning support on all Blink platforms. This will allow Android to run programs that use SIMD. 




GraalVM 21.3 Released

GraalVM, the runtime that compiles Java bytecode into native self-contained executables and capable of running programs in different languages other than Java, has reached version 21.3.

Why We Need An Hour Of Code

Today marks the start of Computer Science Education Week when schools and coding clubs around the globe will be staging Hour of Code Events. This year, however, the Hour of Code is more part of the ro [ ... ]

More News






or email your comment to:

Last Updated ( Thursday, 14 October 2021 )