Single Instruction Multiple Dated

Single Multiple Instruction one Dated , or SIMD , indicates an operating process of the Ordinateur S equipped with several calculating units functioning in parallel. In this mode, the same instruction is applied simultaneously to several data to produce several results. One uses this abbreviation in opposition to SISD (Single Individual Instruction one Dated), traditional operation, and MIMD (Multiple Multiple Instructions one Dated), operation with several independent processors. Model SIMD is appropriate particularly well for the treatments of which the structure is very regular, as it is the case for the matrix algebra.

Instructions SIMD were added to the modern processors to be able to improve the speed of implying treatment on calculations of the numbers in floating point. Instructions SIMD are in particular made up of the instruction sets:

Processors X86: MX, 3DNow!, SE, SE, SE, SSSE3 and SE
Processors PowerPC: AltiVec and VMX

There exist two types of SIMD:

Vectorial (Relating to the data):

The same instructions in parallel are treated, then one passes to following calculation when both finished.

Parallel (In the process of disappearance because called upon specific processors):

In the case of the parallel SIMD, one treats an instruction in asynchronous execution

Their uses require many work and thorough knowledge in programming, assembler X86 or PowerPC, and data processing.

In general, one starts by developing a generic code which will function everywhere. When the algorithm is correct and that the code functions well, one writes a version specialized for an extension of a given processor. The optimized programs are those which require few resources processors: Data compression, codec for the reading of sound and/or video, calculation on great integers (Cryptography in particular), etc

Random links: