Codeplay Software Limited is a compiler and software tools developer based in Edinburgh, Scotland. Codeplay develop C/C++ compilers for multi-core and special purpose processor architectures, offering their Sieve C++ Multicore Programming System as a means to easily write C/C++ code for acceleration on multicore chips. They also build bespoke C/C++ compilers for customers, using their VectorC Retargetable Compiler Engine.
Codeplay is a contributing member of the Multicore Association and the Khronos Group.
Contents |
Codeplay was set up in 2002 by Andrew Richards, an experienced games developer (formerly head of Software Development at Eutechnyx) who believed there was a need for development tools that could keep up with advances in processor technology. Initial funding for the company was provided by Jez San OBE, founder of Argonaut Games and ARC International. Jez San is also currently the majority shareholder. Codeplay began its business developing compilers aimed at games programmers, releasing its VectorC compiler for PlayStation 2 and x86 Windows PC in 2001. In 2005 Codeplay shifted focus to developing bespoke C/C++ compilers for processor designers.
Codeplay have also conducted extensive research and development into building tools in aiding multicore software development.
Codeplay's current product lineup is focused on providing bespoke C/C++ compilers and tools to processor designers. Codeplay's core technologies include:
The Sieve C++ Multicore Programming System is a tool set that allow programmers to quickly and easily adapt existing C and C++ software, or write new software, to take advantage of new multi-core processor architectures.
VectorC is a Retargetable C/C++ compiler engine used to create C and C++ compilers for custom-built processor architectures.[1] Codeplay describes VectorC as being optimized for SIMD processors, DSP's, vector units and multi core processors.
VectorC, as its name suggests, was designed from the outset to exploit vectorization. It can automatically take advantage of advanced CPU 'vector' units via SIMD instruction such as MMX, SSE/SSE2, 3D Now and Altivec. VectorC is based around a common optimization core so that all appropriate existing optimizations can be used on the new target architecture it is being customized for. The Optimizer can query the back-end to see what optimizations are possible or suitable on the target.
VectorC also supports C++ to C compilation through a ANSI C compliant back-end. This enables the compilation of C++ code, with or without additional language extensions, to any architecture where a C compiler already exists. This feature is used in Codeplay's Sieve C++ Parallel Programming System to generate parallelized code that can be ported to any multicore architecture with a native C compiler.
Codeplay state that VectorC performs all the standard non-vector optimizations expected from high-performance compilers.
The VectorC engine has been designed to cope with multiple memory spaces, distributed latencies and an optimize data-movement using double-buffered streaming DMA when required, in addition to standard write combining and prefetching functions. There is support for Software pipelining, automatically parallelizing loops on pipelined architectures.
VectorC has extensive support for portable intrinsic functions, letting users write complex vector operations as built in functions. These are converted into the best combination of instructions on each architecture that the source code is compiled for.
Codeplay continues to offer VectorC based custom compilers to clients. VectorC is also described by Codeplay as being the basis of Sieve C++ Multicore Programming System.
Based on VectorC technology, the Retargetable Shader Compiler offers customized high performance shader language compilers for use in programmable GPU's and embedded devices.