Mentor Graphics Announces Algorithmic C Datatypes

Mentor Graphics Corporation has announced the availability of Algorithmic C datatypes, new high-speed datatypes based on ANSI C++. These arbitrary-bit-width datatypes enable algorithm, system and hardware designers to precisely model bit-true behavior in C++ specifications while accelerating simulation speeds by 10-200x.

Algorithm and system designers need to specify bit widths to create bit-accurate models used for system and algorithm validation. Previous bit-width specifications require one of two compromises: designers must either use native processor precision (64-bits for integer), requiring them to truncate the results and thus introduce artificial corner cases; or they must create a model using an arbitrary bit width resulting in much slower simulation speeds.

Based on universal standard ANSI C++, the new integer and fixed-point Algorithmic C datatypes offer the best of both worlds, allowing algorithm and system designers to specify arbitrary bit widths while improving simulation performance by 10-200x versus other datatypes such as SystemC. Arbitrary bit widths are also critical for high-level synthesis, as they allow hardware designers to make tradeoffs between hardware size and numerical precision. For example, a hardware designer would use arbitrary bit widths to explore tradeoffs in area, performance, or power versus image quality in a video application.

The new Algorithmic C datatypes also solve problems of semantic consistency inherent in previous datatypes. The semantics of Algorithmic C datatypes are intuitive for users familiar with ANSI C++, enabling designers to become proficient in just one day. In contrast, there is a semantic divide in SystemC between the limited precision datatypes (sc_int or sc_fixed_fast), and the arbitrary precision datatypes (sc_bigint or sc_fixed). Proficiency in one datatype does not ensure familiarity with the others, making it error prone to adjust bit-width in a given design. Consequently, one must invest significant effort to master the coding style of both datatypes in order to employ them in an optimal design implementation.

write your comments about the article :: 2006 Computing News :: home page