Section: Software

FLIP: Floating-point Library for Integer Processors

Participants : Claude-Pierre Jeannerod, Jingyan Jourdan-Lu.

FLIP is a C library for the efficient software support of binary32 IEEE 754-2008 floating-point arithmetic on processors without floating-point hardware units, such as VLIW or DSP processors for embedded applications. The current target architecture is the VLIW ST200 family from STMicroelectronics (especially the ST231 cores). This year, we have mostly worked on improving the design and implementation of the following operators with correct rounding “to nearest even”: DP2 (fused dot product in dimension two) and sum of two squares. The impact of the DP2 operator has been evaluated on the UTDSP benchmark, and on some kernels speed-ups of 1.46 have been observed. On the other hand, specializing DP2 to a sum of squares brings a speed-up of 2.

URL: http://flip.gforge.inria.fr/

• ACM: D.2.2 (Software libraries), G.4 (Mathematical software)

• AMS: 26-04 Real Numbers, Explicit machine computation and programs.

• APP: IDDN.FR.001.230018.S.A.2010.000.10000

• License: CeCILL v2

• Type of human computer interaction: C library callable, from any C program.

• OS/Middleware: any, as long as a C compiler is available.

• Required library or software: none.

• Programming language: C