Automatic Code Transformation to Optimize Accuracy and Speed in Floating-Point Arithmetic
Résumé
We want to allow the standard software developer to automatically transform his/her code in order to increase the accuracy of floating-point computations. This transformation is actually an optimization since we aim to take into account two opposite criteria: accuracy and execution time. A first step towards this automatic optimization is presented in this work. We propose to automatically introduce at the compile-time compensation steps in (parts of) the floating-point computations. We present a tool to parse C codes and to insert compensated floating-point operations. This provides a compensated computation that improves the accuracy of speci c computing patterns.