As reported back in May (CI No 2,177), Digital Equipment Corp was showing its developers a translator that took Intel Corp iAPX-86 Windows binary code and spat out binaries suitable for the Alpha chip running Windows NT. For the simple Mandelbrot application demonstrated at Spring Comdex, the translated binaries ran around four times as fast as the equivalent Intel binaries running under the Insignia Solutions Ltd-built Windows emulator and brought the performance of DEC’s AXP 150 up to that of a 60MHz Pentium. Its use? Either for companies that want a quick conversion for the chip, without re-compiling the source code, or alternatively, end-users that may not have the source for their applications, but want to get their existing Intel applications running on the new architecture. Since then the DEC AXP Migration Tools Group has been trying to decide what to do with it. Andy Riebs, who works in the team, describes the reaction of developers as extremely positive but adds that there is significant disagreement over what problems we should be addressing. In particular DEC is still trying to decide whether it has a developers or an end-user tool on its hands. Other issues surround performance. Translators tend to produce ‘dirty’ code full of redundant instructions, but by increasing the complexity of the translator and thereby the time that it takes to run, it is possible to optimise the results. So the question that DEC faces now is whether the translator should run as quickly as possible and produce an application with adequate performance, or whether the translation process should take all the time that it needs to give the fastest possible application. The migration tools group has yet to make up its mind, so in lieu of a fixed direction, Riebs says we’re pushing forward on a number of fronts, but we still haven’t committed to do a product yet.
