On Wed, 23 Jul 2008 10:16:20 +0200 "litlle girl" <little.linux.girl@xxxxxxxxx> wrote: >Ok, >maybe turn ./configure more SMP will be easier? >What about split ./configure to n files > >(n=2 for dualcore) >./configure ==> ./configure1 + ./configure2 >and then >/bin/sh ./configure1 & >/bin/sh ./configure2 & How do you split existing configure into 2 parts: configrue1 and configure2? A multi-thread or SMP programmer have to divide a task into parallelly-executable subtasks and define the point to gather the results from subtask. 20 or 30 years ago, software scientists and engineers were trying to achieve an automatic optimization of a source code designed for single CPU & without thread, to an improved source code for multi CPU & thread. When they restricted their scope to highly restricted scientific numerical calculation without I/O (like Monte-Carlo simulations), they could achive good optimizer. But when they tried to office-oriented softwares, the automatic optimizer is not so good. Today, most scientists and software engineers are forced to insert the additional informations for parallel execution, or rewrite the algorithm from scratch. Autoconf does not have an optimizer for such parallel execution (do you know any optimizer designed for scriptic languages?). The indepth survey of configure.ac by its author (not autoconf developer) is required, I'm afraid. Make is rather easier, because it describe the dependency of the target files, so the detection of independent tasks are possible. But how could we detect the independent task in sh script? Regards, mpsuzuki _______________________________________________ Autoconf mailing list Autoconf@xxxxxxx http://lists.gnu.org/mailman/listinfo/autoconf
