Poul-Henning Kamp wrote: > David O'Brien writes: > >Actually I'd really love for a way to do the outter loop in parallel. > >Due to -j messing up error output too much, I'd like to have the work for > >each platform going in parallel instead. E.g.: > > > > make universe for i386 only & > > make universe for powerpc only & > > make universe for sparc64 only & > > make universe for amd64 only & > > make universe for alpha only & > > make universe for ia64 only & > > True, but I can't get that working no matter how much I've tried. > OK, here it comes. The attached patch unrolls the main universe loop, making it possible for "universe" subtargets corresponding to individual architectures to be built in parallel. EXAMPLES make -j4 universe will build four architectures in parallel, each will be built using four parallel make's. make -j8 universe JFLAG=-j4 will build eight architectures in parallel, each will be built using four parallel make's. make -j8 universe JFLAG=-B will build eight architectures in parallel, each will be built using one make. Please test. Points considered: since "buildkernels" is implemented using the make's .for loop, kernels for a given architecture are built sequentially (each can still be built in parallel). I didn't see much point in parallelizing this task, but this is easily doable. (One redundant JFLAG when calling buildkernels was removed by the patch.) Cheers, -- Ruslan Ermilov ru_at_FreeBSD.org FreeBSD committerReceived on Tue Aug 17 2004 - 06:10:37 UTC
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:38:06 UTC