2012/12/1 Peter Gavin <[email protected]>: > On Fri, Nov 30, 2012 at 9:05 PM, Matthew Hicks <[email protected]> wrote: >> >> I much prefer the nop hacks as it made debugging things much easier. >> I could just insert a special nop-based sequence and find out what was >> going on. The nop hacks are great because they don't require any >> library support, or extra hardware, and they don't perturb the system. > > > Fair point. Plus I suppose a testsuite could just call a function that > chooses which method to use. > > There ought to be a testsuite that's maintained separately from any of the > implementations, and I figure such a testsuite ought not to use the nop > hack. I figure it could make it optional, though, and allow some > implementations to support it. > > -Pete > > _______________________________________________ > OpenRISC mailing list > [email protected] > http://lists.openrisc.net/listinfo/openrisc >
We have been talking about centralizing the test suites. I think it would be great to be able to move all the CPU-specific tests from or1ksim, orpsoc, (minsoc?), mor1kx-dev-env and more into a separate component. Meanwhile, I would say that passing the orpsoc or1200 tests is a minimum for submitting trivial patches, passing the or1ksim tests when there are functional changes and the gcc tests when we do more proper releases. Can we have consensus on that? I've been playing around a bit with newlib and libgloss, and think we provide a default putc/getc that uses nops, but use weak pointers so that we can override this and other hardware specific functions with another lib. This is basically what libboard does today, but a little more flexible -- Olof Kindgren ______________________________________________ ORSoC Website: www.orsoc.se Email: [email protected] ______________________________________________ FPGA, ASIC, DSP - embedded SoC design _______________________________________________ OpenRISC mailing list [email protected] http://lists.openrisc.net/listinfo/openrisc
