[CakeML-Dev] other targets benchmarks
Ramana.Kumar at cl.cam.ac.uk
Fri Oct 7 06:49:00 UTC 2016
FYI the infinite loop seems to run fine, but the other two crash. They're
also disturbingly slow (admittedly it's software simulation, but simple C
programs aren't as slow). The FFI test program crashes before printing
anything, so I think crashes must be happening before calls to C functions
(whether they be FFI or exit). I'll investigate this more on Monday (or if
I find time before then).
On 7 October 2016 at 15:37, Magnus Myreen <magnus.myreen at gmail.com> wrote:
> Three simple tests: the first test you want to run is a simple infinite
> fun f x = f x;
> val y = f 0;
> since this test doesn't depend on a good handover at the end to exit.
> The second test could be:
> val v = 0;
> And the third test could be the hello program which prints a short message.
> On 7 October 2016 at 05:32, Ramana Kumar <Ramana.Kumar at cl.cam.ac.uk>
> > Status update:
> > I've tried to set the avoid_regs and register-name mappings correctly for
> > all targets. The riscv branch (now named too specifically) is where this
> > work lives.
> > I have .S files for all the benchmarks on this branch. Magnus, do you
> > me to send you the ARM-32 ones?
> > I've got a RISC-V setup working on my machine (I can compile and run
> > C programs). I've tried to run the CakeML fib benchmark for RISC-V.
> > Currently, it takes a really long time in the simulator (a couple of
> > minutes?) and then segfaults.
> > It would be very good to have some benchmarks that (a) produce output or
> > are very simple/fast, just to test that things are working at all. Yong
> > Kiam, were you going to add output to the benchmarks? I'll try producing
> > very simple test benchmark myself...
> > _______________________________________________
> > Developers mailing list
> > Developers at cakeml.org
> > https://lists.cakeml.org/listinfo/developers
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Developers