For arc-elf32, I only want to run C and C++ tests, so the runtime of
fortran tests is irrelevant for this purpose.
On the other hand, I run the tests eight-way multilibbed.
Currently, I run the check-gcc on eight hosts (or execution slots on multicore
servers), and the check-g++ tests on eight other hosts, and all the
libstdc++-v3 multilibs on a seventeenth hosts - we still lack multilibbing
support here. At about one hour runtime, the C test take much longer
than the C++ tests, but generally provide the best feedback on issues
with the target compiler and/or libgcc.
On one hand, I want to improve disk locality if I can get some SMP hosts,
on the other hand, I don't want to block hosts with by scheduling a parallel
job t oa huge amount of hosts where only some have work to do all the time.
So I schedule check-gcc as one job for eight execution slots, then check-g++
as another one, and finally check-target-libstdc++-v3 .
I would like to speed this up by farming the C tests out to more hosts
in parallel. Also, support to split the check-libstdc++-v3 load into
different multilib chunks would be nice.
make -j would be a non-starter, since it only works on a single host.
sometimes I only get a bunch of single-core single-processor hosts, so it
is important to have jobs that can run independently on multiple hosts,
but can share the same build directory via NFS.
